diff options
34 files changed, 4358 insertions, 640 deletions
diff --git a/SOURCES/0001-add-ally-x-dmi-quirk-for-controller-suspend.patch b/SOURCES/0001-add-ally-x-dmi-quirk-for-controller-suspend.patch index e0753ae..bffa8c5 100644 --- a/SOURCES/0001-add-ally-x-dmi-quirk-for-controller-suspend.patch +++ b/SOURCES/0001-add-ally-x-dmi-quirk-for-controller-suspend.patch @@ -1,48 +1,28 @@ -From 0000000000000000000000000000000000000000 Mon Sep 17 00:00:00 2001 -From: Jan200101 <sentrycraft123@gmail.com> -Date: Mon, 15 Jul 2024 19:31:40 +0200 +From 79d958eccfa4a1cfbb552032e9542f03333005e7 Mon Sep 17 00:00:00 2001 +From: antheas <antheas@users.noreply.github.com> +Date: Mon, 15 Jul 2024 00:00:45 +0300 Subject: [PATCH] add ally x dmi quirk for controller suspend -Signed-off-by: Jan200101 <sentrycraft123@gmail.com> --- - drivers/platform/x86/asus-wmi.c | 3 +-- - include/linux/platform_data/x86/asus-wmi.h | 14 ++++++++++++++ - 2 files changed, 15 insertions(+), 2 deletions(-) + drivers/platform/x86/asus-wmi.c | 6 ++++-- + 1 file changed, 4 insertions(+), 2 deletions(-) diff --git a/drivers/platform/x86/asus-wmi.c b/drivers/platform/x86/asus-wmi.c -index 4b78c4f4e..a8060e7fb 100644 +index 3f9b6285c9a6..8e362726b703 100644 --- a/drivers/platform/x86/asus-wmi.c +++ b/drivers/platform/x86/asus-wmi.c -@@ -4646,8 +4646,7 @@ static int asus_wmi_add(struct platform_device *pdev) +@@ -4645,8 +4645,10 @@ static int asus_wmi_add(struct platform_device *pdev) asus->egpu_enable_available = asus_wmi_dev_is_present(asus, ASUS_WMI_DEVID_EGPU); asus->dgpu_disable_available = asus_wmi_dev_is_present(asus, ASUS_WMI_DEVID_DGPU); asus->kbd_rgb_state_available = asus_wmi_dev_is_present(asus, ASUS_WMI_DEVID_TUF_RGB_STATE); - asus->ally_mcu_usb_switch = acpi_has_method(NULL, ASUS_USB0_PWR_EC0_CSEE) - && dmi_match(DMI_BOARD_NAME, "RC71L"); -+ asus->ally_mcu_usb_switch = acpi_has_method(NULL, ASUS_USB0_PWR_EC0_CSEE) && dmi_check_system(ally_usb_switch_dmi_ids); ++ asus->ally_mcu_usb_switch = ++ acpi_has_method(NULL, ASUS_USB0_PWR_EC0_CSEE) && ++ (dmi_match(DMI_BOARD_NAME, "RC71L") || ++ dmi_match(DMI_BOARD_NAME, "RC72LA")); if (asus_wmi_dev_is_present(asus, ASUS_WMI_DEVID_MINI_LED_MODE)) asus->mini_led_dev_id = ASUS_WMI_DEVID_MINI_LED_MODE; -diff --git a/include/linux/platform_data/x86/asus-wmi.h b/include/linux/platform_data/x86/asus-wmi.h -index 870f4bb57..3e8edc07e 100644 ---- a/include/linux/platform_data/x86/asus-wmi.h -+++ b/include/linux/platform_data/x86/asus-wmi.h -@@ -225,4 +225,18 @@ static const struct dmi_system_id asus_use_hid_led_dmi_ids[] = { - NULL, - }; - -+static const struct dmi_system_id ally_usb_switch_dmi_ids[] = { -+ { -+ .matches = { -+ DMI_MATCH(DMI_BOARD_NAME, "RC71L"), -+ }, -+ }, -+ { -+ .matches = { -+ DMI_MATCH(DMI_BOARD_NAME, "RC72LA"), -+ }, -+ }, -+ NULL, -+}; -+ - #endif /* __PLATFORM_DATA_X86_ASUS_WMI_H */ +-- +2.45.2 diff --git a/SOURCES/0001-amd-hdr.patch b/SOURCES/0001-amd-hdr.patch deleted file mode 100644 index 8116a3c..0000000 --- a/SOURCES/0001-amd-hdr.patch +++ /dev/null @@ -1,208 +0,0 @@ -From 2f33df4d9d3e86546d7c67453ae799306d55b7f5 Mon Sep 17 00:00:00 2001 -From: Melissa Wen <mwen@igalia.com> -Date: Sat, 22 Apr 2023 14:08:47 -0100 -Subject: [PATCH] HACK: add KConfig to enable driver-specific color mgmt props - -We are enabling a large set of color calibration features to enhance KMS -color mgmt but these properties are specific of AMD display HW, and -cannot be provided by other vendors. Therefore, set a config option to -enable AMD driver-private properties used on Steam Deck color mgmt -pipeline. Replace the agreed name `AMD_PRIVATE_COLOR` with -our downstream version `CONFIG_DRM_AMD_COLOR_STEAMDECK`. - -Signed-off-by: Melissa Wen <mwen@igalia.com> ---- - drivers/gpu/drm/amd/display/Kconfig | 7 +++++++ - drivers/gpu/drm/amd/display/amdgpu_dm/amdgpu_dm.c | 2 +- - drivers/gpu/drm/amd/display/amdgpu_dm/amdgpu_dm_color.c | 2 +- - drivers/gpu/drm/amd/display/amdgpu_dm/amdgpu_dm_crtc.c | 6 +++--- - drivers/gpu/drm/amd/display/amdgpu_dm/amdgpu_dm_plane.c | 6 +++--- - 5 files changed, 15 insertions(+), 8 deletions(-) - -diff --git a/drivers/gpu/drm/amd/display/Kconfig b/drivers/gpu/drm/amd/display/Kconfig -index 901d1961b7392..49523fa82f92a 100644 ---- a/drivers/gpu/drm/amd/display/Kconfig -+++ b/drivers/gpu/drm/amd/display/Kconfig -@@ -51,4 +51,11 @@ config DRM_AMD_SECURE_DISPLAY - This option enables the calculation of crc of specific region via - debugfs. Cooperate with specific DMCU FW. - -+config DRM_AMD_COLOR_STEAMDECK -+ bool "Enable color calibration features for Steam Deck" -+ depends on DRM_AMD_DC -+ help -+ Choose this option if you want to use AMDGPU features for broader -+ color management support on Steam Deck. -+ - endmenu -diff --git a/drivers/gpu/drm/amd/display/amdgpu_dm/amdgpu_dm.c b/drivers/gpu/drm/amd/display/amdgpu_dm/amdgpu_dm.c -index 5853cf0229176..1718ddfe75083 100644 ---- a/drivers/gpu/drm/amd/display/amdgpu_dm/amdgpu_dm.c -+++ b/drivers/gpu/drm/amd/display/amdgpu_dm/amdgpu_dm.c -@@ -4074,7 +4074,7 @@ static int amdgpu_dm_mode_config_init(struct amdgpu_device *adev) - return r; - } - --#ifdef AMD_PRIVATE_COLOR -+#ifdef CONFIG_DRM_AMD_COLOR_STEAMDECK - if (amdgpu_dm_create_color_properties(adev)) - return -ENOMEM; - #endif -diff --git a/drivers/gpu/drm/amd/display/amdgpu_dm/amdgpu_dm_color.c b/drivers/gpu/drm/amd/display/amdgpu_dm/amdgpu_dm_color.c -index 21e0efc42a34c..8f8d2a8fb2921 100644 ---- a/drivers/gpu/drm/amd/display/amdgpu_dm/amdgpu_dm_color.c -+++ b/drivers/gpu/drm/amd/display/amdgpu_dm/amdgpu_dm_color.c -@@ -97,7 +97,7 @@ static inline struct fixed31_32 amdgpu_dm_fixpt_from_s3132(__u64 x) - return val; - } - --#ifdef AMD_PRIVATE_COLOR -+#ifdef CONFIG_DRM_AMD_COLOR_STEAMDECK - /* Pre-defined Transfer Functions (TF) - * - * AMD driver supports pre-defined mathematical functions for transferring -diff --git a/drivers/gpu/drm/amd/display/amdgpu_dm/amdgpu_dm_crtc.c b/drivers/gpu/drm/amd/display/amdgpu_dm/amdgpu_dm_crtc.c -index 6e715ef3a5566..ab9992f24ae21 100644 ---- a/drivers/gpu/drm/amd/display/amdgpu_dm/amdgpu_dm_crtc.c -+++ b/drivers/gpu/drm/amd/display/amdgpu_dm/amdgpu_dm_crtc.c -@@ -290,7 +290,7 @@ static int amdgpu_dm_crtc_late_register(struct drm_crtc *crtc) - } - #endif - --#ifdef AMD_PRIVATE_COLOR -+#ifdef CONFIG_DRM_AMD_COLOR_STEAMDECK - /** - * dm_crtc_additional_color_mgmt - enable additional color properties - * @crtc: DRM CRTC -@@ -372,7 +372,7 @@ static const struct drm_crtc_funcs amdgpu_dm_crtc_funcs = { - #if defined(CONFIG_DEBUG_FS) - .late_register = amdgpu_dm_crtc_late_register, - #endif --#ifdef AMD_PRIVATE_COLOR -+#ifdef CONFIG_DRM_AMD_COLOR_STEAMDECK - .atomic_set_property = amdgpu_dm_atomic_crtc_set_property, - .atomic_get_property = amdgpu_dm_atomic_crtc_get_property, - #endif -@@ -551,7 +551,7 @@ int amdgpu_dm_crtc_init(struct amdgpu_display_manager *dm, - - drm_mode_crtc_set_gamma_size(&acrtc->base, MAX_COLOR_LEGACY_LUT_ENTRIES); - --#ifdef AMD_PRIVATE_COLOR -+#ifdef CONFIG_DRM_AMD_COLOR_STEAMDECK - dm_crtc_additional_color_mgmt(&acrtc->base); - #endif - return 0; -diff --git a/drivers/gpu/drm/amd/display/amdgpu_dm/amdgpu_dm_plane.c b/drivers/gpu/drm/amd/display/amdgpu_dm/amdgpu_dm_plane.c -index 8a4c40b4c27e4..5d87c24f0461f 100644 ---- a/drivers/gpu/drm/amd/display/amdgpu_dm/amdgpu_dm_plane.c -+++ b/drivers/gpu/drm/amd/display/amdgpu_dm/amdgpu_dm_plane.c -@@ -1468,7 +1468,7 @@ static void amdgpu_dm_plane_drm_plane_destroy_state(struct drm_plane *plane, - drm_atomic_helper_plane_destroy_state(plane, state); - } - --#ifdef AMD_PRIVATE_COLOR -+#ifdef CONFIG_DRM_AMD_COLOR_STEAMDECK - static void - dm_atomic_plane_attach_color_mgmt_properties(struct amdgpu_display_manager *dm, - struct drm_plane *plane) -@@ -1659,7 +1659,7 @@ static const struct drm_plane_funcs dm_plane_funcs = { - .atomic_duplicate_state = amdgpu_dm_plane_drm_plane_duplicate_state, - .atomic_destroy_state = amdgpu_dm_plane_drm_plane_destroy_state, - .format_mod_supported = amdgpu_dm_plane_format_mod_supported, --#ifdef AMD_PRIVATE_COLOR -+#ifdef CONFIG_DRM_AMD_COLOR_STEAMDECK - .atomic_set_property = dm_atomic_plane_set_property, - .atomic_get_property = dm_atomic_plane_get_property, - #endif -@@ -1742,7 +1742,7 @@ int amdgpu_dm_plane_init(struct amdgpu_display_manager *dm, - - drm_plane_helper_add(plane, &dm_plane_helper_funcs); - --#ifdef AMD_PRIVATE_COLOR -+#ifdef CONFIG_DRM_AMD_COLOR_STEAMDECK - dm_atomic_plane_attach_color_mgmt_properties(dm, plane); - #endif - /* Create (reset) the plane state */ --- -GitLab - -From b938468f07222b4faab5ae5cf5391eccd9532bb0 Mon Sep 17 00:00:00 2001 -From: Bouke Sybren Haarsma <boukehaarsma23@gmail.com> -Date: Fri, 15 Dec 2023 11:14:58 +0100 -Subject: [PATCH] Don't create color_mgmt_properties on asics < SIENNA_CICHLID - ---- - drivers/gpu/drm/amd/display/amdgpu_dm/amdgpu_dm_plane.c | 3 ++- - 1 file changed, 2 insertions(+), 1 deletion(-) - -diff --git a/drivers/gpu/drm/amd/display/amdgpu_dm/amdgpu_dm_plane.c b/drivers/gpu/drm/amd/display/amdgpu_dm/amdgpu_dm_plane.c -index 2ed20e6e439bb5..65ee8745e96540 100644 ---- a/drivers/gpu/drm/amd/display/amdgpu_dm/amdgpu_dm_plane.c -+++ b/drivers/gpu/drm/amd/display/amdgpu_dm/amdgpu_dm_plane.c -@@ -1742,7 +1742,8 @@ int amdgpu_dm_plane_init(struct amdgpu_display_manager *dm, - drm_plane_helper_add(plane, &dm_plane_helper_funcs); - - #ifdef CONFIG_DRM_AMD_COLOR_STEAMDECK -- dm_atomic_plane_attach_color_mgmt_properties(dm, plane); -+ if (dm->adev->asic_type >= CHIP_SIENNA_CICHLID) -+ dm_atomic_plane_attach_color_mgmt_properties(dm, plane); - #endif - /* Create (reset) the plane state */ - if (plane->funcs->reset) --- -2.43.0 - -From: Joshua Ashton @ 2024-04-10 22:13 UTC (permalink / raw) - To: amd-gfx; +Cc: Joshua Ashton, Harry Wentland, Xaver Hugl, Melissa Wen, Ethan Lee - -The comment here states "no OGAM in DPP since DCN1", yet that is not -true. - -Testing on an RX 7900XTX (dcn32), it actually does exist in hardware and -works fine. -My best guess is the comment is confused with OGAM ROM for DPP, rather -than OGAM RAM. - -I did not test dcn35/351 as I do not have that hardware, but I assume -the same follows there given the seemingly erroneous comment. -Someone at AMD should check that before merging this commit. - -Signed-off-by: Joshua Ashton <joshua at froggi.es> - -Cc: Harry Wentland <harry.wentland at amd.com> -Cc: Xaver Hugl <xaver.hugl at gmail.com> -Cc: Melissa Wen <mwen at igalia.com> -Cc: Ethan Lee <flibitijibibo at gmail.com> ---- - drivers/gpu/drm/amd/display/dc/resource/dcn32/dcn32_resource.c | 2 +- - drivers/gpu/drm/amd/display/dc/resource/dcn35/dcn35_resource.c | 2 +- - 2 files changed, 2 insertions(+), 2 deletions(-) - -diff --git a/drivers/gpu/drm/amd/display/dc/resource/dcn32/dcn32_resource.c b/drivers/gpu/drm/amd/display/dc/resource/dcn32/dcn32_resource.c -index 9aa39bd25be9..94f5d2b5aadf 100644 ---- a/drivers/gpu/drm/amd/display/dc/resource/dcn32/dcn32_resource.c -+++ b/drivers/gpu/drm/amd/display/dc/resource/dcn32/dcn32_resource.c -@@ -2098,7 +2098,7 @@ - dc->caps.color.dpp.dgam_rom_for_yuv = 0; - - dc->caps.color.dpp.hw_3d_lut = 1; -- dc->caps.color.dpp.ogam_ram = 0; // no OGAM in DPP since DCN1 -+ dc->caps.color.dpp.ogam_ram = 1; - // no OGAM ROM on DCN2 and later ASICs - dc->caps.color.dpp.ogam_rom_caps.srgb = 0; - dc->caps.color.dpp.ogam_rom_caps.bt2020 = 0; -diff --git a/drivers/gpu/drm/amd/display/dc/resource/dcn35/dcn35_resource.c b/drivers/gpu/drm/amd/display/dc/resource/dcn35/dcn35_resource.c -index 25ac450944e7..708d63cc3f7f 100644 ---- a/drivers/gpu/drm/amd/display/dc/resource/dcn35/dcn35_resource.c -+++ b/drivers/gpu/drm/amd/display/dc/resource/dcn35/dcn35_resource.c -@@ -1858,7 +1858,7 @@ - dc->caps.color.dpp.dgam_rom_for_yuv = 0; - - dc->caps.color.dpp.hw_3d_lut = 1; -- dc->caps.color.dpp.ogam_ram = 0; // no OGAM in DPP since DCN1 -+ dc->caps.color.dpp.ogam_ram = 1; - // no OGAM ROM on DCN301 - dc->caps.color.dpp.ogam_rom_caps.srgb = 0; - dc->caps.color.dpp.ogam_rom_caps.bt2020 = 0; --- -2.44.0 diff --git a/SOURCES/0001-amd-pstate.patch b/SOURCES/0001-amd-pstate.patch index f654f25..72d5cc6 100644 --- a/SOURCES/0001-amd-pstate.patch +++ b/SOURCES/0001-amd-pstate.patch @@ -1,27 +1,26 @@ -From b36ccc153f8a2ee04f046a4e10220dbe308f37ef Mon Sep 17 00:00:00 2001 +From b2368823f6d79235b4e706263855373ff2fed1e5 Mon Sep 17 00:00:00 2001 From: Peter Jung <admin@ptr1337.dev> -Date: Tue, 25 Jun 2024 13:29:13 +0200 +Date: Fri, 5 Jul 2024 10:31:29 +0200 Subject: [PATCH 02/10] amd-pstate Signed-off-by: Peter Jung <admin@ptr1337.dev> --- - Documentation/admin-guide/pm/amd-pstate.rst | 40 +- + Documentation/admin-guide/pm/amd-pstate.rst | 18 +- arch/x86/include/asm/cpufeatures.h | 1 + arch/x86/include/asm/msr-index.h | 2 + arch/x86/kernel/cpu/scattered.c | 1 + - drivers/cpufreq/acpi-cpufreq.c | 2 - - drivers/cpufreq/amd-pstate-ut.c | 2 +- - drivers/cpufreq/amd-pstate.c | 463 +++++++++++++++----- - drivers/cpufreq/amd-pstate.h | 29 +- - drivers/cpufreq/cpufreq.c | 25 +- - include/linux/cpufreq.h | 2 + - 10 files changed, 436 insertions(+), 131 deletions(-) + drivers/cpufreq/Kconfig.x86 | 1 + + drivers/cpufreq/acpi-cpufreq.c | 3 +- + drivers/cpufreq/amd-pstate.c | 356 ++++++++++++++------ + drivers/cpufreq/amd-pstate.h | 16 +- + drivers/cpufreq/cpufreq.c | 14 +- + 9 files changed, 292 insertions(+), 120 deletions(-) diff --git a/Documentation/admin-guide/pm/amd-pstate.rst b/Documentation/admin-guide/pm/amd-pstate.rst -index 1e0d101b020a0..57995f54f0c88 100644 +index 1e0d101b020a..d0324d44f548 100644 --- a/Documentation/admin-guide/pm/amd-pstate.rst +++ b/Documentation/admin-guide/pm/amd-pstate.rst -@@ -281,6 +281,27 @@ integer values defined between 0 to 255 when EPP feature is enabled by platform +@@ -281,6 +281,22 @@ integer values defined between 0 to 255 when EPP feature is enabled by platform firmware, if EPP feature is disabled, driver will ignore the written value This attribute is read-write. @@ -41,15 +40,10 @@ index 1e0d101b020a0..57995f54f0c88 100644 +boost or `1` to enable it, for the respective CPU using the sysfs path +`/sys/devices/system/cpu/cpuX/cpufreq/boost`, where `X` represents the CPU number. + -+It is important to note that modifying the global variable -+`/sys/devices/system/cpu/amd_pstate/cpb_boost` will override the individual CPU -+settings. -+ -+ Other performance and frequency values can be read back from ``/sys/devices/system/cpu/cpuX/acpi_cppc/``, see :ref:`cppc_sysfs`. -@@ -406,7 +427,7 @@ control its functionality at the system level. They are located in the +@@ -406,7 +422,7 @@ control its functionality at the system level. They are located in the ``/sys/devices/system/cpu/amd_pstate/`` directory and affect all CPUs. ``status`` @@ -58,32 +52,8 @@ index 1e0d101b020a0..57995f54f0c88 100644 "active" The driver is functional and in the ``active mode`` -@@ -440,6 +461,23 @@ control its functionality at the system level. They are located in the - This attribute is read-only to check the state of preferred core set - by the kernel parameter. - -+``cpb_boost`` -+ Specifies whether core performance boost is requested to be enabled or disabled -+ If core performance boost is disabled while a core is in a boosted P-state, the -+ core automatically transitions to the highest performance non-boosted P-state. -+ AMD Core Performance Boost(CPB) is controlled by this attribute file which allows -+ user to change all cores frequency boosting state. It supports all amd-pstate modes. -+ -+ States of the driver "/sys/devices/system/cpu/amd_pstate/cpb_boost" -+ "disabled" Core Performance Boosting Disabled. -+ "enabled" Core Performance Boosting Enabled. -+ -+ To enable core performance boost: -+ # echo "enabled" > /sys/devices/system/cpu/amd_pstate/cpb_boost -+ -+ To disable core performance boost: -+ # echo "disabled" > /sys/devices/system/cpu/amd_pstate/cpb_boost -+ - ``cpupower`` tool support for ``amd-pstate`` - =============================================== - diff --git a/arch/x86/include/asm/cpufeatures.h b/arch/x86/include/asm/cpufeatures.h -index 3c7434329661c..6c128d463a143 100644 +index 3c7434329661..6c128d463a14 100644 --- a/arch/x86/include/asm/cpufeatures.h +++ b/arch/x86/include/asm/cpufeatures.h @@ -470,6 +470,7 @@ @@ -95,7 +65,7 @@ index 3c7434329661c..6c128d463a143 100644 /* * BUG word(s) diff --git a/arch/x86/include/asm/msr-index.h b/arch/x86/include/asm/msr-index.h -index e72c2b8729579..8738a7b3917d8 100644 +index e72c2b872957..8738a7b3917d 100644 --- a/arch/x86/include/asm/msr-index.h +++ b/arch/x86/include/asm/msr-index.h @@ -782,6 +782,8 @@ @@ -108,7 +78,7 @@ index e72c2b8729579..8738a7b3917d8 100644 /* K6 MSRs */ #define MSR_K6_WHCR 0xc0000082 diff --git a/arch/x86/kernel/cpu/scattered.c b/arch/x86/kernel/cpu/scattered.c -index af5aa2c754c22..c84c30188fdf2 100644 +index af5aa2c754c2..c84c30188fdf 100644 --- a/arch/x86/kernel/cpu/scattered.c +++ b/arch/x86/kernel/cpu/scattered.c @@ -45,6 +45,7 @@ static const struct cpuid_bit cpuid_bits[] = { @@ -119,8 +89,20 @@ index af5aa2c754c22..c84c30188fdf2 100644 { X86_FEATURE_MBA, CPUID_EBX, 6, 0x80000008, 0 }, { X86_FEATURE_SMBA, CPUID_EBX, 2, 0x80000020, 0 }, { X86_FEATURE_BMEC, CPUID_EBX, 3, 0x80000020, 0 }, +diff --git a/drivers/cpufreq/Kconfig.x86 b/drivers/cpufreq/Kconfig.x86 +index 438c9e75a04d..97c2d4f15d76 100644 +--- a/drivers/cpufreq/Kconfig.x86 ++++ b/drivers/cpufreq/Kconfig.x86 +@@ -71,6 +71,7 @@ config X86_AMD_PSTATE_DEFAULT_MODE + config X86_AMD_PSTATE_UT + tristate "selftest for AMD Processor P-State driver" + depends on X86 && ACPI_PROCESSOR ++ depends on X86_AMD_PSTATE + default n + help + This kernel module is used for testing. It's safe to say M here. diff --git a/drivers/cpufreq/acpi-cpufreq.c b/drivers/cpufreq/acpi-cpufreq.c -index 37f1cdf46d291..2fc82831bddd5 100644 +index 37f1cdf46d29..094f1f91a9b3 100644 --- a/drivers/cpufreq/acpi-cpufreq.c +++ b/drivers/cpufreq/acpi-cpufreq.c @@ -50,8 +50,6 @@ enum { @@ -132,21 +114,16 @@ index 37f1cdf46d291..2fc82831bddd5 100644 struct acpi_cpufreq_data { unsigned int resume; unsigned int cpu_feature; -diff --git a/drivers/cpufreq/amd-pstate-ut.c b/drivers/cpufreq/amd-pstate-ut.c -index fc275d41d51e9..b528f198f4c36 100644 ---- a/drivers/cpufreq/amd-pstate-ut.c -+++ b/drivers/cpufreq/amd-pstate-ut.c -@@ -227,7 +227,7 @@ static void amd_pstate_ut_check_freq(u32 index) - goto skip_test; - } +@@ -139,6 +137,7 @@ static int set_boost(struct cpufreq_policy *policy, int val) + (void *)(long)val, 1); + pr_debug("CPU %*pbl: Core Boosting %s.\n", + cpumask_pr_args(policy->cpus), str_enabled_disabled(val)); ++ policy->boost_enabled = val; -- if (cpudata->boost_supported) { -+ if (amd_pstate_global_params.cpb_boost) { - if ((policy->max == cpudata->max_freq) || - (policy->max == cpudata->nominal_freq)) - amd_pstate_ut_cases[index].result = AMD_PSTATE_UT_RESULT_PASS; + return 0; + } diff --git a/drivers/cpufreq/amd-pstate.c b/drivers/cpufreq/amd-pstate.c -index 6af175e6c08ac..6eeba793bf442 100644 +index 6af175e6c08a..80eaa58f1405 100644 --- a/drivers/cpufreq/amd-pstate.c +++ b/drivers/cpufreq/amd-pstate.c @@ -51,6 +51,7 @@ @@ -173,36 +150,7 @@ index 6af175e6c08ac..6eeba793bf442 100644 static struct cpufreq_driver *current_pstate_driver; static struct cpufreq_driver amd_pstate_driver; static struct cpufreq_driver amd_pstate_epp_driver; -@@ -102,6 +94,11 @@ static bool cppc_enabled; - static bool amd_pstate_prefcore = true; - static struct quirk_entry *quirks; - -+/* export the amd_pstate_global_params for unit test */ -+struct amd_pstate_global_params amd_pstate_global_params; -+EXPORT_SYMBOL_GPL(amd_pstate_global_params); -+static int amd_pstate_cpu_boost(int cpu, bool state); -+ - /* - * AMD Energy Preference Performance (EPP) - * The EPP is used in the CCLK DPM controller to drive -@@ -143,6 +140,16 @@ static unsigned int epp_values[] = { - [EPP_INDEX_POWERSAVE] = AMD_CPPC_EPP_POWERSAVE, - }; - -+enum CPB_STATE_INDEX { -+ CPB_STATE_DISABLED = 0, -+ CPB_STATE_ENABLED = 1, -+}; -+ -+static const char * const cpb_state[] = { -+ [CPB_STATE_DISABLED] = "disabled", -+ [CPB_STATE_ENABLED] = "enabled", -+}; -+ - typedef int (*cppc_mode_transition_fn)(int); - - static struct quirk_entry quirk_amd_7k62 = { -@@ -157,7 +164,7 @@ static int __init dmi_matched_7k62_bios_bug(const struct dmi_system_id *dmi) +@@ -157,7 +149,7 @@ static int __init dmi_matched_7k62_bios_bug(const struct dmi_system_id *dmi) * broken BIOS lack of nominal_freq and lowest_freq capabilities * definition in ACPI tables */ @@ -211,7 +159,7 @@ index 6af175e6c08ac..6eeba793bf442 100644 quirks = dmi->driver_data; pr_info("Overriding nominal and lowest frequencies for %s\n", dmi->ident); return 1; -@@ -199,7 +206,7 @@ static s16 amd_pstate_get_epp(struct amd_cpudata *cpudata, u64 cppc_req_cached) +@@ -199,7 +191,7 @@ static s16 amd_pstate_get_epp(struct amd_cpudata *cpudata, u64 cppc_req_cached) u64 epp; int ret; @@ -220,7 +168,7 @@ index 6af175e6c08ac..6eeba793bf442 100644 if (!cppc_req_cached) { epp = rdmsrl_on_cpu(cpudata->cpu, MSR_AMD_CPPC_REQ, &cppc_req_cached); -@@ -252,7 +259,7 @@ static int amd_pstate_set_epp(struct amd_cpudata *cpudata, u32 epp) +@@ -252,7 +244,7 @@ static int amd_pstate_set_epp(struct amd_cpudata *cpudata, u32 epp) int ret; struct cppc_perf_ctrls perf_ctrls; @@ -229,7 +177,7 @@ index 6af175e6c08ac..6eeba793bf442 100644 u64 value = READ_ONCE(cpudata->cppc_req_cached); value &= ~GENMASK_ULL(31, 24); -@@ -281,10 +288,8 @@ static int amd_pstate_set_energy_pref_index(struct amd_cpudata *cpudata, +@@ -281,10 +273,8 @@ static int amd_pstate_set_energy_pref_index(struct amd_cpudata *cpudata, int epp = -EINVAL; int ret; @@ -242,7 +190,7 @@ index 6af175e6c08ac..6eeba793bf442 100644 if (epp == -EINVAL) epp = epp_values[pref_index]; -@@ -521,7 +526,10 @@ static inline bool amd_pstate_sample(struct amd_cpudata *cpudata) +@@ -521,7 +511,10 @@ static inline bool amd_pstate_sample(struct amd_cpudata *cpudata) static void amd_pstate_update(struct amd_cpudata *cpudata, u32 min_perf, u32 des_perf, u32 max_perf, bool fast_switch, int gov_flags) { @@ -253,7 +201,7 @@ index 6af175e6c08ac..6eeba793bf442 100644 u64 value = prev; min_perf = clamp_t(unsigned long, min_perf, cpudata->min_limit_perf, -@@ -530,6 +538,9 @@ static void amd_pstate_update(struct amd_cpudata *cpudata, u32 min_perf, +@@ -530,6 +523,9 @@ static void amd_pstate_update(struct amd_cpudata *cpudata, u32 min_perf, cpudata->max_limit_perf); des_perf = clamp_t(unsigned long, des_perf, min_perf, max_perf); @@ -263,18 +211,18 @@ index 6af175e6c08ac..6eeba793bf442 100644 if ((cppc_state == AMD_PSTATE_GUIDED) && (gov_flags & CPUFREQ_GOV_DYNAMIC_SWITCHING)) { min_perf = des_perf; des_perf = 0; -@@ -541,6 +552,10 @@ static void amd_pstate_update(struct amd_cpudata *cpudata, u32 min_perf, +@@ -541,6 +537,10 @@ static void amd_pstate_update(struct amd_cpudata *cpudata, u32 min_perf, value &= ~AMD_CPPC_DES_PERF(~0L); value |= AMD_CPPC_DES_PERF(des_perf); + /* limit the max perf when core performance boost feature is disabled */ -+ if (!amd_pstate_global_params.cpb_boost) ++ if (!cpudata->boost_supported) + max_perf = min_t(unsigned long, nominal_perf, max_perf); + value &= ~AMD_CPPC_MAX_PERF(~0L); value |= AMD_CPPC_MAX_PERF(max_perf); -@@ -651,10 +666,9 @@ static void amd_pstate_adjust_perf(unsigned int cpu, +@@ -651,10 +651,9 @@ static void amd_pstate_adjust_perf(unsigned int cpu, unsigned long capacity) { unsigned long max_perf, min_perf, des_perf, @@ -286,7 +234,7 @@ index 6af175e6c08ac..6eeba793bf442 100644 if (policy->min != cpudata->min_limit_freq || policy->max != cpudata->max_limit_freq) amd_pstate_update_min_max_limit(policy); -@@ -662,7 +676,6 @@ static void amd_pstate_adjust_perf(unsigned int cpu, +@@ -662,7 +661,6 @@ static void amd_pstate_adjust_perf(unsigned int cpu, cap_perf = READ_ONCE(cpudata->highest_perf); lowest_nonlinear_perf = READ_ONCE(cpudata->lowest_nonlinear_perf); @@ -294,7 +242,7 @@ index 6af175e6c08ac..6eeba793bf442 100644 des_perf = cap_perf; if (target_perf < capacity) -@@ -680,34 +693,12 @@ static void amd_pstate_adjust_perf(unsigned int cpu, +@@ -680,32 +678,57 @@ static void amd_pstate_adjust_perf(unsigned int cpu, max_perf = min_perf; des_perf = clamp_t(unsigned long, des_perf, min_perf, max_perf); @@ -307,66 +255,100 @@ index 6af175e6c08ac..6eeba793bf442 100644 } -static int amd_get_min_freq(struct amd_cpudata *cpudata) --{ ++static int amd_pstate_cpu_boost_update(struct cpufreq_policy *policy, bool on) + { - return READ_ONCE(cpudata->min_freq); -} -- ++ struct amd_cpudata *cpudata = policy->driver_data; ++ struct cppc_perf_ctrls perf_ctrls; ++ u32 highest_perf, nominal_perf, nominal_freq, max_freq; ++ int ret; + -static int amd_get_max_freq(struct amd_cpudata *cpudata) -{ - return READ_ONCE(cpudata->max_freq); -} -- ++ highest_perf = READ_ONCE(cpudata->highest_perf); ++ nominal_perf = READ_ONCE(cpudata->nominal_perf); ++ nominal_freq = READ_ONCE(cpudata->nominal_freq); ++ max_freq = READ_ONCE(cpudata->max_freq); + -static int amd_get_nominal_freq(struct amd_cpudata *cpudata) -{ - return READ_ONCE(cpudata->nominal_freq); -} -- ++ if (boot_cpu_has(X86_FEATURE_CPPC)) { ++ u64 value = READ_ONCE(cpudata->cppc_req_cached); + -static int amd_get_lowest_nonlinear_freq(struct amd_cpudata *cpudata) -{ - return READ_ONCE(cpudata->lowest_nonlinear_freq); --} -- ++ value &= ~GENMASK_ULL(7, 0); ++ value |= on ? highest_perf : nominal_perf; ++ WRITE_ONCE(cpudata->cppc_req_cached, value); ++ ++ wrmsrl_on_cpu(cpudata->cpu, MSR_AMD_CPPC_REQ, value); ++ } else { ++ perf_ctrls.max_perf = on ? highest_perf : nominal_perf; ++ ret = cppc_set_perf(cpudata->cpu, &perf_ctrls); ++ if (ret) { ++ cpufreq_cpu_release(policy); ++ pr_debug("Failed to set max perf on CPU:%d. ret:%d\n", ++ cpudata->cpu, ret); ++ return ret; ++ } ++ } ++ ++ if (on) ++ policy->cpuinfo.max_freq = max_freq; ++ else if (policy->cpuinfo.max_freq > nominal_freq * 1000) ++ policy->cpuinfo.max_freq = nominal_freq * 1000; ++ ++ policy->max = policy->cpuinfo.max_freq; ++ ++ if (cppc_state == AMD_PSTATE_PASSIVE) { ++ ret = freq_qos_update_request(&cpudata->req[1], policy->cpuinfo.max_freq); ++ if (ret < 0) ++ pr_debug("Failed to update freq constraint: CPU%d\n", cpudata->cpu); ++ } ++ ++ return ret < 0 ? ret : 0; + } + static int amd_pstate_set_boost(struct cpufreq_policy *policy, int state) - { - struct amd_cpudata *cpudata = policy->driver_data; -@@ -715,36 +706,53 @@ static int amd_pstate_set_boost(struct cpufreq_policy *policy, int state) +@@ -715,36 +738,51 @@ static int amd_pstate_set_boost(struct cpufreq_policy *policy, int state) if (!cpudata->boost_supported) { pr_err("Boost mode is not supported by this processor or SBIOS\n"); - return -EINVAL; -+ return -ENOTSUPP; ++ return -EOPNOTSUPP; } + mutex_lock(&amd_pstate_driver_lock); -+ ret = amd_pstate_cpu_boost(policy->cpu, state); ++ ret = amd_pstate_cpu_boost_update(policy, state); ++ WRITE_ONCE(cpudata->boost_state, !ret ? state : false); ++ policy->boost_enabled = !ret ? state : false; ++ refresh_frequency_limits(policy); + mutex_unlock(&amd_pstate_driver_lock); - if (state) - policy->cpuinfo.max_freq = cpudata->max_freq; - else - policy->cpuinfo.max_freq = cpudata->nominal_freq * 1000; -- -- policy->max = policy->cpuinfo.max_freq; -- -- ret = freq_qos_update_request(&cpudata->req[1], -- policy->cpuinfo.max_freq); -- if (ret < 0) -- return ret; -- -- return 0; -+ return ret < 0 ? ret : 0; - } ++ return ret; ++} --static void amd_pstate_boost_init(struct amd_cpudata *cpudata) +- policy->max = policy->cpuinfo.max_freq; +static int amd_pstate_init_boost_support(struct amd_cpudata *cpudata) - { -- u32 highest_perf, nominal_perf; ++{ + u64 boost_val; + int ret = -1; -- highest_perf = READ_ONCE(cpudata->highest_perf); -- nominal_perf = READ_ONCE(cpudata->nominal_perf); +- ret = freq_qos_update_request(&cpudata->req[1], +- policy->cpuinfo.max_freq); +- if (ret < 0) +- return ret; + /* -+ * If platform has no CPB support or disble it, initialize current driver ++ * If platform has no CPB support or disable it, initialize current driver + * boost_enabled state to be false, it is not an error for cpufreq core to handle. + */ + if (!cpu_feature_enabled(X86_FEATURE_CPB)) { @@ -375,37 +357,39 @@ index 6af175e6c08ac..6eeba793bf442 100644 + goto exit_err; + } -- if (highest_perf <= nominal_perf) -- return; +- return 0; +-} ++ /* at least one CPU supports CPB, even if others fail later on to set up */ ++ current_pstate_driver->boost_enabled = true; + +-static void amd_pstate_boost_init(struct amd_cpudata *cpudata) +-{ +- u32 highest_perf, nominal_perf; + ret = rdmsrl_on_cpu(cpudata->cpu, MSR_K7_HWCR, &boost_val); + if (ret) { + pr_err_once("failed to read initial CPU boost state!