dri-devel September 2021

dri-devel@lists.freedesktop.org

348 participants
817 discussions

[PATCH V4 00/10] PCI/VGA: Rework default VGA device selection
by Huacai Chen 11 Sep '21

11 Sep '21

My original work is at [1]. Bjorn do some rework and extension in V2. It moves the VGA arbiter to the PCI subsystem, fixes a few nits, and breaks a few pieces to make the main patch a little smaller. V3 rewrite the commit log of the last patch (which is also summarized by Bjorn). V4 split the last patch to two steps. All comments welcome! [1] https://lore.kernel.org/dri-devel/20210705100503.1120643-1-chenhuacai@loong… Bjorn Helgaas (4): PCI/VGA: Move vgaarb to drivers/pci PCI/VGA: Replace full MIT license text with SPDX identifier PCI/VGA: Use unsigned format string to print lock counts PCI/VGA: Remove empty vga_arb_device_card_gone() Huacai Chen (6): PCI/VGA: Move vga_arb_integrated_gpu() earlier in file PCI/VGA: Prefer vga_default_device() PCI/VGA: Split out vga_arb_update_default_device() PCI/VGA: Log bridge control messages when adding devices PCI/VGA: Rework default VGA device selection (Step 1) PCI/VGA: Rework default VGA device selection (Step 2) Signed-off-by: Huacai Chen <chenhuacai(a)loongson.cn> Signed-off-by: Bjorn Helgaas <bhelgaas(a)google.com> --- drivers/gpu/vga/Kconfig | 19 --- drivers/gpu/vga/Makefile | 1 - drivers/pci/Kconfig | 19 +++ drivers/pci/Makefile | 1 + drivers/{gpu/vga => pci}/vgaarb.c | 269 ++++++++++++------------------ 5 files changed, 126 insertions(+), 183 deletions(-) rename drivers/{gpu/vga => pci}/vgaarb.c (90%) -- 2.27.0

3 12

[PATCH] drm/i915: Add ww context to intel_dpt_pin
by Maarten Lankhorst 11 Sep '21

11 Sep '21

Ensure i915_vma_pin_iomap and vma_unpin are done with dpt->obj lock held. I don't think there's much of a point in merging intel_dpt_pin() with intel_pin_fb_obj_dpt(), they touch different objects. Signed-off-by: Maarten Lankhorst <maarten.lankhorst(a)linux.intel.com> --- drivers/gpu/drm/i915/display/intel_dpt.c | 40 +++++++++++++++--------- 1 file changed, 25 insertions(+), 15 deletions(-) diff --git a/drivers/gpu/drm/i915/display/intel_dpt.c b/drivers/gpu/drm/i915/display/intel_dpt.c index de62bd77b15e..edd6f1aa2626 100644 --- a/drivers/gpu/drm/i915/display/intel_dpt.c +++ b/drivers/gpu/drm/i915/display/intel_dpt.c @@ -121,32 +121,42 @@ struct i915_vma *intel_dpt_pin(struct i915_address_space *vm) intel_wakeref_t wakeref; struct i915_vma *vma; void __iomem *iomem; + struct i915_gem_ww_ctx ww; + int err; wakeref = intel_runtime_pm_get(&i915->runtime_pm); atomic_inc(&i915->gpu_error.pending_fb_pin); - vma = i915_gem_object_ggtt_pin(dpt->obj, NULL, 0, 4096, - HAS_LMEM(i915) ? 0 : PIN_MAPPABLE); - if (IS_ERR(vma)) - goto err; + for_i915_gem_ww(&ww, err, true) { + err = i915_gem_object_lock(dpt->obj, &ww); + if (err) + continue; - iomem = i915_vma_pin_iomap(vma); - i915_vma_unpin(vma); - if (IS_ERR(iomem)) { - vma = ERR_CAST(iomem); - goto err; - } + vma = i915_gem_object_ggtt_pin_ww(dpt->obj, &ww, NULL, 0, 4096, + HAS_LMEM(i915) ? 0 : PIN_MAPPABLE); + if (IS_ERR(vma)) { + err = PTR_ERR(vma); + continue; + } + + iomem = i915_vma_pin_iomap(vma); + i915_vma_unpin(vma); - dpt->vma = vma; - dpt->iomem = iomem; + if (IS_ERR(iomem)) { + err = PTR_ERR(vma); + continue; + } - i915_vma_get(vma); + dpt->vma = vma; + dpt->iomem = iomem; + + i915_vma_get(vma); + } -err: atomic_dec(&i915->gpu_error.pending_fb_pin); intel_runtime_pm_put(&i915->runtime_pm, wakeref); - return vma; + return err ? ERR_PTR(err) : vma; } void intel_dpt_unpin(struct i915_address_space *vm) -- 2.33.0

