[Git][xorg-team/driver/xserver-xorg-video-intel][debian-unstable] 15 commits: sna/gen7: Prefer blitter for plain copies on Haswell

Timo Aaltonen pushed to branch debian-unstable at X Strike Force / driver / xserver-xorg-video-intel

Commits:

67f15b36

by Chris Wilson at 2020-08-20T09:30:19+01:00

sna/gen7: Prefer blitter for plain copies on Haswell

Since the clear-residuals security fix on gen7, context switches are
very slow. If X is being used with DRI clients, those clients will
typically be using the 3D engine for themselves and every frame
presented will then be copied by X, causing at least a couple of context
switches per frame. That greatly diminishes throughput, but if we prefer
to use the blitter engine for X, we can mostly keep off the render engine
avoiding the context thrash.

Reported-by: Rafael Ristovski <rafael.ristovski@gmail.com>
Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk>

4e8ae05f

by Chris Wilson at 2020-11-16T15:11:19+00:00

Track ABI changes for ABI_VIDEODRV_VERSION 25.2

Several structs had fields renamed, and so we must update our usage.

Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk>

ad5540f6

by Chris Wilson at 2020-11-16T15:23:28+00:00

More ABI changes for ABI_VIDEODRV_VERSION 25.2

Descending down into the naming changes, onto the next struct in the
chain.

Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk>

9236c582

by Chris Wilson at 2020-11-26T20:19:48+00:00

sna: Defer fbGetWindowPixmap() for sna_composite

Mike Lothian ran into a surprising situation where compScreenUpdate was
calling CompositePicture without a pixmap attached to the destination
Window, and so we found ourselves chasing a NULL PixmapPtr.

  #1  to_sna_from_pixmap (pixmap=0x0) at sna.h:521
  #2  sna_composite (op=<optimized out>, src="" mask=0x0,
      dst=0x55b3346c1d50, src_x=<optimized out>, src_y=<optimized out>, mask_x=0,
      mask_y=0, dst_x=0, dst_y=0, width=3840, height=2160) at sna_composite.c:652
  #3  0x000055b33202c208 in damageComposite (op=<optimized out>,
      pSrc=0x55b3346c1420, pMask=0x0, pDst=0x55b3346c1d50, xSrc=<optimized out>,
      ySrc=<optimized out>, xMask=0, yMask=0, xDst=0, yDst=0, width=3840,
      height=2160) at damage.c:513
  #4  0x000055b33201820c in CompositePicture (op=<optimized out>,
      op@entry=1 '\001', pSrc=pSrc@entry=0x55b3346c1420, pMask=pMask@entry=0x0,
      pDst=pDst@entry=0x55b3346c1d50, xSrc=xSrc@entry=0, ySrc=ySrc@entry=0,
      xMask=0, yMask=0, xDst=0, yDst=0, width=3840, height=2160) at picture.c:1547
  #5  0x000055b331fc85d3 in compWindowUpdateAutomatic (
      pWin=pWin@entry=0x55b3343a6bc0) at compwindow.c:705
  #6  0x000055b331fca029 in compPaintWindowToParent (pWin=pWin@entry=0x55b3343a6bc0)
      at compwindow.c:729
  #7  0x000055b331fc9fbb in compPaintChildrenToWindow (pWin=0x55b333e77b50)
      at compwindow.c:744
  #8  0x000055b331fca59f in compScreenUpdate (pClient=<optimized out>,
      closure=<optimized out>) at compalloc.c:57
  #9  0x000055b331f3abf4 in ProcessWorkQueue () at dixutils.c:536
  #10 0x000055b3320aaa51 in WaitForSomething (are_ready=<optimized out>)
      at WaitFor.c:192
  #11 0x000055b331f361a9 in Dispatch () at dispatch.c:421
  #12 0x000055b331f39cec in dix_main (argc=13, argv=0x7ffcf273f538,
      envp=<optimized out>) at main.c:276
  #13 0x000055b331f247de in main (argc=<optimized out>, argv=<optimized out>,
      envp=<optimized out>) at stubmain.c:34

Fortuitously, that drawable was also fully clipped so that it took an
early exit and so we can hide the segfault by delaying querying the
pixmap until after the clip check.

