All of lore.kernel.org
 help / color / mirror / Atom feed
* [Intel-gfx] [PATCH 01/21] Revert "drm/i915/gem: Drop relocation slowpath"
@ 2020-03-24 16:51 Maarten Lankhorst
  2020-03-24 16:51 ` [Intel-gfx] [PATCH 02/21] drm/i915: Add an implementation for i915_gem_ww_ctx locking, v2 Maarten Lankhorst
                   ` (22 more replies)
  0 siblings, 23 replies; 25+ messages in thread
From: Maarten Lankhorst @ 2020-03-24 16:51 UTC (permalink / raw)
  To: intel-gfx; +Cc: Matthew Auld

This reverts commit 7dc8f1143778 ("drm/i915/gem: Drop relocation
slowpath"). We need the slowpath relocation for taking ww-mutex
inside the page fault handler, and we will take this mutex when
pinning all objects.

Cc: Chris Wilson <chris@chris-wilson.co.uk>
Cc: Matthew Auld <matthew.auld@intel.com>
Signed-off-by: Maarten Lankhorst <maarten.lankhorst@linux.intel.com>
---
 .../gpu/drm/i915/gem/i915_gem_execbuffer.c    | 239 +++++++++++++++++-
 1 file changed, 235 insertions(+), 4 deletions(-)

diff --git a/drivers/gpu/drm/i915/gem/i915_gem_execbuffer.c b/drivers/gpu/drm/i915/gem/i915_gem_execbuffer.c
index 6b3013d20851..248b0889416b 100644
--- a/drivers/gpu/drm/i915/gem/i915_gem_execbuffer.c
+++ b/drivers/gpu/drm/i915/gem/i915_gem_execbuffer.c
@@ -1480,7 +1480,9 @@ static int eb_relocate_vma(struct i915_execbuffer *eb, struct eb_vma *ev)
 		 * we would try to acquire the struct mutex again. Obviously
 		 * this is bad and so lockdep complains vehemently.
 		 */
-		copied = __copy_from_user(r, urelocs, count * sizeof(r[0]));
+		pagefault_disable();
+		copied = __copy_from_user_inatomic(r, urelocs, count * sizeof(r[0]));
+		pagefault_enable();
 		if (unlikely(copied)) {
 			remain = -EFAULT;
 			goto out;
@@ -1530,6 +1532,236 @@ static int eb_relocate_vma(struct i915_execbuffer *eb, struct eb_vma *ev)
 	return remain;
 }
 