\n"); + ret = -EIO; + goto exit_err; + } -+ -+ amd_pstate_global_params.cpb_supported = !(boost_val & MSR_K7_HWCR_CPB_DIS); -+ if (amd_pstate_global_params.cpb_supported) { -+ current_pstate_driver->boost_enabled = true; + +- highest_perf = READ_ONCE(cpudata->highest_perf); +- nominal_perf = READ_ONCE(cpudata->nominal_perf); ++ if (!(boost_val & MSR_K7_HWCR_CPB_DIS)) + cpudata->boost_supported = true; -+ cpudata->boost_state = true; -+ } + +- if (highest_perf <= nominal_perf) +- return; ++ return 0; - cpudata->boost_supported = true; - current_pstate_driver->boost_enabled = true; -+ amd_pstate_global_params.cpb_boost = amd_pstate_global_params.cpb_supported; -+ return 0; -+ +exit_err: + cpudata->boost_supported = false; -+ cpudata->boost_state = false; -+ current_pstate_driver->boost_enabled = false; -+ amd_pstate_global_params.cpb_boost = false; + return ret; } static void amd_perf_ctl_reset(unsigned int cpu) -@@ -773,7 +781,7 @@ static int amd_pstate_get_highest_perf(int cpu, u32 *highest_perf) +@@ -773,7 +811,7 @@ static int amd_pstate_get_highest_perf(int cpu, u32 *highest_perf) { int ret; @@ -414,7 +398,7 @@ index 6af175e6c08ac..6eeba793bf442 100644 u64 cap1; ret = rdmsrl_safe_on_cpu(cpu, MSR_AMD_CPPC_CAP1, &cap1); -@@ -860,7 +868,41 @@ static void amd_pstate_update_limits(unsigned int cpu) +@@ -860,7 +898,41 @@ static void amd_pstate_update_limits(unsigned int cpu) mutex_unlock(&amd_pstate_driver_lock); } @@ -457,7 +441,7 @@ index 6af175e6c08ac..6eeba793bf442 100644 * amd_pstate_init_freq: Initialize the max_freq, min_freq, * nominal_freq and lowest_nonlinear_freq for * the @cpudata object. -@@ -881,7 +923,6 @@ static int amd_pstate_init_freq(struct amd_cpudata *cpudata) +@@ -881,7 +953,6 @@ static int amd_pstate_init_freq(struct amd_cpudata *cpudata) u32 boost_ratio, lowest_nonlinear_ratio; struct cppc_perf_caps cppc_perf; @@ -465,7 +449,7 @@ index 6af175e6c08ac..6eeba793bf442 100644 ret = cppc_get_perf_caps(cpudata->cpu, &cppc_perf); if (ret) return ret; -@@ -912,12 +953,30 @@ static int amd_pstate_init_freq(struct amd_cpudata *cpudata) +@@ -912,12 +983,30 @@ static int amd_pstate_init_freq(struct amd_cpudata *cpudata) WRITE_ONCE(cpudata->nominal_freq, nominal_freq); WRITE_ONCE(cpudata->max_freq, max_freq); @@ -497,7 +481,7 @@ index 6af175e6c08ac..6eeba793bf442 100644 struct device *dev; struct amd_cpudata *cpudata; -@@ -946,20 +1005,11 @@ static int amd_pstate_cpu_init(struct cpufreq_policy *policy) +@@ -946,20 +1035,15 @@ static int amd_pstate_cpu_init(struct cpufreq_policy *policy) if (ret) goto free_cpudata1; @@ -510,19 +494,27 @@ index 6af175e6c08ac..6eeba793bf442 100644 - dev_err(dev, "min_freq(%d) or max_freq(%d) value is incorrect\n", - min_freq, max_freq); - ret = -EINVAL; -- goto free_cpudata1; ++ ret = amd_pstate_init_boost_support(cpudata); ++ if (ret) + goto free_cpudata1; - } -+ min_freq = READ_ONCE(cpudata->min_freq); -+ max_freq = READ_ONCE(cpudata->max_freq); - policy->cpuinfo.transition_latency = AMD_PSTATE_TRANSITION_LATENCY; - policy->transition_delay_us = AMD_PSTATE_TRANSITION_DELAY; ++ min_freq = READ_ONCE(cpudata->min_freq); ++ max_freq = READ_ONCE(cpudata->max_freq); ++ + policy->cpuinfo.transition_latency = amd_pstate_get_transition_latency(policy->cpu); + policy->transition_delay_us = amd_pstate_get_transition_delay_us(policy->cpu); policy->min = min_freq; policy->max = max_freq; -@@ -970,7 +1020,7 @@ static int amd_pstate_cpu_init(struct cpufreq_policy *policy) +@@ -967,10 +1051,12 @@ static int amd_pstate_cpu_init(struct cpufreq_policy *policy) + policy->cpuinfo.min_freq = min_freq; + policy->cpuinfo.max_freq = max_freq; + ++ policy->boost_enabled = READ_ONCE(cpudata->boost_supported); ++ /* It will be updated by governor */ policy->cur = policy->cpuinfo.min_freq; @@ -531,7 +523,7 @@ index 6af175e6c08ac..6eeba793bf442 100644 policy->fast_switch_possible = true; ret = freq_qos_add_request(&policy->constraints, &cpudata->req[0], -@@ -992,7 +1042,6 @@ static int amd_pstate_cpu_init(struct cpufreq_policy *policy) +@@ -992,7 +1078,6 @@ static int amd_pstate_cpu_init(struct cpufreq_policy *policy) policy->driver_data = cpudata; @@ -539,7 +531,7 @@ index 6af175e6c08ac..6eeba793bf442 100644 if (!current_pstate_driver->adjust_perf) current_pstate_driver->adjust_perf = amd_pstate_adjust_perf; -@@ -1052,7 +1101,7 @@ static ssize_t show_amd_pstate_max_freq(struct cpufreq_policy *policy, +@@ -1052,7 +1137,7 @@ static ssize_t show_amd_pstate_max_freq(struct cpufreq_policy *policy, int max_freq; struct amd_cpudata *cpudata = policy->driver_data; @@ -548,7 +540,7 @@ index 6af175e6c08ac..6eeba793bf442 100644 if (max_freq < 0) return max_freq; -@@ -1065,7 +1114,7 @@ static ssize_t show_amd_pstate_lowest_nonlinear_freq(struct cpufreq_policy *poli +@@ -1065,7 +1150,7 @@ static ssize_t show_amd_pstate_lowest_nonlinear_freq(struct cpufreq_policy *poli int freq; struct amd_cpudata *cpudata = policy->driver_data; @@ -557,7 +549,7 @@ index 6af175e6c08ac..6eeba793bf442 100644 if (freq < 0) return freq; -@@ -1203,7 +1252,7 @@ static int amd_pstate_change_mode_without_dvr_change(int mode) +@@ -1203,7 +1288,7 @@ static int amd_pstate_change_mode_without_dvr_change(int mode) cppc_state = mode; @@ -566,160 +558,8 @@ index 6af175e6c08ac..6eeba793bf442 100644 return 0; for_each_present_cpu(cpu) { -@@ -1312,6 +1361,118 @@ static ssize_t prefcore_show(struct device *dev, - return sysfs_emit(buf, "%s\n", str_enabled_disabled(amd_pstate_prefcore)); - } - -+static int amd_pstate_cpu_boost_update(struct cpufreq_policy *policy, bool on) -+{ -+ struct amd_cpudata *cpudata = policy->driver_data; -+ struct cppc_perf_ctrls perf_ctrls; -+ u32 highest_perf, nominal_perf, nominal_freq, max_freq; -+ int ret; -+ -+ highest_perf = READ_ONCE(cpudata->highest_perf); -+ nominal_perf = READ_ONCE(cpudata->nominal_perf); -+ nominal_freq = READ_ONCE(cpudata->nominal_freq); -+ max_freq = READ_ONCE(cpudata->max_freq); -+ -+ if (boot_cpu_has(X86_FEATURE_CPPC)) { -+ u64 value = READ_ONCE(cpudata->cppc_req_cached); -+ -+ value &= ~GENMASK_ULL(7, 0); -+ value |= on ? highest_perf : nominal_perf; -+ WRITE_ONCE(cpudata->cppc_req_cached, value); -+ -+ wrmsrl_on_cpu(cpudata->cpu, MSR_AMD_CPPC_REQ, value); -+ } else { -+ perf_ctrls.max_perf = on ? highest_perf : nominal_perf; -+ ret = cppc_set_perf(cpudata->cpu, &perf_ctrls); -+ if (ret) { -+ cpufreq_cpu_release(policy); -+ pr_debug("Failed to set max perf on CPU:%d. ret:%d\n", -+ cpudata->cpu, ret); -+ return ret; -+ } -+ } -+ -+ if (on) -+ policy->cpuinfo.max_freq = max_freq; -+ else -+ policy->cpuinfo.max_freq = nominal_freq * 1000; -+ -+ policy->max = policy->cpuinfo.max_freq; -+ -+ if (cppc_state == AMD_PSTATE_PASSIVE) { -+ ret = freq_qos_update_request(&cpudata->req[1], policy->cpuinfo.max_freq); -+ if (ret < 0) -+ pr_debug("Failed to update freq constraint: CPU%d\n", cpudata->cpu); -+ } -+ -+ return ret < 0 ? ret : 0; -+} -+ -+static int amd_pstate_cpu_boost(int cpu, bool state) -+{ -+ int ret; -+ struct cpufreq_policy *policy = cpufreq_cpu_get(cpu); -+ struct amd_cpudata *cpudata = policy->driver_data; -+ -+ if (!policy) { -+ pr_err("policy is NULL\n"); -+ ret = -ENODATA; -+ goto err_exit; -+ } -+ -+ ret = amd_pstate_cpu_boost_update(policy, state); -+ refresh_frequency_limits(policy); -+ WRITE_ONCE(cpudata->boost_state, state); -+ policy->boost_enabled = state; -+ -+err_exit: -+ cpufreq_cpu_put(policy); -+ return ret < 0 ? ret : 0; -+} -+ -+static ssize_t cpb_boost_show(struct device *dev, -+ struct device_attribute *attr, char *buf) -+{ -+ bool cpb_idx; -+ -+ cpb_idx = amd_pstate_global_params.cpb_boost; -+ -+ return sysfs_emit(buf, "%s\n", cpb_state[cpb_idx]); -+} -+ -+static ssize_t cpb_boost_store(struct device *dev, struct device_attribute *b, -+ const char *buf, size_t count) -+{ -+ bool new_state; -+ ssize_t ret; -+ int cpu, cpb_idx; -+ -+ if (!amd_pstate_global_params.cpb_supported) { -+ pr_err("Boost mode is not supported by this processor or SBIOS\n"); -+ return -EINVAL; -+ } -+ -+ cpb_idx = sysfs_match_string(cpb_state, buf); -+ if (cpb_idx < 0) -+ return -EINVAL; -+ -+ new_state = cpb_idx; -+ -+ mutex_lock(&amd_pstate_driver_lock); -+ for_each_present_cpu(cpu) { -+ ret = amd_pstate_cpu_boost(cpu, new_state); -+ if (ret < 0) { -+ pr_warn("failed to update cpu boost for CPU%d (%zd)\n", cpu, ret); -+ goto err_exit; -+ } -+ } -+ amd_pstate_global_params.cpb_boost = !!new_state; -+ -+err_exit: -+ mutex_unlock(&amd_pstate_driver_lock); -+ return ret < 0 ? ret : count; -+} -+ - cpufreq_freq_attr_ro(amd_pstate_max_freq); - cpufreq_freq_attr_ro(amd_pstate_lowest_nonlinear_freq); - -@@ -1322,6 +1483,7 @@ cpufreq_freq_attr_rw(energy_performance_preference); - cpufreq_freq_attr_ro(energy_performance_available_preferences); - static DEVICE_ATTR_RW(status); - static DEVICE_ATTR_RO(prefcore); -+static DEVICE_ATTR_RW(cpb_boost); - - static struct freq_attr *amd_pstate_attr[] = { - &amd_pstate_max_freq, -@@ -1346,6 +1508,7 @@ static struct freq_attr *amd_pstate_epp_attr[] = { - static struct attribute *pstate_global_attributes[] = { - &dev_attr_status.attr, - &dev_attr_prefcore.attr, -+ &dev_attr_cpb_boost.attr, - NULL - }; +@@ -1376,7 +1461,7 @@ static bool amd_pstate_acpi_pm_profile_undefined(void) -@@ -1374,9 +1537,24 @@ static bool amd_pstate_acpi_pm_profile_undefined(void) - return false; - } - -+static int amd_pstate_init_boost(struct cpufreq_policy *policy) -+{ -+ struct amd_cpudata *cpudata = policy->driver_data; -+ int ret; -+ -+ /* initialize cpu cores boot state */ -+ ret = amd_pstate_init_boost_support(cpudata); -+ if (ret) -+ return ret; -+ -+ policy->boost_enabled = READ_ONCE(cpudata->boost_state); -+ -+ return 0; -+} -+ static int amd_pstate_epp_cpu_init(struct cpufreq_policy *policy) { - int min_freq, max_freq, nominal_freq, lowest_nonlinear_freq, ret; @@ -727,7 +567,7 @@ index 6af175e6c08ac..6eeba793bf442 100644 struct amd_cpudata *cpudata; struct device *dev; u64 value; -@@ -1407,16 +1585,8 @@ static int amd_pstate_epp_cpu_init(struct cpufreq_policy *policy) +@@ -1407,16 +1492,12 @@ static int amd_pstate_epp_cpu_init(struct cpufreq_policy *policy) if (ret) goto free_cpudata1; @@ -739,14 +579,17 @@ index 6af175e6c08ac..6eeba793bf442 100644 - dev_err(dev, "min_freq(%d) or max_freq(%d) value is incorrect\n", - min_freq, max_freq); - ret = -EINVAL; -- goto free_cpudata1; ++ ret = amd_pstate_init_boost_support(cpudata); ++ if (ret) + goto free_cpudata1; - } ++ + min_freq = READ_ONCE(cpudata->min_freq); + max_freq = READ_ONCE(cpudata->max_freq); policy->cpuinfo.min_freq = min_freq; policy->cpuinfo.max_freq = max_freq; -@@ -1425,7 +1595,7 @@ static int amd_pstate_epp_cpu_init(struct cpufreq_policy *policy) +@@ -1425,11 +1506,13 @@ static int amd_pstate_epp_cpu_init(struct cpufreq_policy *policy) policy->driver_data = cpudata; @@ -755,7 +598,13 @@ index 6af175e6c08ac..6eeba793bf442 100644 policy->min = policy->cpuinfo.min_freq; policy->max = policy->cpuinfo.max_freq; -@@ -1440,7 +1610,7 @@ static int amd_pstate_epp_cpu_init(struct cpufreq_policy *policy) + ++ policy->boost_enabled = READ_ONCE(cpudata->boost_supported); ++ + /* + * Set the policy to provide a valid fallback value in case + * the default cpufreq governor is neither powersave nor performance. +@@ -1440,7 +1523,7 @@ static int amd_pstate_epp_cpu_init(struct cpufreq_policy *policy) else policy->policy = CPUFREQ_POLICY_POWERSAVE; @@ -764,7 +613,7 @@ index 6af175e6c08ac..6eeba793bf442 100644 ret = rdmsrl_on_cpu(cpudata->cpu, MSR_AMD_CPPC_REQ, &value); if (ret) return ret; -@@ -1451,7 +1621,6 @@ static int amd_pstate_epp_cpu_init(struct cpufreq_policy *policy) +@@ -1451,7 +1534,6 @@ static int amd_pstate_epp_cpu_init(struct cpufreq_policy *policy) return ret; WRITE_ONCE(cpudata->cppc_cap1_cached, value); } @@ -772,7 +621,7 @@ index 6af175e6c08ac..6eeba793bf442 100644 return 0; -@@ -1530,7 +1699,7 @@ static void amd_pstate_epp_update_limit(struct cpufreq_policy *policy) +@@ -1530,7 +1612,7 @@ static void amd_pstate_epp_update_limit(struct cpufreq_policy *policy) epp = 0; /* Set initial EPP value */ @@ -781,7 +630,7 @@ index 6af175e6c08ac..6eeba793bf442 100644 value &= ~GENMASK_ULL(31, 24); value |= (u64)epp << 24; } -@@ -1553,6 +1722,12 @@ static int amd_pstate_epp_set_policy(struct cpufreq_policy *policy) +@@ -1553,6 +1635,12 @@ static int amd_pstate_epp_set_policy(struct cpufreq_policy *policy) amd_pstate_epp_update_limit(policy); @@ -794,7 +643,7 @@ index 6af175e6c08ac..6eeba793bf442 100644 return 0; } -@@ -1569,7 +1744,7 @@ static void amd_pstate_epp_reenable(struct amd_cpudata *cpudata) +@@ -1569,7 +1657,7 @@ static void amd_pstate_epp_reenable(struct amd_cpudata *cpudata) value = READ_ONCE(cpudata->cppc_req_cached); max_perf = READ_ONCE(cpudata->highest_perf); @@ -803,7 +652,7 @@ index 6af175e6c08ac..6eeba793bf442 100644 wrmsrl_on_cpu(cpudata->cpu, MSR_AMD_CPPC_REQ, value); } else { perf_ctrls.max_perf = max_perf; -@@ -1603,7 +1778,7 @@ static void amd_pstate_epp_offline(struct cpufreq_policy *policy) +@@ -1603,7 +1691,7 @@ static void amd_pstate_epp_offline(struct cpufreq_policy *policy) value = READ_ONCE(cpudata->cppc_req_cached); mutex_lock(&amd_pstate_limits_lock); @@ -812,24 +661,15 @@ index 6af175e6c08ac..6eeba793bf442 100644 cpudata->epp_policy = CPUFREQ_POLICY_UNKNOWN; /* Set max perf same as min perf */ -@@ -1690,6 +1865,7 @@ static struct cpufreq_driver amd_pstate_driver = { - .exit = amd_pstate_cpu_exit, - .suspend = amd_pstate_cpu_suspend, - .resume = amd_pstate_cpu_resume, -+ .init_boost = amd_pstate_init_boost, - .set_boost = amd_pstate_set_boost, - .update_limits = amd_pstate_update_limits, - .name = "amd-pstate", -@@ -1707,6 +1883,8 @@ static struct cpufreq_driver amd_pstate_epp_driver = { +@@ -1707,6 +1795,7 @@ static struct cpufreq_driver amd_pstate_epp_driver = { .suspend = amd_pstate_epp_suspend, .resume = amd_pstate_epp_resume, .update_limits = amd_pstate_update_limits, -+ .init_boost = amd_pstate_init_boost, + .set_boost = amd_pstate_set_boost, .name = "amd-pstate-epp", .attr = amd_pstate_epp_attr, }; -@@ -1730,6 +1908,46 @@ static int __init amd_pstate_set_driver(int mode_idx) +@@ -1730,6 +1819,46 @@ static int __init amd_pstate_set_driver(int mode_idx) return -EINVAL; } @@ -876,7 +716,7 @@ index 6af175e6c08ac..6eeba793bf442 100644 static int __init amd_pstate_init(void) { struct device *dev_root; -@@ -1738,6 +1956,11 @@ static int __init amd_pstate_init(void) +@@ -1738,6 +1867,11 @@ static int __init amd_pstate_init(void) if (boot_cpu_data.x86_vendor != X86_VENDOR_AMD) return -ENODEV; @@ -888,7 +728,7 @@ index 6af175e6c08ac..6eeba793bf442 100644 if (!acpi_cpc_valid()) { pr_warn_once("the _CPC object is not present in SBIOS or ACPI disabled\n"); return -ENODEV; -@@ -1752,35 +1975,43 @@ static int __init amd_pstate_init(void) +@@ -1752,35 +1886,43 @@ static int __init amd_pstate_init(void) /* check if this machine need CPPC quirks */ dmi_check_system(amd_pstate_quirks_table); @@ -942,7 +782,7 @@ index 6af175e6c08ac..6eeba793bf442 100644 pr_debug("AMD CPPC MSR based functionality is supported\n"); if (cppc_state != AMD_PSTATE_ACTIVE) current_pstate_driver->adjust_perf = amd_pstate_adjust_perf; -@@ -1794,13 +2025,15 @@ static int __init amd_pstate_init(void) +@@ -1794,13 +1936,15 @@ static int __init amd_pstate_init(void) /* enable amd pstate feature */ ret = amd_pstate_enable(true); if (ret) { @@ -953,24 +793,24 @@ index 6af175e6c08ac..6eeba793bf442 100644 ret = cpufreq_register_driver(current_pstate_driver); - if (ret) -+ if (ret) { ++ if (ret) { pr_err("failed to register with return %d\n", ret); -+ goto disable_driver; -+ } ++ goto disable_driver; ++ } dev_root = bus_get_dev_root(&cpu_subsys); if (dev_root) { -@@ -1816,6 +2049,8 @@ static int __init amd_pstate_init(void) +@@ -1816,6 +1960,8 @@ static int __init amd_pstate_init(void) global_attr_free: cpufreq_unregister_driver(current_pstate_driver); +disable_driver: -+ amd_pstate_enable(false); ++ amd_pstate_enable(false); return ret; } device_initcall(amd_pstate_init); diff --git a/drivers/cpufreq/amd-pstate.h b/drivers/cpufreq/amd-pstate.h -index bc341f35908d7..fb240a8702892 100644 +index bc341f35908d..cc8bb2bc325a 100644 --- a/drivers/cpufreq/amd-pstate.h +++ b/drivers/cpufreq/amd-pstate.h @@ -42,13 +42,17 @@ struct amd_aperf_mperf { @@ -996,7 +836,7 @@ index bc341f35908d7..fb240a8702892 100644 * @boost_supported: check whether the Processor or SBIOS supports boost mode * @hw_prefcore: check whether HW supports preferred core featue. * Only when hw_prefcore and early prefcore param are true, -@@ -95,6 +99,21 @@ struct amd_cpudata { +@@ -95,6 +99,8 @@ struct amd_cpudata { u32 policy; u64 cppc_cap1_cached; bool suspended; @@ -1004,22 +844,9 @@ index bc341f35908d7..fb240a8702892 100644 + bool boost_state; }; -+/** -+ * struct amd_pstate_global_params - Global parameters, mostly tunable via sysfs. -+ * @cpb_boost: Whether or not to use boost CPU P-states. -+ * @cpb_supported: Whether or not CPU boost P-states are available -+ * based on the MSR_K7_HWCR bit[25] state -+ */ -+struct amd_pstate_global_params { -+ bool cpb_boost; -+ bool cpb_supported; -+}; -+ -+extern struct amd_pstate_global_params amd_pstate_global_params; -+ #endif /* _LINUX_AMD_PSTATE_H */ diff --git a/drivers/cpufreq/cpufreq.c b/drivers/cpufreq/cpufreq.c -index fd9c3ed21f49c..35296d8f9cdfa 100644 +index fd9c3ed21f49..bbbeb8b90313 100644 --- a/drivers/cpufreq/cpufreq.c +++ b/drivers/cpufreq/cpufreq.c @@ -614,10 +614,9 @@ static ssize_t show_boost(struct kobject *kobj, @@ -1049,40 +876,16 @@ index fd9c3ed21f49c..35296d8f9cdfa 100644 return -EINVAL; if (!cpufreq_driver->boost_enabled) -@@ -1430,8 +1429,18 @@ static int cpufreq_online(unsigned int cpu) - goto out_free_policy; +@@ -1431,7 +1430,8 @@ static int cpufreq_online(unsigned int cpu) } -- /* Let the per-policy boost flag mirror the cpufreq_driver boost during init */ + /* Let the per-policy boost flag mirror the cpufreq_driver boost during init */ - policy->boost_enabled = cpufreq_boost_enabled() && policy_has_boost_freq(policy); -+ /* init boost state to prepare set_boost callback for each CPU */ -+ if (cpufreq_driver->init_boost) { -+ ret = cpufreq_driver->init_boost(policy); -+ if (ret) { -+ pr_debug("%s: %d: boost initialization failed\n", __func__, -+ __LINE__); -+ goto out_offline_policy; -+ } -+ } else { -+ /* Let the per-policy boost flag mirror the cpufreq_driver boost during init */ -+ policy->boost_enabled = cpufreq_boost_enabled() && policy_has_boost_freq(policy); -+ } ++ if (cpufreq_boost_enabled() && policy_has_boost_freq(policy)) ++ policy->boost_enabled = true; /* * The initialization has succeeded and the policy is online. -diff --git a/include/linux/cpufreq.h b/include/linux/cpufreq.h -index 9956afb9acc23..3eb5ce712c587 100644 ---- a/include/linux/cpufreq.h -+++ b/include/linux/cpufreq.h -@@ -399,6 +399,8 @@ struct cpufreq_driver { - bool boost_enabled; - int (*set_boost)(struct cpufreq_policy *policy, int state); - -+ /* initialize boost state to be consistent before calling set_boost */ -+ int (*init_boost)(struct cpufreq_policy *policy); - /* - * Set by drivers that want to register with the energy model after the - * policy is properly initialized, but before the governor is started. -- 2.45.2 diff --git a/SOURCES/0001-ntsync.patch b/SOURCES/0001-ntsync.patch index 4f3d773..00a3538 100644 --- a/SOURCES/0001-ntsync.patch +++ b/SOURCES/0001-ntsync.patch @@ -1,6 +1,6 @@ -From 3aa16b2e2770401eaac82fcd8f279ee3d0ab409c Mon Sep 17 00:00:00 2001 +From 0f079f29cb143deb96685a5d7e5c40ee3e709cf8 Mon Sep 17 00:00:00 2001 From: Peter Jung <admin@ptr1337.dev> -Date: Thu, 27 Jun 2024 16:53:18 +0200 +Date: Fri, 5 Jul 2024 10:32:47 +0200 Subject: [PATCH 08/10] ntsync Signed-off-by: Peter Jung <admin@ptr1337.dev> @@ -28,7 +28,7 @@ Signed-off-by: Peter Jung <admin@ptr1337.dev> create mode 100644 tools/testing/selftests/drivers/ntsync/ntsync.c diff --git a/Documentation/userspace-api/index.rst b/Documentation/userspace-api/index.rst -index afecfe3cc4a86..d5745a500fa7b 100644 +index afecfe3cc4a8..d5745a500fa7 100644 --- a/Documentation/userspace-api/index.rst +++ b/Documentation/userspace-api/index.rst @@ -62,6 +62,7 @@ Everything else @@ -40,7 +40,7 @@ index afecfe3cc4a86..d5745a500fa7b 100644 .. only:: subproject and html diff --git a/Documentation/userspace-api/ioctl/ioctl-number.rst b/Documentation/userspace-api/ioctl/ioctl-number.rst -index c472423412bf2..a141e8e65c5d3 100644 +index c472423412bf..a141e8e65c5d 100644 --- a/Documentation/userspace-api/ioctl/ioctl-number.rst +++ b/Documentation/userspace-api/ioctl/ioctl-number.rst @@ -174,6 +174,8 @@ Code Seq# Include File Comments @@ -54,7 +54,7 @@ index c472423412bf2..a141e8e65c5d3 100644 'P' 60-6F sound/sscape_ioctl.h conflict! diff --git a/Documentation/userspace-api/ntsync.rst b/Documentation/userspace-api/ntsync.rst new file mode 100644 -index 0000000000000..767844637a7df +index 000000000000..767844637a7d --- /dev/null +++ b/Documentation/userspace-api/ntsync.rst @@ -0,0 +1,398 @@ @@ -457,7 +457,7 @@ index 0000000000000..767844637a7df + ``objs`` and in ``alert``. If this is attempted, the function fails + with ``EINVAL``. diff --git a/MAINTAINERS b/MAINTAINERS -index 3121709d99e3b..baa28e4151aa6 100644 +index 3121709d99e3..baa28e4151aa 100644 --- a/MAINTAINERS +++ b/MAINTAINERS @@ -15720,6 +15720,15 @@ T: git https://github.com/Paragon-Software-Group/linux-ntfs3.git @@ -477,7 +477,7 @@ index 3121709d99e3b..baa28e4151aa6 100644 M: Finn Thain <fthain@linux-m68k.org> L: linux-m68k@lists.linux-m68k.org diff --git a/drivers/misc/Kconfig b/drivers/misc/Kconfig -index 4fb291f0bf7c8..801ed229ed7dc 100644 +index 4fb291f0bf7c..801ed229ed7d 100644 --- a/drivers/misc/Kconfig +++ b/drivers/misc/Kconfig @@ -506,6 +506,17 @@ config OPEN_DICE @@ -499,7 +499,7 @@ index 4fb291f0bf7c8..801ed229ed7dc 100644 tristate "Guest vCPU stall detector" depends on OF && HAS_IOMEM diff --git a/drivers/misc/Makefile b/drivers/misc/Makefile -index ea6ea5bbbc9c6..153a3f4837e8c 100644 +index ea6ea5bbbc9c..153a3f4837e8 100644 --- a/drivers/misc/Makefile +++ b/drivers/misc/Makefile @@ -59,6 +59,7 @@ obj-$(CONFIG_PVPANIC) += pvpanic/ @@ -512,7 +512,7 @@ index ea6ea5bbbc9c6..153a3f4837e8c 100644 obj-$(CONFIG_GP_PCI1XXXX) += mchp_pci1xxxx/ diff --git a/drivers/misc/ntsync.c b/drivers/misc/ntsync.c new file mode 100644 -index 0000000000000..87a24798a5c7b +index 000000000000..87a24798a5c7 --- /dev/null +++ b/drivers/misc/ntsync.c @@ -0,0 +1,1232 @@ @@ -1750,7 +1750,7 @@ index 0000000000000..87a24798a5c7b +MODULE_LICENSE("GPL"); diff --git a/include/uapi/linux/ntsync.h b/include/uapi/linux/ntsync.h new file mode 100644 -index 0000000000000..4a8095a3fc34c +index 000000000000..4a8095a3fc34 --- /dev/null +++ b/include/uapi/linux/ntsync.h @@ -0,0 +1,62 @@ @@ -1817,7 +1817,7 @@ index 0000000000000..4a8095a3fc34c + +#endif diff --git a/tools/testing/selftests/Makefile b/tools/testing/selftests/Makefile -index e1504833654db..6f95206325e1f 100644 +index e1504833654d..6f95206325e1 100644 --- a/tools/testing/selftests/Makefile +++ b/tools/testing/selftests/Makefile @@ -16,6 +16,7 @@ TARGETS += damon @@ -1830,14 +1830,14 @@ index e1504833654db..6f95206325e1f 100644 TARGETS += drivers/net/team diff --git a/tools/testing/selftests/drivers/ntsync/.gitignore b/tools/testing/selftests/drivers/ntsync/.gitignore new file mode 100644 -index 0000000000000..848573a3d3eaf +index 000000000000..848573a3d3ea --- /dev/null +++ b/tools/testing/selftests/drivers/ntsync/.gitignore @@ -0,0 +1 @@ +ntsync diff --git a/tools/testing/selftests/drivers/ntsync/Makefile b/tools/testing/selftests/drivers/ntsync/Makefile new file mode 100644 -index 0000000000000..dbf2b055c0b28 +index 000000000000..dbf2b055c0b2 --- /dev/null +++ b/tools/testing/selftests/drivers/ntsync/Makefile @@ -0,0 +1,7 @@ @@ -1850,14 +1850,14 @@ index 0000000000000..dbf2b055c0b28 +include ../../lib.mk diff --git a/tools/testing/selftests/drivers/ntsync/config b/tools/testing/selftests/drivers/ntsync/config new file mode 100644 -index 0000000000000..60539c826d062 +index 000000000000..60539c826d06 --- /dev/null +++ b/tools/testing/selftests/drivers/ntsync/config @@ -0,0 +1 @@ +CONFIG_WINESYNC=y diff --git a/tools/testing/selftests/drivers/ntsync/ntsync.c b/tools/testing/selftests/drivers/ntsync/ntsync.c new file mode 100644 -index 0000000000000..5fa2c9a0768c0 +index 000000000000..5fa2c9a0768c --- /dev/null +++ b/tools/testing/selftests/drivers/ntsync/ntsync.c @@ -0,0 +1,1407 @@ diff --git a/SOURCES/cachy-bbr3.patch b/SOURCES/cachy-bbr3.patch new file mode 100644 index 0000000..c714f9d --- /dev/null +++ b/SOURCES/cachy-bbr3.patch @@ -0,0 +1,3386 @@ +From 3b7e1abbffd80645ae13f4fb033fa0f49a641e27 Mon Sep 17 00:00:00 2001 +From: Peter Jung <admin@ptr1337.dev> +Date: Fri, 5 Jul 2024 10:31:38 +0200 +Subject: [PATCH 03/10] bbr3 + +Signed-off-by: Peter Jung <admin@ptr1337.dev> +--- + include/linux/tcp.h | 4 +- + include/net/inet_connection_sock.h | 4 +- + include/net/tcp.h | 72 +- + include/uapi/linux/inet_diag.h | 23 + + include/uapi/linux/rtnetlink.h | 4 +- + include/uapi/linux/tcp.h | 1 + + net/ipv4/Kconfig | 21 +- + net/ipv4/bpf_tcp_ca.c | 9 +- + net/ipv4/tcp.c | 3 + + net/ipv4/tcp_bbr.c | 2230 +++++++++++++++++++++------- + net/ipv4/tcp_cong.c | 1 + + net/ipv4/tcp_input.c | 40 +- + net/ipv4/tcp_minisocks.c | 2 + + net/ipv4/tcp_output.c | 48 +- + net/ipv4/tcp_rate.c | 30 +- + net/ipv4/tcp_timer.c | 1 + + 16 files changed, 1940 insertions(+), 553 deletions(-) + +diff --git a/include/linux/tcp.h b/include/linux/tcp.h +index 55399ee2a57e..c6193d87a8e6 100644 +--- a/include/linux/tcp.h ++++ b/include/linux/tcp.h +@@ -368,7 +368,9 @@ struct tcp_sock { + u8 compressed_ack; + u8 dup_ack_counter:2, + tlp_retrans:1, /* TLP is a retransmission */ +- unused:5; ++ fast_ack_mode:2, /* which fast ack mode ? */ ++ tlp_orig_data_app_limited:1, /* app-limited before TLP rtx? */ ++ unused:2; + u8 thin_lto : 1,/* Use linear timeouts for thin streams */ + recvmsg_inq : 1,/* Indicate # of bytes in queue upon recvmsg */ + fastopen_connect:1, /* FASTOPEN_CONNECT sockopt */ +diff --git a/include/net/inet_connection_sock.h b/include/net/inet_connection_sock.h +index 146ece8563ca..0217922aa100 100644 +--- a/include/net/inet_connection_sock.h ++++ b/include/net/inet_connection_sock.h +@@ -137,8 +137,8 @@ struct inet_connection_sock { + u32 icsk_probes_tstamp; + u32 icsk_user_timeout; + +- u64 icsk_ca_priv[104 / sizeof(u64)]; +-#define ICSK_CA_PRIV_SIZE sizeof_field(struct inet_connection_sock, icsk_ca_priv) ++#define ICSK_CA_PRIV_SIZE (144) ++ u64 icsk_ca_priv[ICSK_CA_PRIV_SIZE / sizeof(u64)]; + }; + + #define ICSK_TIME_RETRANS 1 /* Retransmit timer */ +diff --git a/include/net/tcp.h b/include/net/tcp.h +index 2bcf30381d75..54e1af0789b8 100644 +--- a/include/net/tcp.h ++++ b/include/net/tcp.h +@@ -381,6 +381,8 @@ static inline void tcp_dec_quickack_mode(struct sock *sk) + #define TCP_ECN_QUEUE_CWR 2 + #define TCP_ECN_DEMAND_CWR 4 + #define TCP_ECN_SEEN 8 ++#define TCP_ECN_LOW 16 ++#define TCP_ECN_ECT_PERMANENT 32 + + enum tcp_tw_status { + TCP_TW_SUCCESS = 0, +@@ -782,6 +784,15 @@ static inline void tcp_fast_path_check(struct sock *sk) + + u32 tcp_delack_max(const struct sock *sk); + ++static inline void tcp_set_ecn_low_from_dst(struct sock *sk, ++ const struct dst_entry *dst) ++{ ++ struct tcp_sock *tp = tcp_sk(sk); ++ ++ if (dst_feature(dst, RTAX_FEATURE_ECN_LOW)) ++ tp->ecn_flags |= TCP_ECN_LOW; ++} ++ + /* Compute the actual rto_min value */ + static inline u32 tcp_rto_min(const struct sock *sk) + { +@@ -887,6 +898,11 @@ static inline u32 tcp_stamp_us_delta(u64 t1, u64 t0) + return max_t(s64, t1 - t0, 0); + } + ++static inline u32 tcp_stamp32_us_delta(u32 t1, u32 t0) ++{ ++ return max_t(s32, t1 - t0, 0); ++} ++ + /* provide the departure time in us unit */ + static inline u64 tcp_skb_timestamp_us(const struct sk_buff *skb) + { +@@ -975,9 +991,14 @@ struct tcp_skb_cb { + /* pkts S/ACKed so far upon tx of skb, incl retrans: */ + __u32 delivered; + /* start of send pipeline phase */ +- u64 first_tx_mstamp; ++ u32 first_tx_mstamp; + /* when we reached the "delivered" count */ +- u64 delivered_mstamp; ++ u32 delivered_mstamp; ++#define TCPCB_IN_FLIGHT_BITS 20 ++#define TCPCB_IN_FLIGHT_MAX ((1U << TCPCB_IN_FLIGHT_BITS) - 1) ++ u32 in_flight:20, /* packets in flight at transmit */ ++ unused2:12; ++ u32 lost; /* packets lost so far upon tx of skb */ + } tx; /* only used for outgoing skbs */ + union { + struct inet_skb_parm h4; +@@ -1081,6 +1102,7 @@ enum tcp_ca_event { + CA_EVENT_LOSS, /* loss timeout */ + CA_EVENT_ECN_NO_CE, /* ECT set, but not CE marked */ + CA_EVENT_ECN_IS_CE, /* received CE marked IP packet */ ++ CA_EVENT_TLP_RECOVERY, /* a lost segment was repaired by TLP probe */ + }; + + /* Information about inbound ACK, passed to cong_ops->in_ack_event() */ +@@ -1103,7 +1125,11 @@ enum tcp_ca_ack_event_flags { + #define TCP_CONG_NON_RESTRICTED 0x1 + /* Requires ECN/ECT set on all packets */ + #define TCP_CONG_NEEDS_ECN 0x2 +-#define TCP_CONG_MASK (TCP_CONG_NON_RESTRICTED | TCP_CONG_NEEDS_ECN) ++/* Wants notification of CE events (CA_EVENT_ECN_IS_CE, CA_EVENT_ECN_NO_CE). */ ++#define TCP_CONG_WANTS_CE_EVENTS 0x4 ++#define TCP_CONG_MASK (TCP_CONG_NON_RESTRICTED | \ ++ TCP_CONG_NEEDS_ECN | \ ++ TCP_CONG_WANTS_CE_EVENTS) + + union tcp_cc_info; + +@@ -1123,10 +1149,13 @@ struct ack_sample { + */ + struct rate_sample { + u64 prior_mstamp; /* starting timestamp for interval */ ++ u32 prior_lost; /* tp->lost at "prior_mstamp" */ + u32 prior_delivered; /* tp->delivered at "prior_mstamp" */ + u32 prior_delivered_ce;/* tp->delivered_ce at "prior_mstamp" */ ++ u32 tx_in_flight; /* packets in flight at starting timestamp */ ++ s32 lost; /* number of packets lost over interval */ + s32 delivered; /* number of packets delivered over interval */ +- s32 delivered_ce; /* number of packets delivered w/ CE marks*/ ++ s32 delivered_ce; /* packets delivered w/ CE mark over interval */ + long interval_us; /* time for tp->delivered to incr "delivered" */ + u32 snd_interval_us; /* snd interval for delivered packets */ + u32 rcv_interval_us; /* rcv interval for delivered packets */ +@@ -1137,7 +1166,9 @@ struct rate_sample { + u32 last_end_seq; /* end_seq of most recently ACKed packet */ + bool is_app_limited; /* is sample from packet with bubble in pipe? */ + bool is_retrans; /* is sample from retransmission? */ ++ bool is_acking_tlp_retrans_seq; /* ACKed a TLP retransmit sequence? */ + bool is_ack_delayed; /* is this (likely) a delayed ACK? */ ++ bool is_ece; /* did this ACK have ECN marked? */ + }; + + struct tcp_congestion_ops { +@@ -1161,8 +1192,11 @@ struct tcp_congestion_ops { + /* hook for packet ack accounting (optional) */ + void (*pkts_acked)(struct sock *sk, const struct ack_sample *sample); + +- /* override sysctl_tcp_min_tso_segs */ +- u32 (*min_tso_segs)(struct sock *sk); ++ /* pick target number of segments per TSO/GSO skb (optional): */ ++ u32 (*tso_segs)(struct sock *sk, unsigned int mss_now); ++ ++ /* react to a specific lost skb (optional) */ ++ void (*skb_marked_lost)(struct sock *sk, const struct sk_buff *skb); + + /* call when packets are delivered to update cwnd and pacing rate, + * after all the ca_state processing. (optional) +@@ -1228,6 +1262,14 @@ static inline char *tcp_ca_get_name_by_key(u32 key, char *buffer) + } + #endif + ++static inline bool tcp_ca_wants_ce_events(const struct sock *sk) ++{ ++ const struct inet_connection_sock *icsk = inet_csk(sk); ++ ++ return icsk->icsk_ca_ops->flags & (TCP_CONG_NEEDS_ECN | ++ TCP_CONG_WANTS_CE_EVENTS); ++} ++ + static inline bool tcp_ca_needs_ecn(const struct sock *sk) + { + const struct inet_connection_sock *icsk = inet_csk(sk); +@@ -1247,6 +1289,7 @@ static inline void tcp_ca_event(struct sock *sk, const enum tcp_ca_event event) + void tcp_set_ca_state(struct sock *sk, const u8 ca_state); + + /* From tcp_rate.c */ ++void tcp_set_tx_in_flight(struct sock *sk, struct sk_buff *skb); + void tcp_rate_skb_sent(struct sock *sk, struct sk_buff *skb); + void tcp_rate_skb_delivered(struct sock *sk, struct sk_buff *skb, + struct rate_sample *rs); +@@ -1259,6 +1302,21 @@ static inline bool tcp_skb_sent_after(u64 t1, u64 t2, u32 seq1, u32 seq2) + return t1 > t2 || (t1 == t2 && after(seq1, seq2)); + } + ++/* If a retransmit failed due to local qdisc congestion or other local issues, ++ * then we may have called tcp_set_skb_tso_segs() to increase the number of ++ * segments in the skb without increasing the tx.in_flight. In all other cases, ++ * the tx.in_flight should be at least as big as the pcount of the sk_buff. We ++ * do not have the state to know whether a retransmit failed due to local qdisc ++ * congestion or other local issues, so to avoid spurious warnings we consider ++ * that any skb marked lost may have suffered that fate. ++ */ ++static inline bool tcp_skb_tx_in_flight_is_suspicious(u32 skb_pcount, ++ u32 skb_sacked_flags, ++ u32 tx_in_flight) ++{ ++ return (skb_pcount > tx_in_flight) && !(skb_sacked_flags & TCPCB_LOST); ++} ++ + /* These functions determine how the current flow behaves in respect of SACK + * handling. SACK is negotiated with the peer, and therefore it can vary + * between different flows. +@@ -2417,7 +2475,7 @@ struct tcp_plb_state { + u8 consec_cong_rounds:5, /* consecutive congested rounds */ + unused:3; + u32 pause_until; /* jiffies32 when PLB can resume rerouting */ +-}; ++} __attribute__ ((__packed__)); + + static inline void tcp_plb_init(const struct sock *sk, + struct tcp_plb_state *plb) +diff --git a/include/uapi/linux/inet_diag.h b/include/uapi/linux/inet_diag.h +index 50655de04c9b..82f8bd8f0d16 100644 +--- a/include/uapi/linux/inet_diag.h ++++ b/include/uapi/linux/inet_diag.h +@@ -229,6 +229,29 @@ struct tcp_bbr_info { + __u32 bbr_min_rtt; /* min-filtered RTT in uSec */ + __u32 bbr_pacing_gain; /* pacing gain shifted left 8 bits */ + __u32 bbr_cwnd_gain; /* cwnd gain shifted left 8 bits */ ++ __u32 bbr_bw_hi_lsb; /* lower 32 bits of bw_hi */ ++ __u32 bbr_bw_hi_msb; /* upper 32 bits of bw_hi */ ++ __u32 bbr_bw_lo_lsb; /* lower 32 bits of bw_lo */ ++ __u32 bbr_bw_lo_msb; /* upper 32 bits of bw_lo */ ++ __u8 bbr_mode; /* current bbr_mode in state machine */ ++ __u8 bbr_phase; /* current state machine phase */ ++ __u8 unused1; /* alignment padding; not used yet */ ++ __u8 bbr_version; /* BBR algorithm version */ ++ __u32 bbr_inflight_lo; /* lower short-term data volume bound */ ++ __u32 bbr_inflight_hi; /* higher long-term data volume bound */ ++ __u32 bbr_extra_acked; /* max excess packets ACKed in epoch */ ++}; ++ ++/* TCP BBR congestion control bbr_phase as reported in netlink/ss stats. */ ++enum tcp_bbr_phase { ++ BBR_PHASE_INVALID = 0, ++ BBR_PHASE_STARTUP = 1, ++ BBR_PHASE_DRAIN = 2, ++ BBR_PHASE_PROBE_RTT = 3, ++ BBR_PHASE_PROBE_BW_UP = 4, ++ BBR_PHASE_PROBE_BW_DOWN = 5, ++ BBR_PHASE_PROBE_BW_CRUISE = 6, ++ BBR_PHASE_PROBE_BW_REFILL = 7, + }; + + union tcp_cc_info { +diff --git a/include/uapi/linux/rtnetlink.h b/include/uapi/linux/rtnetlink.h +index 3b687d20c9ed..a7c30c243b54 100644 +--- a/include/uapi/linux/rtnetlink.h ++++ b/include/uapi/linux/rtnetlink.h +@@ -507,12 +507,14 @@ enum { + #define RTAX_FEATURE_TIMESTAMP (1 << 2) /* unused */ + #define RTAX_FEATURE_ALLFRAG (1 << 3) /* unused */ + #define RTAX_FEATURE_TCP_USEC_TS (1 << 4) ++#define RTAX_FEATURE_ECN_LOW (1 << 5) + + #define RTAX_FEATURE_MASK (RTAX_FEATURE_ECN | \ + RTAX_FEATURE_SACK | \ + RTAX_FEATURE_TIMESTAMP | \ + RTAX_FEATURE_ALLFRAG | \ +- RTAX_FEATURE_TCP_USEC_TS) ++ RTAX_FEATURE_TCP_USEC_TS | \ ++ RTAX_FEATURE_ECN_LOW) + + struct rta_session { + __u8 proto; +diff --git a/include/uapi/linux/tcp.h b/include/uapi/linux/tcp.h +index c07e9f90c084..5c88336ced60 100644 +--- a/include/uapi/linux/tcp.h ++++ b/include/uapi/linux/tcp.h +@@ -176,6 +176,7 @@ enum tcp_fastopen_client_fail { + #define TCPI_OPT_ECN_SEEN 16 /* we received at least one packet with ECT */ + #define TCPI_OPT_SYN_DATA 32 /* SYN-ACK acked data in SYN sent or rcvd */ + #define TCPI_OPT_USEC_TS 64 /* usec timestamps */ ++#define TCPI_OPT_ECN_LOW 128 /* Low-latency ECN configured at init */ + + /* + * Sender's congestion state indicating normal or abnormal situations +diff --git a/net/ipv4/Kconfig b/net/ipv4/Kconfig +index 8e94ed7c56a0..50dc9970cad2 100644 +--- a/net/ipv4/Kconfig ++++ b/net/ipv4/Kconfig +@@ -668,15 +668,18 @@ config TCP_CONG_BBR + default n + help + +- BBR (Bottleneck Bandwidth and RTT) TCP congestion control aims to +- maximize network utilization and minimize queues. It builds an explicit +- model of the bottleneck delivery rate and path round-trip propagation +- delay. It tolerates packet loss and delay unrelated to congestion. It +- can operate over LAN, WAN, cellular, wifi, or cable modem links. It can +- coexist with flows that use loss-based congestion control, and can +- operate with shallow buffers, deep buffers, bufferbloat, policers, or +- AQM schemes that do not provide a delay signal. It requires the fq +- ("Fair Queue") pacing packet scheduler. ++ BBR (Bottleneck Bandwidth and RTT) TCP congestion control is a ++ model-based congestion control algorithm that aims to maximize ++ network utilization, keep queues and retransmit rates low, and to be ++ able to coexist with Reno/CUBIC in common scenarios. It builds an ++ explicit model of the network path. It tolerates a targeted degree ++ of random packet loss and delay. It can operate over LAN, WAN, ++ cellular, wifi, or cable modem links, and can use shallow-threshold ++ ECN signals. It can coexist to some degree with flows that use ++ loss-based congestion control, and can operate with shallow buffers, ++ deep buffers, bufferbloat, policers, or AQM schemes that do not ++ provide a delay signal. It requires pacing, using either TCP internal ++ pacing or the fq ("Fair Queue") pacing packet scheduler. + + choice + prompt "Default TCP congestion control" +diff --git a/net/ipv4/bpf_tcp_ca.c b/net/ipv4/bpf_tcp_ca.c +index 7f518ea5f4ac..8f9b677c20e1 100644 +--- a/net/ipv4/bpf_tcp_ca.c ++++ b/net/ipv4/bpf_tcp_ca.c +@@ -302,11 +302,15 @@ static void bpf_tcp_ca_pkts_acked(struct sock *sk, const struct ack_sample *samp + { + } + +-static u32 bpf_tcp_ca_min_tso_segs(struct sock *sk) ++static u32 bpf_tcp_ca_tso_segs(struct sock *sk, unsigned int mss_now) + { + return 0; + } + ++static void bpf_tcp_ca_skb_marked_lost(struct sock *sk, const struct sk_buff *skb) ++{ ++} ++ + static void bpf_tcp_ca_cong_control(struct sock *sk, const struct rate_sample *rs) + { + } +@@ -336,7 +340,8 @@ static struct tcp_congestion_ops __bpf_ops_tcp_congestion_ops = { + .cwnd_event = bpf_tcp_ca_cwnd_event, + .in_ack_event = bpf_tcp_ca_in_ack_event, + .pkts_acked = bpf_tcp_ca_pkts_acked, +- .min_tso_segs = bpf_tcp_ca_min_tso_segs, ++ .tso_segs = bpf_tcp_ca_tso_segs, ++ .skb_marked_lost = bpf_tcp_ca_skb_marked_lost, + .cong_control = bpf_tcp_ca_cong_control, + .undo_cwnd = bpf_tcp_ca_undo_cwnd, + .sndbuf_expand = bpf_tcp_ca_sndbuf_expand, +diff --git a/net/ipv4/tcp.c b/net/ipv4/tcp.c +index 77ee1eda3fd8..b3d43b2b8617 100644 +--- a/net/ipv4/tcp.c ++++ b/net/ipv4/tcp.c +@@ -3099,6 +3099,7 @@ int tcp_disconnect(struct sock *sk, int flags) + tp->rx_opt.dsack = 0; + tp->rx_opt.num_sacks = 0; + tp->rcv_ooopack = 0; ++ tp->fast_ack_mode = 0; + + + /* Clean up fastopen related fields */ +@@ -3825,6 +3826,8 @@ void tcp_get_info(struct sock *sk, struct tcp_info *info) + info->tcpi_options |= TCPI_OPT_ECN; + if (tp->ecn_flags & TCP_ECN_SEEN) + info->tcpi_options |= TCPI_OPT_ECN_SEEN; ++ if (tp->ecn_flags & TCP_ECN_LOW) ++ info->tcpi_options |= TCPI_OPT_ECN_LOW; + if (tp->syn_data_acked) + info->tcpi_options |= TCPI_OPT_SYN_DATA; + if (tp->tcp_usec_ts) +diff --git a/net/ipv4/tcp_bbr.c b/net/ipv4/tcp_bbr.c +index 05dc2d05bc7c..61ed3e549350 100644 +--- a/net/ipv4/tcp_bbr.c ++++ b/net/ipv4/tcp_bbr.c +@@ -1,18 +1,19 @@ +-/* Bottleneck Bandwidth and RTT (BBR) congestion control ++/* BBR (Bottleneck Bandwidth and RTT) congestion control + * +- * BBR congestion control computes the sending rate based on the delivery +- * rate (throughput) estimated from ACKs. In a nutshell: ++ * BBR is a model-based congestion control algorithm that aims for low queues, ++ * low loss, and (bounded) Reno/CUBIC coexistence. To maintain a model of the ++ * network path, it uses measurements of bandwidth and RTT, as well as (if they ++ * occur) packet loss and/or shallow-threshold ECN signals. Note that although ++ * it can use ECN or loss signals explicitly, it does not require either; it ++ * can bound its in-flight data based on its estimate of the BDP. + * +- * On each ACK, update our model of the network path: +- * bottleneck_bandwidth = windowed_max(delivered / elapsed, 10 round trips) +- * min_rtt = windowed_min(rtt, 10 seconds) +- * pacing_rate = pacing_gain * bottleneck_bandwidth +- * cwnd = max(cwnd_gain * bottleneck_bandwidth * min_rtt, 4) +- * +- * The core algorithm does not react directly to packet losses or delays, +- * although BBR may adjust the size of next send per ACK when loss is +- * observed, or adjust the sending rate if it estimates there is a +- * traffic policer, in order to keep the drop rate reasonable. ++ * The model has both higher and lower bounds for the operating range: ++ * lo: bw_lo, inflight_lo: conservative short-term lower bound ++ * hi: bw_hi, inflight_hi: robust long-term upper bound ++ * The bandwidth-probing time scale is (a) extended dynamically based on ++ * estimated BDP to improve coexistence with Reno/CUBIC; (b) bounded by ++ * an interactive wall-clock time-scale to be more scalable and responsive ++ * than Reno and CUBIC. + * + * Here is a state transition diagram for BBR: + * +@@ -65,6 +66,13 @@ + #include <linux/random.h> + #include <linux/win_minmax.h> + ++#include <trace/events/tcp.h> ++#include "tcp_dctcp.h" ++ ++#define BBR_VERSION 3 ++ ++#define bbr_param(sk,name) (bbr_ ## name) ++ + /* Scale factor for rate in pkt/uSec unit to avoid truncation in bandwidth + * estimation. The rate unit ~= (1500 bytes / 1 usec / 2^24) ~= 715 bps. + * This handles bandwidths from 0.06pps (715bps) to 256Mpps (3Tbps) in a u32. +@@ -85,36 +93,41 @@ enum bbr_mode { + BBR_PROBE_RTT, /* cut inflight to min to probe min_rtt */ + }; + ++/* How does the incoming ACK stream relate to our bandwidth probing? */ ++enum bbr_ack_phase { ++ BBR_ACKS_INIT, /* not probing; not getting probe feedback */ ++ BBR_ACKS_REFILLING, /* sending at est. bw to fill pipe */ ++ BBR_ACKS_PROBE_STARTING, /* inflight rising to probe bw */ ++ BBR_ACKS_PROBE_FEEDBACK, /* getting feedback from bw probing */ ++ BBR_ACKS_PROBE_STOPPING, /* stopped probing; still getting feedback */ ++}; ++ + /* BBR congestion control block */ + struct bbr { + u32 min_rtt_us; /* min RTT in min_rtt_win_sec window */ + u32 min_rtt_stamp; /* timestamp of min_rtt_us */ + u32 probe_rtt_done_stamp; /* end time for BBR_PROBE_RTT mode */ +- struct minmax bw; /* Max recent delivery rate in pkts/uS << 24 */ +- u32 rtt_cnt; /* count of packet-timed rounds elapsed */ ++ u32 probe_rtt_min_us; /* min RTT in probe_rtt_win_ms win */ ++ u32 probe_rtt_min_stamp; /* timestamp of probe_rtt_min_us*/ + u32 next_rtt_delivered; /* scb->tx.delivered at end of round */ + u64 cycle_mstamp; /* time of this cycle phase start */ +- u32 mode:3, /* current bbr_mode in state machine */ ++ u32 mode:2, /* current bbr_mode in state machine */ + prev_ca_state:3, /* CA state on previous ACK */ +- packet_conservation:1, /* use packet conservation? */ + round_start:1, /* start of packet-timed tx->ack round? */ ++ ce_state:1, /* If most recent data has CE bit set */ ++ bw_probe_up_rounds:5, /* cwnd-limited rounds in PROBE_UP */ ++ try_fast_path:1, /* can we take fast path? */ + idle_restart:1, /* restarting after idle? */ + probe_rtt_round_done:1, /* a BBR_PROBE_RTT round at 4 pkts? */ +- unused:13, +- lt_is_sampling:1, /* taking long-term ("LT") samples now? */ +- lt_rtt_cnt:7, /* round trips in long-term interval */ +- lt_use_bw:1; /* use lt_bw as our bw estimate? */ +- u32 lt_bw; /* LT est delivery rate in pkts/uS << 24 */ +- u32 lt_last_delivered; /* LT intvl start: tp->delivered */ +- u32 lt_last_stamp; /* LT intvl start: tp->delivered_mstamp */ +- u32 lt_last_lost; /* LT intvl start: tp->lost */ ++ init_cwnd:7, /* initial cwnd */ ++ unused_1:10; + u32 pacing_gain:10, /* current gain for setting pacing rate */ + cwnd_gain:10, /* current gain for setting cwnd */ + full_bw_reached:1, /* reached full bw in Startup? */ + full_bw_cnt:2, /* number of rounds without large bw gains */ +- cycle_idx:3, /* current index in pacing_gain cycle array */ ++ cycle_idx:2, /* current index in pacing_gain cycle array */ + has_seen_rtt:1, /* have we seen an RTT sample yet? */ +- unused_b:5; ++ unused_2:6; + u32 prior_cwnd; /* prior cwnd upon entering loss recovery */ + u32 full_bw; /* recent bw, to estimate if pipe is full */ + +@@ -124,19 +137,67 @@ struct bbr { + u32 ack_epoch_acked:20, /* packets (S)ACKed in sampling epoch */ + extra_acked_win_rtts:5, /* age of extra_acked, in round trips */ + extra_acked_win_idx:1, /* current index in extra_acked array */ +- unused_c:6; ++ /* BBR v3 state: */ ++ full_bw_now:1, /* recently reached full bw plateau? */ ++ startup_ecn_rounds:2, /* consecutive hi ECN STARTUP rounds */ ++ loss_in_cycle:1, /* packet loss in this cycle? */ ++ ecn_in_cycle:1, /* ECN in this cycle? */ ++ unused_3:1; ++ u32 loss_round_delivered; /* scb->tx.delivered ending loss round */ ++ u32 undo_bw_lo; /* bw_lo before latest losses */ ++ u32 undo_inflight_lo; /* inflight_lo before latest losses */ ++ u32 undo_inflight_hi; /* inflight_hi before latest losses */ ++ u32 bw_latest; /* max delivered bw in last round trip */ ++ u32 bw_lo; /* lower bound on sending bandwidth */ ++ u32 bw_hi[2]; /* max recent measured bw sample */ ++ u32 inflight_latest; /* max delivered data in last round trip */ ++ u32 inflight_lo; /* lower bound of inflight data range */ ++ u32 inflight_hi; /* upper bound of inflight data range */ ++ u32 bw_probe_up_cnt; /* packets delivered per inflight_hi incr */ ++ u32 bw_probe_up_acks; /* packets (S)ACKed since inflight_hi incr */ ++ u32 probe_wait_us; /* PROBE_DOWN until next clock-driven probe */ ++ u32 prior_rcv_nxt; /* tp->rcv_nxt when CE state last changed */ ++ u32 ecn_eligible:1, /* sender can use ECN (RTT, handshake)? */ ++ ecn_alpha:9, /* EWMA delivered_ce/delivered; 0..256 */ ++ bw_probe_samples:1, /* rate samples reflect bw probing? */ ++ prev_probe_too_high:1, /* did last PROBE_UP go too high? */ ++ stopped_risky_probe:1, /* last PROBE_UP stopped due to risk? */ ++ rounds_since_probe:8, /* packet-timed rounds since probed bw */ ++ loss_round_start:1, /* loss_round_delivered round trip? */ ++ loss_in_round:1, /* loss marked in this round trip? */ ++ ecn_in_round:1, /* ECN marked in this round trip? */ ++ ack_phase:3, /* bbr_ack_phase: meaning of ACKs */ ++ loss_events_in_round:4,/* losses in STARTUP round */ ++ initialized:1; /* has bbr_init() been called? */ ++ u32 alpha_last_delivered; /* tp->delivered at alpha update */ ++ u32 alpha_last_delivered_ce; /* tp->delivered_ce at alpha update */ ++ ++ u8 unused_4; /* to preserve alignment */ ++ struct tcp_plb_state plb; + }; + +-#define CYCLE_LEN 8 /* number of phases in a pacing gain cycle */ ++struct bbr_context { ++ u32 sample_bw; ++}; + +-/* Window length of bw filter (in rounds): */ +-static const int bbr_bw_rtts = CYCLE_LEN + 2; + /* Window length of min_rtt filter (in sec): */ + static const u32 bbr_min_rtt_win_sec = 10; + /* Minimum time (in ms) spent at bbr_cwnd_min_target in BBR_PROBE_RTT mode: */ + static const u32 bbr_probe_rtt_mode_ms = 200; +-/* Skip TSO below the following bandwidth (bits/sec): */ +-static const int bbr_min_tso_rate = 1200000; ++/* Window length of probe_rtt_min_us filter (in ms), and consequently the ++ * typical interval between PROBE_RTT mode entries. The default is 5000ms. ++ * Note that bbr_probe_rtt_win_ms must be <= bbr_min_rtt_win_sec * MSEC_PER_SEC ++ */ ++static const u32 bbr_probe_rtt_win_ms = 5000; ++/* Proportion of cwnd to estimated BDP in PROBE_RTT, in units of BBR_UNIT: */ ++static const u32 bbr_probe_rtt_cwnd_gain = BBR_UNIT * 1 / 2; ++ ++/* Use min_rtt to help adapt TSO burst size, with smaller min_rtt resulting ++ * in bigger TSO bursts. We cut the RTT-based allowance in half ++ * for every 2^9 usec (aka 512 us) of RTT, so that the RTT-based allowance ++ * is below 1500 bytes after 6 * ~500 usec = 3ms. ++ */ ++static const u32 bbr_tso_rtt_shift = 9; + + /* Pace at ~1% below estimated bw, on average, to reduce queue at bottleneck. + * In order to help drive the network toward lower queues and low latency while +@@ -146,13 +207,15 @@ static const int bbr_min_tso_rate = 1200000; + */ + static const int bbr_pacing_margin_percent = 1; + +-/* We use a high_gain value of 2/ln(2) because it's the smallest pacing gain ++/* We use a startup_pacing_gain of 4*ln(2) because it's the smallest value + * that will allow a smoothly increasing pacing rate that will double each RTT + * and send the same number of packets per RTT that an un-paced, slow-starting + * Reno or CUBIC flow would: + */ +-static const int bbr_high_gain = BBR_UNIT * 2885 / 1000 + 1; +-/* The pacing gain of 1/high_gain in BBR_DRAIN is calculated to typically drain ++static const int bbr_startup_pacing_gain = BBR_UNIT * 277 / 100 + 1; ++/* The gain for deriving startup cwnd: */ ++static const int bbr_startup_cwnd_gain = BBR_UNIT * 2; ++/* The pacing gain in BBR_DRAIN is calculated to typically drain + * the queue created in BBR_STARTUP in a single round: + */ + static const int bbr_drain_gain = BBR_UNIT * 1000 / 2885; +@@ -160,13 +223,17 @@ static const int bbr_drain_gain = BBR_UNIT * 1000 / 2885; + static const int bbr_cwnd_gain = BBR_UNIT * 2; + /* The pacing_gain values for the PROBE_BW gain cycle, to discover/share bw: */ + static const int bbr_pacing_gain[] = { +- BBR_UNIT * 5 / 4, /* probe for more available bw */ +- BBR_UNIT * 3 / 4, /* drain queue and/or yield bw to other flows */ +- BBR_UNIT, BBR_UNIT, BBR_UNIT, /* cruise at 1.0*bw to utilize pipe, */ +- BBR_UNIT, BBR_UNIT, BBR_UNIT /* without creating excess queue... */ ++ BBR_UNIT * 5 / 4, /* UP: probe for more available bw */ ++ BBR_UNIT * 91 / 100, /* DOWN: drain queue and/or yield bw */ ++ BBR_UNIT, /* CRUISE: try to use pipe w/ some headroom */ ++ BBR_UNIT, /* REFILL: refill pipe to estimated 100% */ ++}; ++enum bbr_pacing_gain_phase { ++ BBR_BW_PROBE_UP = 0, /* push up inflight to probe for bw/vol */ ++ BBR_BW_PROBE_DOWN = 1, /* drain excess inflight from the queue */ ++ BBR_BW_PROBE_CRUISE = 2, /* use pipe, w/ headroom in queue/pipe */ ++ BBR_BW_PROBE_REFILL = 3, /* v2: refill the pipe again to 100% */ + }; +-/* Randomize the starting gain cycling phase over N phases: */ +-static const u32 bbr_cycle_rand = 7; + + /* Try to keep at least this many packets in flight, if things go smoothly. For + * smooth functioning, a sliding window protocol ACKing every other packet +@@ -174,24 +241,12 @@ static const u32 bbr_cycle_rand = 7; + */ + static const u32 bbr_cwnd_min_target = 4; + +-/* To estimate if BBR_STARTUP mode (i.e. high_gain) has filled pipe... */ ++/* To estimate if BBR_STARTUP or BBR_BW_PROBE_UP has filled pipe... */ + /* If bw has increased significantly (1.25x), there may be more bw available: */ + static const u32 bbr_full_bw_thresh = BBR_UNIT * 5 / 4; + /* But after 3 rounds w/o significant bw growth, estimate pipe is full: */ + static const u32 bbr_full_bw_cnt = 3; + +-/* "long-term" ("LT") bandwidth estimator parameters... */ +-/* The minimum number of rounds in an LT bw sampling interval: */ +-static const u32 bbr_lt_intvl_min_rtts = 4; +-/* If lost/delivered ratio > 20%, interval is "lossy" and we may be policed: */ +-static const u32 bbr_lt_loss_thresh = 50; +-/* If 2 intervals have a bw ratio <= 1/8, their bw is "consistent": */ +-static const u32 bbr_lt_bw_ratio = BBR_UNIT / 8; +-/* If 2 intervals have a bw diff <= 4 Kbit/sec their bw is "consistent": */ +-static const u32 bbr_lt_bw_diff = 4000 / 8; +-/* If we estimate we're policed, use lt_bw for this many round trips: */ +-static const u32 bbr_lt_bw_max_rtts = 48; +- + /* Gain factor for adding extra_acked to target cwnd: */ + static const int bbr_extra_acked_gain = BBR_UNIT; + /* Window length of extra_acked window. */ +@@ -201,8 +256,121 @@ static const u32 bbr_ack_epoch_acked_reset_thresh = 1U << 20; + /* Time period for clamping cwnd increment due to ack aggregation */ + static const u32 bbr_extra_acked_max_us = 100 * 1000; + ++/* Flags to control BBR ECN-related behavior... */ ++ ++/* Ensure ACKs only ACK packets with consistent ECN CE status? */ ++static const bool bbr_precise_ece_ack = true; ++ ++/* Max RTT (in usec) at which to use sender-side ECN logic. ++ * Disabled when 0 (ECN allowed at any RTT). ++ */ ++static const u32 bbr_ecn_max_rtt_us = 5000; ++ ++/* On losses, scale down inflight and pacing rate by beta scaled by BBR_SCALE. ++ * No loss response when 0. ++ */ ++static const u32 bbr_beta = BBR_UNIT * 30 / 100; ++ ++/* Gain factor for ECN mark ratio samples, scaled by BBR_SCALE (1/16 = 6.25%) */ ++static const u32 bbr_ecn_alpha_gain = BBR_UNIT * 1 / 16; ++ ++/* The initial value for ecn_alpha; 1.0 allows a flow to respond quickly ++ * to congestion if the bottleneck is congested when the flow starts up. ++ */ ++static const u32 bbr_ecn_alpha_init = BBR_UNIT; ++ ++/* On ECN, cut inflight_lo to (1 - ecn_factor * ecn_alpha) scaled by BBR_SCALE. ++ * No ECN based bounding when 0. ++ */ ++static const u32 bbr_ecn_factor = BBR_UNIT * 1 / 3; /* 1/3 = 33% */ ++ ++/* Estimate bw probing has gone too far if CE ratio exceeds this threshold. ++ * Scaled by BBR_SCALE. Disabled when 0. ++ */ ++static const u32 bbr_ecn_thresh = BBR_UNIT * 1 / 2; /* 1/2 = 50% */ ++ ++/* If non-zero, if in a cycle with no losses but some ECN marks, after ECN ++ * clears then make the first round's increment to inflight_hi the following ++ * fraction of inflight_hi. ++ */ ++static const u32 bbr_ecn_reprobe_gain = BBR_UNIT * 1 / 2; ++ ++/* Estimate bw probing has gone too far if loss rate exceeds this level. */ ++static const u32 bbr_loss_thresh = BBR_UNIT * 2 / 100; /* 2% loss */ ++ ++/* Slow down for a packet loss recovered by TLP? */ ++static const bool bbr_loss_probe_recovery = true; ++ ++/* Exit STARTUP if number of loss marking events in a Recovery round is >= N, ++ * and loss rate is higher than bbr_loss_thresh. ++ * Disabled if 0. ++ */ ++static const u32 bbr_full_loss_cnt = 6; ++ ++/* Exit STARTUP if number of round trips with ECN mark rate above ecn_thresh ++ * meets this count. ++ */ ++static const u32 bbr_full_ecn_cnt = 2; ++ ++/* Fraction of unutilized headroom to try to leave in path upon high loss. */ ++static const u32 bbr_inflight_headroom = BBR_UNIT * 15 / 100; ++ ++/* How much do we increase cwnd_gain when probing for bandwidth in ++ * BBR_BW_PROBE_UP? This specifies the increment in units of ++ * BBR_UNIT/4. The default is 1, meaning 0.25. ++ * The min value is 0 (meaning 0.0); max is 3 (meaning 0.75). ++ */ ++static const u32 bbr_bw_probe_cwnd_gain = 1; ++ ++/* Max number of packet-timed rounds to wait before probing for bandwidth. If ++ * we want to tolerate 1% random loss per round, and not have this cut our ++ * inflight too much, we must probe for bw periodically on roughly this scale. ++ * If low, limits Reno/CUBIC coexistence; if high, limits loss tolerance. ++ * We aim to be fair with Reno/CUBIC up to a BDP of at least: ++ * BDP = 25Mbps * .030sec /(1514bytes) = 61.9 packets ++ */ ++static const u32 bbr_bw_probe_max_rounds = 63; ++ ++/* Max amount of randomness to inject in round counting for Reno-coexistence. ++ */ ++static const u32 bbr_bw_probe_rand_rounds = 2; ++ ++/* Use BBR-native probe time scale starting at this many usec. ++ * We aim to be fair with Reno/CUBIC up to an inter-loss time epoch of at least: ++ * BDP*RTT = 25Mbps * .030sec /(1514bytes) * 0.030sec = 1.9 secs ++ */ ++static const u32 bbr_bw_probe_base_us = 2 * USEC_PER_SEC; /* 2 secs */ ++ ++/* Use BBR-native probes spread over this many usec: */ ++static const u32 bbr_bw_probe_rand_us = 1 * USEC_PER_SEC; /* 1 secs */ ++ ++/* Use fast path if app-limited, no loss/ECN, and target cwnd was reached? */ ++static const bool bbr_fast_path = true; ++ ++/* Use fast ack mode? */ ++static const bool bbr_fast_ack_mode = true; ++ ++static u32 bbr_max_bw(const struct sock *sk); ++static u32 bbr_bw(const struct sock *sk); ++static void bbr_exit_probe_rtt(struct sock *sk); ++static void bbr_reset_congestion_signals(struct sock *sk); ++static void bbr_run_loss_probe_recovery(struct sock *sk); ++ + static void bbr_check_probe_rtt_done(struct sock *sk); + ++/* This connection can use ECN if both endpoints have signaled ECN support in ++ * the handshake and the per-route settings indicated this is a ++ * shallow-threshold ECN environment, meaning both: ++ * (a) ECN CE marks indicate low-latency/shallow-threshold congestion, and ++ * (b) TCP endpoints provide precise ACKs that only ACK data segments ++ * with consistent ECN CE status ++ */ ++static bool bbr_can_use_ecn(const struct sock *sk) ++{ ++ return (tcp_sk(sk)->ecn_flags & TCP_ECN_OK) && ++ (tcp_sk(sk)->ecn_flags & TCP_ECN_LOW); ++} ++ + /* Do we estimate that STARTUP filled the pipe? */ + static bool bbr_full_bw_reached(const struct sock *sk) + { +@@ -214,17 +382,17 @@ static bool bbr_full_bw_reached(const struct sock *sk) + /* Return the windowed max recent bandwidth sample, in pkts/uS << BW_SCALE. */ + static u32 bbr_max_bw(const struct sock *sk) + { +- struct bbr *bbr = inet_csk_ca(sk); ++ const struct bbr *bbr = inet_csk_ca(sk); + +- return minmax_get(&bbr->bw); ++ return max(bbr->bw_hi[0], bbr->bw_hi[1]); + } + + /* Return the estimated bandwidth of the path, in pkts/uS << BW_SCALE. */ + static u32 bbr_bw(const struct sock *sk) + { +- struct bbr *bbr = inet_csk_ca(sk); ++ const struct bbr *bbr = inet_csk_ca(sk); + +- return bbr->lt_use_bw ? bbr->lt_bw : bbr_max_bw(sk); ++ return min(bbr_max_bw(sk), bbr->bw_lo); + } + + /* Return maximum extra acked in past k-2k round trips, +@@ -241,15 +409,23 @@ static u16 bbr_extra_acked(const struct sock *sk) + * The order here is chosen carefully to avoid overflow of u64. This should + * work for input rates of up to 2.9Tbit/sec and gain of 2.89x. + */ +-static u64 bbr_rate_bytes_per_sec(struct sock *sk, u64 rate, int gain) ++static u64 bbr_rate_bytes_per_sec(struct sock *sk, u64 rate, int gain, ++ int margin) + { + unsigned int mss = tcp_sk(sk)->mss_cache; + + rate *= mss; + rate *= gain; + rate >>= BBR_SCALE; +- rate *= USEC_PER_SEC / 100 * (100 - bbr_pacing_margin_percent); +- return rate >> BW_SCALE; ++ rate *= USEC_PER_SEC / 100 * (100 - margin); ++ rate >>= BW_SCALE; ++ rate = max(rate, 1ULL); ++ return rate; ++} ++ ++static u64 bbr_bw_bytes_per_sec(struct sock *sk, u64 rate) ++{ ++ return bbr_rate_bytes_per_sec(sk, rate, BBR_UNIT, 0); + } + + /* Convert a BBR bw and gain factor to a pacing rate in bytes per second. */ +@@ -257,12 +433,13 @@ static unsigned long bbr_bw_to_pacing_rate(struct sock *sk, u32 bw, int gain) + { + u64 rate = bw; + +- rate = bbr_rate_bytes_per_sec(sk, rate, gain); ++ rate = bbr_rate_bytes_per_sec(sk, rate, gain, ++ bbr_pacing_margin_percent); + rate = min_t(u64, rate, READ_ONCE(sk->sk_max_pacing_rate)); + return rate; + } + +-/* Initialize pacing rate to: high_gain * init_cwnd / RTT. */ ++/* Initialize pacing rate to: startup_pacing_gain * init_cwnd / RTT. */ + static void bbr_init_pacing_rate_from_rtt(struct sock *sk) + { + struct tcp_sock *tp = tcp_sk(sk); +@@ -279,7 +456,7 @@ static void bbr_init_pacing_rate_from_rtt(struct sock *sk) + bw = (u64)tcp_snd_cwnd(tp) * BW_UNIT; + do_div(bw, rtt_us); + WRITE_ONCE(sk->sk_pacing_rate, +- bbr_bw_to_pacing_rate(sk, bw, bbr_high_gain)); ++ bbr_bw_to_pacing_rate(sk, bw, bbr_param(sk, startup_pacing_gain))); + } + + /* Pace using current bw estimate and a gain factor. */ +@@ -295,26 +472,48 @@ static void bbr_set_pacing_rate(struct sock *sk, u32 bw, int gain) + WRITE_ONCE(sk->sk_pacing_rate, rate); + } + +-/* override sysctl_tcp_min_tso_segs */ +-__bpf_kfunc static u32 bbr_min_tso_segs(struct sock *sk) ++/* Return the number of segments BBR would like in a TSO/GSO skb, given a ++ * particular max gso size as a constraint. TODO: make this simpler and more ++ * consistent by switching bbr to just call tcp_tso_autosize(). ++ */ ++static u32 bbr_tso_segs_generic(struct sock *sk, unsigned int mss_now, ++ u32 gso_max_size) ++{ ++ struct bbr *bbr = inet_csk_ca(sk); ++ u32 segs, r; ++ u64 bytes; ++ ++ /* Budget a TSO/GSO burst size allowance based on bw (pacing_rate). */ ++ bytes = READ_ONCE(sk->sk_pacing_rate) >> READ_ONCE(sk->sk_pacing_shift); ++ ++ /* Budget a TSO/GSO burst size allowance based on min_rtt. For every ++ * K = 2^tso_rtt_shift microseconds of min_rtt, halve the burst. ++ * The min_rtt-based burst allowance is: 64 KBytes / 2^(min_rtt/K) ++ */ ++ if (bbr_param(sk, tso_rtt_shift)) { ++ r = bbr->min_rtt_us >> bbr_param(sk, tso_rtt_shift); ++ if (r < BITS_PER_TYPE(u32)) /* prevent undefined behavior */ ++ bytes += GSO_LEGACY_MAX_SIZE >> r; ++ } ++ ++ bytes = min_t(u32, bytes, gso_max_size - 1 - MAX_TCP_HEADER); ++ segs = max_t(u32, bytes / mss_now, ++ sock_net(sk)->ipv4.sysctl_tcp_min_tso_segs); ++ return segs; ++} ++ ++/* Custom tcp_tso_autosize() for BBR, used at transmit time to cap skb size. */ ++__bpf_kfunc static u32 bbr_tso_segs(struct sock *sk, unsigned int mss_now) + { +- return READ_ONCE(sk->sk_pacing_rate) < (bbr_min_tso_rate >> 3) ? 1 : 2; ++ return bbr_tso_segs_generic(sk, mss_now, sk->sk_gso_max_size); + } + ++/* Like bbr_tso_segs(), using mss_cache, ignoring driver's sk_gso_max_size. */ + static u32 bbr_tso_segs_goal(struct sock *sk) + { + struct tcp_sock *tp = tcp_sk(sk); +- u32 segs, bytes; +- +- /* Sort of tcp_tso_autosize() but ignoring +- * driver provided sk_gso_max_size. +- */ +- bytes = min_t(unsigned long, +- READ_ONCE(sk->sk_pacing_rate) >> READ_ONCE(sk->sk_pacing_shift), +- GSO_LEGACY_MAX_SIZE - 1 - MAX_TCP_HEADER); +- segs = max_t(u32, bytes / tp->mss_cache, bbr_min_tso_segs(sk)); + +- return min(segs, 0x7FU); ++ return bbr_tso_segs_generic(sk, tp->mss_cache, GSO_LEGACY_MAX_SIZE); + } + + /* Save "last known good" cwnd so we can restore it after losses or PROBE_RTT */ +@@ -334,7 +533,9 @@ __bpf_kfunc static void bbr_cwnd_event(struct sock *sk, enum tcp_ca_event event) + struct tcp_sock *tp = tcp_sk(sk); + struct bbr *bbr = inet_csk_ca(sk); + +- if (event == CA_EVENT_TX_START && tp->app_limited) { ++ if (event == CA_EVENT_TX_START) { ++ if (!tp->app_limited) ++ return; + bbr->idle_restart = 1; + bbr->ack_epoch_mstamp = tp->tcp_mstamp; + bbr->ack_epoch_acked = 0; +@@ -345,6 +546,16 @@ __bpf_kfunc static void bbr_cwnd_event(struct sock *sk, enum tcp_ca_event event) + bbr_set_pacing_rate(sk, bbr_bw(sk), BBR_UNIT); + else if (bbr->mode == BBR_PROBE_RTT) + bbr_check_probe_rtt_done(sk); ++ } else if ((event == CA_EVENT_ECN_IS_CE || ++ event == CA_EVENT_ECN_NO_CE) && ++ bbr_can_use_ecn(sk) && ++ bbr_param(sk, precise_ece_ack)) { ++ u32 state = bbr->ce_state; ++ dctcp_ece_ack_update(sk, event, &bbr->prior_rcv_nxt, &state); ++ bbr->ce_state = state; ++ } else if (event == CA_EVENT_TLP_RECOVERY && ++ bbr_param(sk, loss_probe_recovery)) { ++ bbr_run_loss_probe_recovery(sk); + } + } + +@@ -367,10 +578,10 @@ static u32 bbr_bdp(struct sock *sk, u32 bw, int gain) + * default. This should only happen when the connection is not using TCP + * timestamps and has retransmitted all of the SYN/SYNACK/data packets + * ACKed so far. In this case, an RTO can cut cwnd to 1, in which +- * case we need to slow-start up toward something safe: TCP_INIT_CWND. ++ * case we need to slow-start up toward something safe: initial cwnd. + */ + if (unlikely(bbr->min_rtt_us == ~0U)) /* no valid RTT samples yet? */ +- return TCP_INIT_CWND; /* be safe: cap at default initial cwnd*/ ++ return bbr->init_cwnd; /* be safe: cap at initial cwnd */ + + w = (u64)bw * bbr->min_rtt_us; + +@@ -387,23 +598,23 @@ static u32 bbr_bdp(struct sock *sk, u32 bw, int gain) + * - one skb in sending host Qdisc, + * - one skb in sending host TSO/GSO engine + * - one skb being received by receiver host LRO/GRO/delayed-ACK engine +- * Don't worry, at low rates (bbr_min_tso_rate) this won't bloat cwnd because +- * in such cases tso_segs_goal is 1. The minimum cwnd is 4 packets, ++ * Don't worry, at low rates this won't bloat cwnd because ++ * in such cases tso_segs_goal is small. The minimum cwnd is 4 packets, + * which allows 2 outstanding 2-packet sequences, to try to keep pipe + * full even with ACK-every-other-packet delayed ACKs. + */ + static u32 bbr_quantization_budget(struct sock *sk, u32 cwnd) + { + struct bbr *bbr = inet_csk_ca(sk); ++ u32 tso_segs_goal; + +- /* Allow enough full-sized skbs in flight to utilize end systems. */ +- cwnd += 3 * bbr_tso_segs_goal(sk); +- +- /* Reduce delayed ACKs by rounding up cwnd to the next even number. */ +- cwnd = (cwnd + 1) & ~1U; ++ tso_segs_goal = 3 * bbr_tso_segs_goal(sk); + ++ /* Allow enough full-sized skbs in flight to utilize end systems. */ ++ cwnd = max_t(u32, cwnd, tso_segs_goal); ++ cwnd = max_t(u32, cwnd, bbr_param(sk, cwnd_min_target)); + /* Ensure gain cycling gets inflight above BDP even for small BDPs. */ +- if (bbr->mode == BBR_PROBE_BW && bbr->cycle_idx == 0) ++ if (bbr->mode == BBR_PROBE_BW && bbr->cycle_idx == BBR_BW_PROBE_UP) + cwnd += 2; + + return cwnd; +@@ -458,10 +669,10 @@ static u32 bbr_ack_aggregation_cwnd(struct sock *sk) + { + u32 max_aggr_cwnd, aggr_cwnd = 0; + +- if (bbr_extra_acked_gain && bbr_full_bw_reached(sk)) { ++ if (bbr_param(sk, extra_acked_gain)) { + max_aggr_cwnd = ((u64)bbr_bw(sk) * bbr_extra_acked_max_us) + / BW_UNIT; +- aggr_cwnd = (bbr_extra_acked_gain * bbr_extra_acked(sk)) ++ aggr_cwnd = (bbr_param(sk, extra_acked_gain) * bbr_extra_acked(sk)) + >> BBR_SCALE; + aggr_cwnd = min(aggr_cwnd, max_aggr_cwnd); + } +@@ -469,66 +680,27 @@ static u32 bbr_ack_aggregation_cwnd(struct sock *sk) + return aggr_cwnd; + } + +-/* An optimization in BBR to reduce losses: On the first round of recovery, we +- * follow the packet conservation principle: send P packets per P packets acked. +- * After that, we slow-start and send at most 2*P packets per P packets acked. +- * After recovery finishes, or upon undo, we restore the cwnd we had when +- * recovery started (capped by the target cwnd based on estimated BDP). +- * +- * TODO(ycheng/ncardwell): implement a rate-based approach. +- */ +-static bool bbr_set_cwnd_to_recover_or_restore( +- struct sock *sk, const struct rate_sample *rs, u32 acked, u32 *new_cwnd) ++/* Returns the cwnd for PROBE_RTT mode. */ ++static u32 bbr_probe_rtt_cwnd(struct sock *sk) + { +- struct tcp_sock *tp = tcp_sk(sk); +- struct bbr *bbr = inet_csk_ca(sk); +- u8 prev_state = bbr->prev_ca_state, state = inet_csk(sk)->icsk_ca_state; +- u32 cwnd = tcp_snd_cwnd(tp); +- +- /* An ACK for P pkts should release at most 2*P packets. We do this +- * in two steps. First, here we deduct the number of lost packets. +- * Then, in bbr_set_cwnd() we slow start up toward the target cwnd. +- */ +- if (rs->losses > 0) +- cwnd = max_t(s32, cwnd - rs->losses, 1); +- +- if (state == TCP_CA_Recovery && prev_state != TCP_CA_Recovery) { +- /* Starting 1st round of Recovery, so do packet conservation. */ +- bbr->packet_conservation = 1; +- bbr->next_rtt_delivered = tp->delivered; /* start round now */ +- /* Cut unused cwnd from app behavior, TSQ, or TSO deferral: */ +- cwnd = tcp_packets_in_flight(tp) + acked; +- } else if (prev_state >= TCP_CA_Recovery && state < TCP_CA_Recovery) { +- /* Exiting loss recovery; restore cwnd saved before recovery. */ +- cwnd = max(cwnd, bbr->prior_cwnd); +- bbr->packet_conservation = 0; +- } +- bbr->prev_ca_state = state; +- +- if (bbr->packet_conservation) { +- *new_cwnd = max(cwnd, tcp_packets_in_flight(tp) + acked); +- return true; /* yes, using packet conservation */ +- } +- *new_cwnd = cwnd; +- return false; ++ return max_t(u32, bbr_param(sk, cwnd_min_target), ++ bbr_bdp(sk, bbr_bw(sk), bbr_param(sk, probe_rtt_cwnd_gain))); + } + + /* Slow-start up toward target cwnd (if bw estimate is growing, or packet loss + * has drawn us down below target), or snap down to target if we're above it. + */ + static void bbr_set_cwnd(struct sock *sk, const struct rate_sample *rs, +- u32 acked, u32 bw, int gain) ++ u32 acked, u32 bw, int gain, u32 cwnd, ++ struct bbr_context *ctx) + { + struct tcp_sock *tp = tcp_sk(sk); + struct bbr *bbr = inet_csk_ca(sk); +- u32 cwnd = tcp_snd_cwnd(tp), target_cwnd = 0; ++ u32 target_cwnd = 0; + + if (!acked) + goto done; /* no packet fully ACKed; just apply caps */ + +- if (bbr_set_cwnd_to_recover_or_restore(sk, rs, acked, &cwnd)) +- goto done; +- + target_cwnd = bbr_bdp(sk, bw, gain); + + /* Increment the cwnd to account for excess ACKed data that seems +@@ -537,74 +709,26 @@ static void bbr_set_cwnd(struct sock *sk, const struct rate_sample *rs, + target_cwnd += bbr_ack_aggregation_cwnd(sk); + target_cwnd = bbr_quantization_budget(sk, target_cwnd); + +- /* If we're below target cwnd, slow start cwnd toward target cwnd. */ +- if (bbr_full_bw_reached(sk)) /* only cut cwnd if we filled the pipe */ +- cwnd = min(cwnd + acked, target_cwnd); +- else if (cwnd < target_cwnd || tp->delivered < TCP_INIT_CWND) +- cwnd = cwnd + acked; +- cwnd = max(cwnd, bbr_cwnd_min_target); ++ /* Update cwnd and enable fast path if cwnd reaches target_cwnd. */ ++ bbr->try_fast_path = 0; ++ if (bbr_full_bw_reached(sk)) { /* only cut cwnd if we filled the pipe */ ++ cwnd += acked; ++ if (cwnd >= target_cwnd) { ++ cwnd = target_cwnd; ++ bbr->try_fast_path = 1; ++ } ++ } else if (cwnd < target_cwnd || cwnd < 2 * bbr->init_cwnd) { ++ cwnd += acked; ++ } else { ++ bbr->try_fast_path = 1; ++ } + ++ cwnd = max_t(u32, cwnd, bbr_param(sk, cwnd_min_target)); + done: +- tcp_snd_cwnd_set(tp, min(cwnd, tp->snd_cwnd_clamp)); /* apply global cap */ ++ tcp_snd_cwnd_set(tp, min(cwnd, tp->snd_cwnd_clamp)); /* global cap */ + if (bbr->mode == BBR_PROBE_RTT) /* drain queue, refresh min_rtt */ +- tcp_snd_cwnd_set(tp, min(tcp_snd_cwnd(tp), bbr_cwnd_min_target)); +-} +- +-/* End cycle phase if it's time and/or we hit the phase's in-flight target. */ +-static bool bbr_is_next_cycle_phase(struct sock *sk, +- const struct rate_sample *rs) +-{ +- struct tcp_sock *tp = tcp_sk(sk); +- struct bbr *bbr = inet_csk_ca(sk); +- bool is_full_length = +- tcp_stamp_us_delta(tp->delivered_mstamp, bbr->cycle_mstamp) > +- bbr->min_rtt_us; +- u32 inflight, bw; +- +- /* The pacing_gain of 1.0 paces at the estimated bw to try to fully +- * use the pipe without increasing the queue. +- */ +- if (bbr->pacing_gain == BBR_UNIT) +- return is_full_length; /* just use wall clock time */ +- +- inflight = bbr_packets_in_net_at_edt(sk, rs->prior_in_flight); +- bw = bbr_max_bw(sk); +- +- /* A pacing_gain > 1.0 probes for bw by trying to raise inflight to at +- * least pacing_gain*BDP; this may take more than min_rtt if min_rtt is +- * small (e.g. on a LAN). We do not persist if packets are lost, since +- * a path with small buffers may not hold that much. +- */ +- if (bbr->pacing_gain > BBR_UNIT) +- return is_full_length && +- (rs->losses || /* perhaps pacing_gain*BDP won't fit */ +- inflight >= bbr_inflight(sk, bw, bbr->pacing_gain)); +- +- /* A pacing_gain < 1.0 tries to drain extra queue we added if bw +- * probing didn't find more bw. If inflight falls to match BDP then we +- * estimate queue is drained; persisting would underutilize the pipe. +- */ +- return is_full_length || +- inflight <= bbr_inflight(sk, bw, BBR_UNIT); +-} +- +-static void bbr_advance_cycle_phase(struct sock *sk) +-{ +- struct tcp_sock *tp = tcp_sk(sk); +- struct bbr *bbr = inet_csk_ca(sk); +- +- bbr->cycle_idx = (bbr->cycle_idx + 1) & (CYCLE_LEN - 1); +- bbr->cycle_mstamp = tp->delivered_mstamp; +-} +- +-/* Gain cycling: cycle pacing gain to converge to fair share of available bw. */ +-static void bbr_update_cycle_phase(struct sock *sk, +- const struct rate_sample *rs) +-{ +- struct bbr *bbr = inet_csk_ca(sk); +- +- if (bbr->mode == BBR_PROBE_BW && bbr_is_next_cycle_phase(sk, rs)) +- bbr_advance_cycle_phase(sk); ++ tcp_snd_cwnd_set(tp, min_t(u32, tcp_snd_cwnd(tp), ++ bbr_probe_rtt_cwnd(sk))); + } + + static void bbr_reset_startup_mode(struct sock *sk) +@@ -614,191 +738,49 @@ static void bbr_reset_startup_mode(struct sock *sk) + bbr->mode = BBR_STARTUP; + } + +-static void bbr_reset_probe_bw_mode(struct sock *sk) +-{ +- struct bbr *bbr = inet_csk_ca(sk); +- +- bbr->mode = BBR_PROBE_BW; +- bbr->cycle_idx = CYCLE_LEN - 1 - get_random_u32_below(bbr_cycle_rand); +- bbr_advance_cycle_phase(sk); /* flip to next phase of gain cycle */ +-} +- +-static void bbr_reset_mode(struct sock *sk) +-{ +- if (!bbr_full_bw_reached(sk)) +- bbr_reset_startup_mode(sk); +- else +- bbr_reset_probe_bw_mode(sk); +-} +- +-/* Start a new long-term sampling interval. */ +-static void bbr_reset_lt_bw_sampling_interval(struct sock *sk) +-{ +- struct tcp_sock *tp = tcp_sk(sk); +- struct bbr *bbr = inet_csk_ca(sk); +- +- bbr->lt_last_stamp = div_u64(tp->delivered_mstamp, USEC_PER_MSEC); +- bbr->lt_last_delivered = tp->delivered; +- bbr->lt_last_lost = tp->lost; +- bbr->lt_rtt_cnt = 0; +-} +- +-/* Completely reset long-term bandwidth sampling. */ +-static void bbr_reset_lt_bw_sampling(struct sock *sk) +-{ +- struct bbr *bbr = inet_csk_ca(sk); +- +- bbr->lt_bw = 0; +- bbr->lt_use_bw = 0; +- bbr->lt_is_sampling = false; +- bbr_reset_lt_bw_sampling_interval(sk); +-} +- +-/* Long-term bw sampling interval is done. Estimate whether we're policed. */ +-static void bbr_lt_bw_interval_done(struct sock *sk, u32 bw) +-{ +- struct bbr *bbr = inet_csk_ca(sk); +- u32 diff; +- +- if (bbr->lt_bw) { /* do we have bw from a previous interval? */ +- /* Is new bw close to the lt_bw from the previous interval? */ +- diff = abs(bw - bbr->lt_bw); +- if ((diff * BBR_UNIT <= bbr_lt_bw_ratio * bbr->lt_bw) || +- (bbr_rate_bytes_per_sec(sk, diff, BBR_UNIT) <= +- bbr_lt_bw_diff)) { +- /* All criteria are met; estimate we're policed. */ +- bbr->lt_bw = (bw + bbr->lt_bw) >> 1; /* avg 2 intvls */ +- bbr->lt_use_bw = 1; +- bbr->pacing_gain = BBR_UNIT; /* try to avoid drops */ +- bbr->lt_rtt_cnt = 0; +- return; +- } +- } +- bbr->lt_bw = bw; +- bbr_reset_lt_bw_sampling_interval(sk); +-} +- +-/* Token-bucket traffic policers are common (see "An Internet-Wide Analysis of +- * Traffic Policing", SIGCOMM 2016). BBR detects token-bucket policers and +- * explicitly models their policed rate, to reduce unnecessary losses. We +- * estimate that we're policed if we see 2 consecutive sampling intervals with +- * consistent throughput and high packet loss. If we think we're being policed, +- * set lt_bw to the "long-term" average delivery rate from those 2 intervals. ++/* See if we have reached next round trip. Upon start of the new round, ++ * returns packets delivered since previous round start plus this ACK. + */ +-static void bbr_lt_bw_sampling(struct sock *sk, const struct rate_sample *rs) +-{ +- struct tcp_sock *tp = tcp_sk(sk); +- struct bbr *bbr = inet_csk_ca(sk); +- u32 lost, delivered; +- u64 bw; +- u32 t; +- +- if (bbr->lt_use_bw) { /* already using long-term rate, lt_bw? */ +- if (bbr->mode == BBR_PROBE_BW && bbr->round_start && +- ++bbr->lt_rtt_cnt >= bbr_lt_bw_max_rtts) { +- bbr_reset_lt_bw_sampling(sk); /* stop using lt_bw */ +- bbr_reset_probe_bw_mode(sk); /* restart gain cycling */ +- } +- return; +- } +- +- /* Wait for the first loss before sampling, to let the policer exhaust +- * its tokens and estimate the steady-state rate allowed by the policer. +- * Starting samples earlier includes bursts that over-estimate the bw. +- */ +- if (!bbr->lt_is_sampling) { +- if (!rs->losses) +- return; +- bbr_reset_lt_bw_sampling_interval(sk); +- bbr->lt_is_sampling = true; +- } +- +- /* To avoid underestimates, reset sampling if we run out of data. */ +- if (rs->is_app_limited) { +- bbr_reset_lt_bw_sampling(sk); +- return; +- } +- +- if (bbr->round_start) +- bbr->lt_rtt_cnt++; /* count round trips in this interval */ +- if (bbr->lt_rtt_cnt < bbr_lt_intvl_min_rtts) +- return; /* sampling interval needs to be longer */ +- if (bbr->lt_rtt_cnt > 4 * bbr_lt_intvl_min_rtts) { +- bbr_reset_lt_bw_sampling(sk); /* interval is too long */ +- return; +- } +- +- /* End sampling interval when a packet is lost, so we estimate the +- * policer tokens were exhausted. Stopping the sampling before the +- * tokens are exhausted under-estimates the policed rate. +- */ +- if (!rs->losses) +- return; +- +- /* Calculate packets lost and delivered in sampling interval. */ +- lost = tp->lost - bbr->lt_last_lost; +- delivered = tp->delivered - bbr->lt_last_delivered; +- /* Is loss rate (lost/delivered) >= lt_loss_thresh? If not, wait. */ +- if (!delivered || (lost << BBR_SCALE) < bbr_lt_loss_thresh * delivered) +- return; +- +- /* Find average delivery rate in this sampling interval. */ +- t = div_u64(tp->delivered_mstamp, USEC_PER_MSEC) - bbr->lt_last_stamp; +- if ((s32)t < 1) +- return; /* interval is less than one ms, so wait */ +- /* Check if can multiply without overflow */ +- if (t >= ~0U / USEC_PER_MSEC) { +- bbr_reset_lt_bw_sampling(sk); /* interval too long; reset */ +- return; +- } +- t *= USEC_PER_MSEC; +- bw = (u64)delivered * BW_UNIT; +- do_div(bw, t); +- bbr_lt_bw_interval_done(sk, bw); +-} +- +-/* Estimate the bandwidth based on how fast packets are delivered */ +-static void bbr_update_bw(struct sock *sk, const struct rate_sample *rs) ++static u32 bbr_update_round_start(struct sock *sk, ++ const struct rate_sample *rs, struct bbr_context *ctx) + { + struct tcp_sock *tp = tcp_sk(sk); + struct bbr *bbr = inet_csk_ca(sk); +- u64 bw; ++ u32 round_delivered = 0; + + bbr->round_start = 0; +- if (rs->delivered < 0 || rs->interval_us <= 0) +- return; /* Not a valid observation */ + + /* See if we've reached the next RTT */ +- if (!before(rs->prior_delivered, bbr->next_rtt_delivered)) { ++ if (rs->interval_us > 0 && ++ !before(rs->prior_delivered, bbr->next_rtt_delivered)) { ++ round_delivered = tp->delivered - bbr->next_rtt_delivered; + bbr->next_rtt_delivered = tp->delivered; +- bbr->rtt_cnt++; + bbr->round_start = 1; +- bbr->packet_conservation = 0; + } ++ return round_delivered; ++} + +- bbr_lt_bw_sampling(sk, rs); ++/* Calculate the bandwidth based on how fast packets are delivered */ ++static void bbr_calculate_bw_sample(struct sock *sk, ++ const struct rate_sample *rs, struct bbr_context *ctx) ++{ ++ u64 bw = 0; + + /* Divide delivered by the interval to find a (lower bound) bottleneck + * bandwidth sample. Delivered is in packets and interval_us in uS and + * ratio will be <<1 for most connections. So delivered is first scaled. ++ * Round up to allow growth at low rates, even with integer division. + */ +- bw = div64_long((u64)rs->delivered * BW_UNIT, rs->interval_us); +- +- /* If this sample is application-limited, it is likely to have a very +- * low delivered count that represents application behavior rather than +- * the available network rate. Such a sample could drag down estimated +- * bw, causing needless slow-down. Thus, to continue to send at the +- * last measured network rate, we filter out app-limited samples unless +- * they describe the path bw at least as well as our bw model. +- * +- * So the goal during app-limited phase is to proceed with the best +- * network rate no matter how long. We automatically leave this +- * phase when app writes faster than the network can deliver :) +- */ +- if (!rs->is_app_limited || bw >= bbr_max_bw(sk)) { +- /* Incorporate new sample into our max bw filter. */ +- minmax_running_max(&bbr->bw, bbr_bw_rtts, bbr->rtt_cnt, bw); ++ if (rs->interval_us > 0) { ++ if (WARN_ONCE(rs->delivered < 0, ++ "negative delivered: %d interval_us: %ld\n", ++ rs->delivered, rs->interval_us)) ++ return; ++ ++ bw = DIV_ROUND_UP_ULL((u64)rs->delivered * BW_UNIT, rs->interval_us); + } ++ ++ ctx->sample_bw = bw; + } + + /* Estimates the windowed max degree of ack aggregation. +@@ -812,7 +794,7 @@ static void bbr_update_bw(struct sock *sk, const struct rate_sample *rs) + * + * Max extra_acked is clamped by cwnd and bw * bbr_extra_acked_max_us (100 ms). + * Max filter is an approximate sliding window of 5-10 (packet timed) round +- * trips. ++ * trips for non-startup phase, and 1-2 round trips for startup. + */ + static void bbr_update_ack_aggregation(struct sock *sk, + const struct rate_sample *rs) +@@ -820,15 +802,19 @@ static void bbr_update_ack_aggregation(struct sock *sk, + u32 epoch_us, expected_acked, extra_acked; + struct bbr *bbr = inet_csk_ca(sk); + struct tcp_sock *tp = tcp_sk(sk); ++ u32 extra_acked_win_rtts_thresh = bbr_param(sk, extra_acked_win_rtts); + +- if (!bbr_extra_acked_gain || rs->acked_sacked <= 0 || ++ if (!bbr_param(sk, extra_acked_gain) || rs->acked_sacked <= 0 || + rs->delivered < 0 || rs->interval_us <= 0) + return; + + if (bbr->round_start) { + bbr->extra_acked_win_rtts = min(0x1F, + bbr->extra_acked_win_rtts + 1); +- if (bbr->extra_acked_win_rtts >= bbr_extra_acked_win_rtts) { ++ if (!bbr_full_bw_reached(sk)) ++ extra_acked_win_rtts_thresh = 1; ++ if (bbr->extra_acked_win_rtts >= ++ extra_acked_win_rtts_thresh) { + bbr->extra_acked_win_rtts = 0; + bbr->extra_acked_win_idx = bbr->extra_acked_win_idx ? + 0 : 1; +@@ -862,49 +848,6 @@ static void bbr_update_ack_aggregation(struct sock *sk, + bbr->extra_acked[bbr->extra_acked_win_idx] = extra_acked; + } + +-/* Estimate when the pipe is full, using the change in delivery rate: BBR +- * estimates that STARTUP filled the pipe if the estimated bw hasn't changed by +- * at least bbr_full_bw_thresh (25%) after bbr_full_bw_cnt (3) non-app-limited +- * rounds. Why 3 rounds: 1: rwin autotuning grows the rwin, 2: we fill the +- * higher rwin, 3: we get higher delivery rate samples. Or transient +- * cross-traffic or radio noise can go away. CUBIC Hystart shares a similar +- * design goal, but uses delay and inter-ACK spacing instead of bandwidth. +- */ +-static void bbr_check_full_bw_reached(struct sock *sk, +- const struct rate_sample *rs) +-{ +- struct bbr *bbr = inet_csk_ca(sk); +- u32 bw_thresh; +- +- if (bbr_full_bw_reached(sk) || !bbr->round_start || rs->is_app_limited) +- return; +- +- bw_thresh = (u64)bbr->full_bw * bbr_full_bw_thresh >> BBR_SCALE; +- if (bbr_max_bw(sk) >= bw_thresh) { +- bbr->full_bw = bbr_max_bw(sk); +- bbr->full_bw_cnt = 0; +- return; +- } +- ++bbr->full_bw_cnt; +- bbr->full_bw_reached = bbr->full_bw_cnt >= bbr_full_bw_cnt; +-} +- +-/* If pipe is probably full, drain the queue and then enter steady-state. */ +-static void bbr_check_drain(struct sock *sk, const struct rate_sample *rs) +-{ +- struct bbr *bbr = inet_csk_ca(sk); +- +- if (bbr->mode == BBR_STARTUP && bbr_full_bw_reached(sk)) { +- bbr->mode = BBR_DRAIN; /* drain queue we created */ +- tcp_sk(sk)->snd_ssthresh = +- bbr_inflight(sk, bbr_max_bw(sk), BBR_UNIT); +- } /* fall through to check if in-flight is already small: */ +- if (bbr->mode == BBR_DRAIN && +- bbr_packets_in_net_at_edt(sk, tcp_packets_in_flight(tcp_sk(sk))) <= +- bbr_inflight(sk, bbr_max_bw(sk), BBR_UNIT)) +- bbr_reset_probe_bw_mode(sk); /* we estimate queue is drained */ +-} +- + static void bbr_check_probe_rtt_done(struct sock *sk) + { + struct tcp_sock *tp = tcp_sk(sk); +@@ -914,9 +857,9 @@ static void bbr_check_probe_rtt_done(struct sock *sk) + after(tcp_jiffies32, bbr->probe_rtt_done_stamp))) + return; + +- bbr->min_rtt_stamp = tcp_jiffies32; /* wait a while until PROBE_RTT */ ++ bbr->probe_rtt_min_stamp = tcp_jiffies32; /* schedule next PROBE_RTT */ + tcp_snd_cwnd_set(tp, max(tcp_snd_cwnd(tp), bbr->prior_cwnd)); +- bbr_reset_mode(sk); ++ bbr_exit_probe_rtt(sk); + } + + /* The goal of PROBE_RTT mode is to have BBR flows cooperatively and +@@ -942,23 +885,35 @@ static void bbr_update_min_rtt(struct sock *sk, const struct rate_sample *rs) + { + struct tcp_sock *tp = tcp_sk(sk); + struct bbr *bbr = inet_csk_ca(sk); +- bool filter_expired; ++ bool probe_rtt_expired, min_rtt_expired; ++ u32 expire; + +- /* Track min RTT seen in the min_rtt_win_sec filter window: */ +- filter_expired = after(tcp_jiffies32, +- bbr->min_rtt_stamp + bbr_min_rtt_win_sec * HZ); ++ /* Track min RTT in probe_rtt_win_ms to time next PROBE_RTT state. */ ++ expire = bbr->probe_rtt_min_stamp + ++ msecs_to_jiffies(bbr_param(sk, probe_rtt_win_ms)); ++ probe_rtt_expired = after(tcp_jiffies32, expire); + if (rs->rtt_us >= 0 && +- (rs->rtt_us < bbr->min_rtt_us || +- (filter_expired && !rs->is_ack_delayed))) { +- bbr->min_rtt_us = rs->rtt_us; +- bbr->min_rtt_stamp = tcp_jiffies32; ++ (rs->rtt_us < bbr->probe_rtt_min_us || ++ (probe_rtt_expired && !rs->is_ack_delayed))) { ++ bbr->probe_rtt_min_us = rs->rtt_us; ++ bbr->probe_rtt_min_stamp = tcp_jiffies32; ++ } ++ /* Track min RTT seen in the min_rtt_win_sec filter window: */ ++ expire = bbr->min_rtt_stamp + bbr_param(sk, min_rtt_win_sec) * HZ; ++ min_rtt_expired = after(tcp_jiffies32, expire); ++ if (bbr->probe_rtt_min_us <= bbr->min_rtt_us || ++ min_rtt_expired) { ++ bbr->min_rtt_us = bbr->probe_rtt_min_us; ++ bbr->min_rtt_stamp = bbr->probe_rtt_min_stamp; + } + +- if (bbr_probe_rtt_mode_ms > 0 && filter_expired && ++ if (bbr_param(sk, probe_rtt_mode_ms) > 0 && probe_rtt_expired && + !bbr->idle_restart && bbr->mode != BBR_PROBE_RTT) { + bbr->mode = BBR_PROBE_RTT; /* dip, drain queue */ + bbr_save_cwnd(sk); /* note cwnd so we can restore it */ + bbr->probe_rtt_done_stamp = 0; ++ bbr->ack_phase = BBR_ACKS_PROBE_STOPPING; ++ bbr->next_rtt_delivered = tp->delivered; + } + + if (bbr->mode == BBR_PROBE_RTT) { +@@ -967,9 +922,9 @@ static void bbr_update_min_rtt(struct sock *sk, const struct rate_sample *rs) + (tp->delivered + tcp_packets_in_flight(tp)) ? : 1; + /* Maintain min packets in flight for max(200 ms, 1 round). */ + if (!bbr->probe_rtt_done_stamp && +- tcp_packets_in_flight(tp) <= bbr_cwnd_min_target) { ++ tcp_packets_in_flight(tp) <= bbr_probe_rtt_cwnd(sk)) { + bbr->probe_rtt_done_stamp = tcp_jiffies32 + +- msecs_to_jiffies(bbr_probe_rtt_mode_ms); ++ msecs_to_jiffies(bbr_param(sk, probe_rtt_mode_ms)); + bbr->probe_rtt_round_done = 0; + bbr->next_rtt_delivered = tp->delivered; + } else if (bbr->probe_rtt_done_stamp) { +@@ -990,18 +945,20 @@ static void bbr_update_gains(struct sock *sk) + + switch (bbr->mode) { + case BBR_STARTUP: +- bbr->pacing_gain = bbr_high_gain; +- bbr->cwnd_gain = bbr_high_gain; ++ bbr->pacing_gain = bbr_param(sk, startup_pacing_gain); ++ bbr->cwnd_gain = bbr_param(sk, startup_cwnd_gain); + break; + case BBR_DRAIN: +- bbr->pacing_gain = bbr_drain_gain; /* slow, to drain */ +- bbr->cwnd_gain = bbr_high_gain; /* keep cwnd */ ++ bbr->pacing_gain = bbr_param(sk, drain_gain); /* slow, to drain */ ++ bbr->cwnd_gain = bbr_param(sk, startup_cwnd_gain); /* keep cwnd */ + break; + case BBR_PROBE_BW: +- bbr->pacing_gain = (bbr->lt_use_bw ? +- BBR_UNIT : +- bbr_pacing_gain[bbr->cycle_idx]); +- bbr->cwnd_gain = bbr_cwnd_gain; ++ bbr->pacing_gain = bbr_pacing_gain[bbr->cycle_idx]; ++ bbr->cwnd_gain = bbr_param(sk, cwnd_gain); ++ if (bbr_param(sk, bw_probe_cwnd_gain) && ++ bbr->cycle_idx == BBR_BW_PROBE_UP) ++ bbr->cwnd_gain += ++ BBR_UNIT * bbr_param(sk, bw_probe_cwnd_gain) / 4; + break; + case BBR_PROBE_RTT: + bbr->pacing_gain = BBR_UNIT; +@@ -1013,144 +970,1387 @@ static void bbr_update_gains(struct sock *sk) + } + } + +-static void bbr_update_model(struct sock *sk, const struct rate_sample *rs) ++__bpf_kfunc static u32 bbr_sndbuf_expand(struct sock *sk) + { +- bbr_update_bw(sk, rs); +- bbr_update_ack_aggregation(sk, rs); +- bbr_update_cycle_phase(sk, rs); +- bbr_check_full_bw_reached(sk, rs); +- bbr_check_drain(sk, rs); +- bbr_update_min_rtt(sk, rs); +- bbr_update_gains(sk); ++ /* Provision 3 * cwnd since BBR may slow-start even during recovery. */ ++ return 3; + } + +-__bpf_kfunc static void bbr_main(struct sock *sk, const struct rate_sample *rs) ++/* Incorporate a new bw sample into the current window of our max filter. */ ++static void bbr_take_max_bw_sample(struct sock *sk, u32 bw) + { + struct bbr *bbr = inet_csk_ca(sk); +- u32 bw; +- +- bbr_update_model(sk, rs); + +- bw = bbr_bw(sk); +- bbr_set_pacing_rate(sk, bw, bbr->pacing_gain); +- bbr_set_cwnd(sk, rs, rs->acked_sacked, bw, bbr->cwnd_gain); ++ bbr->bw_hi[1] = max(bw, bbr->bw_hi[1]); + } + +-__bpf_kfunc static void bbr_init(struct sock *sk) ++/* Keep max of last 1-2 cycles. Each PROBE_BW cycle, flip filter window. */ ++static void bbr_advance_max_bw_filter(struct sock *sk) + { +- struct tcp_sock *tp = tcp_sk(sk); + struct bbr *bbr = inet_csk_ca(sk); + +- bbr->prior_cwnd = 0; +- tp->snd_ssthresh = TCP_INFINITE_SSTHRESH; +- bbr->rtt_cnt = 0; +- bbr->next_rtt_delivered = tp->delivered; +- bbr->prev_ca_state = TCP_CA_Open; +- bbr->packet_conservation = 0; +- +- bbr->probe_rtt_done_stamp = 0; +- bbr->probe_rtt_round_done = 0; +- bbr->min_rtt_us = tcp_min_rtt(tp); +- bbr->min_rtt_stamp = tcp_jiffies32; +- +- minmax_reset(&bbr->bw, bbr->rtt_cnt, 0); /* init max bw to 0 */ ++ if (!bbr->bw_hi[1]) ++ return; /* no samples in this window; remember old window */ ++ bbr->bw_hi[0] = bbr->bw_hi[1]; ++ bbr->bw_hi[1] = 0; ++} + +- bbr->has_seen_rtt = 0; +- bbr_init_pacing_rate_from_rtt(sk); ++/* Reset the estimator for reaching full bandwidth based on bw plateau. */ ++static void bbr_reset_full_bw(struct sock *sk) ++{ ++ struct bbr *bbr = inet_csk_ca(sk); + +- bbr->round_start = 0; +- bbr->idle_restart = 0; +- bbr->full_bw_reached = 0; + bbr->full_bw = 0; + bbr->full_bw_cnt = 0; +- bbr->cycle_mstamp = 0; +- bbr->cycle_idx = 0; +- bbr_reset_lt_bw_sampling(sk); +- bbr_reset_startup_mode(sk); ++ bbr->full_bw_now = 0; ++} + +- bbr->ack_epoch_mstamp = tp->tcp_mstamp; +- bbr->ack_epoch_acked = 0; +- bbr->extra_acked_win_rtts = 0; +- bbr->extra_acked_win_idx = 0; +- bbr->extra_acked[0] = 0; +- bbr->extra_acked[1] = 0; ++/* How much do we want in flight? Our BDP, unless congestion cut cwnd. */ ++static u32 bbr_target_inflight(struct sock *sk) ++{ ++ u32 bdp = bbr_inflight(sk, bbr_bw(sk), BBR_UNIT); + +- cmpxchg(&sk->sk_pacing_status, SK_PACING_NONE, SK_PACING_NEEDED); ++ return min(bdp, tcp_sk(sk)->snd_cwnd); + } + +-__bpf_kfunc static u32 bbr_sndbuf_expand(struct sock *sk) ++static bool bbr_is_probing_bandwidth(struct sock *sk) + { +- /* Provision 3 * cwnd since BBR may slow-start even during recovery. */ +- return 3; ++ struct bbr *bbr = inet_csk_ca(sk); ++ ++ return (bbr->mode == BBR_STARTUP) || ++ (bbr->mode == BBR_PROBE_BW && ++ (bbr->cycle_idx == BBR_BW_PROBE_REFILL || ++ bbr->cycle_idx == BBR_BW_PROBE_UP)); ++} ++ ++/* Has the given amount of time elapsed since we marked the phase start? */ ++static bool bbr_has_elapsed_in_phase(const struct sock *sk, u32 interval_us) ++{ ++ const struct tcp_sock *tp = tcp_sk(sk); ++ const struct bbr *bbr = inet_csk_ca(sk); ++ ++ return tcp_stamp_us_delta(tp->tcp_mstamp, ++ bbr->cycle_mstamp + interval_us) > 0; ++} ++ ++static void bbr_handle_queue_too_high_in_startup(struct sock *sk) ++{ ++ struct bbr *bbr = inet_csk_ca(sk); ++ u32 bdp; /* estimated BDP in packets, with quantization budget */ ++ ++ bbr->full_bw_reached = 1; ++ ++ bdp = bbr_inflight(sk, bbr_max_bw(sk), BBR_UNIT); ++ bbr->inflight_hi = max(bdp, bbr->inflight_latest); ++} ++ ++/* Exit STARTUP upon N consecutive rounds with ECN mark rate > ecn_thresh. */ ++static void bbr_check_ecn_too_high_in_startup(struct sock *sk, u32 ce_ratio) ++{ ++ struct bbr *bbr = inet_csk_ca(sk); ++ ++ if (bbr_full_bw_reached(sk) || !bbr->ecn_eligible || ++ !bbr_param(sk, full_ecn_cnt) || !bbr_param(sk, ecn_thresh)) ++ return; ++ ++ if (ce_ratio >= bbr_param(sk, ecn_thresh)) ++ bbr->startup_ecn_rounds++; ++ else ++ bbr->startup_ecn_rounds = 0; ++ ++ if (bbr->startup_ecn_rounds >= bbr_param(sk, full_ecn_cnt)) { ++ bbr_handle_queue_too_high_in_startup(sk); ++ return; ++ } ++} ++ ++/* Updates ecn_alpha and returns ce_ratio. -1 if not available. */ ++static int bbr_update_ecn_alpha(struct sock *sk) ++{ ++ struct tcp_sock *tp = tcp_sk(sk); ++ struct net *net = sock_net(sk); ++ struct bbr *bbr = inet_csk_ca(sk); ++ s32 delivered, delivered_ce; ++ u64 alpha, ce_ratio; ++ u32 gain; ++ bool want_ecn_alpha; ++ ++ /* See if we should use ECN sender logic for this connection. */ ++ if (!bbr->ecn_eligible && bbr_can_use_ecn(sk) && ++ bbr_param(sk, ecn_factor) && ++ (bbr->min_rtt_us <= bbr_ecn_max_rtt_us || ++ !bbr_ecn_max_rtt_us)) ++ bbr->ecn_eligible = 1; ++ ++ /* Skip updating alpha only if not ECN-eligible and PLB is disabled. */ ++ want_ecn_alpha = (bbr->ecn_eligible || ++ (bbr_can_use_ecn(sk) && ++ READ_ONCE(net->ipv4.sysctl_tcp_plb_enabled))); ++ if (!want_ecn_alpha) ++ return -1; ++ ++ delivered = tp->delivered - bbr->alpha_last_delivered; ++ delivered_ce = tp->delivered_ce - bbr->alpha_last_delivered_ce; ++ ++ if (delivered == 0 || /* avoid divide by zero */ ++ WARN_ON_ONCE(delivered < 0 || delivered_ce < 0)) /* backwards? */ ++ return -1; ++ ++ BUILD_BUG_ON(BBR_SCALE != TCP_PLB_SCALE); ++ ce_ratio = (u64)delivered_ce << BBR_SCALE; ++ do_div(ce_ratio, delivered); ++ ++ gain = bbr_param(sk, ecn_alpha_gain); ++ alpha = ((BBR_UNIT - gain) * bbr->ecn_alpha) >> BBR_SCALE; ++ alpha += (gain * ce_ratio) >> BBR_SCALE; ++ bbr->ecn_alpha = min_t(u32, alpha, BBR_UNIT); ++ ++ bbr->alpha_last_delivered = tp->delivered; ++ bbr->alpha_last_delivered_ce = tp->delivered_ce; ++ ++ bbr_check_ecn_too_high_in_startup(sk, ce_ratio); ++ return (int)ce_ratio; + } + +-/* In theory BBR does not need to undo the cwnd since it does not +- * always reduce cwnd on losses (see bbr_main()). Keep it for now. ++/* Protective Load Balancing (PLB). PLB rehashes outgoing data (to a new IPv6 ++ * flow label) if it encounters sustained congestion in the form of ECN marks. + */ +-__bpf_kfunc static u32 bbr_undo_cwnd(struct sock *sk) ++static void bbr_plb(struct sock *sk, const struct rate_sample *rs, int ce_ratio) ++{ ++ struct bbr *bbr = inet_csk_ca(sk); ++ ++ if (bbr->round_start && ce_ratio >= 0) ++ tcp_plb_update_state(sk, &bbr->plb, ce_ratio); ++ ++ tcp_plb_check_rehash(sk, &bbr->plb); ++} ++ ++/* Each round trip of BBR_BW_PROBE_UP, double volume of probing data. */ ++static void bbr_raise_inflight_hi_slope(struct sock *sk) ++{ ++ struct tcp_sock *tp = tcp_sk(sk); ++ struct bbr *bbr = inet_csk_ca(sk); ++ u32 growth_this_round, cnt; ++ ++ /* Calculate "slope": packets S/Acked per inflight_hi increment. */ ++ growth_this_round = 1 << bbr->bw_probe_up_rounds; ++ bbr->bw_probe_up_rounds = min(bbr->bw_probe_up_rounds + 1, 30); ++ cnt = tcp_snd_cwnd(tp) / growth_this_round; ++ cnt = max(cnt, 1U); ++ bbr->bw_probe_up_cnt = cnt; ++} ++ ++/* In BBR_BW_PROBE_UP, not seeing high loss/ECN/queue, so raise inflight_hi. */ ++static void bbr_probe_inflight_hi_upward(struct sock *sk, ++ const struct rate_sample *rs) ++{ ++ struct tcp_sock *tp = tcp_sk(sk); ++ struct bbr *bbr = inet_csk_ca(sk); ++ u32 delta; ++ ++ if (!tp->is_cwnd_limited || tcp_snd_cwnd(tp) < bbr->inflight_hi) ++ return; /* not fully using inflight_hi, so don't grow it */ ++ ++ /* For each bw_probe_up_cnt packets ACKed, increase inflight_hi by 1. */ ++ bbr->bw_probe_up_acks += rs->acked_sacked; ++ if (bbr->bw_probe_up_acks >= bbr->bw_probe_up_cnt) { ++ delta = bbr->bw_probe_up_acks / bbr->bw_probe_up_cnt; ++ bbr->bw_probe_up_acks -= delta * bbr->bw_probe_up_cnt; ++ bbr->inflight_hi += delta; ++ bbr->try_fast_path = 0; /* Need to update cwnd */ ++ } ++ ++ if (bbr->round_start) ++ bbr_raise_inflight_hi_slope(sk); ++} ++ ++/* Does loss/ECN rate for this sample say inflight is "too high"? ++ * This is used by both the bbr_check_loss_too_high_in_startup() function, ++ * which can be used in either v1 or v2, and the PROBE_UP phase of v2, which ++ * uses it to notice when loss/ECN rates suggest inflight is too high. ++ */ ++static bool bbr_is_inflight_too_high(const struct sock *sk, ++ const struct rate_sample *rs) ++{ ++ const struct bbr *bbr = inet_csk_ca(sk); ++ u32 loss_thresh, ecn_thresh; ++ ++ if (rs->lost > 0 && rs->tx_in_flight) { ++ loss_thresh = (u64)rs->tx_in_flight * bbr_param(sk, loss_thresh) >> ++ BBR_SCALE; ++ if (rs->lost > loss_thresh) { ++ return true; ++ } ++ } ++ ++ if (rs->delivered_ce > 0 && rs->delivered > 0 && ++ bbr->ecn_eligible && bbr_param(sk, ecn_thresh)) { ++ ecn_thresh = (u64)rs->delivered * bbr_param(sk, ecn_thresh) >> ++ BBR_SCALE; ++ if (rs->delivered_ce > ecn_thresh) { ++ return true; ++ } ++ } ++ ++ return false; ++} ++ ++/* Calculate the tx_in_flight level that corresponded to excessive loss. ++ * We find "lost_prefix" segs of the skb where loss rate went too high, ++ * by solving for "lost_prefix" in the following equation: ++ * lost / inflight >= loss_thresh ++ * (lost_prev + lost_prefix) / (inflight_prev + lost_prefix) >= loss_thresh ++ * Then we take that equation, convert it to fixed point, and ++ * round up to the nearest packet. ++ */ ++static u32 bbr_inflight_hi_from_lost_skb(const struct sock *sk, ++ const struct rate_sample *rs, ++ const struct sk_buff *skb) ++{ ++ const struct tcp_sock *tp = tcp_sk(sk); ++ u32 loss_thresh = bbr_param(sk, loss_thresh); ++ u32 pcount, divisor, inflight_hi; ++ s32 inflight_prev, lost_prev; ++ u64 loss_budget, lost_prefix; ++ ++ pcount = tcp_skb_pcount(skb); ++ ++ /* How much data was in flight before this skb? */ ++ inflight_prev = rs->tx_in_flight - pcount; ++ if (inflight_prev < 0) { ++ WARN_ONCE(tcp_skb_tx_in_flight_is_suspicious( ++ pcount, ++ TCP_SKB_CB(skb)->sacked, ++ rs->tx_in_flight), ++ "tx_in_flight: %u pcount: %u reneg: %u", ++ rs->tx_in_flight, pcount, tcp_sk(sk)->is_sack_reneg); ++ return ~0U; ++ } ++ ++ /* How much inflight data was marked lost before this skb? */ ++ lost_prev = rs->lost - pcount; ++ if (WARN_ONCE(lost_prev < 0, ++ "cwnd: %u ca: %d out: %u lost: %u pif: %u " ++ "tx_in_flight: %u tx.lost: %u tp->lost: %u rs->lost: %d " ++ "lost_prev: %d pcount: %d seq: %u end_seq: %u reneg: %u", ++ tcp_snd_cwnd(tp), inet_csk(sk)->icsk_ca_state, ++ tp->packets_out, tp->lost_out, tcp_packets_in_flight(tp), ++ rs->tx_in_flight, TCP_SKB_CB(skb)->tx.lost, tp->lost, ++ rs->lost, lost_prev, pcount, ++ TCP_SKB_CB(skb)->seq, TCP_SKB_CB(skb)->end_seq, ++ tp->is_sack_reneg)) ++ return ~0U; ++ ++ /* At what prefix of this lost skb did losss rate exceed loss_thresh? */ ++ loss_budget = (u64)inflight_prev * loss_thresh + BBR_UNIT - 1; ++ loss_budget >>= BBR_SCALE; ++ if (lost_prev >= loss_budget) { ++ lost_prefix = 0; /* previous losses crossed loss_thresh */ ++ } else { ++ lost_prefix = loss_budget - lost_prev; ++ lost_prefix <<= BBR_SCALE; ++ divisor = BBR_UNIT - loss_thresh; ++ if (WARN_ON_ONCE(!divisor)) /* loss_thresh is 8 bits */ ++ return ~0U; ++ do_div(lost_prefix, divisor); ++ } ++ ++ inflight_hi = inflight_prev + lost_prefix; ++ return inflight_hi; ++} ++ ++/* If loss/ECN rates during probing indicated we may have overfilled a ++ * buffer, return an operating point that tries to leave unutilized headroom in ++ * the path for other flows, for fairness convergence and lower RTTs and loss. ++ */ ++static u32 bbr_inflight_with_headroom(const struct sock *sk) ++{ ++ struct bbr *bbr = inet_csk_ca(sk); ++ u32 headroom, headroom_fraction; ++ ++ if (bbr->inflight_hi == ~0U) ++ return ~0U; ++ ++ headroom_fraction = bbr_param(sk, inflight_headroom); ++ headroom = ((u64)bbr->inflight_hi * headroom_fraction) >> BBR_SCALE; ++ headroom = max(headroom, 1U); ++ return max_t(s32, bbr->inflight_hi - headroom, ++ bbr_param(sk, cwnd_min_target)); ++} ++ ++/* Bound cwnd to a sensible level, based on our current probing state ++ * machine phase and model of a good inflight level (inflight_lo, inflight_hi). ++ */ ++static void bbr_bound_cwnd_for_inflight_model(struct sock *sk) ++{ ++ struct tcp_sock *tp = tcp_sk(sk); ++ struct bbr *bbr = inet_csk_ca(sk); ++ u32 cap; ++ ++ /* tcp_rcv_synsent_state_process() currently calls tcp_ack() ++ * and thus cong_control() without first initializing us(!). ++ */ ++ if (!bbr->initialized) ++ return; ++ ++ cap = ~0U; ++ if (bbr->mode == BBR_PROBE_BW && ++ bbr->cycle_idx != BBR_BW_PROBE_CRUISE) { ++ /* Probe to see if more packets fit in the path. */ ++ cap = bbr->inflight_hi; ++ } else { ++ if (bbr->mode == BBR_PROBE_RTT || ++ (bbr->mode == BBR_PROBE_BW && ++ bbr->cycle_idx == BBR_BW_PROBE_CRUISE)) ++ cap = bbr_inflight_with_headroom(sk); ++ } ++ /* Adapt to any loss/ECN since our last bw probe. */ ++ cap = min(cap, bbr->inflight_lo); ++ ++ cap = max_t(u32, cap, bbr_param(sk, cwnd_min_target)); ++ tcp_snd_cwnd_set(tp, min(cap, tcp_snd_cwnd(tp))); ++} ++ ++/* How should we multiplicatively cut bw or inflight limits based on ECN? */ ++static u32 bbr_ecn_cut(struct sock *sk) ++{ ++ struct bbr *bbr = inet_csk_ca(sk); ++ ++ return BBR_UNIT - ++ ((bbr->ecn_alpha * bbr_param(sk, ecn_factor)) >> BBR_SCALE); ++} ++ ++/* Init lower bounds if have not inited yet. */ ++static void bbr_init_lower_bounds(struct sock *sk, bool init_bw) ++{ ++ struct tcp_sock *tp = tcp_sk(sk); ++ struct bbr *bbr = inet_csk_ca(sk); ++ ++ if (init_bw && bbr->bw_lo == ~0U) ++ bbr->bw_lo = bbr_max_bw(sk); ++ if (bbr->inflight_lo == ~0U) ++ bbr->inflight_lo = tcp_snd_cwnd(tp); ++} ++ ++/* Reduce bw and inflight to (1 - beta). */ ++static void bbr_loss_lower_bounds(struct sock *sk, u32 *bw, u32 *inflight) ++{ ++ struct bbr* bbr = inet_csk_ca(sk); ++ u32 loss_cut = BBR_UNIT - bbr_param(sk, beta); ++ ++ *bw = max_t(u32, bbr->bw_latest, ++ (u64)bbr->bw_lo * loss_cut >> BBR_SCALE); ++ *inflight = max_t(u32, bbr->inflight_latest, ++ (u64)bbr->inflight_lo * loss_cut >> BBR_SCALE); ++} ++ ++/* Reduce inflight to (1 - alpha*ecn_factor). */ ++static void bbr_ecn_lower_bounds(struct sock *sk, u32 *inflight) ++{ ++ struct bbr *bbr = inet_csk_ca(sk); ++ u32 ecn_cut = bbr_ecn_cut(sk); ++ ++ *inflight = (u64)bbr->inflight_lo * ecn_cut >> BBR_SCALE; ++} ++ ++/* Estimate a short-term lower bound on the capacity available now, based ++ * on measurements of the current delivery process and recent history. When we ++ * are seeing loss/ECN at times when we are not probing bw, then conservatively ++ * move toward flow balance by multiplicatively cutting our short-term ++ * estimated safe rate and volume of data (bw_lo and inflight_lo). We use a ++ * multiplicative decrease in order to converge to a lower capacity in time ++ * logarithmic in the magnitude of the decrease. ++ * ++ * However, we do not cut our short-term estimates lower than the current rate ++ * and volume of delivered data from this round trip, since from the current ++ * delivery process we can estimate the measured capacity available now. ++ * ++ * Anything faster than that approach would knowingly risk high loss, which can ++ * cause low bw for Reno/CUBIC and high loss recovery latency for ++ * request/response flows using any congestion control. ++ */ ++static void bbr_adapt_lower_bounds(struct sock *sk, ++ const struct rate_sample *rs) ++{ ++ struct bbr *bbr = inet_csk_ca(sk); ++ u32 ecn_inflight_lo = ~0U; ++ ++ /* We only use lower-bound estimates when not probing bw. ++ * When probing we need to push inflight higher to probe bw. ++ */ ++ if (bbr_is_probing_bandwidth(sk)) ++ return; ++ ++ /* ECN response. */ ++ if (bbr->ecn_in_round && bbr_param(sk, ecn_factor)) { ++ bbr_init_lower_bounds(sk, false); ++ bbr_ecn_lower_bounds(sk, &ecn_inflight_lo); ++ } ++ ++ /* Loss response. */ ++ if (bbr->loss_in_round) { ++ bbr_init_lower_bounds(sk, true); ++ bbr_loss_lower_bounds(sk, &bbr->bw_lo, &bbr->inflight_lo); ++ } ++ ++ /* Adjust to the lower of the levels implied by loss/ECN. */ ++ bbr->inflight_lo = min(bbr->inflight_lo, ecn_inflight_lo); ++ bbr->bw_lo = max(1U, bbr->bw_lo); ++} ++ ++/* Reset any short-term lower-bound adaptation to congestion, so that we can ++ * push our inflight up. ++ */ ++static void bbr_reset_lower_bounds(struct sock *sk) ++{ ++ struct bbr *bbr = inet_csk_ca(sk); ++ ++ bbr->bw_lo = ~0U; ++ bbr->inflight_lo = ~0U; ++} ++ ++/* After bw probing (STARTUP/PROBE_UP), reset signals before entering a state ++ * machine phase where we adapt our lower bound based on congestion signals. ++ */ ++static void bbr_reset_congestion_signals(struct sock *sk) ++{ ++ struct bbr *bbr = inet_csk_ca(sk); ++ ++ bbr->loss_in_round = 0; ++ bbr->ecn_in_round = 0; ++ bbr->loss_in_cycle = 0; ++ bbr->ecn_in_cycle = 0; ++ bbr->bw_latest = 0; ++ bbr->inflight_latest = 0; ++} ++ ++static void bbr_exit_loss_recovery(struct sock *sk) ++{ ++ struct tcp_sock *tp = tcp_sk(sk); ++ struct bbr *bbr = inet_csk_ca(sk); ++ ++ tcp_snd_cwnd_set(tp, max(tcp_snd_cwnd(tp), bbr->prior_cwnd)); ++ bbr->try_fast_path = 0; /* bound cwnd using latest model */ ++} ++ ++/* Update rate and volume of delivered data from latest round trip. */ ++static void bbr_update_latest_delivery_signals( ++ struct sock *sk, const struct rate_sample *rs, struct bbr_context *ctx) ++{ ++ struct tcp_sock *tp = tcp_sk(sk); ++ struct bbr *bbr = inet_csk_ca(sk); ++ ++ bbr->loss_round_start = 0; ++ if (rs->interval_us <= 0 || !rs->acked_sacked) ++ return; /* Not a valid observation */ ++ ++ bbr->bw_latest = max_t(u32, bbr->bw_latest, ctx->sample_bw); ++ bbr->inflight_latest = max_t(u32, bbr->inflight_latest, rs->delivered); ++ ++ if (!before(rs->prior_delivered, bbr->loss_round_delivered)) { ++ bbr->loss_round_delivered = tp->delivered; ++ bbr->loss_round_start = 1; /* mark start of new round trip */ ++ } ++} ++ ++/* Once per round, reset filter for latest rate and volume of delivered data. */ ++static void bbr_advance_latest_delivery_signals( ++ struct sock *sk, const struct rate_sample *rs, struct bbr_context *ctx) ++{ ++ struct bbr *bbr = inet_csk_ca(sk); ++ ++ /* If ACK matches a TLP retransmit, persist the filter. If we detect ++ * that a TLP retransmit plugged a tail loss, we'll want to remember ++ * how much data the path delivered before the tail loss. ++ */ ++ if (bbr->loss_round_start && !rs->is_acking_tlp_retrans_seq) { ++ bbr->bw_latest = ctx->sample_bw; ++ bbr->inflight_latest = rs->delivered; ++ } ++} ++ ++/* Update (most of) our congestion signals: track the recent rate and volume of ++ * delivered data, presence of loss, and EWMA degree of ECN marking. ++ */ ++static void bbr_update_congestion_signals( ++ struct sock *sk, const struct rate_sample *rs, struct bbr_context *ctx) + { + struct bbr *bbr = inet_csk_ca(sk); ++ u64 bw; ++ ++ if (rs->interval_us <= 0 || !rs->acked_sacked) ++ return; /* Not a valid observation */ ++ bw = ctx->sample_bw; + +- bbr->full_bw = 0; /* spurious slow-down; reset full pipe detection */ ++ if (!rs->is_app_limited || bw >= bbr_max_bw(sk)) ++ bbr_take_max_bw_sample(sk, bw); ++ ++ bbr->loss_in_round |= (rs->losses > 0); ++ ++ if (!bbr->loss_round_start) ++ return; /* skip the per-round-trip updates */ ++ /* Now do per-round-trip updates. */ ++ bbr_adapt_lower_bounds(sk, rs); ++ ++ bbr->loss_in_round = 0; ++ bbr->ecn_in_round = 0; ++} ++ ++/* Bandwidth probing can cause loss. To help coexistence with loss-based ++ * congestion control we spread out our probing in a Reno-conscious way. Due to ++ * the shape of the Reno sawtooth, the time required between loss epochs for an ++ * idealized Reno flow is a number of round trips that is the BDP of that ++ * flow. We count packet-timed round trips directly, since measured RTT can ++ * vary widely, and Reno is driven by packet-timed round trips. ++ */ ++static bool bbr_is_reno_coexistence_probe_time(struct sock *sk) ++{ ++ struct bbr *bbr = inet_csk_ca(sk); ++ u32 rounds; ++ ++ /* Random loss can shave some small percentage off of our inflight ++ * in each round. To survive this, flows need robust periodic probes. ++ */ ++ rounds = min_t(u32, bbr_param(sk, bw_probe_max_rounds), bbr_target_inflight(sk)); ++ return bbr->rounds_since_probe >= rounds; ++} ++ ++/* How long do we want to wait before probing for bandwidth (and risking ++ * loss)? We randomize the wait, for better mixing and fairness convergence. ++ * ++ * We bound the Reno-coexistence inter-bw-probe time to be 62-63 round trips. ++ * This is calculated to allow fairness with a 25Mbps, 30ms Reno flow, ++ * (eg 4K video to a broadband user): ++ * BDP = 25Mbps * .030sec /(1514bytes) = 61.9 packets ++ * ++ * We bound the BBR-native inter-bw-probe wall clock time to be: ++ * (a) higher than 2 sec: to try to avoid causing loss for a long enough time ++ * to allow Reno at 30ms to get 4K video bw, the inter-bw-probe time must ++ * be at least: 25Mbps * .030sec / (1514bytes) * 0.030sec = 1.9secs ++ * (b) lower than 3 sec: to ensure flows can start probing in a reasonable ++ * amount of time to discover unutilized bw on human-scale interactive ++ * time-scales (e.g. perhaps traffic from a web page download that we ++ * were competing with is now complete). ++ */ ++static void bbr_pick_probe_wait(struct sock *sk) ++{ ++ struct bbr *bbr = inet_csk_ca(sk); ++ ++ /* Decide the random round-trip bound for wait until probe: */ ++ bbr->rounds_since_probe = ++ get_random_u32_below(bbr_param(sk, bw_probe_rand_rounds)); ++ /* Decide the random wall clock bound for wait until probe: */ ++ bbr->probe_wait_us = bbr_param(sk, bw_probe_base_us) + ++ get_random_u32_below(bbr_param(sk, bw_probe_rand_us)); ++} ++ ++static void bbr_set_cycle_idx(struct sock *sk, int cycle_idx) ++{ ++ struct bbr *bbr = inet_csk_ca(sk); ++ ++ bbr->cycle_idx = cycle_idx; ++ /* New phase, so need to update cwnd and pacing rate. */ ++ bbr->try_fast_path = 0; ++} ++ ++/* Send at estimated bw to fill the pipe, but not queue. We need this phase ++ * before PROBE_UP, because as soon as we send faster than the available bw ++ * we will start building a queue, and if the buffer is shallow we can cause ++ * loss. If we do not fill the pipe before we cause this loss, our bw_hi and ++ * inflight_hi estimates will underestimate. ++ */ ++static void bbr_start_bw_probe_refill(struct sock *sk, u32 bw_probe_up_rounds) ++{ ++ struct tcp_sock *tp = tcp_sk(sk); ++ struct bbr *bbr = inet_csk_ca(sk); ++ ++ bbr_reset_lower_bounds(sk); ++ bbr->bw_probe_up_rounds = bw_probe_up_rounds; ++ bbr->bw_probe_up_acks = 0; ++ bbr->stopped_risky_probe = 0; ++ bbr->ack_phase = BBR_ACKS_REFILLING; ++ bbr->next_rtt_delivered = tp->delivered; ++ bbr_set_cycle_idx(sk, BBR_BW_PROBE_REFILL); ++} ++ ++/* Now probe max deliverable data rate and volume. */ ++static void bbr_start_bw_probe_up(struct sock *sk, struct bbr_context *ctx) ++{ ++ struct tcp_sock *tp = tcp_sk(sk); ++ struct bbr *bbr = inet_csk_ca(sk); ++ ++ bbr->ack_phase = BBR_ACKS_PROBE_STARTING; ++ bbr->next_rtt_delivered = tp->delivered; ++ bbr->cycle_mstamp = tp->tcp_mstamp; ++ bbr_reset_full_bw(sk); ++ bbr->full_bw = ctx->sample_bw; ++ bbr_set_cycle_idx(sk, BBR_BW_PROBE_UP); ++ bbr_raise_inflight_hi_slope(sk); ++} ++ ++/* Start a new PROBE_BW probing cycle of some wall clock length. Pick a wall ++ * clock time at which to probe beyond an inflight that we think to be ++ * safe. This will knowingly risk packet loss, so we want to do this rarely, to ++ * keep packet loss rates low. Also start a round-trip counter, to probe faster ++ * if we estimate a Reno flow at our BDP would probe faster. ++ */ ++static void bbr_start_bw_probe_down(struct sock *sk) ++{ ++ struct tcp_sock *tp = tcp_sk(sk); ++ struct bbr *bbr = inet_csk_ca(sk); ++ ++ bbr_reset_congestion_signals(sk); ++ bbr->bw_probe_up_cnt = ~0U; /* not growing inflight_hi any more */ ++ bbr_pick_probe_wait(sk); ++ bbr->cycle_mstamp = tp->tcp_mstamp; /* start wall clock */ ++ bbr->ack_phase = BBR_ACKS_PROBE_STOPPING; ++ bbr->next_rtt_delivered = tp->delivered; ++ bbr_set_cycle_idx(sk, BBR_BW_PROBE_DOWN); ++} ++ ++/* Cruise: maintain what we estimate to be a neutral, conservative ++ * operating point, without attempting to probe up for bandwidth or down for ++ * RTT, and only reducing inflight in response to loss/ECN signals. ++ */ ++static void bbr_start_bw_probe_cruise(struct sock *sk) ++{ ++ struct bbr *bbr = inet_csk_ca(sk); ++ ++ if (bbr->inflight_lo != ~0U) ++ bbr->inflight_lo = min(bbr->inflight_lo, bbr->inflight_hi); ++ ++ bbr_set_cycle_idx(sk, BBR_BW_PROBE_CRUISE); ++} ++ ++/* Loss and/or ECN rate is too high while probing. ++ * Adapt (once per bw probe) by cutting inflight_hi and then restarting cycle. ++ */ ++static void bbr_handle_inflight_too_high(struct sock *sk, ++ const struct rate_sample *rs) ++{ ++ struct bbr *bbr = inet_csk_ca(sk); ++ const u32 beta = bbr_param(sk, beta); ++ ++ bbr->prev_probe_too_high = 1; ++ bbr->bw_probe_samples = 0; /* only react once per probe */ ++ /* If we are app-limited then we are not robustly ++ * probing the max volume of inflight data we think ++ * might be safe (analogous to how app-limited bw ++ * samples are not known to be robustly probing bw). ++ */ ++ if (!rs->is_app_limited) { ++ bbr->inflight_hi = max_t(u32, rs->tx_in_flight, ++ (u64)bbr_target_inflight(sk) * ++ (BBR_UNIT - beta) >> BBR_SCALE); ++ } ++ if (bbr->mode == BBR_PROBE_BW && bbr->cycle_idx == BBR_BW_PROBE_UP) ++ bbr_start_bw_probe_down(sk); ++} ++ ++/* If we're seeing bw and loss samples reflecting our bw probing, adapt ++ * using the signals we see. If loss or ECN mark rate gets too high, then adapt ++ * inflight_hi downward. If we're able to push inflight higher without such ++ * signals, push higher: adapt inflight_hi upward. ++ */ ++static bool bbr_adapt_upper_bounds(struct sock *sk, ++ const struct rate_sample *rs, ++ struct bbr_context *ctx) ++{ ++ struct bbr *bbr = inet_csk_ca(sk); ++ ++ /* Track when we'll see bw/loss samples resulting from our bw probes. */ ++ if (bbr->ack_phase == BBR_ACKS_PROBE_STARTING && bbr->round_start) ++ bbr->ack_phase = BBR_ACKS_PROBE_FEEDBACK; ++ if (bbr->ack_phase == BBR_ACKS_PROBE_STOPPING && bbr->round_start) { ++ /* End of samples from bw probing phase. */ ++ bbr->bw_probe_samples = 0; ++ bbr->ack_phase = BBR_ACKS_INIT; ++ /* At this point in the cycle, our current bw sample is also ++ * our best recent chance at finding the highest available bw ++ * for this flow. So now is the best time to forget the bw ++ * samples from the previous cycle, by advancing the window. ++ */ ++ if (bbr->mode == BBR_PROBE_BW && !rs->is_app_limited) ++ bbr_advance_max_bw_filter(sk); ++ /* If we had an inflight_hi, then probed and pushed inflight all ++ * the way up to hit that inflight_hi without seeing any ++ * high loss/ECN in all the resulting ACKs from that probing, ++ * then probe up again, this time letting inflight persist at ++ * inflight_hi for a round trip, then accelerating beyond. ++ */ ++ if (bbr->mode == BBR_PROBE_BW && ++ bbr->stopped_risky_probe && !bbr->prev_probe_too_high) { ++ bbr_start_bw_probe_refill(sk, 0); ++ return true; /* yes, decided state transition */ ++ } ++ } ++ if (bbr_is_inflight_too_high(sk, rs)) { ++ if (bbr->bw_probe_samples) /* sample is from bw probing? */ ++ bbr_handle_inflight_too_high(sk, rs); ++ } else { ++ /* Loss/ECN rate is declared safe. Adjust upper bound upward. */ ++ ++ if (bbr->inflight_hi == ~0U) ++ return false; /* no excess queue signals yet */ ++ ++ /* To be resilient to random loss, we must raise bw/inflight_hi ++ * if we observe in any phase that a higher level is safe. ++ */ ++ if (rs->tx_in_flight > bbr->inflight_hi) { ++ bbr->inflight_hi = rs->tx_in_flight; ++ } ++ ++ if (bbr->mode == BBR_PROBE_BW && ++ bbr->cycle_idx == BBR_BW_PROBE_UP) ++ bbr_probe_inflight_hi_upward(sk, rs); ++ } ++ ++ return false; ++} ++ ++/* Check if it's time to probe for bandwidth now, and if so, kick it off. */ ++static bool bbr_check_time_to_probe_bw(struct sock *sk, ++ const struct rate_sample *rs) ++{ ++ struct bbr *bbr = inet_csk_ca(sk); ++ u32 n; ++ ++ /* If we seem to be at an operating point where we are not seeing loss ++ * but we are seeing ECN marks, then when the ECN marks cease we reprobe ++ * quickly (in case cross-traffic has ceased and freed up bw). ++ */ ++ if (bbr_param(sk, ecn_reprobe_gain) && bbr->ecn_eligible && ++ bbr->ecn_in_cycle && !bbr->loss_in_cycle && ++ inet_csk(sk)->icsk_ca_state == TCP_CA_Open) { ++ /* Calculate n so that when bbr_raise_inflight_hi_slope() ++ * computes growth_this_round as 2^n it will be roughly the ++ * desired volume of data (inflight_hi*ecn_reprobe_gain). ++ */ ++ n = ilog2((((u64)bbr->inflight_hi * ++ bbr_param(sk, ecn_reprobe_gain)) >> BBR_SCALE)); ++ bbr_start_bw_probe_refill(sk, n); ++ return true; ++ } ++ ++ if (bbr_has_elapsed_in_phase(sk, bbr->probe_wait_us) || ++ bbr_is_reno_coexistence_probe_time(sk)) { ++ bbr_start_bw_probe_refill(sk, 0); ++ return true; ++ } ++ return false; ++} ++ ++/* Is it time to transition from PROBE_DOWN to PROBE_CRUISE? */ ++static bool bbr_check_time_to_cruise(struct sock *sk, u32 inflight, u32 bw) ++{ ++ /* Always need to pull inflight down to leave headroom in queue. */ ++ if (inflight > bbr_inflight_with_headroom(sk)) ++ return false; ++ ++ return inflight <= bbr_inflight(sk, bw, BBR_UNIT); ++} ++ ++/* PROBE_BW state machine: cruise, refill, probe for bw, or drain? */ ++static void bbr_update_cycle_phase(struct sock *sk, ++ const struct rate_sample *rs, ++ struct bbr_context *ctx) ++{ ++ struct tcp_sock *tp = tcp_sk(sk); ++ struct bbr *bbr = inet_csk_ca(sk); ++ bool is_bw_probe_done = false; ++ u32 inflight, bw; ++ ++ if (!bbr_full_bw_reached(sk)) ++ return; ++ ++ /* In DRAIN, PROBE_BW, or PROBE_RTT, adjust upper bounds. */ ++ if (bbr_adapt_upper_bounds(sk, rs, ctx)) ++ return; /* already decided state transition */ ++ ++ if (bbr->mode != BBR_PROBE_BW) ++ return; ++ ++ inflight = bbr_packets_in_net_at_edt(sk, rs->prior_in_flight); ++ bw = bbr_max_bw(sk); ++ ++ switch (bbr->cycle_idx) { ++ /* First we spend most of our time cruising with a pacing_gain of 1.0, ++ * which paces at the estimated bw, to try to fully use the pipe ++ * without building queue. If we encounter loss/ECN marks, we adapt ++ * by slowing down. ++ */ ++ case BBR_BW_PROBE_CRUISE: ++ if (bbr_check_time_to_probe_bw(sk, rs)) ++ return; /* already decided state transition */ ++ break; ++ ++ /* After cruising, when it's time to probe, we first "refill": we send ++ * at the estimated bw to fill the pipe, before probing higher and ++ * knowingly risking overflowing the bottleneck buffer (causing loss). ++ */ ++ case BBR_BW_PROBE_REFILL: ++ if (bbr->round_start) { ++ /* After one full round trip of sending in REFILL, we ++ * start to see bw samples reflecting our REFILL, which ++ * may be putting too much data in flight. ++ */ ++ bbr->bw_probe_samples = 1; ++ bbr_start_bw_probe_up(sk, ctx); ++ } ++ break; ++ ++ /* After we refill the pipe, we probe by using a pacing_gain > 1.0, to ++ * probe for bw. If we have not seen loss/ECN, we try to raise inflight ++ * to at least pacing_gain*BDP; note that this may take more than ++ * min_rtt if min_rtt is small (e.g. on a LAN). ++ * ++ * We terminate PROBE_UP bandwidth probing upon any of the following: ++ * ++ * (1) We've pushed inflight up to hit the inflight_hi target set in the ++ * most recent previous bw probe phase. Thus we want to start ++ * draining the queue immediately because it's very likely the most ++ * recently sent packets will fill the queue and cause drops. ++ * (2) If inflight_hi has not limited bandwidth growth recently, and ++ * yet delivered bandwidth has not increased much recently ++ * (bbr->full_bw_now). ++ * (3) Loss filter says loss rate is "too high". ++ * (4) ECN filter says ECN mark rate is "too high". ++ * ++ * (1) (2) checked here, (3) (4) checked in bbr_is_inflight_too_high() ++ */ ++ case BBR_BW_PROBE_UP: ++ if (bbr->prev_probe_too_high && ++ inflight >= bbr->inflight_hi) { ++ bbr->stopped_risky_probe = 1; ++ is_bw_probe_done = true; ++ } else { ++ if (tp->is_cwnd_limited && ++ tcp_snd_cwnd(tp) >= bbr->inflight_hi) { ++ /* inflight_hi is limiting bw growth */ ++ bbr_reset_full_bw(sk); ++ bbr->full_bw = ctx->sample_bw; ++ } else if (bbr->full_bw_now) { ++ /* Plateau in estimated bw. Pipe looks full. */ ++ is_bw_probe_done = true; ++ } ++ } ++ if (is_bw_probe_done) { ++ bbr->prev_probe_too_high = 0; /* no loss/ECN (yet) */ ++ bbr_start_bw_probe_down(sk); /* restart w/ down */ ++ } ++ break; ++ ++ /* After probing in PROBE_UP, we have usually accumulated some data in ++ * the bottleneck buffer (if bw probing didn't find more bw). We next ++ * enter PROBE_DOWN to try to drain any excess data from the queue. To ++ * do this, we use a pacing_gain < 1.0. We hold this pacing gain until ++ * our inflight is less then that target cruising point, which is the ++ * minimum of (a) the amount needed to leave headroom, and (b) the ++ * estimated BDP. Once inflight falls to match the target, we estimate ++ * the queue is drained; persisting would underutilize the pipe. ++ */ ++ case BBR_BW_PROBE_DOWN: ++ if (bbr_check_time_to_probe_bw(sk, rs)) ++ return; /* already decided state transition */ ++ if (bbr_check_time_to_cruise(sk, inflight, bw)) ++ bbr_start_bw_probe_cruise(sk); ++ break; ++ ++ default: ++ WARN_ONCE(1, "BBR invalid cycle index %u\n", bbr->cycle_idx); ++ } ++} ++ ++/* Exiting PROBE_RTT, so return to bandwidth probing in STARTUP or PROBE_BW. */ ++static void bbr_exit_probe_rtt(struct sock *sk) ++{ ++ struct bbr *bbr = inet_csk_ca(sk); ++ ++ bbr_reset_lower_bounds(sk); ++ if (bbr_full_bw_reached(sk)) { ++ bbr->mode = BBR_PROBE_BW; ++ /* Raising inflight after PROBE_RTT may cause loss, so reset ++ * the PROBE_BW clock and schedule the next bandwidth probe for ++ * a friendly and randomized future point in time. ++ */ ++ bbr_start_bw_probe_down(sk); ++ /* Since we are exiting PROBE_RTT, we know inflight is ++ * below our estimated BDP, so it is reasonable to cruise. ++ */ ++ bbr_start_bw_probe_cruise(sk); ++ } else { ++ bbr->mode = BBR_STARTUP; ++ } ++} ++ ++/* Exit STARTUP based on loss rate > 1% and loss gaps in round >= N. Wait until ++ * the end of the round in recovery to get a good estimate of how many packets ++ * have been lost, and how many we need to drain with a low pacing rate. ++ */ ++static void bbr_check_loss_too_high_in_startup(struct sock *sk, ++ const struct rate_sample *rs) ++{ ++ struct bbr *bbr = inet_csk_ca(sk); ++ ++ if (bbr_full_bw_reached(sk)) ++ return; ++ ++ /* For STARTUP exit, check the loss rate at the end of each round trip ++ * of Recovery episodes in STARTUP. We check the loss rate at the end ++ * of the round trip to filter out noisy/low loss and have a better ++ * sense of inflight (extent of loss), so we can drain more accurately. ++ */ ++ if (rs->losses && bbr->loss_events_in_round < 0xf) ++ bbr->loss_events_in_round++; /* update saturating counter */ ++ if (bbr_param(sk, full_loss_cnt) && bbr->loss_round_start && ++ inet_csk(sk)->icsk_ca_state == TCP_CA_Recovery && ++ bbr->loss_events_in_round >= bbr_param(sk, full_loss_cnt) && ++ bbr_is_inflight_too_high(sk, rs)) { ++ bbr_handle_queue_too_high_in_startup(sk); ++ return; ++ } ++ if (bbr->loss_round_start) ++ bbr->loss_events_in_round = 0; ++} ++ ++/* Estimate when the pipe is full, using the change in delivery rate: BBR ++ * estimates bw probing filled the pipe if the estimated bw hasn't changed by ++ * at least bbr_full_bw_thresh (25%) after bbr_full_bw_cnt (3) non-app-limited ++ * rounds. Why 3 rounds: 1: rwin autotuning grows the rwin, 2: we fill the ++ * higher rwin, 3: we get higher delivery rate samples. Or transient ++ * cross-traffic or radio noise can go away. CUBIC Hystart shares a similar ++ * design goal, but uses delay and inter-ACK spacing instead of bandwidth. ++ */ ++static void bbr_check_full_bw_reached(struct sock *sk, ++ const struct rate_sample *rs, ++ struct bbr_context *ctx) ++{ ++ struct bbr *bbr = inet_csk_ca(sk); ++ u32 bw_thresh, full_cnt, thresh; ++ ++ if (bbr->full_bw_now || rs->is_app_limited) ++ return; ++ ++ thresh = bbr_param(sk, full_bw_thresh); ++ full_cnt = bbr_param(sk, full_bw_cnt); ++ bw_thresh = (u64)bbr->full_bw * thresh >> BBR_SCALE; ++ if (ctx->sample_bw >= bw_thresh) { ++ bbr_reset_full_bw(sk); ++ bbr->full_bw = ctx->sample_bw; ++ return; ++ } ++ if (!bbr->round_start) ++ return; ++ ++bbr->full_bw_cnt; ++ bbr->full_bw_now = bbr->full_bw_cnt >= full_cnt; ++ bbr->full_bw_reached |= bbr->full_bw_now; ++} ++ ++/* If pipe is probably full, drain the queue and then enter steady-state. */ ++static void bbr_check_drain(struct sock *sk, const struct rate_sample *rs, ++ struct bbr_context *ctx) ++{ ++ struct bbr *bbr = inet_csk_ca(sk); ++ ++ if (bbr->mode == BBR_STARTUP && bbr_full_bw_reached(sk)) { ++ bbr->mode = BBR_DRAIN; /* drain queue we created */ ++ /* Set ssthresh to export purely for monitoring, to signal ++ * completion of initial STARTUP by setting to a non- ++ * TCP_INFINITE_SSTHRESH value (ssthresh is not used by BBR). ++ */ ++ tcp_sk(sk)->snd_ssthresh = ++ bbr_inflight(sk, bbr_max_bw(sk), BBR_UNIT); ++ bbr_reset_congestion_signals(sk); ++ } /* fall through to check if in-flight is already small: */ ++ if (bbr->mode == BBR_DRAIN && ++ bbr_packets_in_net_at_edt(sk, tcp_packets_in_flight(tcp_sk(sk))) <= ++ bbr_inflight(sk, bbr_max_bw(sk), BBR_UNIT)) { ++ bbr->mode = BBR_PROBE_BW; ++ bbr_start_bw_probe_down(sk); ++ } ++} ++ ++static void bbr_update_model(struct sock *sk, const struct rate_sample *rs, ++ struct bbr_context *ctx) ++{ ++ bbr_update_congestion_signals(sk, rs, ctx); ++ bbr_update_ack_aggregation(sk, rs); ++ bbr_check_loss_too_high_in_startup(sk, rs); ++ bbr_check_full_bw_reached(sk, rs, ctx); ++ bbr_check_drain(sk, rs, ctx); ++ bbr_update_cycle_phase(sk, rs, ctx); ++ bbr_update_min_rtt(sk, rs); ++} ++ ++/* Fast path for app-limited case. ++ * ++ * On each ack, we execute bbr state machine, which primarily consists of: ++ * 1) update model based on new rate sample, and ++ * 2) update control based on updated model or state change. ++ * ++ * There are certain workload/scenarios, e.g. app-limited case, where ++ * either we can skip updating model or we can skip update of both model ++ * as well as control. This provides signifcant softirq cpu savings for ++ * processing incoming acks. ++ * ++ * In case of app-limited, if there is no congestion (loss/ecn) and ++ * if observed bw sample is less than current estimated bw, then we can ++ * skip some of the computation in bbr state processing: ++ * ++ * - if there is no rtt/mode/phase change: In this case, since all the ++ * parameters of the network model are constant, we can skip model ++ * as well control update. ++ * ++ * - else we can skip rest of the model update. But we still need to ++ * update the control to account for the new rtt/mode/phase. ++ * ++ * Returns whether we can take fast path or not. ++ */ ++static bool bbr_run_fast_path(struct sock *sk, bool *update_model, ++ const struct rate_sample *rs, struct bbr_context *ctx) ++{ ++ struct bbr *bbr = inet_csk_ca(sk); ++ u32 prev_min_rtt_us, prev_mode; ++ ++ if (bbr_param(sk, fast_path) && bbr->try_fast_path && ++ rs->is_app_limited && ctx->sample_bw < bbr_max_bw(sk) && ++ !bbr->loss_in_round && !bbr->ecn_in_round ) { ++ prev_mode = bbr->mode; ++ prev_min_rtt_us = bbr->min_rtt_us; ++ bbr_check_drain(sk, rs, ctx); ++ bbr_update_cycle_phase(sk, rs, ctx); ++ bbr_update_min_rtt(sk, rs); ++ ++ if (bbr->mode == prev_mode && ++ bbr->min_rtt_us == prev_min_rtt_us && ++ bbr->try_fast_path) { ++ return true; ++ } ++ ++ /* Skip model update, but control still needs to be updated */ ++ *update_model = false; ++ } ++ return false; ++} ++ ++__bpf_kfunc static void bbr_main(struct sock *sk, const struct rate_sample *rs) ++{ ++ struct tcp_sock *tp = tcp_sk(sk); ++ struct bbr *bbr = inet_csk_ca(sk); ++ struct bbr_context ctx = { 0 }; ++ bool update_model = true; ++ u32 bw, round_delivered; ++ int ce_ratio = -1; ++ ++ round_delivered = bbr_update_round_start(sk, rs, &ctx); ++ if (bbr->round_start) { ++ bbr->rounds_since_probe = ++ min_t(s32, bbr->rounds_since_probe + 1, 0xFF); ++ ce_ratio = bbr_update_ecn_alpha(sk); ++ } ++ bbr_plb(sk, rs, ce_ratio); ++ ++ bbr->ecn_in_round |= (bbr->ecn_eligible && rs->is_ece); ++ bbr_calculate_bw_sample(sk, rs, &ctx); ++ bbr_update_latest_delivery_signals(sk, rs, &ctx); ++ ++ if (bbr_run_fast_path(sk, &update_model, rs, &ctx)) ++ goto out; ++ ++ if (update_model) ++ bbr_update_model(sk, rs, &ctx); ++ ++ bbr_update_gains(sk); ++ bw = bbr_bw(sk); ++ bbr_set_pacing_rate(sk, bw, bbr->pacing_gain); ++ bbr_set_cwnd(sk, rs, rs->acked_sacked, bw, bbr->cwnd_gain, ++ tcp_snd_cwnd(tp), &ctx); ++ bbr_bound_cwnd_for_inflight_model(sk); ++ ++out: ++ bbr_advance_latest_delivery_signals(sk, rs, &ctx); ++ bbr->prev_ca_state = inet_csk(sk)->icsk_ca_state; ++ bbr->loss_in_cycle |= rs->lost > 0; ++ bbr->ecn_in_cycle |= rs->delivered_ce > 0; ++} ++ ++__bpf_kfunc static void bbr_init(struct sock *sk) ++{ ++ struct tcp_sock *tp = tcp_sk(sk); ++ struct bbr *bbr = inet_csk_ca(sk); ++ ++ bbr->initialized = 1; ++ ++ bbr->init_cwnd = min(0x7FU, tcp_snd_cwnd(tp)); ++ bbr->prior_cwnd = tp->prior_cwnd; ++ tp->snd_ssthresh = TCP_INFINITE_SSTHRESH; ++ bbr->next_rtt_delivered = tp->delivered; ++ bbr->prev_ca_state = TCP_CA_Open; ++ ++ bbr->probe_rtt_done_stamp = 0; ++ bbr->probe_rtt_round_done = 0; ++ bbr->probe_rtt_min_us = tcp_min_rtt(tp); ++ bbr->probe_rtt_min_stamp = tcp_jiffies32; ++ bbr->min_rtt_us = tcp_min_rtt(tp); ++ bbr->min_rtt_stamp = tcp_jiffies32; ++ ++ bbr->has_seen_rtt = 0; ++ bbr_init_pacing_rate_from_rtt(sk); ++ ++ bbr->round_start = 0; ++ bbr->idle_restart = 0; ++ bbr->full_bw_reached = 0; ++ bbr->full_bw = 0; + bbr->full_bw_cnt = 0; +- bbr_reset_lt_bw_sampling(sk); +- return tcp_snd_cwnd(tcp_sk(sk)); ++ bbr->cycle_mstamp = 0; ++ bbr->cycle_idx = 0; ++ ++ bbr_reset_startup_mode(sk); ++ ++ bbr->ack_epoch_mstamp = tp->tcp_mstamp; ++ bbr->ack_epoch_acked = 0; ++ bbr->extra_acked_win_rtts = 0; ++ bbr->extra_acked_win_idx = 0; ++ bbr->extra_acked[0] = 0; ++ bbr->extra_acked[1] = 0; ++ ++ bbr->ce_state = 0; ++ bbr->prior_rcv_nxt = tp->rcv_nxt; ++ bbr->try_fast_path = 0; ++ ++ cmpxchg(&sk->sk_pacing_status, SK_PACING_NONE, SK_PACING_NEEDED); ++ ++ /* Start sampling ECN mark rate after first full flight is ACKed: */ ++ bbr->loss_round_delivered = tp->delivered + 1; ++ bbr->loss_round_start = 0; ++ bbr->undo_bw_lo = 0; ++ bbr->undo_inflight_lo = 0; ++ bbr->undo_inflight_hi = 0; ++ bbr->loss_events_in_round = 0; ++ bbr->startup_ecn_rounds = 0; ++ bbr_reset_congestion_signals(sk); ++ bbr->bw_lo = ~0U; ++ bbr->bw_hi[0] = 0; ++ bbr->bw_hi[1] = 0; ++ bbr->inflight_lo = ~0U; ++ bbr->inflight_hi = ~0U; ++ bbr_reset_full_bw(sk); ++ bbr->bw_probe_up_cnt = ~0U; ++ bbr->bw_probe_up_acks = 0; ++ bbr->bw_probe_up_rounds = 0; ++ bbr->probe_wait_us = 0; ++ bbr->stopped_risky_probe = 0; ++ bbr->ack_phase = BBR_ACKS_INIT; ++ bbr->rounds_since_probe = 0; ++ bbr->bw_probe_samples = 0; ++ bbr->prev_probe_too_high = 0; ++ bbr->ecn_eligible = 0; ++ bbr->ecn_alpha = bbr_param(sk, ecn_alpha_init); ++ bbr->alpha_last_delivered = 0; ++ bbr->alpha_last_delivered_ce = 0; ++ bbr->plb.pause_until = 0; ++ ++ tp->fast_ack_mode = bbr_fast_ack_mode ? 1 : 0; ++ ++ if (bbr_can_use_ecn(sk)) ++ tp->ecn_flags |= TCP_ECN_ECT_PERMANENT; ++} ++ ++/* BBR marks the current round trip as a loss round. */ ++static void bbr_note_loss(struct sock *sk) ++{ ++ struct tcp_sock *tp = tcp_sk(sk); ++ struct bbr *bbr = inet_csk_ca(sk); ++ ++ /* Capture "current" data over the full round trip of loss, to ++ * have a better chance of observing the full capacity of the path. ++ */ ++ if (!bbr->loss_in_round) /* first loss in this round trip? */ ++ bbr->loss_round_delivered = tp->delivered; /* set round trip */ ++ bbr->loss_in_round = 1; ++ bbr->loss_in_cycle = 1; + } + +-/* Entering loss recovery, so save cwnd for when we exit or undo recovery. */ ++/* Core TCP stack informs us that the given skb was just marked lost. */ ++__bpf_kfunc static void bbr_skb_marked_lost(struct sock *sk, ++ const struct sk_buff *skb) ++{ ++ struct tcp_sock *tp = tcp_sk(sk); ++ struct bbr *bbr = inet_csk_ca(sk); ++ struct tcp_skb_cb *scb = TCP_SKB_CB(skb); ++ struct rate_sample rs = {}; ++ ++ bbr_note_loss(sk); ++ ++ if (!bbr->bw_probe_samples) ++ return; /* not an skb sent while probing for bandwidth */ ++ if (unlikely(!scb->tx.delivered_mstamp)) ++ return; /* skb was SACKed, reneged, marked lost; ignore it */ ++ /* We are probing for bandwidth. Construct a rate sample that ++ * estimates what happened in the flight leading up to this lost skb, ++ * then see if the loss rate went too high, and if so at which packet. ++ */ ++ rs.tx_in_flight = scb->tx.in_flight; ++ rs.lost = tp->lost - scb->tx.lost; ++ rs.is_app_limited = scb->tx.is_app_limited; ++ if (bbr_is_inflight_too_high(sk, &rs)) { ++ rs.tx_in_flight = bbr_inflight_hi_from_lost_skb(sk, &rs, skb); ++ bbr_handle_inflight_too_high(sk, &rs); ++ } ++} ++ ++static void bbr_run_loss_probe_recovery(struct sock *sk) ++{ ++ struct tcp_sock *tp = tcp_sk(sk); ++ struct bbr *bbr = inet_csk_ca(sk); ++ struct rate_sample rs = {0}; ++ ++ bbr_note_loss(sk); ++ ++ if (!bbr->bw_probe_samples) ++ return; /* not sent while probing for bandwidth */ ++ /* We are probing for bandwidth. Construct a rate sample that ++ * estimates what happened in the flight leading up to this ++ * loss, then see if the loss rate went too high. ++ */ ++ rs.lost = 1; /* TLP probe repaired loss of a single segment */ ++ rs.tx_in_flight = bbr->inflight_latest + rs.lost; ++ rs.is_app_limited = tp->tlp_orig_data_app_limited; ++ if (bbr_is_inflight_too_high(sk, &rs)) ++ bbr_handle_inflight_too_high(sk, &rs); ++} ++ ++/* Revert short-term model if current loss recovery event was spurious. */ ++__bpf_kfunc static u32 bbr_undo_cwnd(struct sock *sk) ++{ ++ struct bbr *bbr = inet_csk_ca(sk); ++ ++ bbr_reset_full_bw(sk); /* spurious slow-down; reset full bw detector */ ++ bbr->loss_in_round = 0; ++ ++ /* Revert to cwnd and other state saved before loss episode. */ ++ bbr->bw_lo = max(bbr->bw_lo, bbr->undo_bw_lo); ++ bbr->inflight_lo = max(bbr->inflight_lo, bbr->undo_inflight_lo); ++ bbr->inflight_hi = max(bbr->inflight_hi, bbr->undo_inflight_hi); ++ bbr->try_fast_path = 0; /* take slow path to set proper cwnd, pacing */ ++ return bbr->prior_cwnd; ++} ++ ++/* Entering loss recovery, so save state for when we undo recovery. */ + __bpf_kfunc static u32 bbr_ssthresh(struct sock *sk) + { ++ struct bbr *bbr = inet_csk_ca(sk); ++ + bbr_save_cwnd(sk); ++ /* For undo, save state that adapts based on loss signal. */ ++ bbr->undo_bw_lo = bbr->bw_lo; ++ bbr->undo_inflight_lo = bbr->inflight_lo; ++ bbr->undo_inflight_hi = bbr->inflight_hi; + return tcp_sk(sk)->snd_ssthresh; + } + ++static enum tcp_bbr_phase bbr_get_phase(struct bbr *bbr) ++{ ++ switch (bbr->mode) { ++ case BBR_STARTUP: ++ return BBR_PHASE_STARTUP; ++ case BBR_DRAIN: ++ return BBR_PHASE_DRAIN; ++ case BBR_PROBE_BW: ++ break; ++ case BBR_PROBE_RTT: ++ return BBR_PHASE_PROBE_RTT; ++ default: ++ return BBR_PHASE_INVALID; ++ } ++ switch (bbr->cycle_idx) { ++ case BBR_BW_PROBE_UP: ++ return BBR_PHASE_PROBE_BW_UP; ++ case BBR_BW_PROBE_DOWN: ++ return BBR_PHASE_PROBE_BW_DOWN; ++ case BBR_BW_PROBE_CRUISE: ++ return BBR_PHASE_PROBE_BW_CRUISE; ++ case BBR_BW_PROBE_REFILL: ++ return BBR_PHASE_PROBE_BW_REFILL; ++ default: ++ return BBR_PHASE_INVALID; ++ } ++} ++ + static size_t bbr_get_info(struct sock *sk, u32 ext, int *attr, +- union tcp_cc_info *info) ++ union tcp_cc_info *info) + { + if (ext & (1 << (INET_DIAG_BBRINFO - 1)) || + ext & (1 << (INET_DIAG_VEGASINFO - 1))) { +- struct tcp_sock *tp = tcp_sk(sk); + struct bbr *bbr = inet_csk_ca(sk); +- u64 bw = bbr_bw(sk); +- +- bw = bw * tp->mss_cache * USEC_PER_SEC >> BW_SCALE; +- memset(&info->bbr, 0, sizeof(info->bbr)); +- info->bbr.bbr_bw_lo = (u32)bw; +- info->bbr.bbr_bw_hi = (u32)(bw >> 32); +- info->bbr.bbr_min_rtt = bbr->min_rtt_us; +- info->bbr.bbr_pacing_gain = bbr->pacing_gain; +- info->bbr.bbr_cwnd_gain = bbr->cwnd_gain; ++ u64 bw = bbr_bw_bytes_per_sec(sk, bbr_bw(sk)); ++ u64 bw_hi = bbr_bw_bytes_per_sec(sk, bbr_max_bw(sk)); ++ u64 bw_lo = bbr->bw_lo == ~0U ? ++ ~0ULL : bbr_bw_bytes_per_sec(sk, bbr->bw_lo); ++ struct tcp_bbr_info *bbr_info = &info->bbr; ++ ++ memset(bbr_info, 0, sizeof(*bbr_info)); ++ bbr_info->bbr_bw_lo = (u32)bw; ++ bbr_info->bbr_bw_hi = (u32)(bw >> 32); ++ bbr_info->bbr_min_rtt = bbr->min_rtt_us; ++ bbr_info->bbr_pacing_gain = bbr->pacing_gain; ++ bbr_info->bbr_cwnd_gain = bbr->cwnd_gain; ++ bbr_info->bbr_bw_hi_lsb = (u32)bw_hi; ++ bbr_info->bbr_bw_hi_msb = (u32)(bw_hi >> 32); ++ bbr_info->bbr_bw_lo_lsb = (u32)bw_lo; ++ bbr_info->bbr_bw_lo_msb = (u32)(bw_lo >> 32); ++ bbr_info->bbr_mode = bbr->mode; ++ bbr_info->bbr_phase = (__u8)bbr_get_phase(bbr); ++ bbr_info->bbr_version = (__u8)BBR_VERSION; ++ bbr_info->bbr_inflight_lo = bbr->inflight_lo; ++ bbr_info->bbr_inflight_hi = bbr->inflight_hi; ++ bbr_info->bbr_extra_acked = bbr_extra_acked(sk); + *attr = INET_DIAG_BBRINFO; +- return sizeof(info->bbr); ++ return sizeof(*bbr_info); + } + return 0; + } + + __bpf_kfunc static void bbr_set_state(struct sock *sk, u8 new_state) + { ++ struct tcp_sock *tp = tcp_sk(sk); + struct bbr *bbr = inet_csk_ca(sk); + + if (new_state == TCP_CA_Loss) { +- struct rate_sample rs = { .losses = 1 }; + + bbr->prev_ca_state = TCP_CA_Loss; +- bbr->full_bw = 0; +- bbr->round_start = 1; /* treat RTO like end of a round */ +- bbr_lt_bw_sampling(sk, &rs); ++ tcp_plb_update_state_upon_rto(sk, &bbr->plb); ++ /* The tcp_write_timeout() call to sk_rethink_txhash() likely ++ * repathed this flow, so re-learn the min network RTT on the ++ * new path: ++ */ ++ bbr_reset_full_bw(sk); ++ if (!bbr_is_probing_bandwidth(sk) && bbr->inflight_lo == ~0U) { ++ /* bbr_adapt_lower_bounds() needs cwnd before ++ * we suffered an RTO, to update inflight_lo: ++ */ ++ bbr->inflight_lo = ++ max(tcp_snd_cwnd(tp), bbr->prior_cwnd); ++ } ++ } else if (bbr->prev_ca_state == TCP_CA_Loss && ++ new_state != TCP_CA_Loss) { ++ bbr_exit_loss_recovery(sk); + } + } + ++ + static struct tcp_congestion_ops tcp_bbr_cong_ops __read_mostly = { +- .flags = TCP_CONG_NON_RESTRICTED, ++ .flags = TCP_CONG_NON_RESTRICTED | TCP_CONG_WANTS_CE_EVENTS, + .name = "bbr", + .owner = THIS_MODULE, + .init = bbr_init, + .cong_control = bbr_main, + .sndbuf_expand = bbr_sndbuf_expand, ++ .skb_marked_lost = bbr_skb_marked_lost, + .undo_cwnd = bbr_undo_cwnd, + .cwnd_event = bbr_cwnd_event, + .ssthresh = bbr_ssthresh, +- .min_tso_segs = bbr_min_tso_segs, ++ .tso_segs = bbr_tso_segs, + .get_info = bbr_get_info, + .set_state = bbr_set_state, + }; +@@ -1161,10 +2361,11 @@ BTF_KFUNCS_START(tcp_bbr_check_kfunc_ids) + BTF_ID_FLAGS(func, bbr_init) + BTF_ID_FLAGS(func, bbr_main) + BTF_ID_FLAGS(func, bbr_sndbuf_expand) ++BTF_ID_FLAGS(func, bbr_skb_marked_lost) + BTF_ID_FLAGS(func, bbr_undo_cwnd) + BTF_ID_FLAGS(func, bbr_cwnd_event) + BTF_ID_FLAGS(func, bbr_ssthresh) +-BTF_ID_FLAGS(func, bbr_min_tso_segs) ++BTF_ID_FLAGS(func, bbr_tso_segs) + BTF_ID_FLAGS(func, bbr_set_state) + #endif + #endif +@@ -1199,5 +2400,12 @@ MODULE_AUTHOR("Van Jacobson <vanj@google.com>"); + MODULE_AUTHOR("Neal Cardwell <ncardwell@google.com>"); + MODULE_AUTHOR("Yuchung Cheng <ycheng@google.com>"); + MODULE_AUTHOR("Soheil Hassas Yeganeh <soheil@google.com>"); ++MODULE_AUTHOR("Priyaranjan Jha <priyarjha@google.com>"); ++MODULE_AUTHOR("Yousuk Seung <ysseung@google.com>"); ++MODULE_AUTHOR("Kevin Yang <yyd@google.com>"); ++MODULE_AUTHOR("Arjun Roy <arjunroy@google.com>"); ++MODULE_AUTHOR("David Morley <morleyd@google.com>"); ++ + MODULE_LICENSE("Dual BSD/GPL"); + MODULE_DESCRIPTION("TCP BBR (Bottleneck Bandwidth and RTT)"); ++MODULE_VERSION(__stringify(BBR_VERSION)); +diff --git a/net/ipv4/tcp_cong.c b/net/ipv4/tcp_cong.c +index 28ffcfbeef14..7b13915ba288 100644 +--- a/net/ipv4/tcp_cong.c ++++ b/net/ipv4/tcp_cong.c +@@ -237,6 +237,7 @@ void tcp_init_congestion_control(struct sock *sk) + struct inet_connection_sock *icsk = inet_csk(sk); + + tcp_sk(sk)->prior_ssthresh = 0; ++ tcp_sk(sk)->fast_ack_mode = 0; + if (icsk->icsk_ca_ops->init) + icsk->icsk_ca_ops->init(sk); + if (tcp_ca_needs_ecn(sk)) +diff --git a/net/ipv4/tcp_input.c b/net/ipv4/tcp_input.c +index 7b692bcb61d4..650db0ec7222 100644 +--- a/net/ipv4/tcp_input.c ++++ b/net/ipv4/tcp_input.c +@@ -364,7 +364,7 @@ static void __tcp_ecn_check_ce(struct sock *sk, const struct sk_buff *skb) + tcp_enter_quickack_mode(sk, 2); + break; + case INET_ECN_CE: +- if (tcp_ca_needs_ecn(sk)) ++ if (tcp_ca_wants_ce_events(sk)) + tcp_ca_event(sk, CA_EVENT_ECN_IS_CE); + + if (!(tp->ecn_flags & TCP_ECN_DEMAND_CWR)) { +@@ -375,7 +375,7 @@ static void __tcp_ecn_check_ce(struct sock *sk, const struct sk_buff *skb) + tp->ecn_flags |= TCP_ECN_SEEN; + break; + default: +- if (tcp_ca_needs_ecn(sk)) ++ if (tcp_ca_wants_ce_events(sk)) + tcp_ca_event(sk, CA_EVENT_ECN_NO_CE); + tp->ecn_flags |= TCP_ECN_SEEN; + break; +@@ -1112,7 +1112,12 @@ static void tcp_verify_retransmit_hint(struct tcp_sock *tp, struct sk_buff *skb) + */ + static void tcp_notify_skb_loss_event(struct tcp_sock *tp, const struct sk_buff *skb) + { ++ struct sock *sk = (struct sock *)tp; ++ const struct tcp_congestion_ops *ca_ops = inet_csk(sk)->icsk_ca_ops; ++ + tp->lost += tcp_skb_pcount(skb); ++ if (ca_ops->skb_marked_lost) ++ ca_ops->skb_marked_lost(sk, skb); + } + + void tcp_mark_skb_lost(struct sock *sk, struct sk_buff *skb) +@@ -1493,6 +1498,17 @@ static bool tcp_shifted_skb(struct sock *sk, struct sk_buff *prev, + WARN_ON_ONCE(tcp_skb_pcount(skb) < pcount); + tcp_skb_pcount_add(skb, -pcount); + ++ /* Adjust tx.in_flight as pcount is shifted from skb to prev. */ ++ if (WARN_ONCE(TCP_SKB_CB(skb)->tx.in_flight < pcount, ++ "prev in_flight: %u skb in_flight: %u pcount: %u", ++ TCP_SKB_CB(prev)->tx.in_flight, ++ TCP_SKB_CB(skb)->tx.in_flight, ++ pcount)) ++ TCP_SKB_CB(skb)->tx.in_flight = 0; ++ else ++ TCP_SKB_CB(skb)->tx.in_flight -= pcount; ++ TCP_SKB_CB(prev)->tx.in_flight += pcount; ++ + /* When we're adding to gso_segs == 1, gso_size will be zero, + * in theory this shouldn't be necessary but as long as DSACK + * code can come after this skb later on it's better to keep +@@ -3778,7 +3794,8 @@ static void tcp_replace_ts_recent(struct tcp_sock *tp, u32 seq) + /* This routine deals with acks during a TLP episode and ends an episode by + * resetting tlp_high_seq. Ref: TLP algorithm in draft-ietf-tcpm-rack + */ +-static void tcp_process_tlp_ack(struct sock *sk, u32 ack, int flag) ++static void tcp_process_tlp_ack(struct sock *sk, u32 ack, int flag, ++ struct rate_sample *rs) + { + struct tcp_sock *tp = tcp_sk(sk); + +@@ -3795,6 +3812,7 @@ static void tcp_process_tlp_ack(struct sock *sk, u32 ack, int flag) + /* ACK advances: there was a loss, so reduce cwnd. Reset + * tlp_high_seq in tcp_init_cwnd_reduction() + */ ++ tcp_ca_event(sk, CA_EVENT_TLP_RECOVERY); + tcp_init_cwnd_reduction(sk); + tcp_set_ca_state(sk, TCP_CA_CWR); + tcp_end_cwnd_reduction(sk); +@@ -3805,6 +3823,11 @@ static void tcp_process_tlp_ack(struct sock *sk, u32 ack, int flag) + FLAG_NOT_DUP | FLAG_DATA_SACKED))) { + /* Pure dupack: original and TLP probe arrived; no loss */ + tp->tlp_high_seq = 0; ++ } else { ++ /* This ACK matches a TLP retransmit. We cannot yet tell if ++ * this ACK is for the original or the TLP retransmit. ++ */ ++ rs->is_acking_tlp_retrans_seq = 1; + } + } + +@@ -3913,6 +3936,7 @@ static int tcp_ack(struct sock *sk, const struct sk_buff *skb, int flag) + + prior_fack = tcp_is_sack(tp) ? tcp_highest_sack_seq(tp) : tp->snd_una; + rs.prior_in_flight = tcp_packets_in_flight(tp); ++ tcp_rate_check_app_limited(sk); + + /* ts_recent update must be made after we are sure that the packet + * is in window. +@@ -3987,7 +4011,7 @@ static int tcp_ack(struct sock *sk, const struct sk_buff *skb, int flag) + tcp_rack_update_reo_wnd(sk, &rs); + + if (tp->tlp_high_seq) +- tcp_process_tlp_ack(sk, ack, flag); ++ tcp_process_tlp_ack(sk, ack, flag, &rs); + + if (tcp_ack_is_dubious(sk, flag)) { + if (!(flag & (FLAG_SND_UNA_ADVANCED | +@@ -4011,6 +4035,7 @@ static int tcp_ack(struct sock *sk, const struct sk_buff *skb, int flag) + delivered = tcp_newly_delivered(sk, delivered, flag); + lost = tp->lost - lost; /* freshly marked lost */ + rs.is_ack_delayed = !!(flag & FLAG_ACK_MAYBE_DELAYED); ++ rs.is_ece = !!(flag & FLAG_ECE); + tcp_rate_gen(sk, delivered, lost, is_sack_reneg, sack_state.rate); + tcp_cong_control(sk, ack, delivered, flag, sack_state.rate); + tcp_xmit_recovery(sk, rexmit); +@@ -4030,7 +4055,7 @@ static int tcp_ack(struct sock *sk, const struct sk_buff *skb, int flag) + tcp_ack_probe(sk); + + if (tp->tlp_high_seq) +- tcp_process_tlp_ack(sk, ack, flag); ++ tcp_process_tlp_ack(sk, ack, flag, &rs); + return 1; + + old_ack: +@@ -5688,13 +5713,14 @@ static void __tcp_ack_snd_check(struct sock *sk, int ofo_possible) + + /* More than one full frame received... */ + if (((tp->rcv_nxt - tp->rcv_wup) > inet_csk(sk)->icsk_ack.rcv_mss && ++ (tp->fast_ack_mode == 1 || + /* ... and right edge of window advances far enough. + * (tcp_recvmsg() will send ACK otherwise). + * If application uses SO_RCVLOWAT, we want send ack now if + * we have not received enough bytes to satisfy the condition. + */ +- (tp->rcv_nxt - tp->copied_seq < sk->sk_rcvlowat || +- __tcp_select_window(sk) >= tp->rcv_wnd)) || ++ (tp->rcv_nxt - tp->copied_seq < sk->sk_rcvlowat || ++ __tcp_select_window(sk) >= tp->rcv_wnd))) || + /* We ACK each frame or... */ + tcp_in_quickack_mode(sk) || + /* Protocol state mandates a one-time immediate ACK */ +diff --git a/net/ipv4/tcp_minisocks.c b/net/ipv4/tcp_minisocks.c +index f0761f060a83..90a9389499e6 100644 +--- a/net/ipv4/tcp_minisocks.c ++++ b/net/ipv4/tcp_minisocks.c +@@ -460,6 +460,8 @@ void tcp_ca_openreq_child(struct sock *sk, const struct dst_entry *dst) + u32 ca_key = dst_metric(dst, RTAX_CC_ALGO); + bool ca_got_dst = false; + ++ tcp_set_ecn_low_from_dst(sk, dst); ++ + if (ca_key != TCP_CA_UNSPEC) { + const struct tcp_congestion_ops *ca; + +diff --git a/net/ipv4/tcp_output.c b/net/ipv4/tcp_output.c +index 02caeb7bcf63..0e445bc5cc60 100644 +--- a/net/ipv4/tcp_output.c ++++ b/net/ipv4/tcp_output.c +@@ -332,10 +332,9 @@ static void tcp_ecn_send_syn(struct sock *sk, struct sk_buff *skb) + bool bpf_needs_ecn = tcp_bpf_ca_needs_ecn(sk); + bool use_ecn = READ_ONCE(sock_net(sk)->ipv4.sysctl_tcp_ecn) == 1 || + tcp_ca_needs_ecn(sk) || bpf_needs_ecn; ++ const struct dst_entry *dst = __sk_dst_get(sk); + + if (!use_ecn) { +- const struct dst_entry *dst = __sk_dst_get(sk); +- + if (dst && dst_feature(dst, RTAX_FEATURE_ECN)) + use_ecn = true; + } +@@ -347,6 +346,9 @@ static void tcp_ecn_send_syn(struct sock *sk, struct sk_buff *skb) + tp->ecn_flags = TCP_ECN_OK; + if (tcp_ca_needs_ecn(sk) || bpf_needs_ecn) + INET_ECN_xmit(sk); ++ ++ if (dst) ++ tcp_set_ecn_low_from_dst(sk, dst); + } + } + +@@ -384,7 +386,8 @@ static void tcp_ecn_send(struct sock *sk, struct sk_buff *skb, + th->cwr = 1; + skb_shinfo(skb)->gso_type |= SKB_GSO_TCP_ECN; + } +- } else if (!tcp_ca_needs_ecn(sk)) { ++ } else if (!(tp->ecn_flags & TCP_ECN_ECT_PERMANENT) && ++ !tcp_ca_needs_ecn(sk)) { + /* ACK or retransmitted segment: clear ECT|CE */ + INET_ECN_dontxmit(sk); + } +@@ -1593,7 +1596,7 @@ int tcp_fragment(struct sock *sk, enum tcp_queue tcp_queue, + { + struct tcp_sock *tp = tcp_sk(sk); + struct sk_buff *buff; +- int old_factor; ++ int old_factor, inflight_prev; + long limit; + int nlen; + u8 flags; +@@ -1668,6 +1671,30 @@ int tcp_fragment(struct sock *sk, enum tcp_queue tcp_queue, + + if (diff) + tcp_adjust_pcount(sk, skb, diff); ++ ++ inflight_prev = TCP_SKB_CB(skb)->tx.in_flight - old_factor; ++ if (inflight_prev < 0) { ++ WARN_ONCE(tcp_skb_tx_in_flight_is_suspicious( ++ old_factor, ++ TCP_SKB_CB(skb)->sacked, ++ TCP_SKB_CB(skb)->tx.in_flight), ++ "inconsistent: tx.in_flight: %u " ++ "old_factor: %d mss: %u sacked: %u " ++ "1st pcount: %d 2nd pcount: %d " ++ "1st len: %u 2nd len: %u ", ++ TCP_SKB_CB(skb)->tx.in_flight, old_factor, ++ mss_now, TCP_SKB_CB(skb)->sacked, ++ tcp_skb_pcount(skb), tcp_skb_pcount(buff), ++ skb->len, buff->len); ++ inflight_prev = 0; ++ } ++ /* Set 1st tx.in_flight as if 1st were sent by itself: */ ++ TCP_SKB_CB(skb)->tx.in_flight = inflight_prev + ++ tcp_skb_pcount(skb); ++ /* Set 2nd tx.in_flight with new 1st and 2nd pcounts: */ ++ TCP_SKB_CB(buff)->tx.in_flight = inflight_prev + ++ tcp_skb_pcount(skb) + ++ tcp_skb_pcount(buff); + } + + /* Link BUFF into the send queue. */ +@@ -2025,13 +2052,12 @@ static u32 tcp_tso_autosize(const struct sock *sk, unsigned int mss_now, + static u32 tcp_tso_segs(struct sock *sk, unsigned int mss_now) + { + const struct tcp_congestion_ops *ca_ops = inet_csk(sk)->icsk_ca_ops; +- u32 min_tso, tso_segs; +- +- min_tso = ca_ops->min_tso_segs ? +- ca_ops->min_tso_segs(sk) : +- READ_ONCE(sock_net(sk)->ipv4.sysctl_tcp_min_tso_segs); ++ u32 tso_segs; + +- tso_segs = tcp_tso_autosize(sk, mss_now, min_tso); ++ tso_segs = ca_ops->tso_segs ? ++ ca_ops->tso_segs(sk, mss_now) : ++ tcp_tso_autosize(sk, mss_now, ++ sock_net(sk)->ipv4.sysctl_tcp_min_tso_segs); + return min_t(u32, tso_segs, sk->sk_gso_max_segs); + } + +@@ -2731,6 +2757,7 @@ static bool tcp_write_xmit(struct sock *sk, unsigned int mss_now, int nonagle, + skb_set_delivery_time(skb, tp->tcp_wstamp_ns, true); + list_move_tail(&skb->tcp_tsorted_anchor, &tp->tsorted_sent_queue); + tcp_init_tso_segs(skb, mss_now); ++ tcp_set_tx_in_flight(sk, skb); + goto repair; /* Skip network transmission */ + } + +@@ -2944,6 +2971,7 @@ void tcp_send_loss_probe(struct sock *sk) + if (WARN_ON(!skb || !tcp_skb_pcount(skb))) + goto rearm_timer; + ++ tp->tlp_orig_data_app_limited = TCP_SKB_CB(skb)->tx.is_app_limited; + if (__tcp_retransmit_skb(sk, skb, 1)) + goto rearm_timer; + +diff --git a/net/ipv4/tcp_rate.c b/net/ipv4/tcp_rate.c +index a8f6d9d06f2e..8737f2134648 100644 +--- a/net/ipv4/tcp_rate.c ++++ b/net/ipv4/tcp_rate.c +@@ -34,6 +34,24 @@ + * ready to send in the write queue. + */ + ++void tcp_set_tx_in_flight(struct sock *sk, struct sk_buff *skb) ++{ ++ struct tcp_sock *tp = tcp_sk(sk); ++ u32 in_flight; ++ ++ /* Check, sanitize, and record packets in flight after skb was sent. */ ++ in_flight = tcp_packets_in_flight(tp) + tcp_skb_pcount(skb); ++ if (WARN_ONCE(in_flight > TCPCB_IN_FLIGHT_MAX, ++ "insane in_flight %u cc %s mss %u " ++ "cwnd %u pif %u %u %u %u\n", ++ in_flight, inet_csk(sk)->icsk_ca_ops->name, ++ tp->mss_cache, tp->snd_cwnd, ++ tp->packets_out, tp->retrans_out, ++ tp->sacked_out, tp->lost_out)) ++ in_flight = TCPCB_IN_FLIGHT_MAX; ++ TCP_SKB_CB(skb)->tx.in_flight = in_flight; ++} ++ + /* Snapshot the current delivery information in the skb, to generate + * a rate sample later when the skb is (s)acked in tcp_rate_skb_delivered(). + */ +@@ -66,7 +84,9 @@ void tcp_rate_skb_sent(struct sock *sk, struct sk_buff *skb) + TCP_SKB_CB(skb)->tx.delivered_mstamp = tp->delivered_mstamp; + TCP_SKB_CB(skb)->tx.delivered = tp->delivered; + TCP_SKB_CB(skb)->tx.delivered_ce = tp->delivered_ce; ++ TCP_SKB_CB(skb)->tx.lost = tp->lost; + TCP_SKB_CB(skb)->tx.is_app_limited = tp->app_limited ? 1 : 0; ++ tcp_set_tx_in_flight(sk, skb); + } + + /* When an skb is sacked or acked, we fill in the rate sample with the (prior) +@@ -91,18 +111,21 @@ void tcp_rate_skb_delivered(struct sock *sk, struct sk_buff *skb, + if (!rs->prior_delivered || + tcp_skb_sent_after(tx_tstamp, tp->first_tx_mstamp, + scb->end_seq, rs->last_end_seq)) { ++ rs->prior_lost = scb->tx.lost; + rs->prior_delivered_ce = scb->tx.delivered_ce; + rs->prior_delivered = scb->tx.delivered; + rs->prior_mstamp = scb->tx.delivered_mstamp; + rs->is_app_limited = scb->tx.is_app_limited; + rs->is_retrans = scb->sacked & TCPCB_RETRANS; ++ rs->tx_in_flight = scb->tx.in_flight; + rs->last_end_seq = scb->end_seq; + + /* Record send time of most recently ACKed packet: */ + tp->first_tx_mstamp = tx_tstamp; + /* Find the duration of the "send phase" of this window: */ +- rs->interval_us = tcp_stamp_us_delta(tp->first_tx_mstamp, +- scb->tx.first_tx_mstamp); ++ rs->interval_us = tcp_stamp32_us_delta( ++ tp->first_tx_mstamp, ++ scb->tx.first_tx_mstamp); + + } + /* Mark off the skb delivered once it's sacked to avoid being +@@ -144,6 +167,7 @@ void tcp_rate_gen(struct sock *sk, u32 delivered, u32 lost, + return; + } + rs->delivered = tp->delivered - rs->prior_delivered; ++ rs->lost = tp->lost - rs->prior_lost; + + rs->delivered_ce = tp->delivered_ce - rs->prior_delivered_ce; + /* delivered_ce occupies less than 32 bits in the skb control block */ +@@ -155,7 +179,7 @@ void tcp_rate_gen(struct sock *sk, u32 delivered, u32 lost, + * longer phase. + */ + snd_us = rs->interval_us; /* send phase */ +- ack_us = tcp_stamp_us_delta(tp->tcp_mstamp, ++ ack_us = tcp_stamp32_us_delta(tp->tcp_mstamp, + rs->prior_mstamp); /* ack phase */ + rs->interval_us = max(snd_us, ack_us); + +diff --git a/net/ipv4/tcp_timer.c b/net/ipv4/tcp_timer.c +index f96f68cf7961..af0ce59fbbc0 100644 +--- a/net/ipv4/tcp_timer.c ++++ b/net/ipv4/tcp_timer.c +@@ -682,6 +682,7 @@ void tcp_write_timer_handler(struct sock *sk) + return; + } + ++ tcp_rate_check_app_limited(sk); + tcp_mstamp_refresh(tcp_sk(sk)); + event = icsk->icsk_pending; + +-- +2.45.2 + diff --git a/SOURCES/cachy-fixes.patch b/SOURCES/cachy-fixes.patch new file mode 100644 index 0000000..1c632df --- /dev/null +++ b/SOURCES/cachy-fixes.patch @@ -0,0 +1,573 @@ +diff --git a/drivers/gpu/drm/amd/display/amdgpu_dm/amdgpu_dm.c b/drivers/gpu/drm/amd/display/amdgpu_dm/amdgpu_dm.c +index 371531b2a3d0..ddc064b4fc32 100644 +--- a/drivers/gpu/drm/amd/display/amdgpu_dm/amdgpu_dm.c ++++ b/drivers/gpu/drm/amd/display/amdgpu_dm/amdgpu_dm.c +@@ -8327,6 +8327,122 @@ static inline uint32_t get_mem_type(struct drm_framebuffer *fb) + return abo->tbo.resource ? abo->tbo.resource->mem_type : 0; + } + ++/* TODO remove duplicate */ ++static int amdgpu_dm_plane_get_cursor_position(struct drm_plane *plane, struct drm_crtc *crtc, ++ struct dc_cursor_position *position) ++{ ++ struct amdgpu_crtc *amdgpu_crtc = to_amdgpu_crtc(crtc); ++ int x, y; ++ int xorigin = 0, yorigin = 0; ++ ++ if (!crtc || !plane->state->fb) ++ return 0; ++ ++ if ((plane->state->crtc_w > amdgpu_crtc->max_cursor_width) || ++ (plane->state->crtc_h > amdgpu_crtc->max_cursor_height)) { ++ DRM_ERROR("%s: bad cursor width or height %d x %d\n", ++ __func__, ++ plane->state->crtc_w, ++ plane->state->crtc_h); ++ return -EINVAL; ++ } ++ ++ x = plane->state->crtc_x; ++ y = plane->state->crtc_y; ++ ++ if (x <= -amdgpu_crtc->max_cursor_width || ++ y <= -amdgpu_crtc->max_cursor_height) ++ return 0; ++ ++ if (x < 0) { ++ xorigin = min(-x, amdgpu_crtc->max_cursor_width - 1); ++ x = 0; ++ } ++ if (y < 0) { ++ yorigin = min(-y, amdgpu_crtc->max_cursor_height - 1); ++ y = 0; ++ } ++ position->enable = true; ++ position->translate_by_source = true; ++ position->x = x; ++ position->y = y; ++ position->x_hotspot = xorigin; ++ position->y_hotspot = yorigin; ++ ++ return 0; ++} ++ ++static void amdgpu_dm_update_cursor(struct drm_plane *plane, ++ struct drm_plane_state *old_plane_state, ++ struct dc_stream_update *update) ++{ ++ struct amdgpu_device *adev = drm_to_adev(plane->dev); ++ struct amdgpu_framebuffer *afb = to_amdgpu_framebuffer(plane->state->fb); ++ struct drm_crtc *crtc = afb ? plane->state->crtc : old_plane_state->crtc; ++ struct dm_crtc_state *crtc_state = crtc ? to_dm_crtc_state(crtc->state) : NULL; ++ struct amdgpu_crtc *amdgpu_crtc = to_amdgpu_crtc(crtc); ++ uint64_t address = afb ? afb->address : 0; ++ struct dc_cursor_position position = {0}; ++ struct dc_cursor_attributes attributes; ++ int ret; ++ ++ if (!plane->state->fb && !old_plane_state->fb) ++ return; ++ ++ drm_dbg_atomic(plane->dev, "crtc_id=%d with size %d to %d\n", ++ amdgpu_crtc->crtc_id, plane->state->crtc_w, ++ plane->state->crtc_h); ++ ++ ret = amdgpu_dm_plane_get_cursor_position(plane, crtc, &position); ++ if (ret) ++ return; ++ ++ if (!position.enable) { ++ /* turn off cursor */ ++ if (crtc_state && crtc_state->stream) { ++ dc_stream_set_cursor_position(crtc_state->stream, ++ &position); ++ update->cursor_position = &crtc_state->stream->cursor_position; ++ } ++ return; ++ } ++ ++ amdgpu_crtc->cursor_width = plane->state->crtc_w; ++ amdgpu_crtc->cursor_height = plane->state->crtc_h; ++ ++ memset(&attributes, 0, sizeof(attributes)); ++ attributes.address.high_part = upper_32_bits(address); ++ attributes.address.low_part = lower_32_bits(address); ++ attributes.width = plane->state->crtc_w; ++ attributes.height = plane->state->crtc_h; ++ attributes.color_format = CURSOR_MODE_COLOR_PRE_MULTIPLIED_ALPHA; ++ attributes.rotation_angle = 0; ++ attributes.attribute_flags.value = 0; ++ ++ /* Enable cursor degamma ROM on DCN3+ for implicit sRGB degamma in DRM ++ * legacy gamma setup. ++ */ ++ if (crtc_state->cm_is_degamma_srgb && ++ adev->dm.dc->caps.color.dpp.gamma_corr) ++ attributes.attribute_flags.bits.ENABLE_CURSOR_DEGAMMA = 1; ++ ++ attributes.pitch = afb->base.pitches[0] / afb->base.format->cpp[0]; ++ ++ if (crtc_state->stream) { ++ if (!dc_stream_set_cursor_attributes(crtc_state->stream, ++ &attributes)) ++ DRM_ERROR("DC failed to set cursor attributes\n"); ++ ++ update->cursor_attributes = &crtc_state->stream->cursor_attributes; ++ ++ if (!dc_stream_set_cursor_position(crtc_state->stream, ++ &position)) ++ DRM_ERROR("DC failed to set cursor position\n"); ++ ++ update->cursor_position = &crtc_state->stream->cursor_position; ++ } ++} ++ + static void amdgpu_dm_commit_planes(struct drm_atomic_state *state, + struct drm_device *dev, + struct amdgpu_display_manager *dm, +@@ -8350,6 +8466,7 @@ static void amdgpu_dm_commit_planes(struct drm_atomic_state *state, + bool cursor_update = false; + bool pflip_present = false; + bool dirty_rects_changed = false; ++ bool updated_planes_and_streams = false; + struct { + struct dc_surface_update surface_updates[MAX_SURFACES]; + struct dc_plane_info plane_infos[MAX_SURFACES]; +@@ -8386,8 +8503,10 @@ static void amdgpu_dm_commit_planes(struct drm_atomic_state *state, + /* Cursor plane is handled after stream updates */ + if (plane->type == DRM_PLANE_TYPE_CURSOR) { + if ((fb && crtc == pcrtc) || +- (old_plane_state->fb && old_plane_state->crtc == pcrtc)) ++ (old_plane_state->fb && old_plane_state->crtc == pcrtc)) { + cursor_update = true; ++ amdgpu_dm_update_cursor(plane, old_plane_state, &bundle->stream_update); ++ } + + continue; + } +@@ -8660,6 +8779,7 @@ static void amdgpu_dm_commit_planes(struct drm_atomic_state *state, + acrtc_state->stream, + &bundle->stream_update, + bundle->surface_updates); ++ updated_planes_and_streams = true; + + /** + * Enable or disable the interrupts on the backend. +@@ -8737,7 +8857,7 @@ static void amdgpu_dm_commit_planes(struct drm_atomic_state *state, + * This avoids redundant programming in the case where we're going + * to be disabling a single plane - those pipes are being disabled. + */ +- if (acrtc_state->active_planes) ++ if (acrtc_state->active_planes && !updated_planes_and_streams) + amdgpu_dm_commit_cursors(state); + + cleanup: +@@ -8925,7 +9045,7 @@ static void amdgpu_dm_commit_streams(struct drm_atomic_state *state, + + memset(&position, 0, sizeof(position)); + mutex_lock(&dm->dc_lock); +- dc_stream_set_cursor_position(dm_old_crtc_state->stream, &position); ++ dc_stream_program_cursor_position(dm_old_crtc_state->stream, &position); + mutex_unlock(&dm->dc_lock); + } + +diff --git a/drivers/gpu/drm/amd/display/amdgpu_dm/amdgpu_dm_plane.c b/drivers/gpu/drm/amd/display/amdgpu_dm/amdgpu_dm_plane.c +index 779880c64575..8c8d0d209d7a 100644 +--- a/drivers/gpu/drm/amd/display/amdgpu_dm/amdgpu_dm_plane.c ++++ b/drivers/gpu/drm/amd/display/amdgpu_dm/amdgpu_dm_plane.c +@@ -1254,7 +1254,7 @@ void amdgpu_dm_plane_handle_cursor_update(struct drm_plane *plane, + /* turn off cursor */ + if (crtc_state && crtc_state->stream) { + mutex_lock(&adev->dm.dc_lock); +- dc_stream_set_cursor_position(crtc_state->stream, ++ dc_stream_program_cursor_position(crtc_state->stream, + &position); + mutex_unlock(&adev->dm.dc_lock); + } +@@ -1284,11 +1284,11 @@ void amdgpu_dm_plane_handle_cursor_update(struct drm_plane *plane, + + if (crtc_state->stream) { + mutex_lock(&adev->dm.dc_lock); +- if (!dc_stream_set_cursor_attributes(crtc_state->stream, ++ if (!dc_stream_program_cursor_attributes(crtc_state->stream, + &attributes)) + DRM_ERROR("DC failed to set cursor attributes\n"); + +- if (!dc_stream_set_cursor_position(crtc_state->stream, ++ if (!dc_stream_program_cursor_position(crtc_state->stream, + &position)) + DRM_ERROR("DC failed to set cursor position\n"); + mutex_unlock(&adev->dm.dc_lock); +diff --git a/drivers/gpu/drm/amd/display/dc/core/dc.c b/drivers/gpu/drm/amd/display/dc/core/dc.c +index d68c83e40d4d..575290784c61 100644 +--- a/drivers/gpu/drm/amd/display/dc/core/dc.c ++++ b/drivers/gpu/drm/amd/display/dc/core/dc.c +@@ -3188,6 +3188,83 @@ static bool update_planes_and_stream_state(struct dc *dc, + + } + ++static void program_cursor_attributes( ++ struct dc *dc, ++ struct dc_stream_state *stream) ++{ ++ int i; ++ struct resource_context *res_ctx; ++ struct pipe_ctx *pipe_to_program = NULL; ++ ++ if (!stream) ++ return; ++ ++ res_ctx = &dc->current_state->res_ctx; ++ ++ for (i = 0; i < MAX_PIPES; i++) { ++ struct pipe_ctx *pipe_ctx = &res_ctx->pipe_ctx[i]; ++ ++ if (pipe_ctx->stream != stream) ++ continue; ++ ++ if (!pipe_to_program) { ++ pipe_to_program = pipe_ctx; ++ dc->hwss.cursor_lock(dc, pipe_to_program, true); ++ if (pipe_to_program->next_odm_pipe) ++ dc->hwss.cursor_lock(dc, pipe_to_program->next_odm_pipe, true); ++ } ++ ++ dc->hwss.set_cursor_attribute(pipe_ctx); ++ if (dc->ctx->dmub_srv) ++ dc_send_update_cursor_info_to_dmu(pipe_ctx, i); ++ if (dc->hwss.set_cursor_sdr_white_level) ++ dc->hwss.set_cursor_sdr_white_level(pipe_ctx); ++ } ++ ++ if (pipe_to_program) { ++ dc->hwss.cursor_lock(dc, pipe_to_program, false); ++ if (pipe_to_program->next_odm_pipe) ++ dc->hwss.cursor_lock(dc, pipe_to_program->next_odm_pipe, false); ++ } ++} ++ ++static void program_cursor_position( ++ struct dc *dc, ++ struct dc_stream_state *stream) ++{ ++ int i; ++ struct resource_context *res_ctx; ++ struct pipe_ctx *pipe_to_program = NULL; ++ ++ if (!stream) ++ return; ++ ++ res_ctx = &dc->current_state->res_ctx; ++ ++ for (i = 0; i < MAX_PIPES; i++) { ++ struct pipe_ctx *pipe_ctx = &res_ctx->pipe_ctx[i]; ++ ++ if (pipe_ctx->stream != stream || ++ (!pipe_ctx->plane_res.mi && !pipe_ctx->plane_res.hubp) || ++ !pipe_ctx->plane_state || ++ (!pipe_ctx->plane_res.xfm && !pipe_ctx->plane_res.dpp) || ++ (!pipe_ctx->plane_res.ipp && !pipe_ctx->plane_res.dpp)) ++ continue; ++ ++ if (!pipe_to_program) { ++ pipe_to_program = pipe_ctx; ++ dc->hwss.cursor_lock(dc, pipe_to_program, true); ++ } ++ ++ dc->hwss.set_cursor_position(pipe_ctx); ++ if (dc->ctx->dmub_srv) ++ dc_send_update_cursor_info_to_dmu(pipe_ctx, i); ++ } ++ ++ if (pipe_to_program) ++ dc->hwss.cursor_lock(dc, pipe_to_program, false); ++} ++ + static void commit_planes_do_stream_update(struct dc *dc, + struct dc_stream_state *stream, + struct dc_stream_update *stream_update, +@@ -3248,6 +3325,13 @@ static void commit_planes_do_stream_update(struct dc *dc, + } + } + ++ if (stream_update->cursor_attributes) { ++ program_cursor_attributes(dc, stream); ++ } ++ ++ if (stream_update->cursor_position) { ++ program_cursor_position(dc, stream); ++ } + + /* Full fe update*/ + if (update_type == UPDATE_TYPE_FAST) +diff --git a/drivers/gpu/drm/amd/display/dc/core/dc_stream.c b/drivers/gpu/drm/amd/display/dc/core/dc_stream.c +index 51a970fcb5d0..5601ee8a8d41 100644 +--- a/drivers/gpu/drm/amd/display/dc/core/dc_stream.c ++++ b/drivers/gpu/drm/amd/display/dc/core/dc_stream.c +@@ -278,7 +278,6 @@ bool dc_stream_set_cursor_attributes( + const struct dc_cursor_attributes *attributes) + { + struct dc *dc; +- bool reset_idle_optimizations = false; + + if (NULL == stream) { + dm_error("DC: dc_stream is NULL!