The ongoing mystery is how we ended up in that state in the first place.

Closes: https://gitlab.freedesktop.org/xorg/driver/xf86-video-intel/-/issues/204
Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk>

a511f22c

by Chris Wilson at 2020-12-15T17:00:20+00:00

sna: Enable TearFree by default on recent HW

When there is ample memory bandwidth and we are not fighting for global
resources, enable TearFree by default. Avoiding tearing is much more
pleasant (for direct rendering where the source itself is not being
synchronized to vblank) at negligible power cost; just doubles the
memory footprint of scanout.

References: https://gitlab.freedesktop.org/drm/intel/-/issues/2799
References: https://gitlab.freedesktop.org/drm/intel/-/issues/2763
Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk>

ab906aa0

by Chris Wilson at 2021-01-10T00:23:28+00:00

sna/gen7: Avoid clear-residuals overhead on all gen7

Since not just Haswell will enjoy clear-residuals, be very careful
before using a potential context switch from DRI clients.

Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk>

31486f40

by Chris Wilson at 2021-01-15T20:59:05+00:00

sna: Always validate userptr upon creation

Since not all memory ranges can be mapped by userptr, in particular those
passed by XShmAttachFD, we need to validate the userptr before use. We
would ideally want to continue to lazily populate the pages as often the
userptr is created but never used, but preventing an EFAULT later is
more important.

In https://patchwork.freedesktop.org/series/33449/ we provided a more
efficient method for probing the userptr on construction while
preserving the lazy population of gup-pages. For now, always follow
userptr with set-domain.

Reported-by: Jinoh Kang <jinoh.kang.kr@gmail.com>
Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk>

bf41b9a8

by Timo Aaltonen at 2022-02-03T16:23:33+02:00

Merge branch 'upstream-unstable' into debian-unstable

68700cd7
by Timo Aaltonen at 2022-02-03T16:24:26+02:00
```
version bump
```
d27240c3
by Timo Aaltonen at 2022-02-03T16:49:07+02:00
```
Build against xserver 21.1.
```
d9cc7b1e
by Timo Aaltonen at 2022-02-03T16:49:53+02:00
```
rules: Disable building with lto.
```

64cccc1b

by Timo Aaltonen at 2022-02-09T12:38:36+02:00

control: Bump debhelper-compat to 13, policy to 4.6.0.

c9e7af22
by Timo Aaltonen at 2022-02-09T12:50:15+02:00
```
Update README.source for salsa.
```
ee7e65dd
by Timo Aaltonen at 2022-02-09T12:51:28+02:00
```
watch: Update upstream git url.
```
dbcac90e
by Timo Aaltonen at 2022-02-09T12:51:57+02:00
```
release to sid
```

Changes:

debian/README.source

@@ -25,9 +25,9 @@ Guide To The X Strike Force Packages
  ------------------------------------
  The X Strike Force team maintains X packages in git repositories on
 -git.debian.org in the pkg-xorg subdirectory. Most upstream packages
 +salsa.debian.org in the xorg-team subdirectory. Most upstream packages
  are actually maintained in git repositories as well, so they often
 -just need to be pulled into git.debian.org in a "upstream-*" branch.
 +just need to be pulled into salsa.debian.org in a "upstream-*" branch.
  Otherwise, the upstream sources are manually installed in the Debian
  git repository.

