On Fri, Aug 07, 2015 at 10:45:21AM +0100, Michel Thierry wrote:
Gen8+ supports 48-bit virtual addresses, but some objects must always be
allocated inside the 32-bit address range.
In specific, any resource used with flat/heapless (0x00000000-0xfffff000)
General State Heap (GSH) or Instruction State Heap (ISH) must be in a
32-bit range, because the General State Offset and Instruction State Offset
are limited to 32-bits.
The i915 driver has been modified to provide a flag to set when the 4GB
limit is not necessary in a given bo (EXEC_OBJECT_SUPPORTS_48B_ADDRESS).
48-bit range will only be used when explicitly requested.
Calls to the new drm_intel_bo_emit_reloc_48bit function will have this flag
set automatically, while calls to drm_intel_bo_emit_reloc will clear it.
v2: Make set/clear functions nops on pre-gen8 platforms, and use them
internally in emit_reloc functions (Ben)
s/48BADDRESS/48B_ADDRESS/ (Dave)
v3: Keep set/clear functions internal, no-one needs to use them directly.
References: http://lists.freedesktop.org/archives/intel-gfx/2015-July/072612.html
Cc: Ben Widawsky <ben@xxxxxxxxxxxx>
Cc: Dave Gordon <david.s.gordon@xxxxxxxxx>
Signed-off-by: Michel Thierry <michel.thierry@xxxxxxxxx>
---
include/drm/i915_drm.h | 3 ++-
intel/intel_bufmgr.c | 16 ++++++++++++++
intel/intel_bufmgr.h | 6 +++++-
intel/intel_bufmgr_gem.c | 54 +++++++++++++++++++++++++++++++++++++++++++----
intel/intel_bufmgr_priv.h | 11 ++++++++++
5 files changed, 84 insertions(+), 6 deletions(-)
diff --git a/include/drm/i915_drm.h b/include/drm/i915_drm.h
index ded43b1..426b25c 100644
--- a/include/drm/i915_drm.h
+++ b/include/drm/i915_drm.h
@@ -680,7 +680,8 @@ struct drm_i915_gem_exec_object2 {
#define EXEC_OBJECT_NEEDS_FENCE (1<<0)
#define EXEC_OBJECT_NEEDS_GTT (1<<1)
#define EXEC_OBJECT_WRITE (1<<2)
-#define __EXEC_OBJECT_UNKNOWN_FLAGS -(EXEC_OBJECT_WRITE<<1)
+#define EXEC_OBJECT_SUPPORTS_48B_ADDRESS (1<<3)
+#define __EXEC_OBJECT_UNKNOWN_FLAGS -(EXEC_OBJECT_SUPPORTS_48B_ADDRESS<<1)
__u64 flags;
__u64 rsvd1;
diff --git a/intel/intel_bufmgr.c b/intel/intel_bufmgr.c
index 14ea9f9..0bd5191 100644
--- a/intel/intel_bufmgr.c
+++ b/intel/intel_bufmgr.c
@@ -202,6 +202,22 @@ drm_intel_bo_emit_reloc(drm_intel_bo *bo, uint32_t offset,
drm_intel_bo *target_bo, uint32_t target_offset,
uint32_t read_domains, uint32_t write_domain)
{
+ if (bo->bufmgr->bo_clear_supports_48b_address)
+ bo->bufmgr->bo_clear_supports_48b_address(target_bo);
This is always true - you assign func to this func pointer unconditionally.
Also - why not return some meaningful value if the user does not have
enable_ppgtt=3 set? You can get that from I915_PARAM_HAS_ALIASING_PPGTT right
now. Check for gen >= 8 seems rather inadequate, and even then you're not
returning anything useful to the caller.
+
+ return bo->bufmgr->bo_emit_reloc(bo, offset,
+ target_bo, target_offset,
+ read_domains, write_domain);
+}
+
Using emit_reloc to set a BO flag seems confusing and can be error prone,
emit_reloc can be called many times and caller needs to be careful and call the
right function for each reloc.
+int
+drm_intel_bo_emit_reloc_48bit(drm_intel_bo *bo, uint32_t offset,
+ drm_intel_bo *target_bo, uint32_t target_offset,
+ uint32_t read_domains, uint32_t write_domain)
+{
+ if (bo->bufmgr->bo_set_supports_48b_address)
+ bo->bufmgr->bo_set_supports_48b_address(target_bo);
Same situation as with clear_supports_48b_address.
