From: Ben Widawsky <benjamin.widawsky@intel.com>
To: Intel GFX <intel-gfx@lists.freedesktop.org>
Cc: Ben Widawsky <ben@bwidawsk.net>,
Ben Widawsky <benjamin.widawsky@intel.com>
Subject: [PATCH 55/56] drm/i915: Track userptr VMAs
Date: Fri, 9 May 2014 20:59:50 -0700 [thread overview]
Message-ID: <1399694391-3935-56-git-send-email-benjamin.widawsky@intel.com> (raw)
In-Reply-To: <1399694391-3935-1-git-send-email-benjamin.widawsky@intel.com>
This HACK allows users to reuse the userptr ioctl in order to
pre-reserve the VMA at a specific location. The vma will follow all the
same paths as other userptr objects - only the drm_mm node is actually
allocated.
Again, this patch is a big HACK to get some other people currently using
userptr enabled.
Signed-off-by: Ben Widawsky <ben@bwidawsk.net>
---
drivers/gpu/drm/i915/i915_drv.h | 1 +
drivers/gpu/drm/i915/i915_gem.c | 22 +++++++++++++++++++---
drivers/gpu/drm/i915/i915_gem_execbuffer.c | 3 +++
drivers/gpu/drm/i915/i915_gem_gtt.h | 4 ++++
4 files changed, 27 insertions(+), 3 deletions(-)
diff --git a/drivers/gpu/drm/i915/i915_drv.h b/drivers/gpu/drm/i915/i915_drv.h
index 60513e7..71e39ff 100644
--- a/drivers/gpu/drm/i915/i915_drv.h
+++ b/drivers/gpu/drm/i915/i915_drv.h
@@ -2166,6 +2166,7 @@ void i915_gem_vma_destroy(struct i915_vma *vma);
#define PIN_GLOBAL 0x4
#define PIN_ALIASING 0x8
#define PIN_GLOBAL_ALIASED (PIN_ALIASING | PIN_GLOBAL)
+#define PIN_BOUND 0x10
int __must_check i915_gem_object_pin(struct drm_i915_gem_object *obj,
struct i915_address_space *vm,
uint32_t alignment,
diff --git a/drivers/gpu/drm/i915/i915_gem.c b/drivers/gpu/drm/i915/i915_gem.c
index 287d48e..ff75971 100644
--- a/drivers/gpu/drm/i915/i915_gem.c
+++ b/drivers/gpu/drm/i915/i915_gem.c
@@ -3279,7 +3279,13 @@ i915_gem_object_bind_to_vm(struct drm_i915_gem_object *obj,
if (IS_ERR(vma))
goto err_unpin;
+ if (flags & PIN_BOUND) {
+ WARN_ON(!vma->node.allocated && !vma->obj->userptr.ptr);
+ goto skip_search;
+ }
+
search_free:
+ WARN_ON(vma->node.allocated);
ret = drm_mm_insert_node_in_range_generic(&vm->mm, &vma->node,
size, alignment,
obj->cache_level, 0, gtt_max,
@@ -3293,6 +3299,7 @@ search_free:
goto err_free_vma;
}
+skip_search:
if (WARN_ON(!i915_gem_valid_gtt_space(dev, &vma->node,
obj->cache_level))) {
ret = -EINVAL;
@@ -3329,10 +3336,13 @@ search_free:
i915_gem_vma_bind(vma, obj->cache_level, vma_bind_flags);
i915_gem_verify_gtt(dev);
+ if (flags & PIN_BOUND)
+ vma->uptr_bind=1;
return vma;
err_remove_node:
- drm_mm_remove_node(&vma->node);
+ if ((flags & PIN_BOUND) == 0)
+ drm_mm_remove_node(&vma->node);
err_free_vma:
i915_gem_vma_destroy(vma);
vma = ERR_PTR(ret);
@@ -3875,6 +3885,11 @@ i915_gem_object_pin(struct drm_i915_gem_object *obj,
if (WARN_ON(flags & (PIN_GLOBAL | PIN_MAPPABLE) && !i915_is_ggtt(vm)))
return -EINVAL;
+ if (flags & PIN_BOUND) {
+ if (WARN_ON(flags & ~PIN_BOUND))
+ return -EINVAL;
+ }
+
vma = i915_gem_obj_to_vma(obj, vm);
if (vma) {
if (WARN_ON(vma->pin_count == DRM_I915_GEM_OBJECT_MAX_PIN_COUNT))
@@ -3898,7 +3913,8 @@ i915_gem_object_pin(struct drm_i915_gem_object *obj,
}
}
- if (vma == NULL || !drm_mm_node_allocated(&vma->node)) {
+ if (vma == NULL || !drm_mm_node_allocated(&vma->node) ||
+ ((flags & PIN_BOUND) && !vma->uptr_bind)) {
vma = i915_gem_object_bind_to_vm(obj, vm, alignment, flags);
if (IS_ERR(vma))
return PTR_ERR(vma);
@@ -4265,7 +4281,7 @@ struct i915_vma *i915_gem_obj_to_vma(struct drm_i915_gem_object *obj,
void i915_gem_vma_destroy(struct i915_vma *vma)
{