debian/changelog

 +xserver-xorg-video-intel (2:2.99.917+git20210115-1) unstable; urgency=medium
++
 +  * New upstream snapshot.
 +  * Build against xserver 21.1.
 +  * rules: Disable building with lto.
 +  * control: Bump debhelper-compat to 13, policy to 4.6.0.
 +  * Update README.source for salsa.
 +  * watch: Update upstream git url.
++
 + -- Timo Aaltonen <tjaalton@debian.org>  Wed, 09 Feb 2022 12:51:46 +0200
++
  xserver-xorg-video-intel (2:2.99.917+git20200714-2) unstable; urgency=medium
    [ Timo Aaltonen ]

debian/control

@@ -8,9 +8,9 @@ Uploaders:
   maximilian attems <maks@debian.org>,
   Vincent Cheng <vcheng@debian.org>,
  Build-Depends:
 - debhelper-compat (= 12),
 + debhelper-compat (= 13),
   pkg-config,
 - xserver-xorg-dev (>= 2:1.11.99.901),
 + xserver-xorg-dev (>= 2:21.1),
   x11proto-dev,
   libx11-dev,
   libdrm-dev (>= 2.4.23),
@@ -42,7 +42,7 @@ Build-Depends:
  # tests, benchmarks:
  # libcairo2-dev,
  # libpng12-dev,
 -Standards-Version: 4.5.0
 +Standards-Version: 4.6.0
  Vcs-Git: https://salsa.debian.org/xorg-team/driver/xserver-xorg-video-intel.git
  Vcs-Browser: https://salsa.debian.org/xorg-team/driver/xserver-xorg-video-intel
  Homepage: https://www.x.org/

debian/rules

@@ -6,6 +6,8 @@ BUILDER        := $(shell dpkg-parsechangelog | awk -F': ' '/^Maintainer: / {pri
  DEB_HOST_ARCH_OS ?= $(shell dpkg-architecture -qDEB_HOST_ARCH_OS)
 +export DEB_BUILD_MAINT_OPTIONS = optimize=-lto
++
  ifeq ($(DEB_HOST_ARCH_OS), linux)
  valgrind = --enable-valgrind
  else

debian/watch

 -#git=git://anongit.freedesktop.org/xorg/driver/xf86-video-intel
 +#git=https://gitlab.freedesktop.org/xorg/driver/xf86-video-intel
  version=3
  https://xorg.freedesktop.org/releases/individual/driver/ xf86-video-intel-(.*)\.tar\.gz

src/compat-api.h

@@ -260,4 +260,12 @@ static inline void FreePixmap(PixmapPtr pixmap)
  #define RemoveNotifyFd(fd) RemoveGeneralSocket(fd)
  #endif
 +#if ABI_VIDEODRV_VERSION >= SET_ABI_VERSION(25, 2)
 +#define PixmapDirtyDst(d) ((d)->secondary_dst)
 +#define PixmapDirtyPrimary(d) (PixmapDirtyDst(d)->primary_pixmap)
 +#else
 +#define PixmapDirtyDst(d) ((d)->slave_dst)
 +#define PixmapDirtyPrimary(d) (PixmapDirtyDst(d)->master_pixmap)
 +#endif
++
  #endif

src/sna/gen7_render.c

@@ -2957,42 +2957,10 @@ prefer_blt_copy(struct sna *sna,
  		struct kgem_bo *dst_bo,
  		unsigned flags)
+ {
 -	if (sna->kgem.mode == KGEM_BLT)
 -		return true;
+-
 -	assert((flags & COPY_SYNC) == 0);
+-
 -	if (untiled_tlb_miss(src_bo) ||
 -	    untiled_tlb_miss(dst_bo))
 -		return true;
+-
 -	if (flags & COPY_DRI && !sna->kgem.has_semaphores)
 -		return false;
+-
 -	if (force_blt_ring(sna, dst_bo, src_bo))
 -		return true;
+-
 -	if ((flags & COPY_SMALL ||
 -	     (sna->render_state.gt < 3 && src_bo == dst_bo)) &&
 -            can_switch_to_blt(sna, dst_bo, flags))
 -		return true;
+-
 -	if (kgem_bo_is_render(dst_bo) ||
 -	    kgem_bo_is_render(src_bo))
 -		return false;
+-
 -	if (flags & COPY_LAST &&
 -	    sna->render_state.gt < 3 &&
 -            can_switch_to_blt(sna, dst_bo, flags))
 -		return true;
+-
 -	if (prefer_render_ring(sna, dst_bo))
 -		return false;
+-
 -	if (!prefer_blt_ring(sna, dst_bo, flags))
 +	if (sna->kgem.ring != KGEM_BLT)
  		return false;
 -	return prefer_blt_bo(sna, src_bo, dst_bo);
 +	return true; /* avoid clear-residuals context overhead */
+ }
  static bool