2 1

[git pull] drm fixes for 5.15-rc1
by Dave Airlie 10 Sep '21

10 Sep '21

Hi Linus, Just an initial bunch of fixes for the merge window, amdgpu is most of them with a few ttm fixes and an fbdev avoid multiply overflow fix. Dave. drm-next-2021-09-10: drm fixes for 5.15-rc1 ttm: - Fix ttm_bo_move_memcpy() when ttm_resource is subclassed. - Fix ttm deadlock if target BO isn't idle - ttm build fix - ttm docs fix dma-buf: - config option fixes fbdev: - limit resolutions to avoid int overflow i915: - stddef change. amdgpu: - Misc cleanups, typo fixes - EEPROM fix - Add some new PCI IDs - Scatter/Gather display support for Yellow Carp - PCIe DPM fix for RKL platforms - RAS fix amdkfd: - SVM fix vc4: - static function fix mgag200: - fix uninit var panfrost: - lock_region fixes - Make some dma-buf config options depend on DMA_SHARED_BUFFER. - Handle multiplication overflow of fbdev xres/yres in the core. The following changes since commit 8f0284f190e6a0aa09015090568c03f18288231a: Merge tag 'amd-drm-next-5.15-2021-08-27' of https://gitlab.freedesktop.org/agd5f/linux into drm-next (2021-08-30 09:06:03 +1000) are available in the Git repository at: git://anongit.freedesktop.org/drm/drm tags/drm-next-2021-09-10 for you to fetch changes up to 70982eef4d7eebb47a3b1ef25ec1bc742f3a21cf: drm/ttm: Fix a deadlock if the target BO is not idle during swap (2021-09-10 16:18:53 +1000) ---------------------------------------------------------------- drm fixes for 5.15-rc1 ttm: - Fix ttm_bo_move_memcpy() when ttm_resource is subclassed. - Fix ttm deadlock if target BO isn't idle - ttm build fix - ttm docs fix dma-buf: - config option fixes fbdev: - limit resolutions to avoid int overflow i915: - stddef change. amdgpu: - Misc cleanups, typo fixes - EEPROM fix - Add some new PCI IDs - Scatter/Gather display support for Yellow Carp - PCIe DPM fix for RKL platforms - RAS fix amdkfd: - SVM fix vc4: - static function fix mgag200: - fix uninit var panfrost: - lock_region fixes - Make some dma-buf config options depend on DMA_SHARED_BUFFER. - Handle multiplication overflow of fbdev xres/yres in the core. ---------------------------------------------------------------- Aaron Liu (1): drm/amd/display: setup system context for APUs Alex Deucher (1): drm/amdgpu: add some additional RDNA2 PCI IDs Alex Sierra (1): drm/amdkfd: drop process ref count when xnack disable Alyssa Rosenzweig (3): drm/panfrost: Simplify lock_region calculation drm/panfrost: Use u64 for size in lock_region drm/panfrost: Clamp lock region to Bifrost minimum Angus Wang (1): drm/amd/display: cleanup idents after a revert Anson Jacob (1): drm/amd/display: Fix memory leak reported by coverity Colin Ian King (2): drm/mgag200: Fix uninitialized variable delta drm/amdgpu/swsmu: fix spelling mistake "minimun" -> "minimum" Dave Airlie (3): Merge tag 'amd-drm-next-5.15-2021-09-01' of https://gitlab.freedesktop.org/agd5f/linux into drm-next Merge tag 'drm-misc-next-fixes-2021-09-03' of git://anongit.freedesktop.org/drm/drm-misc into drm-next Merge tag 'drm-misc-next-fixes-2021-09-09' of git://anongit.freedesktop.org/drm/drm-misc into drm-next Evan Quan (1): drm/amdgpu: reenable BACO support for 699F:C7 polaris12 SKU Geert Uytterhoeven (3): dma-buf: DMABUF_MOVE_NOTIFY should depend on DMA_SHARED_BUFFER dma-buf: DMABUF_DEBUG should depend on DMA_SHARED_BUFFER dma-buf: DMABUF_SYSFS_STATS should depend on DMA_SHARED_BUFFER Guchun Chen (1): drm/amdgpu: stop scheduler when calling hw_fini (v2) Jason Ekstrand (2): drm/ttm: ttm_bo_device is