All of lore.kernel.org
 help / color / mirror / Atom feed
From: Matthew Auld <matthew.auld@intel.com>
To: intel-gfx@lists.freedesktop.org
Cc: "Thomas Hellström" <thomas.hellstrom@linux.intel.com>,
	dri-devel@lists.freedesktop.org
Subject: [Intel-gfx] [PATCH v2 13/15] drm/i915/uapi: add NEEDS_CPU_ACCESS hint
Date: Thu, 10 Feb 2022 12:13:11 +0000	[thread overview]
Message-ID: <20220210121313.701004-14-matthew.auld@intel.com> (raw)
In-Reply-To: <20220210121313.701004-1-matthew.auld@intel.com>

If set, force the allocation to be placed in the mappable portion of
LMEM. One big restriction here is that system memory must be given as a
potential placement for the object, that way we can always spill the
object into system memory if we can't make space.

XXX: Still very much WIP and needs IGTs. Including now just for the sake
of having more complete picture.

Signed-off-by: Matthew Auld <matthew.auld@intel.com>
Cc: Thomas Hellström <thomas.hellstrom@linux.intel.com>
---
 drivers/gpu/drm/i915/gem/i915_gem_create.c | 28 ++++++++++++-------
 include/uapi/drm/i915_drm.h                | 31 +++++++++++++++++++++-
 2 files changed, 49 insertions(+), 10 deletions(-)

diff --git a/drivers/gpu/drm/i915/gem/i915_gem_create.c b/drivers/gpu/drm/i915/gem/i915_gem_create.c
index cc9ddb943f96..2737977f81a4 100644
--- a/drivers/gpu/drm/i915/gem/i915_gem_create.c
+++ b/drivers/gpu/drm/i915/gem/i915_gem_create.c
@@ -238,6 +238,7 @@ struct create_ext {
 	struct drm_i915_private *i915;
 	struct intel_memory_region *placements[INTEL_REGION_UNKNOWN];
 	unsigned int n_placements;
+	unsigned int placement_mask;
 	unsigned long flags;
 };
 
@@ -334,6 +335,7 @@ static int set_placements(struct drm_i915_gem_create_ext_memory_regions *args,
 	for (i = 0; i < args->num_regions; i++)
 		ext_data->placements[i] = placements[i];
 
+	ext_data->placement_mask = mask;
 	return 0;
 
 out_dump:
@@ -408,7 +410,7 @@ i915_gem_create_ext_ioctl(struct drm_device *dev, void *data,
 	struct drm_i915_gem_object *obj;
 	int ret;
 
-	if (args->flags)
+	if (args->flags & ~I915_GEM_CREATE_EXT_FLAG_NEEDS_CPU_ACCESS)
 		return -EINVAL;
 
 	ret = i915_user_extensions(u64_to_user_ptr(args->extensions),
@@ -424,14 +426,22 @@ i915_gem_create_ext_ioctl(struct drm_device *dev, void *data,
 		ext_data.n_placements = 1;
 	}
 
-	/*
-	 * TODO: add a userspace hint to force CPU_ACCESS for the object, which
-	 * can override this.
-	 */
-	if (!IS_DG1(i915) && (ext_data.n_placements > 1 ||
-			      ext_data.placements[0]->type !=
-			      INTEL_MEMORY_SYSTEM))
-		ext_data.flags |= I915_BO_ALLOC_GPU_ONLY;
+	if (args->flags & I915_GEM_CREATE_EXT_FLAG_NEEDS_CPU_ACCESS) {
+		if (ext_data.n_placements == 1)
+			return -EINVAL;
+
+		/*
+		 * We always need to be able to spill to system memory, if we
+		 * can't place in the mappable part of LMEM.
+		 */
+		if (!(ext_data.placement_mask & BIT(INTEL_REGION_SMEM)))
+			return -EINVAL;
+	} else {
+		if (!IS_DG1(i915) && (ext_data.n_placements > 1 ||
+				      ext_data.placements[0]->type !=
+				      INTEL_MEMORY_SYSTEM))
+			ext_data.flags |= I915_BO_ALLOC_GPU_ONLY;
+	}
 