src/sna/kgem.c

@@ -520,7 +520,7 @@ static bool gem_set_caching(int fd, uint32_t handle, int caching)
  	return do_ioctl(fd, LOCAL_IOCTL_I915_GEM_SET_CACHING, &arg) == 0;
+ }
 -static uint32_t gem_userptr(int fd, void *ptr, int size, int read_only)
 +static uint32_t gem_userptr(int fd, void *ptr, size_t size, int read_only)
+ {
  	struct local_i915_gem_userptr arg;
@@ -1944,6 +1944,19 @@ static uint64_t get_gtt_size(int fd)
  	return aperture.aper_size;
+ }
 +static int get_gtt_type(int fd)
 +{
 +        struct drm_i915_getparam p;
 +        int val = 0;
++
 +        memset(&p, 0, sizeof(p));
 +        p.param = I915_PARAM_HAS_ALIASING_PPGTT;
 +        p.value = &val;
++
 +	drmIoctl(fd, DRM_IOCTL_I915_GETPARAM, &p);
 +	return val;
 +}
++
  void kgem_init(struct kgem *kgem, int fd, struct pci_device *dev, unsigned gen)
+ {
  	size_t totalram;
@@ -2108,6 +2121,8 @@ void kgem_init(struct kgem *kgem, int fd, struct pci_device *dev, unsigned gen)
  	     !DBG_NO_CPU && (kgem->has_llc | kgem->has_userptr | kgem->has_caching),
  	     kgem->has_llc, kgem->has_caching, kgem->has_userptr));
 +	kgem->has_full_ppgtt = get_gtt_type(fd) > 1;
++
  	gtt_size = get_gtt_size(fd);
  	kgem->aperture_total = gtt_size;
  	kgem->aperture_high = gtt_size * 3/4;
@@ -7016,6 +7031,30 @@ uint32_t kgem_bo_flink(struct kgem *kgem, struct kgem_bo *bo)
  	return flink.name;
+ }
 +static bool probe(struct kgem *kgem, uint32_t handle)
 +{
 +	struct drm_i915_gem_set_domain arg = {
 +		.handle = handle,
 +		.read_domains = I915_GEM_DOMAIN_CPU,
 +	};
++
 +	return do_ioctl(kgem->fd, DRM_IOCTL_I915_GEM_SET_DOMAIN, &arg) == 0;
 +}
++
 +static uint32_t probe_userptr(struct kgem *kgem,
 +			      void *ptr, size_t size, int read_only)
 +{
 +	uint32_t handle;
++
 +	handle = gem_userptr(kgem->fd, ptr, size, read_only);
 +	if (handle && !probe(kgem, handle)) {
 +		gem_close(kgem->fd, handle);
 +		handle = 0;
 +	}
++
 +	return handle;
 +}
++
  struct kgem_bo *kgem_create_map(struct kgem *kgem,
  				void *ptr, uint32_t size,
  				bool read_only)
@@ -7038,30 +7077,16 @@ struct kgem_bo *kgem_create_map(struct kgem *kgem,
  	last_page &= ~(uintptr_t)(PAGE_SIZE-1);
  	assert(last_page > first_page);
 -	handle = gem_userptr(kgem->fd,
 -			     (void *)first_page, last_page-first_page,
 -			     read_only);
 +	handle = probe_userptr(kgem,
 +			       (void *)first_page, last_page-first_page,
 +			       read_only);
 +	if (handle == 0 && read_only && kgem->has_wc_mmap)
 +		handle = probe_userptr(kgem,
 +				       (void *)first_page, last_page-first_page,
 +				       false);
  	if (handle == 0) {
 -		if (read_only && kgem->has_wc_mmap) {
 -			struct drm_i915_gem_set_domain set_domain;
+-
 -			handle = gem_userptr(kgem->fd,
 -					     (void *)first_page, last_page-first_page,
 -					     false);
+-
 -			VG_CLEAR(set_domain);
 -			set_domain.handle = handle;
 -			set_domain.read_domains = I915_GEM_DOMAIN_GTT;
 -			set_domain.write_domain = 0;
 -			if (do_ioctl(kgem->fd, DRM_IOCTL_I915_GEM_SET_DOMAIN, &set_domain)) {
 -				gem_close(kgem->fd, handle);
 -				handle = 0;
 -			}
 -		}
 -		if (handle == 0) {
 -			DBG(("%s: import failed, errno=%d\n", __FUNCTION__, errno));
 -			return NULL;
 -		}
 +		DBG(("%s: import failed, errno=%d\n", __FUNCTION__, errno));
 +		return NULL;
+ 	}
  	bo = __kgem_bo_alloc(handle, (last_page - first_page) / PAGE_SIZE);