+static int
+eb_relocate_vma_slow(struct i915_execbuffer *eb, struct eb_vma *ev)
+{
+	const struct drm_i915_gem_exec_object2 *entry = ev->exec;
+	struct drm_i915_gem_relocation_entry *relocs =
+		u64_to_ptr(typeof(*relocs), entry->relocs_ptr);
+	unsigned int i;
+	int err;
+
+	for (i = 0; i < entry->relocation_count; i++) {
+		u64 offset = eb_relocate_entry(eb, ev, &relocs[i]);
+
+		if ((s64)offset < 0) {
+			err = (int)offset;
+			goto err;
+		}
+	}
+	err = 0;
+err:
+	reloc_cache_reset(&eb->reloc_cache);
+	return err;
+}
+
+static int check_relocations(const struct drm_i915_gem_exec_object2 *entry)
+{
+	const char __user *addr, *end;
+	unsigned long size;
+	char __maybe_unused c;
+
+	size = entry->relocation_count;
+	if (size == 0)
+		return 0;
+
+	if (size > N_RELOC(ULONG_MAX))
+		return -EINVAL;
+
+	addr = u64_to_user_ptr(entry->relocs_ptr);
+	size *= sizeof(struct drm_i915_gem_relocation_entry);
+	if (!access_ok(addr, size))
+		return -EFAULT;
+
+	end = addr + size;
+	for (; addr < end; addr += PAGE_SIZE) {
+		int err = __get_user(c, addr);
+		if (err)
+			return err;
+	}
+	return __get_user(c, end - 1);
+}
+
+static int eb_copy_relocations(const struct i915_execbuffer *eb)
+{
+	struct drm_i915_gem_relocation_entry *relocs;
+	const unsigned int count = eb->buffer_count;
+	unsigned int i;
+	int err;
+
+	for (i = 0; i < count; i++) {
+		const unsigned int nreloc = eb->exec[i].relocation_count;
+		struct drm_i915_gem_relocation_entry __user *urelocs;
+		unsigned long size;
+		unsigned long copied;
+
+		if (nreloc == 0)
+			continue;
+
+		err = check_relocations(&eb->exec[i]);
+		if (err)
+			goto err;
+
+		urelocs = u64_to_user_ptr(eb->exec[i].relocs_ptr);
+		size = nreloc * sizeof(*relocs);
+
+		relocs = kvmalloc_array(size, 1, GFP_KERNEL);
+		if (!relocs) {
+			err = -ENOMEM;
+			goto err;
+		}
+
+		/* copy_from_user is limited to < 4GiB */
+		copied = 0;
+		do {
+			unsigned int len =
+				min_t(u64, BIT_ULL(31), size - copied);
+
+			if (__copy_from_user((char *)relocs + copied,
+					     (char __user *)urelocs + copied,
+					     len))
+				goto end;
+
+			copied += len;
+		} while (copied < size);
+
+		/*
+		 * As we do not update the known relocation offsets after
+		 * relocating (due to the complexities in lock handling),
+		 * we need to mark them as invalid now so that we force the
+		 * relocation processing next time. Just in case the target
+		 * object is evicted and then rebound into its old
+		 * presumed_offset before the next execbuffer - if that
+		 * happened we would make the mistake of assuming that the
+		 * relocations were valid.
+		 */
+		if (!user_access_begin(urelocs, size))
+			goto end;
+
+		for (copied = 0; copied < nreloc; copied++)
+			unsafe_put_user(-1,
+					&urelocs[copied].presumed_offset,
+					end_user);
+		user_access_end();
+
+		eb->exec[i].relocs_ptr = (uintptr_t)relocs;
+	}
+
+	return 0;
+
+end_user:
+	user_access_end();
+end:
+	kvfree(relocs);
+	err = -EFAULT;
+err:
+	while (i--) {
+		relocs = u64_to_ptr(typeof(*relocs), eb->exec[i].relocs_ptr);
+		if (eb->exec[i].relocation_count)
+			kvfree(relocs);
+	}
+	return err;
+}
+
+static int eb_prefault_relocations(const struct i915_execbuffer *eb)
+{
+	const unsigned int count = eb->buffer_count;
+	unsigned int i;
+
+	for (i = 0; i < count; i++) {
+		int err;
+
+		err = check_relocations(&eb->exec[i]);
+		if (err)
+			return err;
+	}
+
+	return 0;
+}
+
+static noinline int eb_relocate_slow(struct i915_execbuffer *eb)
+{
+	bool have_copy = false;
+	struct eb_vma *ev;
+	int err = 0;
+
+repeat:
+	if (signal_pending(current)) {
+		err = -ERESTARTSYS;
+		goto out;
+	}
+
+	/*
+	 * We take 3 passes through the slowpatch.
+	 *
+	 * 1 - we try to just prefault all the user relocation entries and
+	 * then attempt to reuse the atomic pagefault disabled fast path again.
+	 *
+	 * 2 - we copy the user entries to a local buffer here outside of the
+	 * local and allow ourselves to wait upon any rendering before
+	 * relocations
+	 *
+	 * 3 - we already have a local copy of the relocation entries, but
+	 * were interrupted (EAGAIN) whilst waiting for the objects, try again.
+	 */
+	if (!err) {
+		err = eb_prefault_relocations(eb);
+	} else if (!have_copy) {
+		err = eb_copy_relocations(eb);
+		have_copy = err == 0;
+	} else {
+		cond_resched();
+		err = 0;
+	}
+	if (err)
+		goto out;
+
+	list_for_each_entry(ev, &eb->relocs, reloc_link) {
+		if (!have_copy) {
+			pagefault_disable();
+			err = eb_relocate_vma(eb, ev);
+			pagefault_enable();
+			if (err)
+				goto repeat;
+		} else {
+			err = eb_relocate_vma_slow(eb, ev);
+			if (err)
+				goto err;
+		}
+	}
+
+	/*
+	 * Leave the user relocations as are, this is the painfully slow path,
+	 * and we want to avoid the complication of dropping the lock whilst
+	 * having buffers reserved in the aperture and so causing spurious
+	 * ENOSPC for random operations.
+	 */
+
+err:
+	if (err == -EAGAIN)
+		goto repeat;
+
+out:
+	if (have_copy) {
+		const unsigned int count = eb->buffer_count;
+		unsigned int i;
+
+		for (i = 0; i < count; i++) {
+			const struct drm_i915_gem_exec_object2 *entry =
+				&eb->exec[i];
+			struct drm_i915_gem_relocation_entry *relocs;
+
+			if (!entry->relocation_count)
+				continue;
+
+			relocs = u64_to_ptr(typeof(*relocs), entry->relocs_ptr);
+			kvfree(relocs);
+		}
+	}
+
+	return err;
+}
+
 static int eb_relocate(struct i915_execbuffer *eb)
 {
 	int err;
@@ -1549,9 +1781,8 @@ static int eb_relocate(struct i915_execbuffer *eb)
 		struct eb_vma *ev;
 