\n"); +@@ -309,20 +308,36 @@ bool dc_stream_set_cursor_attributes( + + stream->cursor_attributes = *attributes; + +- dc_z10_restore(dc); +- /* disable idle optimizations while updating cursor */ +- if (dc->idle_optimizations_allowed) { +- dc_allow_idle_optimizations(dc, false); +- reset_idle_optimizations = true; +- } ++ return true; ++} + +- program_cursor_attributes(dc, stream, attributes); ++bool dc_stream_program_cursor_attributes( ++ struct dc_stream_state *stream, ++ const struct dc_cursor_attributes *attributes) ++{ ++ struct dc *dc; ++ bool reset_idle_optimizations = false; + +- /* re-enable idle optimizations if necessary */ +- if (reset_idle_optimizations) +- dc_allow_idle_optimizations(dc, true); ++ dc = stream ? stream->ctx->dc : NULL; + +- return true; ++ if (dc_stream_set_cursor_attributes(stream, attributes)) { ++ dc_z10_restore(dc); ++ /* disable idle optimizations while updating cursor */ ++ if (dc->idle_optimizations_allowed) { ++ dc_allow_idle_optimizations(dc, false); ++ reset_idle_optimizations = true; ++ } ++ ++ program_cursor_attributes(dc, stream, attributes); ++ ++ /* re-enable idle optimizations if necessary */ ++ if (reset_idle_optimizations) ++ dc_allow_idle_optimizations(dc, true); ++ ++ return true; ++ } ++ ++ return false; + } + + static void program_cursor_position( +@@ -367,9 +382,6 @@ bool dc_stream_set_cursor_position( + struct dc_stream_state *stream, + const struct dc_cursor_position *position) + { +- struct dc *dc; +- bool reset_idle_optimizations = false; +- + if (NULL == stream) { + dm_error("DC: dc_stream is NULL!\n"); + return false; +@@ -380,24 +392,43 @@ bool dc_stream_set_cursor_position( + return false; + } + ++ stream->cursor_position = *position; ++ ++ return true; ++} ++ ++bool dc_stream_program_cursor_position( ++ struct dc_stream_state *stream, ++ const struct dc_cursor_position *position) ++{ ++ struct dc *dc; ++ bool reset_idle_optimizations = false; ++ const struct dc_cursor_position *old_position; ++ ++ old_position = stream ? &stream->cursor_position : NULL; + dc = stream->ctx->dc; +- dc_z10_restore(dc); + +- /* disable idle optimizations if enabling cursor */ +- if (dc->idle_optimizations_allowed && (!stream->cursor_position.enable || dc->debug.exit_idle_opt_for_cursor_updates) +- && position->enable) { +- dc_allow_idle_optimizations(dc, false); +- reset_idle_optimizations = true; +- } ++ if (dc_stream_set_cursor_position(stream, position)) { ++ dc_z10_restore(dc); + +- stream->cursor_position = *position; ++ /* disable idle optimizations if enabling cursor */ ++ if (dc->idle_optimizations_allowed && ++ (!old_position->enable || dc->debug.exit_idle_opt_for_cursor_updates) && ++ position->enable) { ++ dc_allow_idle_optimizations(dc, false); ++ reset_idle_optimizations = true; ++ } + +- program_cursor_position(dc, stream, position); +- /* re-enable idle optimizations if necessary */ +- if (reset_idle_optimizations) +- dc_allow_idle_optimizations(dc, true); + +- return true; ++ program_cursor_position(dc, stream, position); ++ /* re-enable idle optimizations if necessary */ ++ if (reset_idle_optimizations) ++ dc_allow_idle_optimizations(dc, true); ++ ++ return true; ++ } ++ ++ return false; + } + + bool dc_stream_add_writeback(struct dc *dc, +diff --git a/drivers/gpu/drm/amd/display/dc/dc_stream.h b/drivers/gpu/drm/amd/display/dc/dc_stream.h +index ee10941caa59..df86668f560f 100644 +--- a/drivers/gpu/drm/amd/display/dc/dc_stream.h ++++ b/drivers/gpu/drm/amd/display/dc/dc_stream.h +@@ -327,6 +327,9 @@ struct dc_stream_update { + + struct test_pattern *pending_test_pattern; + struct dc_crtc_timing_adjust *crtc_timing_adjust; ++ ++ struct dc_cursor_attributes *cursor_attributes; ++ struct dc_cursor_position *cursor_position; + }; + + bool dc_is_stream_unchanged( +@@ -478,10 +481,17 @@ bool dc_stream_set_cursor_attributes( + struct dc_stream_state *stream, + const struct dc_cursor_attributes *attributes); + ++bool dc_stream_program_cursor_attributes( ++ struct dc_stream_state *stream, ++ const struct dc_cursor_attributes *attributes); ++ + bool dc_stream_set_cursor_position( + struct dc_stream_state *stream, + const struct dc_cursor_position *position); + ++bool dc_stream_program_cursor_position( ++ struct dc_stream_state *stream, ++ const struct dc_cursor_position *position); + + bool dc_stream_adjust_vmin_vmax(struct dc *dc, + struct dc_stream_state *stream, +diff --git a/drivers/gpu/drm/amd/display/dc/hwss/dcn30/dcn30_hwseq.c b/drivers/gpu/drm/amd/display/dc/hwss/dcn30/dcn30_hwseq.c +index 8bc3d01537bb..f91169c80f7f 100644 +--- a/drivers/gpu/drm/amd/display/dc/hwss/dcn30/dcn30_hwseq.c ++++ b/drivers/gpu/drm/amd/display/dc/hwss/dcn30/dcn30_hwseq.c +@@ -1042,7 +1042,7 @@ bool dcn30_apply_idle_power_optimizations(struct dc *dc, bool enable) + + /* Use copied cursor, and it's okay to not switch back */ + cursor_attr.address.quad_part = cmd.mall.cursor_copy_dst.quad_part; +- dc_stream_set_cursor_attributes(stream, &cursor_attr); ++ dc_stream_program_cursor_attributes(stream, &cursor_attr); + } + + /* Enable MALL */ +diff --git a/fs/btrfs/extent_io.c b/fs/btrfs/extent_io.c +index 1e020620748d..082f87834503 100644 +--- a/fs/btrfs/extent_io.c ++++ b/fs/btrfs/extent_io.c +@@ -4269,6 +4269,13 @@ void set_extent_buffer_uptodate(struct extent_buffer *eb) + } + } + ++static void clear_extent_buffer_reading(struct extent_buffer *eb) ++{ ++ clear_bit(EXTENT_BUFFER_READING, &eb->bflags); ++ smp_mb__after_atomic(); ++ wake_up_bit(&eb->bflags, EXTENT_BUFFER_READING); ++} ++ + static void end_bbio_meta_read(struct btrfs_bio *bbio) + { + struct extent_buffer *eb = bbio->private; +@@ -4277,6 +4284,13 @@ static void end_bbio_meta_read(struct btrfs_bio *bbio) + struct folio_iter fi; + u32 bio_offset = 0; + ++ /* ++ * If the extent buffer is marked UPTODATE before the read operation ++ * completes, other calls to read_extent_buffer_pages() will return ++ * early without waiting for the read to finish, causing data races. ++ */ ++ WARN_ON(test_bit(EXTENT_BUFFER_UPTODATE, &eb->bflags)); ++ + eb->read_mirror = bbio->mirror_num; + + if (uptodate && +@@ -4303,9 +4317,7 @@ static void end_bbio_meta_read(struct btrfs_bio *bbio) + bio_offset += len; + } + +- clear_bit(EXTENT_BUFFER_READING, &eb->bflags); +- smp_mb__after_atomic(); +- wake_up_bit(&eb->bflags, EXTENT_BUFFER_READING); ++ clear_extent_buffer_reading(eb); + free_extent_buffer(eb); + + bio_put(&bbio->bio); +@@ -4339,9 +4351,7 @@ int read_extent_buffer_pages(struct extent_buffer *eb, int wait, int mirror_num, + * will now be set, and we shouldn't read it in again. + */ + if (unlikely(test_bit(EXTENT_BUFFER_UPTODATE, &eb->bflags))) { +- clear_bit(EXTENT_BUFFER_READING, &eb->bflags); +- smp_mb__after_atomic(); +- wake_up_bit(&eb->bflags, EXTENT_BUFFER_READING); ++ clear_extent_buffer_reading(eb); + return 0; + } + +diff --git a/include/linux/filter.h b/include/linux/filter.h +index cf12bfa2a78c..2acf8634008d 100644 +--- a/include/linux/filter.h ++++ b/include/linux/filter.h +@@ -644,14 +644,16 @@ static __always_inline u32 __bpf_prog_run(const struct bpf_prog *prog, + cant_migrate(); + if (static_branch_unlikely(&bpf_stats_enabled_key)) { + struct bpf_prog_stats *stats; +- u64 start = sched_clock(); ++ u64 duration, start = sched_clock(); + unsigned long flags; + + ret = dfunc(ctx, prog->insnsi, prog->bpf_func); ++ ++ duration = sched_clock() - start; + stats = this_cpu_ptr(prog->stats); + flags = u64_stats_update_begin_irqsave(&stats->syncp); + u64_stats_inc(&stats->cnt); +- u64_stats_add(&stats->nsecs, sched_clock() - start); ++ u64_stats_add(&stats->nsecs, duration); + u64_stats_update_end_irqrestore(&stats->syncp, flags); + } else { + ret = dfunc(ctx, prog->insnsi, prog->bpf_func); +diff --git a/kernel/bpf/trampoline.c b/kernel/bpf/trampoline.c +index db7599c59c78..146de96e99b8 100644 +--- a/kernel/bpf/trampoline.c ++++ b/kernel/bpf/trampoline.c +@@ -883,12 +883,13 @@ static void notrace update_prog_stats(struct bpf_prog *prog, + * Hence check that 'start' is valid. + */ + start > NO_START_TIME) { ++ u64 duration = sched_clock() - start; + unsigned long flags; + + stats = this_cpu_ptr(prog->stats); + flags = u64_stats_update_begin_irqsave(&stats->syncp); + u64_stats_inc(&stats->cnt); +- u64_stats_add(&stats->nsecs, sched_clock() - start); ++ u64_stats_add(&stats->nsecs, duration); + u64_stats_update_end_irqrestore(&stats->syncp, flags); + } + } +-- +2.45.2 + diff --git a/SOURCES/kernel-aarch64-16k-debug-fedora.config b/SOURCES/kernel-aarch64-16k-debug-fedora.config index 4c44c22..8b9bf99 100644 --- a/SOURCES/kernel-aarch64-16k-debug-fedora.config +++ b/SOURCES/kernel-aarch64-16k-debug-fedora.config @@ -10104,3 +10104,4 @@ CONFIG_HID_APPLETB_BL=m CONFIG_HID_APPLETB_KBD=m CONFIG_HID_APPLE_MAGIC_BACKLIGHT=m CONFIG_APPLE_BCE=m +CONFIG_AMD_PRIVATE_COLOR=y diff --git a/SOURCES/kernel-aarch64-16k-fedora.config b/SOURCES/kernel-aarch64-16k-fedora.config index 6ef0ca5..c6ccf78 100644 --- a/SOURCES/kernel-aarch64-16k-fedora.config +++ b/SOURCES/kernel-aarch64-16k-fedora.config @@ -10075,3 +10075,4 @@ CONFIG_HID_APPLETB_BL=m CONFIG_HID_APPLETB_KBD=m CONFIG_HID_APPLE_MAGIC_BACKLIGHT=m CONFIG_APPLE_BCE=m +CONFIG_AMD_PRIVATE_COLOR=y diff --git a/SOURCES/kernel-aarch64-64k-debug-rhel.config b/SOURCES/kernel-aarch64-64k-debug-rhel.config index 6b426c3..b1ea174 100644 --- a/SOURCES/kernel-aarch64-64k-debug-rhel.config +++ b/SOURCES/kernel-aarch64-64k-debug-rhel.config @@ -8140,3 +8140,4 @@ CONFIG_HID_APPLETB_BL=m CONFIG_HID_APPLETB_KBD=m CONFIG_HID_APPLE_MAGIC_BACKLIGHT=m CONFIG_APPLE_BCE=m +CONFIG_AMD_PRIVATE_COLOR=y diff --git a/SOURCES/kernel-aarch64-64k-rhel.config b/SOURCES/kernel-aarch64-64k-rhel.config index a0b10ae..30709e2 100644 --- a/SOURCES/kernel-aarch64-64k-rhel.config +++ b/SOURCES/kernel-aarch64-64k-rhel.config @@ -8115,3 +8115,4 @@ CONFIG_HID_APPLETB_BL=m CONFIG_HID_APPLETB_KBD=m CONFIG_HID_APPLE_MAGIC_BACKLIGHT=m CONFIG_APPLE_BCE=m +CONFIG_AMD_PRIVATE_COLOR=y diff --git a/SOURCES/kernel-aarch64-debug-fedora.config b/SOURCES/kernel-aarch64-debug-fedora.config index 6411934..85a7e95 100644 --- a/SOURCES/kernel-aarch64-debug-fedora.config +++ b/SOURCES/kernel-aarch64-debug-fedora.config @@ -10103,3 +10103,4 @@ CONFIG_HID_APPLETB_BL=m CONFIG_HID_APPLETB_KBD=m CONFIG_HID_APPLE_MAGIC_BACKLIGHT=m CONFIG_APPLE_BCE=m +CONFIG_AMD_PRIVATE_COLOR=y diff --git a/SOURCES/kernel-aarch64-debug-rhel.config b/SOURCES/kernel-aarch64-debug-rhel.config index a2357fb..58aa680 100644 --- a/SOURCES/kernel-aarch64-debug-rhel.config +++ b/SOURCES/kernel-aarch64-debug-rhel.config @@ -8136,3 +8136,4 @@ CONFIG_HID_APPLETB_BL=m CONFIG_HID_APPLETB_KBD=m CONFIG_HID_APPLE_MAGIC_BACKLIGHT=m CONFIG_APPLE_BCE=m +CONFIG_AMD_PRIVATE_COLOR=y diff --git a/SOURCES/kernel-aarch64-fedora.config b/SOURCES/kernel-aarch64-fedora.config index 912cde2..45d7c64 100644 --- a/SOURCES/kernel-aarch64-fedora.config +++ b/SOURCES/kernel-aarch64-fedora.config @@ -10074,3 +10074,4 @@ CONFIG_HID_APPLETB_BL=m CONFIG_HID_APPLETB_KBD=m CONFIG_HID_APPLE_MAGIC_BACKLIGHT=m CONFIG_APPLE_BCE=m +CONFIG_AMD_PRIVATE_COLOR=y diff --git a/SOURCES/kernel-aarch64-rhel.config b/SOURCES/kernel-aarch64-rhel.config index 17340d8..8dfe7db 100644 --- a/SOURCES/kernel-aarch64-rhel.config +++ b/SOURCES/kernel-aarch64-rhel.config @@ -8111,3 +8111,4 @@ CONFIG_HID_APPLETB_BL=m CONFIG_HID_APPLETB_KBD=m CONFIG_HID_APPLE_MAGIC_BACKLIGHT=m CONFIG_APPLE_BCE=m +CONFIG_AMD_PRIVATE_COLOR=y diff --git a/SOURCES/kernel-aarch64-rt-debug-rhel.config b/SOURCES/kernel-aarch64-rt-debug-rhel.config index 2230ac1..87cb2c1 100644 --- a/SOURCES/kernel-aarch64-rt-debug-rhel.config +++ b/SOURCES/kernel-aarch64-rt-debug-rhel.config @@ -8196,3 +8196,4 @@ CONFIG_HID_APPLETB_BL=m CONFIG_HID_APPLETB_KBD=m CONFIG_HID_APPLE_MAGIC_BACKLIGHT=m CONFIG_APPLE_BCE=m +CONFIG_AMD_PRIVATE_COLOR=y diff --git a/SOURCES/kernel-aarch64-rt-rhel.config b/SOURCES/kernel-aarch64-rt-rhel.config index d1b854a..2b1ac51 100644 --- a/SOURCES/kernel-aarch64-rt-rhel.config +++ b/SOURCES/kernel-aarch64-rt-rhel.config @@ -8171,3 +8171,4 @@ CONFIG_HID_APPLETB_BL=m CONFIG_HID_APPLETB_KBD=m CONFIG_HID_APPLE_MAGIC_BACKLIGHT=m CONFIG_APPLE_BCE=m +CONFIG_AMD_PRIVATE_COLOR=y diff --git a/SOURCES/kernel-ppc64le-debug-fedora.config b/SOURCES/kernel-ppc64le-debug-fedora.config index da78efe..bacd0fe 100644 --- a/SOURCES/kernel-ppc64le-debug-fedora.config +++ b/SOURCES/kernel-ppc64le-debug-fedora.config @@ -8355,3 +8355,4 @@ CONFIG_HID_APPLETB_BL=m CONFIG_HID_APPLETB_KBD=m CONFIG_HID_APPLE_MAGIC_BACKLIGHT=m CONFIG_APPLE_BCE=m +CONFIG_AMD_PRIVATE_COLOR=y diff --git a/SOURCES/kernel-ppc64le-debug-rhel.config b/SOURCES/kernel-ppc64le-debug-rhel.config index 8d613cf..ba2f4e6 100644 --- a/SOURCES/kernel-ppc64le-debug-rhel.config +++ b/SOURCES/kernel-ppc64le-debug-rhel.config @@ -7604,3 +7604,4 @@ CONFIG_HID_APPLETB_BL=m CONFIG_HID_APPLETB_KBD=m CONFIG_HID_APPLE_MAGIC_BACKLIGHT=m CONFIG_APPLE_BCE=m +CONFIG_AMD_PRIVATE_COLOR=y diff --git a/SOURCES/kernel-ppc64le-fedora.config b/SOURCES/kernel-ppc64le-fedora.config index ffda948..330cf3c 100644 --- a/SOURCES/kernel-ppc64le-fedora.config +++ b/SOURCES/kernel-ppc64le-fedora.config @@ -8324,3 +8324,4 @@ CONFIG_HID_APPLETB_BL=m CONFIG_HID_APPLETB_KBD=m CONFIG_HID_APPLE_MAGIC_BACKLIGHT=m CONFIG_APPLE_BCE=m +CONFIG_AMD_PRIVATE_COLOR=y diff --git a/SOURCES/kernel-ppc64le-rhel.config b/SOURCES/kernel-ppc64le-rhel.config index a809663..5b74759 100644 --- a/SOURCES/kernel-ppc64le-rhel.config +++ b/SOURCES/kernel-ppc64le-rhel.config @@ -7581,3 +7581,4 @@ CONFIG_HID_APPLETB_BL=m CONFIG_HID_APPLETB_KBD=m CONFIG_HID_APPLE_MAGIC_BACKLIGHT=m CONFIG_APPLE_BCE=m +CONFIG_AMD_PRIVATE_COLOR=y diff --git a/SOURCES/kernel-s390x-debug-fedora.config b/SOURCES/kernel-s390x-debug-fedora.config index d1b1dce..c3b15ab 100644 --- a/SOURCES/kernel-s390x-debug-fedora.config +++ b/SOURCES/kernel-s390x-debug-fedora.config @@ -8290,3 +8290,4 @@ CONFIG_HID_APPLETB_BL=m CONFIG_HID_APPLETB_KBD=m CONFIG_HID_APPLE_MAGIC_BACKLIGHT=m CONFIG_APPLE_BCE=m +CONFIG_AMD_PRIVATE_COLOR=y diff --git a/SOURCES/kernel-s390x-debug-rhel.config b/SOURCES/kernel-s390x-debug-rhel.config index 0330690..d9e180e 100644 --- a/SOURCES/kernel-s390x-debug-rhel.config +++ b/SOURCES/kernel-s390x-debug-rhel.config @@ -7586,3 +7586,4 @@ CONFIG_HID_APPLETB_BL=m CONFIG_HID_APPLETB_KBD=m CONFIG_HID_APPLE_MAGIC_BACKLIGHT=m CONFIG_APPLE_BCE=m +CONFIG_AMD_PRIVATE_COLOR=y diff --git a/SOURCES/kernel-s390x-fedora.config b/SOURCES/kernel-s390x-fedora.config index 8153e5e..1d7548b 100644 --- a/SOURCES/kernel-s390x-fedora.config +++ b/SOURCES/kernel-s390x-fedora.config @@ -8259,3 +8259,4 @@ CONFIG_HID_APPLETB_BL=m CONFIG_HID_APPLETB_KBD=m CONFIG_HID_APPLE_MAGIC_BACKLIGHT=m CONFIG_APPLE_BCE=m +CONFIG_AMD_PRIVATE_COLOR=y diff --git a/SOURCES/kernel-s390x-rhel.config b/SOURCES/kernel-s390x-rhel.config index 9caba6f..ce38556 100644 --- a/SOURCES/kernel-s390x-rhel.config +++ b/SOURCES/kernel-s390x-rhel.config @@ -7563,3 +7563,4 @@ CONFIG_HID_APPLETB_BL=m CONFIG_HID_APPLETB_KBD=m CONFIG_HID_APPLE_MAGIC_BACKLIGHT=m CONFIG_APPLE_BCE=m +CONFIG_AMD_PRIVATE_COLOR=y diff --git a/SOURCES/kernel-s390x-zfcpdump-rhel.config b/SOURCES/kernel-s390x-zfcpdump-rhel.config index f39c27f..f987196 100644 --- a/SOURCES/kernel-s390x-zfcpdump-rhel.config +++ b/SOURCES/kernel-s390x-zfcpdump-rhel.config @@ -7586,3 +7586,4 @@ CONFIG_HID_APPLETB_BL=m CONFIG_HID_APPLETB_KBD=m CONFIG_HID_APPLE_MAGIC_BACKLIGHT=m CONFIG_APPLE_BCE=m +CONFIG_AMD_PRIVATE_COLOR=y diff --git a/SOURCES/kernel-x86_64-debug-fedora.config b/SOURCES/kernel-x86_64-debug-fedora.config index d4e1c16..9486803 100644 --- a/SOURCES/kernel-x86_64-debug-fedora.config +++ b/SOURCES/kernel-x86_64-debug-fedora.config @@ -9047,3 +9047,4 @@ CONFIG_SND_SOC_SOF_IPC3=y CONFIG_SND_SOC_SOF_INTEL_IPC4=y CONFIG_SND_SOC_SOF_AMD_COMMON=m CONFIG_SND_SOC_TOPOLOGY=y +CONFIG_AMD_PRIVATE_COLOR=y diff --git a/SOURCES/kernel-x86_64-debug-rhel.config b/SOURCES/kernel-x86_64-debug-rhel.config index d814503..8b78a23 100644 --- a/SOURCES/kernel-x86_64-debug-rhel.config +++ b/SOURCES/kernel-x86_64-debug-rhel.config @@ -8035,3 +8035,4 @@ CONFIG_SND_SOC_SOF_AMD_ACP63=m # CONFIG_SND_AMD_ASOC_REMBRANDT is not set # CONFIG_SND_SOC_AMD_LEGACY_MACH is not set CONFIG_SND_SOC_TOPOLOGY=y +CONFIG_AMD_PRIVATE_COLOR=y diff --git a/SOURCES/kernel-x86_64-fedora.config b/SOURCES/kernel-x86_64-fedora.config index 0c9353f..a4240f3 100644 --- a/SOURCES/kernel-x86_64-fedora.config +++ b/SOURCES/kernel-x86_64-fedora.config @@ -9017,3 +9017,4 @@ CONFIG_SND_SOC_SOF_IPC3=y CONFIG_SND_SOC_SOF_INTEL_IPC4=y CONFIG_SND_SOC_SOF_AMD_COMMON=m CONFIG_SND_SOC_TOPOLOGY=y +CONFIG_AMD_PRIVATE_COLOR=y diff --git a/SOURCES/kernel-x86_64-rhel.config b/SOURCES/kernel-x86_64-rhel.config index 35f7ad5..c7b6ac2 100644 --- a/SOURCES/kernel-x86_64-rhel.config +++ b/SOURCES/kernel-x86_64-rhel.config @@ -8011,3 +8011,4 @@ CONFIG_SND_SOC_SOF_AMD_ACP63=m # CONFIG_SND_AMD_ASOC_REMBRANDT is not set # CONFIG_SND_SOC_AMD_LEGACY_MACH is not set CONFIG_SND_SOC_TOPOLOGY=y +CONFIG_AMD_PRIVATE_COLOR=y diff --git a/SOURCES/kernel-x86_64-rt-debug-rhel.config b/SOURCES/kernel-x86_64-rt-debug-rhel.config index bae0b0e..f9a79f1 100644 --- a/SOURCES/kernel-x86_64-rt-debug-rhel.config +++ b/SOURCES/kernel-x86_64-rt-debug-rhel.config @@ -8096,3 +8096,4 @@ CONFIG_SND_SOC_SOF_AMD_ACP63=m # CONFIG_SND_AMD_ASOC_REMBRANDT is not set # CONFIG_SND_SOC_AMD_LEGACY_MACH is not set CONFIG_SND_SOC_TOPOLOGY=y +CONFIG_AMD_PRIVATE_COLOR=y diff --git a/SOURCES/kernel-x86_64-rt-rhel.config b/SOURCES/kernel-x86_64-rt-rhel.config index 37a5069..5e66ede 100644 --- a/SOURCES/kernel-x86_64-rt-rhel.config +++ b/SOURCES/kernel-x86_64-rt-rhel.config @@ -8072,3 +8072,4 @@ CONFIG_SND_SOC_SOF_AMD_ACP63=m # CONFIG_SND_AMD_ASOC_REMBRANDT is not set # CONFIG_SND_SOC_AMD_LEGACY_MACH is not set CONFIG_SND_SOC_TOPOLOGY=y +CONFIG_AMD_PRIVATE_COLOR=y diff --git a/SOURCES/tkg-misc-additions.patch b/SOURCES/tkg-misc-additions.patch index 31368a7..938ddf3 100644 --- a/SOURCES/tkg-misc-additions.patch +++ b/SOURCES/tkg-misc-additions.patch @@ -97,3 +97,157 @@ index f4b210ab061291..837d0dbb28ea08 100644 depends on HAVE_ARCH_MMAP_RND_COMPAT_BITS help This value can be used to select the number of bits to use to + +diff --git a/drivers/gpu/drm/amd/display/dc/core/dc_state.c b/drivers/gpu/drm/amd/display/dc/core/dc_state.c +index 180ac47868c2..42bf097b36b7 100644 +--- a/drivers/gpu/drm/amd/display/dc/core/dc_state.c ++++ b/drivers/gpu/drm/amd/display/dc/core/dc_state.c +@@ -300,7 +300,6 @@ void dc_state_destruct(struct dc_state *state) + state->phantom_plane_count = 0; + + state->stream_mask = 0; +- memset(&state->res_ctx, 0, sizeof(state->res_ctx)); + memset(&state->pp_display_cfg, 0, sizeof(state->pp_display_cfg)); + memset(&state->dcn_bw_vars, 0, sizeof(state->dcn_bw_vars)); + state->clk_mgr = NULL; + +From 299b81f3e619aea3ceda77d7c42842a496b34a53 Mon Sep 17 00:00:00 2001 +From: Peter Jung <admin@ptr1337.dev> +Date: Thu, 21 Mar 2024 19:00:50 +0100 +Subject: [PATCH] cachy: move AMD_PRIVATE_COLOR to Kconfig + +Co-authored-by: PedroHLC <root@pedrohlc.com> +Signed-off-by: Peter Jung <admin@ptr1337.dev> +--- + drivers/gpu/drm/amd/display/Kconfig | 6 ++++++ + drivers/gpu/drm/amd/display/amdgpu_dm/amdgpu_dm.c | 2 +- + drivers/gpu/drm/amd/display/amdgpu_dm/amdgpu_dm_color.c | 2 +- + drivers/gpu/drm/amd/display/amdgpu_dm/amdgpu_dm_crtc.c | 6 +++--- + drivers/gpu/drm/amd/display/amdgpu_dm/amdgpu_dm_plane.c | 6 +++--- + 5 files changed, 14 insertions(+), 8 deletions(-) + +diff --git a/drivers/gpu/drm/amd/display/Kconfig b/drivers/gpu/drm/amd/display/Kconfig +index 901d1961b73927..05c49141f5802f 100644 +--- a/drivers/gpu/drm/amd/display/Kconfig ++++ b/drivers/gpu/drm/amd/display/Kconfig +@@ -51,4 +51,10 @@ config DRM_AMD_SECURE_DISPLAY + This option enables the calculation of crc of specific region via + debugfs. Cooperate with specific DMCU FW. + ++config AMD_PRIVATE_COLOR ++ bool "Enable KMS color management by AMD for AMD" ++ default n ++ help ++ This option extends the KMS color management API with AMD driver-specific properties to enhance the color management support on AMD Steam Deck. ++ + endmenu +diff --git a/drivers/gpu/drm/amd/display/amdgpu_dm/amdgpu_dm.c b/drivers/gpu/drm/amd/display/amdgpu_dm/amdgpu_dm.c +index 59d2eee72a3297..0a4e75de95c257 100644 +--- a/drivers/gpu/drm/amd/display/amdgpu_dm/amdgpu_dm.c ++++ b/drivers/gpu/drm/amd/display/amdgpu_dm/amdgpu_dm.c +@@ -4078,7 +4078,7 @@ static int amdgpu_dm_mode_config_init(struct amdgpu_device *adev) + return r; + } + +-#ifdef AMD_PRIVATE_COLOR ++#ifdef CONFIG_AMD_PRIVATE_COLOR + if (amdgpu_dm_create_color_properties(adev)) + return -ENOMEM; + #endif +diff --git a/drivers/gpu/drm/amd/display/amdgpu_dm/amdgpu_dm_color.c b/drivers/gpu/drm/amd/display/amdgpu_dm/amdgpu_dm_color.c +index c87b64e464ed5c..6fe07243adc3d5 100644 +--- a/drivers/gpu/drm/amd/display/amdgpu_dm/amdgpu_dm_color.c ++++ b/drivers/gpu/drm/amd/display/amdgpu_dm/amdgpu_dm_color.c +@@ -97,7 +97,7 @@ static inline struct fixed31_32 amdgpu_dm_fixpt_from_s3132(__u64 x) + return val; + } + +-#ifdef AMD_PRIVATE_COLOR ++#ifdef CONFIG_AMD_PRIVATE_COLOR + /* Pre-defined Transfer Functions (TF) + * + * AMD driver supports pre-defined mathematical functions for transferring +diff --git a/drivers/gpu/drm/amd/display/amdgpu_dm/amdgpu_dm_crtc.c b/drivers/gpu/drm/amd/display/amdgpu_dm/amdgpu_dm_crtc.c +index 6e715ef3a5566e..11c7199ec3b348 100644 +--- a/drivers/gpu/drm/amd/display/amdgpu_dm/amdgpu_dm_crtc.c ++++ b/drivers/gpu/drm/amd/display/amdgpu_dm/amdgpu_dm_crtc.c +@@ -290,7 +290,7 @@ static int amdgpu_dm_crtc_late_register(struct drm_crtc *crtc) + } + #endif + +-#ifdef AMD_PRIVATE_COLOR ++#ifdef CONFIG_AMD_PRIVATE_COLOR + /** + * dm_crtc_additional_color_mgmt - enable additional color properties + * @crtc: DRM CRTC +@@ -372,7 +372,7 @@ static const struct drm_crtc_funcs amdgpu_dm_crtc_funcs = { + #if defined(CONFIG_DEBUG_FS) + .late_register = amdgpu_dm_crtc_late_register, + #endif +-#ifdef AMD_PRIVATE_COLOR ++#ifdef CONFIG_AMD_PRIVATE_COLOR + .atomic_set_property = amdgpu_dm_atomic_crtc_set_property, + .atomic_get_property = amdgpu_dm_atomic_crtc_get_property, + #endif +@@ -551,7 +551,7 @@ int amdgpu_dm_crtc_init(struct amdgpu_display_manager *dm, + + drm_mode_crtc_set_gamma_size(&acrtc->base, MAX_COLOR_LEGACY_LUT_ENTRIES); + +-#ifdef AMD_PRIVATE_COLOR ++#ifdef CONFIG_AMD_PRIVATE_COLOR + dm_crtc_additional_color_mgmt(&acrtc->base); + #endif + return 0; +diff --git a/drivers/gpu/drm/amd/display/amdgpu_dm/amdgpu_dm_plane.c b/drivers/gpu/drm/amd/display/amdgpu_dm/amdgpu_dm_plane.c +index 8a4c40b4c27e4f..779880c6457553 100644 +--- a/drivers/gpu/drm/amd/display/amdgpu_dm/amdgpu_dm_plane.c ++++ b/drivers/gpu/drm/amd/display/amdgpu_dm/amdgpu_dm_plane.c +@@ -1468,7 +1468,7 @@ static void amdgpu_dm_plane_drm_plane_destroy_state(struct drm_plane *plane, + drm_atomic_helper_plane_destroy_state(plane, state); + } + +-#ifdef AMD_PRIVATE_COLOR ++#ifdef CONFIG_AMD_PRIVATE_COLOR + static void + dm_atomic_plane_attach_color_mgmt_properties(struct amdgpu_display_manager *dm, + struct drm_plane *plane) +@@ -1659,7 +1659,7 @@ static const struct drm_plane_funcs dm_plane_funcs = { + .atomic_duplicate_state = amdgpu_dm_plane_drm_plane_duplicate_state, + .atomic_destroy_state = amdgpu_dm_plane_drm_plane_destroy_state, + .format_mod_supported = amdgpu_dm_plane_format_mod_supported, +-#ifdef AMD_PRIVATE_COLOR ++#ifdef CONFIG_AMD_PRIVATE_COLOR + .atomic_set_property = dm_atomic_plane_set_property, + .atomic_get_property = dm_atomic_plane_get_property, + #endif +@@ -1742,7 +1742,7 @@ int amdgpu_dm_plane_init(struct amdgpu_display_manager *dm, + + drm_plane_helper_add(plane, &dm_plane_helper_funcs); + +-#ifdef AMD_PRIVATE_COLOR ++#ifdef CONFIG_AMD_PRIVATE_COLOR + dm_atomic_plane_attach_color_mgmt_properties(dm, plane); + #endif + /* Create (reset) the plane state */ + +From f1807682de0edbff6c1e46b19642a517d2e15c57 Mon Sep 17 00:00:00 2001 +From: Tk-Glitch <ti3nou@gmail.com> +Date: Sat, 13 Apr 2024 18:25:35 +0530 +Subject: revert "drm/amd/pm: Fetch current power limit from FW" + +Fetching FW value prevents power limit modification by the user. +The "out-of-band ways" are unclear. +This fixes 7700/7800 XT custom powercap. + +diff --git a/drivers/gpu/drm/amd/pm/swsmu/amdgpu_smu.c b/drivers/gpu/drm/amd/pm/swsmu/amdgpu_smu.c +index a54663f2e2ab9f..7ffad3eb0a0150 100644 +--- a/drivers/gpu/drm/amd/pm/swsmu/amdgpu_smu.c ++++ b/drivers/gpu/drm/amd/pm/swsmu/amdgpu_smu.c +@@ -2703,7 +2703,6 @@ int smu_get_power_limit(void *handle, + case SMU_PPT_LIMIT_CURRENT: + switch (amdgpu_ip_version(adev, MP1_HWIP, 0)) { + case IP_VERSION(13, 0, 2): +- case IP_VERSION(13, 0, 6): + case IP_VERSION(11, 0, 7): + case IP_VERSION(11, 0, 11): + case IP_VERSION(11, 0, 12): diff --git a/SPECS/kernel.spec b/SPECS/kernel.spec index 913814a..94722d8 100644 --- a/SPECS/kernel.spec +++ b/SPECS/kernel.spec @@ -169,7 +169,7 @@ Summary: The Linux kernel # This is needed to do merge window version magic %define patchlevel 9 # This allows pkg_release to have configurable %%{?dist} tag -%define specrelease 202%{?buildid}%{?dist} +%define specrelease 203%{?buildid}%{?dist} # This defines the kabi tarball version %define kabiversion 6.9.9 @@ -984,6 +984,8 @@ Patch211: tkg-misc-additions.patch Patch212: tkg-unprivileged-CLONE_NEWUSER.patch Patch213: cachy-bore.patch Patch214: scx-kernel.patch +Patch215: cachy-bbr3.patch +Patch216: cachy-fixes.patch # device specific patches Patch300: linux-surface.patch @@ -1006,7 +1008,6 @@ Patch337: 0001-ally-x-audio-quirk.patch # hdr: https://github.com/CachyOS/kernel-patches Patch208: 0001-ntsync.patch Patch325: 0001-amd-pstate.patch -Patch326: 0001-amd-hdr.patch Patch327: 0001-add-acpi_call.patch Patch328: uinput.patch @@ -1883,6 +1884,8 @@ ApplyOptionalPatch tkg-misc-additions.patch ApplyOptionalPatch tkg-unprivileged-CLONE_NEWUSER.patch ApplyOptionalPatch cachy-bore.patch ApplyOptionalPatch scx-kernel.patch +ApplyOptionalPatch cachy-bbr3.patch +ApplyOptionalPatch cachy-fixes.patch # device specific patches ApplyOptionalPatch linux-surface.patch @@ -1905,7 +1908,6 @@ ApplyOptionalPatch 0001-ally-x-audio-quirk.patch # https://github.com/CachyOS/kernel-patches ApplyOptionalPatch 0001-ntsync.patch ApplyOptionalPatch 0001-amd-pstate.patch -ApplyOptionalPatch 0001-amd-hdr.patch ApplyOptionalPatch 0001-add-acpi_call.patch ApplyOptionalPatch uinput.patch @@ -4123,6 +4125,9 @@ fi\ # # %changelog +* Thu Jul 18 2024 Jan200101 <sentrycraft123@gmail.com> - 6.9.9-203.fsync +- Update patchsets + * Tue Jul 16 2024 Jan200101 <sentrycraft123@gmail.com> - 6.9.9-202.fsync - kernel-fsync v6.9.9 diff --git a/TOOLS/patch_configs.py b/TOOLS/patch_configs.py index 6838cd6..1a84d21 100755 --- a/TOOLS/patch_configs.py +++ b/TOOLS/patch_configs.py @@ -37,8 +37,8 @@ GENERIC_PATCHES = [ ["SENSORS_SURFACE_FAN", None, MODULE], ["SENSORS_SURFACE_TEMP", None, MODULE], - # Steam Deck / amdgpu HDR Color management - ["DRM_AMD_COLOR_STEAMDECK", None, ENABLE], + # amdgpu HDR Color management + ["AMD_PRIVATE_COLOR", None, ENABLE], # Rog Ally Gyro Fix ["BMI323_I2C", None, MODULE], |