+
return bo->bufmgr->bo_emit_reloc(bo, offset,
target_bo, target_offset,
read_domains, write_domain);
diff --git a/intel/intel_bufmgr.h b/intel/intel_bufmgr.h
index 285919e..8f91ffe 100644
--- a/intel/intel_bufmgr.h
+++ b/intel/intel_bufmgr.h
@@ -87,7 +87,8 @@ struct _drm_intel_bo {
/**
* Last seen card virtual address (offset from the beginning of the
* aperture) for the object. This should be used to fill relocation
- * entries when calling drm_intel_bo_emit_reloc()
+ * entries when calling drm_intel_bo_emit_reloc() or
+ * drm_intel_bo_emit_reloc_48bit()
*/
uint64_t offset64;
};
@@ -147,6 +148,9 @@ int drm_intel_bufmgr_check_aperture_space(drm_intel_bo ** bo_array, int count);
int drm_intel_bo_emit_reloc(drm_intel_bo *bo, uint32_t offset,
drm_intel_bo *target_bo, uint32_t target_offset,
uint32_t read_domains, uint32_t write_domain);
+int drm_intel_bo_emit_reloc_48bit(drm_intel_bo *bo, uint32_t offset,
+ drm_intel_bo *target_bo, uint32_t target_offset,
+ uint32_t read_domains, uint32_t write_domain);
int drm_intel_bo_emit_reloc_fence(drm_intel_bo *bo, uint32_t offset,
drm_intel_bo *target_bo,
uint32_t target_offset,
diff --git a/intel/intel_bufmgr_gem.c b/intel/intel_bufmgr_gem.c
index 41de396..713ed4e 100644
--- a/intel/intel_bufmgr_gem.c
+++ b/intel/intel_bufmgr_gem.c
@@ -140,6 +140,7 @@ typedef struct _drm_intel_bufmgr_gem {
} drm_intel_bufmgr_gem;
#define DRM_INTEL_RELOC_FENCE (1<<0)
+#define DRM_INTEL_RELOC_SUPPORTS_48B_ADDRESS (2<<0)
Why are you duplicating information about support for 48B in both emit reloc
function argument and struct bo_gem?
typedef struct _drm_intel_reloc_target_info {
drm_intel_bo *bo;
@@ -237,6 +238,14 @@ struct _drm_intel_bo_gem {
bool is_userptr;
/**
+ * Boolean of whether this buffer can be in the whole 48-bit address.
+ *
+ * By default, buffers will be keep in a 32-bit range, unless this
+ * flag is explicitly set.
+ */
+ bool supports_48b_address;
+
+ /**
* Size in bytes of this buffer and its relocation descendents.
*
* Used to avoid costly tree walking in
@@ -463,13 +472,18 @@ drm_intel_add_validate_buffer(drm_intel_bo *bo)
}
static void
-drm_intel_add_validate_buffer2(drm_intel_bo *bo, int need_fence)
+drm_intel_add_validate_buffer2(drm_intel_bo *bo, int need_fence,
+ int supports_48b_address)
{
drm_intel_bufmgr_gem *bufmgr_gem = (drm_intel_bufmgr_gem *)bo->bufmgr;
drm_intel_bo_gem *bo_gem = (drm_intel_bo_gem *)bo;
int index;
if (bo_gem->validate_index != -1) {
+ if (supports_48b_address) {
+ bufmgr_gem->exec2_objects[bo_gem->validate_index].flags |=
+ EXEC_OBJECT_SUPPORTS_48B_ADDRESS;
+ }
if (need_fence)
bufmgr_gem->exec2_objects[bo_gem->validate_index].flags |=
EXEC_OBJECT_NEEDS_FENCE;
@@ -508,6 +522,10 @@ drm_intel_add_validate_buffer2(drm_intel_bo *bo, int need_fence)
bufmgr_gem->exec2_objects[index].flags |=
EXEC_OBJECT_NEEDS_FENCE;
}
+ if (supports_48b_address) {
+ bufmgr_gem->exec2_objects[index].flags |=
+ EXEC_OBJECT_SUPPORTS_48B_ADDRESS;
+ }
bufmgr_gem->exec_count++;
}
@@ -1925,11 +1943,33 @@ do_bo_emit_reloc(drm_intel_bo *bo, uint32_t offset,
else
bo_gem->reloc_target_info[bo_gem->reloc_count].flags = 0;
+ if (target_bo_gem->supports_48b_address)
+ bo_gem->reloc_target_info[bo_gem->reloc_count].flags |=
+ DRM_INTEL_RELOC_SUPPORTS_48B_ADDRESS;
+
bo_gem->reloc_count++;
return 0;
}
+static void drm_intel_gem_bo_set_supports_48b_address(drm_intel_bo *bo)
+{
+ drm_intel_bufmgr_gem *bufmgr_gem = (drm_intel_bufmgr_gem *) bo->bufmgr;