 		list_for_each_entry(ev, &eb->relocs, reloc_link) {
-			err = eb_relocate_vma(eb, ev);
-			if (err)
-				return err;
+			if (eb_relocate_vma(eb, ev))
+				return eb_relocate_slow(eb);
 		}
 	}
 
-- 
2.25.1

_______________________________________________
Intel-gfx mailing list
Intel-gfx@lists.freedesktop.org
https://lists.freedesktop.org/mailman/listinfo/intel-gfx

^ permalink raw reply related	[flat|nested] 25+ messages in thread
* [Intel-gfx] [PATCH 01/21] Revert "drm/i915/gem: Drop relocation slowpath"
@ 2020-03-26 11:49 Maarten Lankhorst
  2020-03-26 11:49 ` [Intel-gfx] [PATCH 12/21] drm/i915: Convert i915_gem_object/client_blt.c to use ww locking as well, v2 Maarten Lankhorst
  0 siblings, 1 reply; 25+ messages in thread
From: Maarten Lankhorst @ 2020-03-26 11:49 UTC (permalink / raw)
  To: intel-gfx; +Cc: Matthew Auld

This reverts commit 7dc8f1143778 ("drm/i915/gem: Drop relocation
slowpath"). We need the slowpath relocation for taking ww-mutex
inside the page fault handler, and we will take this mutex when
pinning all objects.

Cc: Chris Wilson <chris@chris-wilson.co.uk>
Cc: Matthew Auld <matthew.auld@intel.com>
Signed-off-by: Maarten Lankhorst <maarten.lankhorst@linux.intel.com>
---
 .../gpu/drm/i915/gem/i915_gem_execbuffer.c    | 239 +++++++++++++++++-
 1 file changed, 235 insertions(+), 4 deletions(-)

diff --git a/drivers/gpu/drm/i915/gem/i915_gem_execbuffer.c b/drivers/gpu/drm/i915/gem/i915_gem_execbuffer.c
index c643eec4dca0..b451aac69211 100644
--- a/drivers/gpu/drm/i915/gem/i915_gem_execbuffer.c
+++ b/drivers/gpu/drm/i915/gem/i915_gem_execbuffer.c
@@ -1480,7 +1480,9 @@ static int eb_relocate_vma(struct i915_execbuffer *eb, struct eb_vma *ev)
 		 * we would try to acquire the struct mutex again. Obviously
 		 * this is bad and so lockdep complains vehemently.
 		 */
-		copied = __copy_from_user(r, urelocs, count * sizeof(r[0]));
+		pagefault_disable();
+		copied = __copy_from_user_inatomic(r, urelocs, count * sizeof(r[0]));
+		pagefault_enable();
 		if (unlikely(copied)) {
 			remain = -EFAULT;
 			goto out;
@@ -1530,6 +1532,236 @@ static int eb_relocate_vma(struct i915_execbuffer *eb, struct eb_vma *ev)
 	return remain;
 }
 