- WARN_ON(vma->node.allocated);
+ WARN_ON(vma->node.allocated && !vma->uptr);
/* Keep the vma as a placeholder in the execbuffer reservation lists */
if (!list_empty(&vma->exec_list))
diff --git a/drivers/gpu/drm/i915/i915_gem_execbuffer.c b/drivers/gpu/drm/i915/i915_gem_execbuffer.c
index 08fde7d..596e51e 100644
--- a/drivers/gpu/drm/i915/i915_gem_execbuffer.c
+++ b/drivers/gpu/drm/i915/i915_gem_execbuffer.c
@@ -566,6 +566,9 @@ i915_gem_execbuffer_reserve_vma(struct i915_vma *vma,
if (entry->flags & EXEC_OBJECT_NEEDS_GTT)
flags |= PIN_GLOBAL;
+ if (vma->uptr)
+ flags |= PIN_BOUND;
+
ret = i915_gem_object_pin(obj, vma->vm, entry->alignment, flags);
if (ret)
return ret;
diff --git a/drivers/gpu/drm/i915/i915_gem_gtt.h b/drivers/gpu/drm/i915/i915_gem_gtt.h
index c265c23..bdb4b05 100644
--- a/drivers/gpu/drm/i915/i915_gem_gtt.h
+++ b/drivers/gpu/drm/i915/i915_gem_gtt.h
@@ -171,6 +171,10 @@ struct i915_vma {
unsigned int pin_count:4;
#define DRM_I915_GEM_OBJECT_MAX_PIN_COUNT 0xf
+ /* FIXME: */
+ unsigned int uptr:1; /* Whether this VMA has been userptr'd */
+ unsigned int uptr_bind:1; /* Whether we've actually bound it */
+
/** Unmap an object from an address space. This usually consists of
* setting the valid PTE entries to a reserved scratch page. */
void (*unbind_vma)(struct i915_vma *vma);
--
1.9.2
next prev parent reply other threads:[~2014-05-10 4:03 UTC|newest]
Thread overview: 58+ messages / expand[flat|nested] mbox.gz Atom feed top
2014-05-10 3:58 [PATCH 00/56] [RFCish] Dynamic page table alloc, 64b, and GPU/CPU mirror Ben Widawsky
2014-05-10 3:58 ` [PATCH 01/56] drm/i915: Fix flush before context switch comment Ben Widawsky
2014-05-10 3:58 ` [PATCH 02/56] Revert "drm/i915: Drop I915_PARAM_HAS_FULL_PPGTT again" Ben Widawsky
2014-05-10 3:58 ` [PATCH 03/56] drm/i915: Prevent signals from interrupting close() Ben Widawsky
2014-05-10 3:58 ` [PATCH 04/56] drm/i915: Wrap VMA binding Ben Widawsky
2014-05-10 3:59 ` [PATCH 05/56] drm/i915: Make pin global flags explicit Ben Widawsky
2014-05-10 3:59 ` [PATCH 06/56] drm/i915: Split out aliasing binds Ben Widawsky
2014-05-10 3:59 ` [PATCH 07/56] drm/i915: fix gtt_total_entries() Ben Widawsky
2014-05-10 3:59 ` [PATCH 08/56] drm/i915: Rename to GEN8_LEGACY_PDPES Ben Widawsky
2014-05-10 3:59 ` [PATCH 09/56] drm/i915: Split out verbose PPGTT dumping Ben Widawsky
2014-05-10 3:59 ` [PATCH 10/56] drm/i915: s/pd/pdpe, s/pt/pde Ben Widawsky
2014-05-10 3:59 ` [PATCH 11/56] drm/i915: rename map/unmap to dma_map/unmap Ben Widawsky
2014-05-10 3:59 ` [PATCH 12/56] drm/i915: Setup less PPGTT on failed pagedir Ben Widawsky
2014-05-10 3:59 ` [PATCH 13/56] drm/i915: clean up PPGTT init error path Ben Widawsky
2014-05-10 3:59 ` [PATCH 14/56] drm/i915: Un-hardcode number of page directories Ben Widawsky
2014-05-10 3:59 ` [PATCH 15/56] drm/i915: Make gen6_write_pdes gen6_map_page_tables Ben Widawsky
2014-05-10 3:59 ` [PATCH 16/56] drm/i915: Range clearing is PPGTT agnostic Ben Widawsky
2014-05-10 3:59 ` [PATCH 17/56] drm/i915: Page table helpers, and define renames Ben Widawsky
2014-05-10 3:59 ` [PATCH 18/56] drm/i915: construct page table abstractions Ben Widawsky
2014-05-10 3:59 ` [PATCH 19/56] drm/i915: Complete page table structures Ben Widawsky
2014-05-10 3:59 ` [PATCH 20/56] drm/i915: Create page table allocators Ben Widawsky
2014-05-10 3:59 ` [PATCH 21/56] drm/i915: Generalize GEN6 mapping Ben Widawsky
2014-05-10 3:59 ` [PATCH 22/56] drm/i915: Clean up pagetable DMA map & unmap Ben Widawsky