now ttm_device drm/ttm: Include pagemap.h from ttm_tt.h Jiapeng Chong (1): drm/vc4: hdmi: make vc4_hdmi_codec_pdata static Jiawei Gu (1): drm/amdgpu: enable more pm sysfs under SRIOV 1-VF mode Jing Yangyang (1): drm:dcn31: fix boolreturn.cocci warnings John Clements (1): drm/amdgpu: Clear RAS interrupt status on aldebaran Kees Cook (1): drm/amd/pm: And destination bounds checking to struct copy Koba Ko (1): drm/amdgpu: Disable PCIE_DPM on Intel RKL Platform Lang Yu (1): drm/amdgpu: show both cmd id and name when psp cmd failed Luben Tuikov (2): drm/amdgpu: Fixes to returning VBIOS RAS EEPROM address drm/amdgpu: Process any VBIOS RAS EEPROM address Michael Strauss (1): drm/amd/display: Initialize lt_settings on instantiation Nicholas Kazlauskas (1): drm/amdgpu: Enable S/G for Yellow Carp Philip Yang (1): drm/amdgpu: fix fdinfo race with process exit Stephen Rothwell (1): drm/i915: use linux/stddef.h due to "isystem: trim/fixup stdarg.h and other headers" Tetsuo Handa (1): fbmem: don't allow too huge resolutions Thomas Hellström (1): drm/ttm: Fix ttm_bo_move_memcpy() for subclassed struct ttm_resource Yifan Zhang (1): drm/amdgpu: correct comments in memory type managers YuBiao Wang (1): drm/amd/amdgpu: Add ready_to_reset resp for vega10 xinhui pan (2): drm/amdgpu: Fix a deadlock if previous GEM object allocation fails drm/ttm: Fix a deadlock if the target BO is not idle during swap Documentation/gpu/drm-mm.rst | 2 +- drivers/dma-buf/Kconfig | 4 +- drivers/gpu/drm/amd/amdgpu/amdgpu_atomfirmware.c | 50 ++++++++++++++-------- drivers/gpu/drm/amd/amdgpu/amdgpu_display.c | 1 + drivers/gpu/drm/amd/amdgpu/amdgpu_drv.c | 17 ++++++++ drivers/gpu/drm/amd/amdgpu/amdgpu_fdinfo.c | 11 ++++- drivers/gpu/drm/amd/amdgpu/amdgpu_fence.c | 8 ++++ drivers/gpu/drm/amd/amdgpu/amdgpu_gem.c | 23 +++++----- drivers/gpu/drm/amd/amdgpu/amdgpu_gtt_mgr.c | 4 +- drivers/gpu/drm/amd/amdgpu/amdgpu_psp.c | 8 ++-- drivers/gpu/drm/amd/amdgpu/amdgpu_ras_eeprom.c | 25 +++++------ drivers/gpu/drm/amd/amdgpu/amdgpu_vram_mgr.c | 6 +-- drivers/gpu/drm/amd/amdgpu/mxgpu_ai.c | 2 + drivers/gpu/drm/amd/amdgpu/mxgpu_ai.h | 1 + drivers/gpu/drm/amd/amdgpu/nbio_v7_4.c | 30 ++++++++++--- drivers/gpu/drm/amd/amdgpu/vi.c | 9 +--- drivers/gpu/drm/amd/amdkfd/kfd_svm.c | 3 +- drivers/gpu/drm/amd/display/amdgpu_dm/amdgpu_dm.c | 2 +- drivers/gpu/drm/amd/display/dc/core/dc_link_dp.c | 6 +-- .../drm/amd/display/dc/dcn303/dcn303_resource.c | 6 ++- .../drm/amd/display/dc/dcn31/dcn31_panel_cntl.c | 4 +- .../display/dc/dml/dcn20/display_mode_vba_20v2.c | 2 +- drivers/gpu/drm/amd/pm/amdgpu_pm.c | 8 ++-- drivers/gpu/drm/amd/pm/inc/amdgpu_smu.h | 24 +++++++++++ .../gpu/drm/amd/pm/powerplay/hwmgr/smu7_hwmgr.c | 17 +++++++- drivers/gpu/drm/amd/pm/swsmu/smu11/arcturus_ppt.c | 6 +-- drivers/gpu/drm/amd/pm/swsmu/smu11/navi10_ppt.c | 8 ++-- drivers/gpu/drm/amd/pm/swsmu/smu11/vangogh_ppt.c | 2 +- drivers/gpu/drm/amd/pm/swsmu/smu12/renoir_ppt.c | 2 +- drivers/gpu/drm/amd/pm/swsmu/smu13/aldebaran_ppt.c | 5 +-- .../gpu/drm/amd/pm/swsmu/smu13/yellow_carp_ppt.c | 2 +- drivers/gpu/drm/i915/gt/intel_gt_requests.h | 2 +- drivers/gpu/drm/mgag200/mgag200_pll.c | 1 + drivers/gpu/drm/panfrost/panfrost_mmu.c | 31 +++++--------- drivers/gpu/drm/panfrost/panfrost_regs.h | 2 + drivers/gpu/drm/ttm/ttm_bo.c | 6 +-- drivers/gpu/drm/ttm/ttm_bo_util.c | 7 ++- drivers/gpu/drm/ttm/ttm_tt.c | 1 - drivers/gpu/drm/vc4/vc4_hdmi.c | 2 +- drivers/video/fbdev/core/fbmem.c | 6 +++ include/drm/ttm/ttm_tt.h | 3 +- 41 files changed, 232 insertions(+), 127 deletions(-)

