Currently, we perform a locked update of the shadow entry when
allocating a page directory entry such that if two clients are
concurrently allocating neighbouring ranges we only insert one new entry
for the pair of them. However, we also need to serialise both clients
wrt to the actual entry in the HW table, or else we may allow one client
or even a third client to proceed ahead of the HW write. My handwave
before was that under the _pathological_ condition we would see the
scratch entry instead of the expected entry, causing a temporary
glitch. That starvation condition will eventually show up in practice, so
fix it.
The reason for the previous cheat was to avoid having to free the extra
allocation while under the spinlock. Now, we keep the extra entry
allocated until the end instead.
v2: Fix error paths for gen6
Fixes: 1d1b5490b91c ("drm/i915/gtt: Replace struct_mutex serialisation for allocation")
Signed-off-by: Chris Wilson <chris@xxxxxxxxxxxxxxxxxx>
Cc: Matthew Auld <matthew.auld@xxxxxxxxx>
Cc: Mika Kuoppala <mika.kuoppala@xxxxxxxxx>
---
drivers/gpu/drm/i915/i915_gem_gtt.c | 130 +++++++++++++++-------------
1 file changed, 72 insertions(+), 58 deletions(-)
diff --git a/drivers/gpu/drm/i915/i915_gem_gtt.c b/drivers/gpu/drm/i915/i915_gem_gtt.c
index 0392a4c4bb9b..0987748d327b 100644
--- a/drivers/gpu/drm/i915/i915_gem_gtt.c
+++ b/drivers/gpu/drm/i915/i915_gem_gtt.c
@@ -1387,82 +1387,88 @@ static int gen8_ppgtt_alloc_pd(struct i915_address_space *vm,
struct i915_page_directory *pd,
u64 start, u64 length)
{
+ struct i915_page_table *alloc = NULL;
struct i915_page_table *pt;
u64 from = start;
unsigned int pde;
+ int ret = 0;
spin_lock(&pd->lock);
gen8_for_each_pde(pt, pd, start, length, pde) {
const int count = gen8_pte_count(start, length);
if (pt == vm->scratch_pt) {
- struct i915_page_table *old;
-
spin_unlock(&pd->lock);
- pt = alloc_pt(vm);
- if (IS_ERR(pt))
+ pt = fetch_and_zero(&alloc);
+ if (!pt)
+ pt = alloc_pt(vm);
+ if (IS_ERR(pt)) {
+ ret = PTR_ERR(pt);
goto unwind;
+ }
if (count < GEN8_PTES || intel_vgpu_active(vm->i915))
gen8_initialize_pt(vm, pt);
- old = cmpxchg(&pd->page_table[pde], vm->scratch_pt, pt);
- if (old == vm->scratch_pt) {
+ spin_lock(&pd->lock);
+ if (pd->page_table[pde] == vm->scratch_pt) {
gen8_ppgtt_set_pde(vm, pd, pt, pde);
+ pd->page_table[pde] = pt;
atomic_inc(&pd->used_pdes);
} else {
- free_pt(vm, pt);
- pt = old;
+ alloc = pt;
+ pt = pd->page_table[pde];
}
-
- spin_lock(&pd->lock);
}
atomic_add(count, &pt->used_ptes);
}
spin_unlock(&pd->lock);
-
- return 0;
+ goto out;
unwind:
gen8_ppgtt_clear_pd(vm, pd, from, start - from);
- return -ENOMEM;
+out:
+ if (alloc)