 	obj = __i915_gem_object_create_user_ext(i915, args->size,
 						ext_data.placements,
diff --git a/include/uapi/drm/i915_drm.h b/include/uapi/drm/i915_drm.h
index 914ebd9290e5..925685bd261e 100644
--- a/include/uapi/drm/i915_drm.h
+++ b/include/uapi/drm/i915_drm.h
@@ -3157,7 +3157,36 @@ struct drm_i915_gem_create_ext {
 	 * Object handles are nonzero.
 	 */
 	__u32 handle;
-	/** @flags: MBZ */
+	/**
+	 * @flags: Optional flags.
+	 *
+	 * Supported values:
+	 *
+	 * I915_GEM_CREATE_EXT_FLAG_NEEDS_CPU_ACCESS - Signal to the kernel that
+	 * the object will need to be accessed via the CPU.
+	 *
+	 * Only valid when placing objects in I915_MEMORY_CLASS_DEVICE, and
+	 * only strictly required on platforms where only some of the device
+	 * memory is directly visible or mappable through the CPU, like on DG2+.
+	 *
+	 * One of the placements MUST also be I915_MEMORY_CLASS_SYSTEM, to
+	 * ensure we can always spill the allocation to system memory, if we
+	 * can't place the object in the mappable part of
+	 * I915_MEMORY_CLASS_DEVICE.
+	 *
+	 * Note that buffers that need to be captured with EXEC_OBJECT_CAPTURE,
+	 * will need to enable this hint, if the object can also be placed in
+	 * I915_MEMORY_CLASS_DEVICE, starting from DG2+. The execbuf call will
+	 * throw an error otherwise. This also means that such objects will need
+	 * I915_MEMORY_CLASS_SYSTEM set as a possible placement.
+	 *
+	 * Without this hint, the kernel will assume that non-mappable
+	 * I915_MEMORY_CLASS_DEVICE is preferred for this object. Note that the
+	 * kernel can still migrate the object to the mappable part, as a last
+	 * resort, if userspace ever CPU faults this object, but this might be
+	 * expensive, and so ideally should be avoided.
+	 */
+#define I915_GEM_CREATE_EXT_FLAG_NEEDS_CPU_ACCESS (1 << 0)
 	__u32 flags;
 	/**
 	 * @extensions: The chain of extensions to apply to this object.
-- 
2.34.1


WARNING: multiple messages have this Message-ID (diff)
From: Matthew Auld <matthew.auld@intel.com>
To: intel-gfx@lists.freedesktop.org
Cc: "Thomas Hellström" <thomas.hellstrom@linux.intel.com>,
	dri-devel@lists.freedesktop.org
Subject: [PATCH v2 13/15] drm/i915/uapi: add NEEDS_CPU_ACCESS hint
Date: Thu, 10 Feb 2022 12:13:11 +0000	[thread overview]
Message-ID: <20220210121313.701004-14-matthew.auld@intel.com> (raw)
In-Reply-To: <20220210121313.701004-1-matthew.auld@intel.com>

If set, force the allocation to be placed in the mappable portion of
LMEM. One big restriction here is that system memory must be given as a
potential placement for the object, that way we can always spill the
object into system memory if we can't make space.

XXX: Still very much WIP and needs IGTs. Including now just for the sake
of having more complete picture.

Signed-off-by: Matthew Auld <matthew.auld@intel.com>
Cc: Thomas Hellström <thomas.hellstrom@linux.intel.com>
---
 drivers/gpu/drm/i915/gem/i915_gem_create.c | 28 ++++++++++++-------
 include/uapi/drm/i915_drm.h                | 31 +++++++++++++++++++++-
 2 files changed, 49 insertions(+), 10 deletions(-)

diff --git a/drivers/gpu/drm/i915/gem/i915_gem_create.c b/drivers/gpu/drm/i915/gem/i915_gem_create.c
index cc9ddb943f96..2737977f81a4 100644
--- a/drivers/gpu/drm/i915/gem/i915_gem_create.c
+++ b/drivers/gpu/drm/i915/gem/i915_gem_create.c
@@ -238,6 +238,7 @@ struct create_ext {
 	struct drm_i915_private *i915;
 	struct intel_memory_region *placements[INTEL_REGION_UNKNOWN];
 	unsigned int n_placements;
+	unsigned int placement_mask;
 	unsigned long flags;
 };
 
@@ -334,6 +335,7 @@ static int set_placements(struct drm_i915_gem_create_ext_memory_regions *args,
 	for (i = 0; i < args->num_regions; i++)
 		ext_data->placements[i] = placements[i];
 
+	ext_data->placement_mask = mask;
 	return 0;
 
 out_dump:
@@ -408,7 +410,7 @@ i915_gem_create_ext_ioctl(struct drm_device *dev, void *data,
 	struct drm_i915_gem_object *obj;
 	int ret;
 
-	if (args->flags)
+	if (args->flags & ~I915_GEM_CREATE_EXT_FLAG_NEEDS_CPU_ACCESS)
 		return -EINVAL;
 
 	ret = i915_user_extensions(u64_to_user_ptr(args->extensions),
@@ -424,14 +426,22 @@ i915_gem_create_ext_ioctl(struct drm_device *dev, void *data,
 		ext_data.n_placements = 1;
 	}
 