2 1

[PATCH] video: fbdev: atyfb: Remove assigned but never used variable statements
by Colin King 10 Sep '21

10 Sep '21

From: Colin Ian King <colin.king(a)canonical.com> There are a couple of statements where local variables are being assigned values that are never read because the function returns immediately after the assignment. Clean up the code by removing them. Addresses-Coverity: ("Unused value") Signed-off-by: Colin Ian King <colin.king(a)canonical.com> --- drivers/video/fbdev/aty/mach64_gx.c | 2 -- 1 file changed, 2 deletions(-) diff --git a/drivers/video/fbdev/aty/mach64_gx.c b/drivers/video/fbdev/aty/mach64_gx.c index 9c37e28fb78b..d06d24830080 100644 --- a/drivers/video/fbdev/aty/mach64_gx.c +++ b/drivers/video/fbdev/aty/mach64_gx.c @@ -352,10 +352,8 @@ static int aty_var_to_pll_18818(const struct fb_info *info, u32 vclk_per, post_divider = 1; if (MHz100 > MAX_FREQ_2595) { - MHz100 = MAX_FREQ_2595; return -EINVAL; } else if (MHz100 < ABS_MIN_FREQ_2595) { - program_bits = 0; /* MHz100 = 257 */ return -EINVAL; } else { while (MHz100 < MIN_FREQ_2595) { -- 2.32.0

1 0

Re: [PATCH v2 5/9] vfio/mdev: Consolidate all the device_api sysfs into the core code
by Jason Gunthorpe 10 Sep '21