+ free_pt(vm, alloc);
+ return ret;
}
static int gen8_ppgtt_alloc_pdp(struct i915_address_space *vm,
struct i915_page_directory_pointer *pdp,
u64 start, u64 length)
{
+ struct i915_page_directory *alloc = NULL;
struct i915_page_directory *pd;
u64 from = start;
unsigned int pdpe;
- int ret;
+ int ret = 0;
spin_lock(&pdp->lock);
gen8_for_each_pdpe(pd, pdp, start, length, pdpe) {
if (pd == vm->scratch_pd) {
- struct i915_page_directory *old;
-
spin_unlock(&pdp->lock);
- pd = alloc_pd(vm);
- if (IS_ERR(pd))
+ pd = fetch_and_zero(&alloc);
+ if (!pd)
+ pd = alloc_pd(vm);
+ if (IS_ERR(pd)) {
+ ret = PTR_ERR(pd);
goto unwind;
+ }
gen8_initialize_pd(vm, pd);
- old = cmpxchg(&pdp->page_directory[pdpe],
- vm->scratch_pd, pd);
- if (old == vm->scratch_pd) {
+ spin_lock(&pdp->lock);
+ if (pdp->page_directory[pdpe] == vm->scratch_pd) {
gen8_ppgtt_set_pdpe(vm, pdp, pd, pdpe);
+ pdp->page_directory[pdpe] = pd;
atomic_inc(&pdp->used_pdpes);
} else {
- free_pd(vm, pd);
- pd = old;
+ alloc = pd;
+ pd = pdp->page_directory[pdpe];
}
-
- spin_lock(&pdp->lock);
}
atomic_inc(&pd->used_pdes);
spin_unlock(&pdp->lock);
@@ -1475,8 +1481,7 @@ static int gen8_ppgtt_alloc_pdp(struct i915_address_space *vm,
atomic_dec(&pd->used_pdes);
}
spin_unlock(&pdp->lock);
-
- return 0;
+ goto out;
unwind_pd:
spin_lock(&pdp->lock);
@@ -1489,7 +1494,10 @@ static int gen8_ppgtt_alloc_pdp(struct i915_address_space *vm,
spin_unlock(&pdp->lock);
unwind:
gen8_ppgtt_clear_pdp(vm, pdp, from, start - from);
- return -ENOMEM;
+out:
+ if (alloc)
+ free_pd(vm, alloc);
+ return ret;
}
static int gen8_ppgtt_alloc_3lvl(struct i915_address_space *vm,
@@ -1504,33 +1512,35 @@ static int gen8_ppgtt_alloc_4lvl(struct i915_address_space *vm,
{
struct i915_ppgtt *ppgtt = i915_vm_to_ppgtt(vm);
struct i915_pml4 *pml4 = &ppgtt->pml4;
+ struct i915_page_directory_pointer *alloc = NULL;
struct i915_page_directory_pointer *pdp;
u64 from = start;
+ int ret = 0;
u32 pml4e;
- int ret;
spin_lock(&pml4->lock);
gen8_for_each_pml4e(pdp, pml4, start, length, pml4e) {
if (pdp == vm->scratch_pdp) {
- struct i915_page_directory_pointer *old;
-
spin_unlock(&pml4->lock);
- pdp = alloc_pdp(vm);
- if (IS_ERR(pdp))
+ pdp = fetch_and_zero(&alloc);
+ if (!pdp)
+ pdp = alloc_pdp(vm);
+ if (IS_ERR(pdp)) {
+ ret = PTR_ERR(pdp);
goto unwind;
+ }
gen8_initialize_pdp(vm, pdp);
- old = cmpxchg(&pml4->pdps[pml4e], vm->scratch_pdp, pdp);
- if (old == vm->scratch_pdp) {
+ spin_lock(&pml4->lock);
+ if (pml4->pdps[pml4e] == vm->scratch_pdp) {
gen8_ppgtt_set_pml4e(pml4, pdp, pml4e);
+ pml4->pdps[pml4e] = pdp;
} else {
- free_pdp(vm, pdp);
- pdp = old;
+ alloc = pdp;
+ pdp = pml4->pdps[pml4e];
}
-
- spin_lock(&pml4->lock);