+static int
+eb_relocate_vma_slow(struct i915_execbuffer *eb, struct eb_vma *ev)
+{
+	const struct drm_i915_gem_exec_object2 *entry = ev->exec;
+	struct drm_i915_gem_relocation_entry *relocs =
+		u64_to_ptr(typeof(*relocs), entry->relocs_ptr);
+	unsigned int i;
+	int err;
+
+	for (i = 0; i < entry->relocation_count; i++) {
+		u64 offset = eb_relocate_entry(eb, ev, &relocs[i]);
+
+		if ((s64)offset < 0) {
+			err = (int)offset;
+			goto err;
+		}
+	}
+	err = 0;
+err:
+	reloc_cache_reset(&eb->reloc_cache);
+	return err;
+}
+
+static int check_relocations(const struct drm_i915_gem_exec_object2 *entry)
+{
+	const char __user *addr, *end;
+	unsigned long size;
+	char __maybe_unused c;
+
+	size = entry->relocation_count;
+	if (size == 0)
+		return 0;
+
+	if (size > N_RELOC(ULONG_MAX))
+		return -EINVAL;
+
+	addr = u64_to_user_ptr(entry->relocs_ptr);
+	size *= sizeof(struct drm_i915_gem_relocation_entry);
+	if (!access_ok(addr, size))
+		return -EFAULT;
+
+	end = addr + size;
+	for (; addr < end; addr += PAGE_SIZE) {
+		int err = __get_user(c, addr);
+		if (err)
+			return err;
+	}
+	return __get_user(c, end - 1);
+}
+
+static int eb_copy_relocations(const struct i915_execbuffer *eb)
+{
+	struct drm_i915_gem_relocation_entry *relocs;
+	const unsigned int count = eb->buffer_count;
+	unsigned int i;
+	int err;
+
+	for (i = 0; i < count; i++) {
+		const unsigned int nreloc = eb->exec[i].relocation_count;
+		struct drm_i915_gem_relocation_entry __user *urelocs;
+		unsigned long size;
+		unsigned long copied;
+
+		if (nreloc == 0)
+			continue;
+
+		err = check_relocations(&eb->exec[i]);
+		if (err)
+			goto err;
+
+		urelocs = u64_to_user_ptr(eb->exec[i].relocs_ptr);
+		size = nreloc * sizeof(*relocs);
+
+		relocs = kvmalloc_array(size, 1, GFP_KERNEL);
+		if (!relocs) {
+			err = -ENOMEM;
+			goto err;
+		}
+
+		/* copy_from_user is limited to < 4GiB */
+		copied = 0;
+		do {
+			unsigned int len =
+				min_t(u64, BIT_ULL(31), size - copied);
+
+			if (__copy_from_user((char *)relocs + copied,
+					     (char __user *)urelocs + copied,
+					     len))
+				goto end;
+
+			copied += len;
+		} while (copied < size);
+
+		/*
+		 * As we do not update the known relocation offsets after
+		 * relocating (due to the complexities in lock handling),
+		 * we need to mark them as invalid now so that we force the
+		 * relocation processing next time. Just in case the target
+		 * object is evicted and then rebound into its old
+		 * presumed_offset before the next execbuffer - if that
+		 * happened we would make the mistake of assuming that the
+		 * relocations were valid.
+		 */
+		if (!user_access_begin(urelocs, size))
+			goto end;
+
+		for (copied = 0; copied < nreloc; copied++)
+			unsafe_put_user(-1,