10 Sep '21

On Fri, Sep 10, 2021 at 01:10:46PM +0100, Christoph Hellwig wrote: > On Thu, Sep 09, 2021 at 04:38:45PM -0300, Jason Gunthorpe wrote: > > Every driver just emits a static string, simply feed it through the ops > > and provide a standard sysfs show function. > > Looks sensible. But can you make the attribute optional and add a > comment marking it deprecated? Because it really is completely useless. > We don't version userspace APIs, userspae has to discover new features > individually by e.g. finding new sysfs files or just trying new ioctls. To be honest I have no idea what side effects that would have.. device code search tells me libvirt reads it and stuffs it into some XML Something called mdevctl touches it, feeds it into some JSON and other stuff.. qemu has some VFIO_DEVICE_API_* constants but it is all dead code I agree it shouldn't have been there in the first place Cornelia? Alex? Any thoughts? Jason

2 1

[RFC PATCH 0/4] Allow to use DRM fbdev emulation layer with CONFIG_FB disabled
by Javier Martinez Canillas 10 Sep '21

10 Sep '21

This patch series splits the fbdev core support in two different Kconfig symbols: FB and FB_CORE. The motivation for this is to allow CONFIG_FB to be disabled, while still using fbcon with the DRM fbdev emulation layer. The reason for doing this is that now with simpledrm we could just boot with simpledrm -> real DRM driver, without needing any legacy fbdev driver (e.g: efifb or simplefb) even for the early console. We want to do that in the Fedora kernel, but currently need to keep option CONFIG_FB enabled and all fbdev drivers explicitly disabled, which makes the configuration harder to maintain. It is a RFC because I'm not that familiar with the fbdev core, but I have tested and works with CONFIG_DRM_FBDEV_EMULATION=y and CONFIG_FB disabled. This config automatically disables all the fbdev drivers that is our goal. Patch 1/4 is just a clean up, patch 2/4 moves a couple of functions out of fbsysfs.o, that are not related to sysfs attributes creation and finally patch 3/4 makes the fbdev split that is mentioned above. Patch 4/4 makes the DRM fbdev emulation depend on the new FB_CORE symbol instead of FB. This could be done as a follow-up but for completeness is also included in this series. Best regards, Javier Javier Martinez Canillas (4): fbdev: Rename fb_*_device() functions names to match what they do fbdev: Move framebuffer_{alloc,release}() functions to fbmem.c fbdev: Split frame buffer support in FB and FB_CORE symbols drm: Make fbdev emulation depend on FB_CORE instead of FB arch/x86/Makefile | 2 +- arch/x86/video/Makefile | 2 +- drivers/gpu/drm/Kconfig | 2 +- drivers/video/console/Kconfig | 2 +- drivers/video/fbdev/Kconfig | 57 +++++++++++++--------- drivers/video/fbdev/core/Makefile | 13 +++-- drivers/video/fbdev/core/fbmem.c | 73 ++++++++++++++++++++++++++-- drivers/video/fbdev/core/fbsysfs.c | 77 +----------------------------- include/linux/fb.h | 18 ++++++- 9 files changed, 134 insertions(+), 112 deletions(-) -- 2.31.1

5 11

[PATCH v2 0/3] Error out if 'pixclock' equals zero
by Zheyu Ma 10 Sep '21

10 Sep '21

Zheyu Ma (3): video: fbdev: asiliantfb: Error out if 'pixclock' equals zero video: fbdev: kyro: Error out if 'pixclock' equals zero video: fbdev: riva: Error out if 'pixclock' equals zero drivers/video/fbdev/asiliantfb.c | 3 +++ drivers/video/fbdev/kyro/fbdev.c | 3 +++ drivers/video/fbdev/riva/fbdev.c | 3 +++ 3 files changed, 9 insertions(+) -- 2.17.6

3 5

[PATCH 0/6] i915: Simplify mmio handling & add new DG2 shadow table
by Matt Roper 10 Sep '21