}
atomic_inc(&pdp->used_pdpes);
spin_unlock(&pml4->lock);
@@ -1543,8 +1553,7 @@ static int gen8_ppgtt_alloc_4lvl(struct i915_address_space *vm,
atomic_dec(&pdp->used_pdpes);
}
spin_unlock(&pml4->lock);
-
- return 0;
+ goto out;
unwind_pdp:
spin_lock(&pml4->lock);
@@ -1555,7 +1564,10 @@ static int gen8_ppgtt_alloc_4lvl(struct i915_address_space *vm,
spin_unlock(&pml4->lock);
unwind:
gen8_ppgtt_clear_4lvl(vm, from, start - from);
- return -ENOMEM;
+out:
+ if (alloc)
+ free_pdp(vm, alloc);
+ return ret;
}
static int gen8_preallocate_top_level_pdp(struct i915_ppgtt *ppgtt)
@@ -1820,11 +1832,13 @@ static int gen6_alloc_va_range(struct i915_address_space *vm,
u64 start, u64 length)
{
struct gen6_ppgtt *ppgtt = to_gen6_ppgtt(i915_vm_to_ppgtt(vm));
+ struct i915_page_table *alloc = NULL;
struct i915_page_table *pt;
intel_wakeref_t wakeref;
u64 from = start;
unsigned int pde;
bool flush = false;
+ int ret = 0;
wakeref = intel_runtime_pm_get(&vm->i915->runtime_pm);
@@ -1833,19 +1847,20 @@ static int gen6_alloc_va_range(struct i915_address_space *vm,
const unsigned int count = gen6_pte_count(start, length);
if (pt == vm->scratch_pt) {
- struct i915_page_table *old;
-
spin_unlock(&ppgtt->base.pd.lock);
- pt = alloc_pt(vm);
- if (IS_ERR(pt))
+ pt = fetch_and_zero(&alloc);
+ if (!pt)
+ pt = alloc_pt(vm);
+ if (IS_ERR(pt)) {
+ ret = PTR_ERR(pt);
goto unwind_out;
+ }
gen6_initialize_pt(vm, pt);
- old = cmpxchg(&ppgtt->base.pd.page_table[pde],
- vm->scratch_pt, pt);
- if (old == vm->scratch_pt) {
+ spin_lock(&ppgtt->base.pd.lock);
+ if (ppgtt->base.pd.page_table[pde] == vm->scratch_pt) {
ppgtt->base.pd.page_table[pde] = pt;
if (i915_vma_is_bound(ppgtt->vma,
I915_VMA_GLOBAL_BIND)) {
@@ -1853,11 +1868,9 @@ static int gen6_alloc_va_range(struct i915_address_space *vm,
flush = true;
}
} else {
- free_pt(vm, pt);
- pt = old;
+ alloc = pt;
+ pt = ppgtt->base.pd.page_table[pde];
}
-
- spin_lock(&ppgtt->base.pd.lock);
}
atomic_add(count, &pt->used_ptes);
@@ -1869,14 +1882,15 @@ static int gen6_alloc_va_range(struct i915_address_space *vm,
gen6_ggtt_invalidate(vm->i915);
}
- intel_runtime_pm_put(&vm->i915->runtime_pm, wakeref);
-
- return 0;
+ goto out;
unwind_out:
- intel_runtime_pm_put(&vm->i915->runtime_pm, wakeref);
gen6_ppgtt_clear_range(vm, from, start - from);
- return -ENOMEM;
+out:
+ if (alloc)
+ free_pt(vm, alloc);
+ intel_runtime_pm_put(&vm->i915->runtime_pm, wakeref);
+ return ret;
}
static int gen6_ppgtt_init_scratch(struct gen6_ppgtt *ppgtt)
--
2.20.1
_______________________________________________
Intel-gfx mailing list
Intel-gfx@xxxxxxxxxxxxxxxxxxxxx
https://lists.freedesktop.org/mailman/listinfo/intel-gfx