+					&urelocs[copied].presumed_offset,
+					end_user);
+		user_access_end();
+
+		eb->exec[i].relocs_ptr = (uintptr_t)relocs;
+	}
+
+	return 0;
+
+end_user:
+	user_access_end();
+end:
+	kvfree(relocs);
+	err = -EFAULT;
+err:
+	while (i--) {
+		relocs = u64_to_ptr(typeof(*relocs), eb->exec[i].relocs_ptr);
+		if (eb->exec[i].relocation_count)
+			kvfree(relocs);
+	}
+	return err;
+}
+
+static int eb_prefault_relocations(const struct i915_execbuffer *eb)
+{
+	const unsigned int count = eb->buffer_count;
+	unsigned int i;
+
+	for (i = 0; i < count; i++) {
+		int err;
+
+		err = check_relocations(&eb->exec[i]);
+		if (err)
+			return err;
+	}
+
+	return 0;
+}
+
+static noinline int eb_relocate_slow(struct i915_execbuffer *eb)
+{
+	bool have_copy = false;
+	struct eb_vma *ev;
+	int err = 0;
+
+repeat:
+	if (signal_pending(current)) {
+		err = -ERESTARTSYS;
+		goto out;
+	}
+
+	/*
+	 * We take 3 passes through the slowpatch.
+	 *
+	 * 1 - we try to just prefault all the user relocation entries and
+	 * then attempt to reuse the atomic pagefault disabled fast path again.
+	 *
+	 * 2 - we copy the user entries to a local buffer here outside of the
+	 * local and allow ourselves to wait upon any rendering before
+	 * relocations
+	 *
+	 * 3 - we already have a local copy of the relocation entries, but
+	 * were interrupted (EAGAIN) whilst waiting for the objects, try again.
+	 */
+	if (!err) {
+		err = eb_prefault_relocations(eb);
+	} else if (!have_copy) {
+		err = eb_copy_relocations(eb);
+		have_copy = err == 0;
+	} else {
+		cond_resched();
+		err = 0;
+	}
+	if (err)
+		goto out;
+
+	list_for_each_entry(ev, &eb->relocs, reloc_link) {
+		if (!have_copy) {
+			pagefault_disable();
+			err = eb_relocate_vma(eb, ev);
+			pagefault_enable();
+			if (err)
+				goto repeat;
+		} else {
+			err = eb_relocate_vma_slow(eb, ev);
+			if (err)
+				goto err;
+		}
+	}
+
+	/*
+	 * Leave the user relocations as are, this is the painfully slow path,
+	 * and we want to avoid the complication of dropping the lock whilst
+	 * having buffers reserved in the aperture and so causing spurious
+	 * ENOSPC for random operations.
+	 */
+
+err:
+	if (err == -EAGAIN)
+		goto repeat;
+
+out:
+	if (have_copy) {
+		const unsigned int count = eb->buffer_count;
+		unsigned int i;
+
+		for (i = 0; i < count; i++) {
+			const struct drm_i915_gem_exec_object2 *entry =
+				&eb->exec[i];
+			struct drm_i915_gem_relocation_entry *relocs;
+
+			if (!entry->relocation_count)
+				continue;
+
+			relocs = u64_to_ptr(typeof(*relocs), entry->relocs_ptr);
+			kvfree(relocs);
+		}
+	}
+
+	return err;
+}
+
 static int eb_relocate(struct i915_execbuffer *eb)
 {
 	int err;
@@ -1549,9 +1781,8 @@ static int eb_relocate(struct i915_execbuffer *eb)
 		struct eb_vma *ev;
 