10 Sep '21

Our uncore MMIO functions for reading/writing registers have become very complicated over time. There's significant macro magic used to generate several nearly-identical functions that only really differ in terms of which platform-specific shadow register table they should check on write operations. We can significantly simplify our MMIO handlers by storing a reference to the current platform's shadow table within the 'struct intel_uncore' the same way we already do for forcewake; this allows us to consolidate the multiple variants of each 'write' function down to just a single 'fwtable' version that gets the shadow table out of the uncore struct rather than hardcoding the name of a specific platform's table. We can do similar consolidation on the MMIO read side by creating a single-entry forcewake table to replace the open-coded range check they had been using previously. The final patch of the series adds a new shadow table for DG2; this becomes quite clean and simple now, given the refactoring in the first five patches. Matt Roper (6): drm/i915/uncore: Convert gen6/gen7 read operations to fwtable drm/i915/uncore: Associate shadow table with uncore drm/i915/uncore: Replace gen8 write functions with general fwtable drm/i915/uncore: Drop gen11/gen12 mmio write handlers drm/i915/uncore: Drop gen11 mmio read handlers drm/i915/dg2: Add DG2-specific shadow register table drivers/gpu/drm/i915/intel_uncore.c | 190 ++++++++++-------- drivers/gpu/drm/i915/intel_uncore.h | 7 + drivers/gpu/drm/i915/selftests/intel_uncore.c | 1 + 3 files changed, 110 insertions(+), 88 deletions(-) -- 2.25.4

2 12

[PATCH] drm/vc4: hdmi: Remove unused struct
by Maxime Ripard 10 Sep '21

10 Sep '21

Commitc7d30623540b ("drm/vc4: hdmi: Remove unused struct") removed the references to the vc4_hdmi_audio_widgets and vc4_hdmi_audio_routes structures, but not the structures themselves resulting in two warnings. Remove it. Fixes: c7d30623540b ("drm/vc4: hdmi: Remove unused struct") Reported-by: kernel test robot <lkp(a)intel.com> Signed-off-by: Maxime Ripard <maxime(a)cerno.tech> --- drivers/gpu/drm/vc4/vc4_hdmi.c | 8 -------- 1 file changed, 8 deletions(-) diff --git a/drivers/gpu/drm/vc4/vc4_hdmi.c b/drivers/gpu/drm/vc4/vc4_hdmi.c index b7dc32a0c9bb..1e2d976e8736 100644 --- a/drivers/gpu/drm/vc4/vc4_hdmi.c +++ b/drivers/gpu/drm/vc4/vc4_hdmi.c @@ -1403,14 +1403,6 @@ static int vc4_hdmi_audio_prepare(struct device *dev, void *data, return 0; } -static const struct snd_soc_dapm_widget vc4_hdmi_audio_widgets[] = { - SND_SOC_DAPM_OUTPUT("TX"), -}; - -static const struct snd_soc_dapm_route vc4_hdmi_audio_routes[] = { - { "TX", NULL, "Playback" }, -}; - static const struct snd_soc_component_driver vc4_hdmi_audio_cpu_dai_comp = { .name = "vc4-hdmi-cpu-dai-component", }; -- 2.31.1

2 1

[PATCH v5] drm/i915: Use Transparent Hugepages when IOMMU is enabled
by Tvrtko Ursulin 10 Sep '21