-	/*
-	 * TODO: add a userspace hint to force CPU_ACCESS for the object, which
-	 * can override this.
-	 */
-	if (!IS_DG1(i915) && (ext_data.n_placements > 1 ||
-			      ext_data.placements[0]->type !=
-			      INTEL_MEMORY_SYSTEM))
-		ext_data.flags |= I915_BO_ALLOC_GPU_ONLY;
+	if (args->flags & I915_GEM_CREATE_EXT_FLAG_NEEDS_CPU_ACCESS) {
+		if (ext_data.n_placements == 1)
+			return -EINVAL;
+
+		/*
+		 * We always need to be able to spill to system memory, if we
+		 * can't place in the mappable part of LMEM.
+		 */
+		if (!(ext_data.placement_mask & BIT(INTEL_REGION_SMEM)))
+			return -EINVAL;
+	} else {
+		if (!IS_DG1(i915) && (ext_data.n_placements > 1 ||
+				      ext_data.placements[0]->type !=
+				      INTEL_MEMORY_SYSTEM))
+			ext_data.flags |= I915_BO_ALLOC_GPU_ONLY;
+	}
 
 	obj = __i915_gem_object_create_user_ext(i915, args->size,
 						ext_data.placements,
diff --git a/include/uapi/drm/i915_drm.h b/include/uapi/drm/i915_drm.h
index 914ebd9290e5..925685bd261e 100644
--- a/include/uapi/drm/i915_drm.h
+++ b/include/uapi/drm/i915_drm.h
@@ -3157,7 +3157,36 @@ struct drm_i915_gem_create_ext {
 	 * Object handles are nonzero.
 	 */
 	__u32 handle;
-	/** @flags: MBZ */
+	/**
+	 * @flags: Optional flags.
+	 *
+	 * Supported values:
+	 *
+	 * I915_GEM_CREATE_EXT_FLAG_NEEDS_CPU_ACCESS - Signal to the kernel that
+	 * the object will need to be accessed via the CPU.
+	 *
+	 * Only valid when placing objects in I915_MEMORY_CLASS_DEVICE, and
+	 * only strictly required on platforms where only some of the device
+	 * memory is directly visible or mappable through the CPU, like on DG2+.
+	 *
+	 * One of the placements MUST also be I915_MEMORY_CLASS_SYSTEM, to
+	 * ensure we can always spill the allocation to system memory, if we
+	 * can't place the object in the mappable part of
+	 * I915_MEMORY_CLASS_DEVICE.
+	 *
+	 * Note that buffers that need to be captured with EXEC_OBJECT_CAPTURE,
+	 * will need to enable this hint, if the object can also be placed in
+	 * I915_MEMORY_CLASS_DEVICE, starting from DG2+. The execbuf call will
+	 * throw an error otherwise. This also means that such objects will need
+	 * I915_MEMORY_CLASS_SYSTEM set as a possible placement.
+	 *
+	 * Without this hint, the kernel will assume that non-mappable
+	 * I915_MEMORY_CLASS_DEVICE is preferred for this object. Note that the
+	 * kernel can still migrate the object to the mappable part, as a last
+	 * resort, if userspace ever CPU faults this object, but this might be
+	 * expensive, and so ideally should be avoided.
+	 */
+#define I915_GEM_CREATE_EXT_FLAG_NEEDS_CPU_ACCESS (1 << 0)
 	__u32 flags;
 	/**
 	 * @extensions: The chain of extensions to apply to this object.
-- 
2.34.1


  parent reply	other threads:[~2022-02-10 12:17 UTC|newest]