2014-05-10 3:59 ` [PATCH 23/56] drm/i915: Always dma map page table allocations Ben Widawsky
2014-05-10 3:59 ` [PATCH 24/56] drm/i915: Consolidate dma mappings Ben Widawsky
2014-05-10 3:59 ` [PATCH 25/56] drm/i915: Always dma map page directory allocations Ben Widawsky
2014-05-10 3:59 ` [PATCH 26/56] drm/i915: Track GEN6 page table usage Ben Widawsky
2014-05-10 3:59 ` [PATCH 27/56] drm/i915: Extract context switch skip logic Ben Widawsky
2014-05-10 3:59 ` [PATCH 28/56] drm/i915: Force pd restore when PDEs change, gen6-7 Ben Widawsky
2014-05-10 3:59 ` [PATCH 29/56] drm/i915: Finish gen6/7 dynamic page table allocation Ben Widawsky
2014-05-10 3:59 ` [PATCH 30/56] drm/i915/bdw: Use dynamic allocation idioms on free Ben Widawsky
2014-05-10 3:59 ` [PATCH 31/56] drm/i915/bdw: pagedirs rework allocation Ben Widawsky
2014-05-10 3:59 ` [PATCH 32/56] drm/i915/bdw: pagetable allocation rework Ben Widawsky
2014-05-10 3:59 ` [PATCH 33/56] drm/i915/bdw: Make the pdp switch a bit less hacky Ben Widawsky
2014-05-10 3:59 ` [PATCH 34/56] drm/i915: num_pd_pages/num_pd_entries isn't useful Ben Widawsky
2014-05-10 3:59 ` [PATCH 35/56] drm/i915: Extract PPGTT param from pagedir alloc Ben Widawsky
2014-05-10 3:59 ` [PATCH 36/56] drm/i915/bdw: Split out mappings Ben Widawsky
2014-05-10 3:59 ` [PATCH 37/56] drm/i915/bdw: begin bitmap tracking Ben Widawsky
2014-05-10 3:59 ` [PATCH 38/56] drm/i915/bdw: Dynamic page table allocations Ben Widawsky
2014-05-10 3:59 ` [PATCH 39/56] drm/i915/bdw: Scratch unused pages Ben Widawsky
2014-05-10 3:59 ` [PATCH 40/56] drm/i915/bdw: Add ppgtt info for dynamic pages Ben Widawsky
2014-05-10 3:59 ` [PATCH 41/56] drm/i915/bdw: Optimize PDP loads Ben Widawsky
2014-05-10 3:59 ` [PATCH 42/56] TESTME: Either drop the last patch or fix it Ben Widawsky
2014-05-10 3:59 ` [PATCH 43/56] drm/i915/bdw: Add dynamic page trace events Ben Widawsky
2014-05-10 3:59 ` [PATCH 44/56] drm/i915/bdw: Make pdp allocation more dynamic Ben Widawsky
2014-05-10 3:59 ` [PATCH 45/56] drm/i915/bdw: Abstract PDP usage Ben Widawsky
2014-05-10 3:59 ` [PATCH 46/56] drm/i915/bdw: implement alloc/teardown for 4lvl Ben Widawsky
2014-05-10 3:59 ` [PATCH 47/56] drm/i915/bdw: 4 level pages tables Ben Widawsky
2014-05-10 3:59 ` [PATCH 48/56] drm/i915: Restructure map vs. insert entries Ben Widawsky
2014-05-10 3:59 ` [PATCH 49/56] drm/i915/bdw: make aliasing PPGTT dynamic Ben Widawsky
2014-05-10 3:59 ` [PATCH 50/56] drm/i915: Expand error state's address width to 64b Ben Widawsky
2014-05-10 3:59 ` [PATCH 51/56] drm/i915/bdw: Flip the 48b switch Ben Widawsky
2014-05-10 3:59 ` [PATCH 52/56] TESTME: GFX_TLB_INVALIDATE_EXPLICIT Ben Widawsky
2014-05-10 3:59 ` [PATCH 53/56] TESTME: Always force invalidate Ben Widawsky
2014-05-10 3:59 ` [PATCH 54/56] drm/i915: Introduce mapping of user pages into video memory (userptr) ioctl Ben Widawsky
2014-05-10 3:59 ` Ben Widawsky [this message]
2014-05-10 3:59 ` [PATCH 56/56] drm/i915/userptr: Mirror GPU addr at ioctl (HACK/POC) Ben Widawsky
2014-05-11 17:33 ` [PATCH 00/56] [RFCish] Dynamic page table alloc, 64b, and GPU/CPU mirror Daniel Vetter
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=1399694391-3935-56-git-send-email-benjamin.widawsky@intel.com \
--to=benjamin.widawsky@intel.com \
--cc=ben@bwidawsk.net \
--cc=intel-gfx@lists.freedesktop.org \
/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 a public inbox, see mirroring instructions
for how to clone and mirror all data and code used for this inbox