10 Sep '21

From: Tvrtko Ursulin <tvrtko.ursulin(a)intel.com> Usage of Transparent Hugepages was disabled in 9987da4b5dcf ("drm/i915: Disable THP until we have a GPU read BW W/A"), but since it appears majority of performance regressions reported with an enabled IOMMU can be almost eliminated by turning them on, lets just do that. To err on the side of safety we keep the current default in cases where IOMMU is not active, and only when it is default to the "huge=within_size" mode. Although there probably would be wins to enable them throughout, more extensive testing across benchmarks and platforms would need to be done. With the patch and IOMMU enabled my local testing on a small Skylake part shows OglVSTangent regression being reduced from ~14% (IOMMU on versus IOMMU off) to ~2% (same comparison but with THP on). More detailed testing done in the below referenced Gitlab issue by Eero: Skylake GT4e: Performance drops from enabling IOMMU: 30-35% SynMark CSDof 20-25% Unigine Heaven, MemBW GPU write, SynMark VSTangent ~20% GLB Egypt (1/2 screen window) 10-15% GLB T-Rex (1/2 screen window) 8-10% GfxBench T-Rex, MemBW GPU blit 7-8% SynMark DeferredAA + TerrainFly* + ZBuffer 6-7% GfxBench Manhattan 3.0 + 3.1, SynMark TexMem128 & CSCloth 5-6% GfxBench CarChase, Unigine Valley 3-5% GfxBench Vulkan & GL AztecRuins + ALU2, MemBW GPU texture, SynMark Fill*, Deferred, TerrainPan* 1-2% Most of the other tests With the patch drops become: 20-25% SynMark TexMem* 15-20% GLB Egypt (1/2 screen window) 10-15% GLB T-Rex (1/2 screen window) 4-7% GfxBench T-Rex, GpuTest Triangle 1-8% GfxBench ALU2 (offscreen 1%, onscreen 8%) 3% GfxBench Manhattan 3.0, SynMark CSDof 2-3% Unigine Heaven + Valley, MemBW GPU texture 1-3 GfxBench Manhattan 3.1 + CarChase + Vulkan & GL AztecRuins Broxton: Performance drops from IOMMU, without patch: 30% MemBW GPU write 25% SynMark ZBuffer + Fill* 20% MemBW GPU blit 15% MemBW GPU blend, GpuTest Triangle 10-15% MemBW GPU texture 10% GLB Egypt, Unigine Heaven (had hangs), SynMark TerrainFly* 7-9% GLB T-Rex, GfxBench Manhattan 3.0 + T-Rex, SynMark Deferred* + TexMem* 6-8% GfxBench CarChase, Unigine Valley, SynMark CSCloth + ShMapVsm + TerrainPan* 5-6% GfxBench Manhattan 3.1 + GL AztecRuins, SynMark CSDof + TexFilterTri 2-4% GfxBench ALU2, SynMark DrvRes + GSCloth + ShMapPcf + Batch[0-5] + TexFilterAniso, GpuTest GiMark + 32-bit Julia And with patch: 15-20% MemBW GPU texture 10% SynMark TexMem* 8-9% GLB Egypt (1/2 screen window) 4-5% GLB T-Rex (1/2 screen window) 3-6% GfxBench Manhattan 3.0, GpuTest FurMark, SynMark Deferred + TexFilterTri 3-4% GfxBench Manhattan 3.1 + T-Rex, SynMark VSInstancing 2-4% GpuTest Triangle, SynMark DeferredAA 2-3% Unigine Heaven + Valley 1-3% SynMark Terrain* 1-2% GfxBench CarChase, SynMark TexFilterAniso + ZBuffer Tigerlake-H: 20-25% MemBW GPU texture 15-20% GpuTest Triangle 13-15% SynMark TerrainFly* + DeferredAA + HdrBloom 8-10% GfxBench Manhattan 3.1, SynMark TerrainPan* + DrvRes 6-7% GfxBench Manhattan 3.0, SynMark TexMem* 4-8% GLB onscreen Fill + T-Rex + Egypt (more in onscreen than offscreen versions of T-Rex/Egypt) 4-6% GfxBench CarChase + GLES AztecRuins + ALU2, GpuTest 32-bit Julia, SynMark CSDof + DrvState 3-5% GfxBench T-Rex + Egypt, Unigine Heaven + Valley, GpuTest Plot3D 1-7% Media tests 2-3% MemBW GPU blit 1-3% Most of the rest of 3D tests With the patch: 6-8% MemBW GPU blend => the only regression in these tests (compared to IOMMU without THP) 4-6% SynMark DrvState (not impacted) + HdrBloom (improved) 3-4% GLB T-Rex ~3% GLB Egypt, SynMark DrvRes 1-3% GfxBench