 		list_for_each_entry(ev, &eb->relocs, reloc_link) {
-			err = eb_relocate_vma(eb, ev);
-			if (err)
-				return err;
+			if (eb_relocate_vma(eb, ev))
+				return eb_relocate_slow(eb);
 		}
 	}
 
-- 
2.25.1

_______________________________________________
Intel-gfx mailing list
Intel-gfx@lists.freedesktop.org
https://lists.freedesktop.org/mailman/listinfo/intel-gfx

^ permalink raw reply related	[flat|nested] 25+ messages in thread
end of thread, other threads:[~2020-03-26 11:49 UTC | newest]

Thread overview: 25+ messages (download: mbox.gz / follow: Atom feed)
-- links below jump to the message on this page --
2020-03-24 16:51 [Intel-gfx] [PATCH 01/21] Revert "drm/i915/gem: Drop relocation slowpath" Maarten Lankhorst
2020-03-24 16:51 ` [Intel-gfx] [PATCH 02/21] drm/i915: Add an implementation for i915_gem_ww_ctx locking, v2 Maarten Lankhorst
2020-03-24 16:51 ` [Intel-gfx] [PATCH 03/21] drm/i915: Remove locking from i915_gem_object_prepare_read/write Maarten Lankhorst
2020-03-24 16:51 ` [Intel-gfx] [PATCH 04/21] drm/i915: Parse command buffer earlier in eb_relocate(slow) Maarten Lankhorst
2020-03-24 16:51 ` [Intel-gfx] [PATCH 05/21] drm/i915: Use per object locking in execbuf, v6 Maarten Lankhorst
2020-03-24 16:51 ` [Intel-gfx] [PATCH 06/21] drm/i915: Use ww locking in intel_renderstate Maarten Lankhorst
2020-03-24 16:51 ` [Intel-gfx] [PATCH 07/21] drm/i915: Add ww context handling to context_barrier_task Maarten Lankhorst
2020-03-24 16:51 ` [Intel-gfx] [PATCH 08/21] drm/i915: Nuke arguments to eb_pin_engine Maarten Lankhorst
2020-03-24 16:51 ` [Intel-gfx] [PATCH 09/21] drm/i915: Pin engine before pinning all objects, v3 Maarten Lankhorst
2020-03-24 16:51 ` [Intel-gfx] [PATCH 10/21] drm/i915: Rework intel_context pinning to do everything outside of pin_mutex Maarten Lankhorst
2020-03-24 16:51 ` [Intel-gfx] [PATCH 11/21] drm/i915: Make sure execbuffer always passes ww state to i915_vma_pin Maarten Lankhorst
2020-03-24 16:51 ` [Intel-gfx] [PATCH 12/21] drm/i915: Convert i915_gem_object/client_blt.c to use ww locking as well, v2 Maarten Lankhorst
2020-03-24 16:51 ` [Intel-gfx] [PATCH 13/21] drm/i915: Kill last user of intel_context_create_request outside of selftests Maarten Lankhorst
2020-03-24 16:51 ` [Intel-gfx] [PATCH 14/21] drm/i915: Convert i915_perf to ww locking as well Maarten Lankhorst
2020-03-24 16:51 ` [Intel-gfx] [PATCH 15/21] drm/i915: Dirty hack to fix selftests locking inversion Maarten Lankhorst
2020-03-24 16:51 ` [Intel-gfx] [PATCH 16/21] drm/i915/selftests: Fix locking inversion in lrc selftest Maarten Lankhorst
2020-03-24 16:51 ` [Intel-gfx] [PATCH 17/21] drm/i915: Use ww pinning for intel_context_create_request() Maarten Lankhorst
2020-03-24 16:51 ` [Intel-gfx] [PATCH 18/21] drm/i915: Move i915_vma_lock in the selftests to avoid lock inversion Maarten Lankhorst
2020-03-24 16:51 ` [Intel-gfx] [PATCH 19/21] drm/i915: Add ww locking to vm_fault_gtt Maarten Lankhorst
2020-03-24 16:51 ` [Intel-gfx] [PATCH 20/21] drm/i915: Add ww locking to pin_to_display_plane Maarten Lankhorst
2020-03-24 16:51 ` [Intel-gfx] [PATCH 21/21] drm/i915: Ensure we hold the pin mutex Maarten Lankhorst
2020-03-24 17:02 ` [Intel-gfx] ✗ Fi.CI.CHECKPATCH: warning for series starting with [01/21] Revert "drm/i915/gem: Drop relocation slowpath" Patchwork
2020-03-24 17:17 ` [Intel-gfx] ✗ Fi.CI.BAT: failure " Patchwork
2020-03-25  9:09 ` [Intel-gfx] ✓ Fi.CI.IGT: success " Patchwork
2020-03-26 11:49 [Intel-gfx] [PATCH 01/21] " Maarten Lankhorst
2020-03-26 11:49 ` [Intel-gfx] [PATCH 12/21] drm/i915: Convert i915_gem_object/client_blt.c to use ww locking as well, v2 Maarten Lankhorst

This is an external index of several public inboxes,
see mirroring instructions on how to clone and mirror
all data and code used by this external index.