Thread overview: 56+ messages / expand[flat|nested]  mbox.gz  Atom feed  top
2022-02-10 12:12 [Intel-gfx] [PATCH v2 00/15] Initial support for small BAR recovery Matthew Auld
2022-02-10 12:12 ` Matthew Auld
2022-02-10 12:12 ` [Intel-gfx] [PATCH v2 01/15] drm/i915: add io_size plumbing Matthew Auld
2022-02-10 12:12   ` Matthew Auld
2022-02-10 12:13 ` [Intel-gfx] [PATCH v2 02/15] drm/i915/ttm: require mappable by default Matthew Auld
2022-02-10 12:13   ` Matthew Auld
2022-02-11  9:20   ` [Intel-gfx] " Thomas Hellström
2022-02-11  9:20     ` Thomas Hellström
2022-02-10 12:13 ` [Intel-gfx] [PATCH v2 03/15] drm/i915: add I915_BO_ALLOC_GPU_ONLY Matthew Auld
2022-02-10 12:13   ` Matthew Auld
2022-02-11  9:23   ` [Intel-gfx] " Thomas Hellström
2022-02-11  9:23     ` Thomas Hellström
2022-02-10 12:13 ` [Intel-gfx] [PATCH v2 04/15] drm/i915/buddy: track available visible size Matthew Auld
2022-02-10 12:13   ` Matthew Auld
2022-02-11  9:25   ` [Intel-gfx] " Thomas Hellström
2022-02-11  9:25     ` Thomas Hellström
2022-02-10 12:13 ` [Intel-gfx] [PATCH v2 05/15] drm/i915/buddy: adjust res->start Matthew Auld
2022-02-10 12:13   ` Matthew Auld
2022-02-10 12:13 ` [Intel-gfx] [PATCH v2 06/15] drm/i915/buddy: tweak 2big check Matthew Auld
2022-02-10 12:13   ` Matthew Auld
2022-02-10 12:13 ` [Intel-gfx] [PATCH v2 07/15] drm/i915/selftests: mock test io_size Matthew Auld
2022-02-10 12:13   ` Matthew Auld
2022-02-10 12:13 ` [Intel-gfx] [PATCH v2 08/15] drm/i915/ttm: make eviction mappable aware Matthew Auld
2022-02-10 12:13   ` Matthew Auld
2022-02-10 12:13 ` [Intel-gfx] [PATCH v2 09/15] drm/i915/ttm: mappable migration on fault Matthew Auld
2022-02-10 12:13   ` Matthew Auld
2022-02-10 12:13 ` [Intel-gfx] [PATCH v2 10/15] drm/i915/selftests: exercise mmap migration Matthew Auld
2022-02-10 12:13   ` Matthew Auld
2022-02-11  9:36   ` [Intel-gfx] " Thomas Hellström
2022-02-11  9:36     ` Thomas Hellström
2022-02-10 12:13 ` [Intel-gfx] [PATCH v2 11/15] drm/i915/selftests: handle allocation failures Matthew Auld
2022-02-10 12:13   ` Matthew Auld
2022-02-11  9:37   ` [Intel-gfx] " Thomas Hellström
2022-02-11  9:37     ` Thomas Hellström
2022-02-10 12:13 ` [Intel-gfx] [PATCH v2 12/15] drm/i915/create: apply ALLOC_GPU_ONLY by default Matthew Auld
2022-02-10 12:13   ` Matthew Auld
2022-02-11  9:49   ` [Intel-gfx] " Thomas Hellström
2022-02-11  9:49     ` Thomas Hellström
2022-02-11  9:52     ` [Intel-gfx] " Matthew Auld
2022-02-11  9:55       ` Thomas Hellström
2022-02-11 10:00         ` Matthew Auld
2022-02-11 10:14           ` Thomas Hellström
2022-02-10 12:13 ` Matthew Auld [this message]
2022-02-10 12:13   ` [PATCH v2 13/15] drm/i915/uapi: add NEEDS_CPU_ACCESS hint Matthew Auld
2022-02-11  9:59   ` [Intel-gfx] " Thomas Hellström
2022-02-11  9:59     ` Thomas Hellström
2022-02-10 12:13 ` [Intel-gfx] [PATCH v2 14/15] drm/i915/uapi: forbid ALLOC_GPU_ONLY for error capture Matthew Auld
2022-02-10 12:13   ` Matthew Auld
2022-02-11 10:06   ` [Intel-gfx] " Thomas Hellström
2022-02-11 10:06     ` Thomas Hellström
2022-02-10 12:13 ` [Intel-gfx] [PATCH v2 15/15] drm/i915/lmem: don't treat small BAR as an error Matthew Auld
2022-02-10 12:13   ` Matthew Auld
2022-02-11 10:08   ` [Intel-gfx] " Thomas Hellström
2022-02-11 10:08     ` Thomas Hellström
2022-02-10 14:40 ` [Intel-gfx] [PATCH v2 00/15] Initial support for small BAR recovery Das, Nirmoy
2022-02-10 15:24 ` [Intel-gfx] ✗ Fi.CI.BUILD: failure for Initial support for small BAR recovery (rev3) Patchwork

Reply instructions:

You may reply publicly to this message via plain-text email
using any one of the following methods:

* Save the following mbox file, import it into your mail client,
  and reply-to-all from there: mbox

  Avoid top-posting and favor interleaved quoting:
  https://en.wikipedia.org/wiki/Posting_style#Interleaved_style

* Reply using the --to, --cc, and --in-reply-to
  switches of git-send-email(1):

  git send-email \
    --in-reply-to=20220210121313.701004-14-matthew.auld@intel.com \
    --to=matthew.auld@intel.com \
    --cc=dri-devel@lists.freedesktop.org \
    --cc=intel-gfx@lists.freedesktop.org \
    --cc=thomas.hellstrom@linux.intel.com \
    /path/to/YOUR_REPLY

  https://kernel.org/pub/software/scm/git/docs/git-send-email.html

* If your mail client supports setting the In-Reply-To header
  via mailto: links, try the mailto: link
Be sure your reply has a Subject: header at the top and a blank line before the message body.
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.