+ drm_intel_bo_gem *bo_gem = (drm_intel_bo_gem *) bo;
+
+ if (bufmgr_gem->gen >= 8)
+ bo_gem->supports_48b_address = 1;
+}
+
+static void drm_intel_gem_bo_clear_supports_48b_address(drm_intel_bo *bo)
+{
+ drm_intel_bufmgr_gem *bufmgr_gem = (drm_intel_bufmgr_gem *) bo->bufmgr;
+ drm_intel_bo_gem *bo_gem = (drm_intel_bo_gem *) bo;
+
+ if (bufmgr_gem->gen >= 8)
+ bo_gem->supports_48b_address = 0;
+}
+
static int
drm_intel_gem_bo_emit_reloc(drm_intel_bo *bo, uint32_t offset,
drm_intel_bo *target_bo, uint32_t target_offset,
@@ -2043,7 +2083,7 @@ drm_intel_gem_bo_process_reloc2(drm_intel_bo *bo)
for (i = 0; i < bo_gem->reloc_count; i++) {
drm_intel_bo *target_bo = bo_gem->reloc_target_info[i].bo;
- int need_fence;
+ int need_fence, supports_48b_addr;
if (target_bo == bo)
continue;
@@ -2056,8 +2096,12 @@ drm_intel_gem_bo_process_reloc2(drm_intel_bo *bo)
need_fence = (bo_gem->reloc_target_info[i].flags &
DRM_INTEL_RELOC_FENCE);
+ supports_48b_addr = (bo_gem->reloc_target_info[i].flags &
+ DRM_INTEL_RELOC_SUPPORTS_48B_ADDRESS);
+
/* Add the target to the validate list */
- drm_intel_add_validate_buffer2(target_bo, need_fence);
+ drm_intel_add_validate_buffer2(target_bo, need_fence,
+ supports_48b_addr);
}
}
@@ -2217,7 +2261,7 @@ do_exec2(drm_intel_bo *bo, int used, drm_intel_context *ctx,
/* Add the batch buffer to the validation list. There are no relocations
* pointing to it.
*/
- drm_intel_add_validate_buffer2(bo, 0);
+ drm_intel_add_validate_buffer2(bo, 0, 0);
memclear(execbuf);
execbuf.buffers_ptr = (uintptr_t)bufmgr_gem->exec2_objects;
@@ -3299,6 +3343,8 @@ drm_intel_bufmgr_gem_init(int fd, int batch_size)
bufmgr_gem->bufmgr.bo_wait_rendering = drm_intel_gem_bo_wait_rendering;
bufmgr_gem->bufmgr.bo_emit_reloc = drm_intel_gem_bo_emit_reloc;
bufmgr_gem->bufmgr.bo_emit_reloc_fence = drm_intel_gem_bo_emit_reloc_fence;
+ bufmgr_gem->bufmgr.bo_set_supports_48b_address = drm_intel_gem_bo_set_supports_48b_address;
+ bufmgr_gem->bufmgr.bo_clear_supports_48b_address = drm_intel_gem_bo_clear_supports_48b_address;
bufmgr_gem->bufmgr.bo_pin = drm_intel_gem_bo_pin;
bufmgr_gem->bufmgr.bo_unpin = drm_intel_gem_bo_unpin;
bufmgr_gem->bufmgr.bo_get_tiling = drm_intel_gem_bo_get_tiling;
diff --git a/intel/intel_bufmgr_priv.h b/intel/intel_bufmgr_priv.h
index 59ebd18..774fa02 100644
--- a/intel/intel_bufmgr_priv.h
+++ b/intel/intel_bufmgr_priv.h
@@ -152,6 +152,17 @@ struct _drm_intel_bufmgr {
void (*destroy) (drm_intel_bufmgr *bufmgr);
/**
+ * Set/Clear 48-bit address support flag in a given bo.
+ *
+ * Any resource used with flat/heapless (0x00000000-0xfffff000)
+ * General State Heap (GSH) or Intructions State Heap (ISH) must
+ * be in a 32-bit range. 48-bit range will only be used when explicitly
+ * requested.
+ */
+ void (*bo_set_supports_48b_address) (drm_intel_bo *bo);
+ void (*bo_clear_supports_48b_address) (drm_intel_bo *bo);
+
+ /**
* Add relocation entry in reloc_buf, which will be updated with the
* target buffer's real offset on on command submission.
*
--
2.5.0
You also haven't updated the dump_validation_list with proper format specifier
(addresses are truncated otherwise).
I really don't like hiding set_supports_48b_address/clear_supports_48b_address
in emit reloc - making this explicit would be a better approach.
Of course this is just my opinion, but let me post a bit different version of
this API - just as an RFC so you can see how it would look with explicit set on
bo and without adding another emit_reloc variant.