src/sna/kgem.h

@@ -191,6 +191,7 @@ struct kgem {
  	uint32_t has_pinned_batches :1;
  	uint32_t has_caching :1;
  	uint32_t has_coherent_mmap_gtt :1;
 +	uint32_t has_full_ppgtt :1;
  	uint32_t has_llc :1;
  	uint32_t has_wt :1;
  	uint32_t has_no_reloc :1;

src/sna/sna_accel.c

@@ -17696,7 +17696,7 @@ static void sna_accel_post_damage(struct sna *sna)
  #else
  		src = dirty->src;
  #endif
 -		dst = dirty->slave_dst->master_pixmap;
 +		dst = PixmapDirtyPrimary(dirty);
  		region.extents.x1 = dirty->x;
  		region.extents.x2 = dirty->x + dst->drawable.width;
@@ -17723,7 +17723,7 @@ static void sna_accel_post_damage(struct sna *sna)
  		dy += dirty->dst_y;
  #endif
  		RegionTranslate(&region, dx, dy);
 -		DamageRegionAppend(&dirty->slave_dst->drawable, &region);
 +		DamageRegionAppend(&PixmapDirtyDst(dirty)->drawable, &region);
  		DBG(("%s: slave:  ((%d, %d), (%d, %d))x%d\n", __FUNCTION__,
  		     region.extents.x1, region.extents.y1,
@@ -17800,7 +17800,7 @@ fallback:
  			kgem_bo_sync__gtt(&sna->kgem, __sna_pixmap_get_bo(dst));
+ 		}
 -		DamageRegionProcessPending(&dirty->slave_dst->drawable);
 +		DamageRegionProcessPending(&PixmapDirtyDst(dirty)->drawable);
  skip:
  		RegionUninit(&region);
  		DamageEmpty(dirty->damage);

src/sna/sna_composite.c

@@ -649,15 +649,15 @@ sna_composite(CARD8 op,
  	      CARD16 width, CARD16 height)
+ {
  	PixmapPtr pixmap = get_drawable_pixmap(dst->pDrawable);
 -	struct sna *sna = to_sna_from_pixmap(pixmap);
  	struct sna_pixmap *priv;
  	struct sna_composite_op tmp;
  	RegionRec region;
 +	struct sna *sna;
  	int dx, dy;
  	DBG(("%s(pixmap=%ld, op=%d, src="" %d), mask=%ld+(%d, %d), dst=%ld+(%d, %d)+(%d, %d), size=(%d, %d)\n",
  	     __FUNCTION__,
 -	     pixmap->drawable.serialNumber, op,
 +	     pixmap ? pixmap->drawable.serialNumber : 0, op,
  	     get_picture_id(src), src_x, src_y,
  	     get_picture_id(mask), mask_x, mask_y,
  	     get_picture_id(dst), dst_x, dst_y,
@@ -672,8 +672,7 @@ sna_composite(CARD8 op,
  	if (op == PictOpClear) {
  		DBG(("%s: discarding source and mask for clear\n", __FUNCTION__));
  		mask = NULL;
 -		if (sna->clear)
 -			src = sna->clear;
 +		src = NULL;
+ 	}
  	if (!sna_compute_composite_region(&region,
@@ -694,11 +693,6 @@ sna_composite(CARD8 op,
  	if (NO_COMPOSITE)
  		goto fallback;
 -	if (wedged(sna)) {
 -		DBG(("%s: fallback -- wedged\n", __FUNCTION__));
 -		goto fallback;
 -	}
+-
  	if (!can_render_to_picture(dst)) {
  		DBG(("%s: fallback due to unhandled picture\n", __FUNCTION__));
  		goto fallback;
@@ -711,6 +705,15 @@ sna_composite(CARD8 op,
  		goto fallback;
+ 	}
 +	sna = to_sna_from_pixmap(pixmap);
 +	if (wedged(sna)) {
 +		DBG(("%s: fallback -- wedged\n", __FUNCTION__));
 +		goto fallback;
 +	}
++
 +	if (op == PictOpClear)
 +		src = sna->clear;
++
  	if (use_cpu(pixmap, priv, op, width, height) &&
  	    !picture_is_gpu(sna, src, PREFER_GPU_RENDER) &&
  	    !picture_is_gpu(sna, mask, PREFER_GPU_RENDER)) {