T-Rex + Egypt, SynMark TexFilterTri 1-2% GfxBench CarChase + GLES AztecRuins, Unigine Valley, GpuTest Triangle ~1% GfxBench Manhattan 3.0/3.1, Unigine Heaven Perf of several tests actually improved with IOMMU + THP, compared to no IOMMU / no THP: 10-15% SynMark Batch[0-3] 5-10% MemBW GPU texture, SynMark ShMapVsm 3-4% SynMark Fill* + Geom* 2-3% SynMark TexMem512 + CSCloth 1-2% SynMark TexMem128 + DeferredAA As a summary across all platforms, these are the benchmarks where enabling THP on top of IOMMU enabled brings regressions: * Skylake GT4e: 20-25% SynMark TexMem* (whereas all MemBW GPU tests either improve or are not affected) * Broxton J4205: 7% MemBW GPU texture 2-3% SynMark TexMem* * Tigerlake-H: 7% MemBW GPU blend Other benchmarks show either lowering of regressions or improvements. v2: * Add Kconfig dependency to transparent hugepages and some help text. * Move to helper for easier handling of kernel build options. v3: * Drop Kconfig. (Daniel) v4: * Add some benchmark results to commit message. v5: * Add explicit regression summary to commit message. (Eero) References: b901bb89324a ("drm/i915/gemfs: enable THP") References: 9987da4b5dcf ("drm/i915: Disable THP until we have a GPU read BW W/A") References: https://gitlab.freedesktop.org/drm/intel/-/issues/430 Co-developed-by: Chris Wilson <chris(a)chris-wilson.co.uk> Signed-off-by: Chris Wilson <chris(a)chris-wilson.co.uk> Cc: Joonas Lahtinen <joonas.lahtinen(a)linux.intel.com> Cc: Matthew Auld <matthew.auld(a)intel.com> Cc: Eero Tamminen <eero.t.tamminen(a)intel.com> Cc: Tvrtko Ursulin <tvrtko.ursulin(a)intel.com> Cc: Rodrigo Vivi <rodrigo.vivi(a)intel.com> Cc: Daniel Vetter <daniel(a)ffwll.ch> Signed-off-by: Tvrtko Ursulin <tvrtko.ursulin(a)intel.com> Reviewed-by: Rodrigo Vivi <rodrigo.vivi(a)intel.com> # v1 --- drivers/gpu/drm/i915/gem/i915_gemfs.c | 22 +++++++++++++++++++--- 1 file changed, 19 insertions(+), 3 deletions(-) diff --git a/drivers/gpu/drm/i915/gem/i915_gemfs.c b/drivers/gpu/drm/i915/gem/i915_gemfs.c index 5e6e8c91ab38..dbdbdc344d87 100644 --- a/drivers/gpu/drm/i915/gem/i915_gemfs.c +++ b/drivers/gpu/drm/i915/gem/i915_gemfs.c @@ -6,7 +6,6 @@ #include <linux/fs.h> #include <linux/mount.h> -#include <linux/pagemap.h> #include "i915_drv.h" #include "i915_gemfs.h" @@ -15,6 +14,7 @@ int i915_gemfs_init(struct drm_i915_private *i915) { struct file_system_type *type; struct vfsmount *gemfs; + char *opts; type = get_fs_type("tmpfs"); if (!type) @@ -26,10 +26,26 @@ int i915_gemfs_init(struct drm_i915_private *i915) * * One example, although it is probably better with a per-file * control, is selecting huge page allocations ("huge=within_size"). - * Currently unused due to bandwidth issues (slow reads) on Broadwell+. + * However, we only do so to offset the overhead of iommu lookups + * due to bandwidth issues (slow reads) on Broadwell+. */ - gemfs = kern_mount(type); + opts = NULL; + if (intel_vtd_active()) { + if (IS_ENABLED(CONFIG_TRANSPARENT_HUGEPAGE)) { + static char huge_opt[] = "huge=within_size"; /* r/w */ + + opts = huge_opt; + drm_info(&i915->drm, + "Transparent Hugepage mode '%s'\n", + opts); + } else { + drm_notice(&i915->drm, + "Transparent Hugepage support is recommended for optimal performance when IOMMU is enabled!\n"); + } + } + + gemfs = vfs_kern_mount(type, SB_KERNMOUNT, type->name, opts); if (IS_ERR(gemfs)) return PTR_ERR(gemfs); -- 2.30.2

2 2

← Newer
1
...
60
61
62
63
64
65
66
...
82
Older →

Jump to page:

2025

2024

2023

2022

2021

2020

2019

2018

2017

2016

2015

2014

2013

2012

2011

2010

dri-devel September 2021