src/sna/sna_driver.c

@@ -459,7 +459,18 @@ static bool enable_tear_free(struct sna *sna)
  	if (sna->flags & SNA_LINEAR_FB)
  		return false;
 -	/* Under certain conditions, we should enable TearFree by default,
 +	/*
 +	 * On recent HW, where the display surfaces are in a seperate GTT
 +	 * to userspace, there is much less contention on global resources
 +	 * and also we can assume there is much more memory bandwidth
 +	 * available (i.e. gen8+). This HW should rarely be under such
 +	 * constaints as to need to disable TearFree, so enable by default.
 +	 */
 +	if (sna->kgem.has_full_ppgtt)
 +		return true;
++
 +	/*
 +	 * Under certain conditions, we should enable TearFree by default,
  	 * for example when the hardware requires pageflipping to run within
  	 * its power/performance budget.
  	 */

src/uxa/intel_driver.c

@@ -624,30 +624,30 @@ redisplay_dirty(ScreenPtr screen, PixmapDirtyUpdatePtr dirty)
  	RegionRec pixregion;
  	int was_blocked;
 -	PixmapRegionInit(&pixregion, dirty->slave_dst->master_pixmap);
 +	PixmapRegionInit(&pixregion, PixmapDirtyPrimary(dirty));
  	RegionTranslate(&pixregion, dirty->x, dirty->y);
  	RegionIntersect(&pixregion, &pixregion, DamageRegion(dirty->damage));
  	RegionTranslate(&pixregion, -dirty->x, -dirty->y);
  	was_blocked = RegionNil(&pixregion);
 -	DamageRegionAppend(&dirty->slave_dst->drawable, &pixregion);
 +	DamageRegionAppend(&PixmapDirtyDst(dirty)->drawable, &pixregion);
  	RegionUninit(&pixregion);
  	if (was_blocked)
  		return;
 -	PixmapRegionInit(&pixregion, dirty->slave_dst->master_pixmap);
 +	PixmapRegionInit(&pixregion, PixmapDirtyPrimary(dirty));
  	PixmapSyncDirtyHelper(dirty, &pixregion);
  	RegionUninit(&pixregion);
          intel_flush(intel);
  	if (!intel->has_prime_vmap_flush) {
 -		drm_intel_bo *bo = intel_uxa_get_pixmap_bo(dirty->slave_dst->master_pixmap);
 +		drm_intel_bo *bo = intel_uxa_get_pixmap_bo(PixmapDirtyPrimary(dirty));
  		was_blocked = xf86BlockSIGIO();
  		drm_intel_bo_map(bo, FALSE);
  		drm_intel_bo_unmap(bo);
  		xf86UnblockSIGIO(was_blocked);
+ 	}
 -	DamageRegionProcessPending(&dirty->slave_dst->drawable);
 +	DamageRegionProcessPending(&PixmapDirtyDst(dirty)->drawable);
  	return;
+ }