@@ -813,6 +813,42 @@ __clear_pd_entry(struct i915_page_directory * const pd,
#define clear_pd_entry(pd, pde, to) \
__clear_pd_entry((pd), (pde), px_base(to), gen8_pde_encode)
+static bool
+release_pt_entry(struct i915_page_directory * const pd,
+ const unsigned short pde,
+ struct i915_page_table * const pt,
+ struct i915_page_table * const scratch)
+{
+ bool free = false;
+
+ spin_lock(&pd->lock);
+ if (atomic_dec_and_test(&pt->used)) {
+ clear_pd_entry(pd, pde, scratch);
+ free = true;
+ }
+ spin_unlock(&pd->lock);
+
+ return free;
+}
+
+static bool
+release_pd_entry(struct i915_page_directory * const pd,
+ const unsigned short pde,
+ struct i915_page_directory * const entry,
+ struct i915_page_directory * const scratch)
+{
+ bool free = false;
+
+ spin_lock(&pd->lock);
+ if (atomic_dec_and_test(&entry->used)) {
+ clear_pd_entry(pd, pde, scratch);
+ free = true;
+ }
+ spin_unlock(&pd->lock);
+
+ return free;
+}
+
/*
* PDE TLBs are a pain to invalidate on GEN8+. When we modify
* the page table structures, we mark them dirty so that
@@ -827,11 +863,11 @@ static void mark_tlbs_dirty(struct i915_ppgtt *ppgtt)
/* Removes entries from a single page table, releasing it if it's empty.
* Caller can use the return value to update higher-level entries.
*/
-static bool gen8_ppgtt_clear_pt(const struct i915_address_space *vm,
+static void gen8_ppgtt_clear_pt(const struct i915_address_space *vm,
struct i915_page_table *pt,
u64 start, u64 length)
{
- unsigned int num_entries = gen8_pte_count(start, length);
+ const unsigned int num_entries = gen8_pte_count(start, length);
gen8_pte_t *vaddr;
vaddr = kmap_atomic_px(pt);
@@ -839,10 +875,11 @@ static bool gen8_ppgtt_clear_pt(const struct i915_address_space *vm,
kunmap_atomic(vaddr);
GEM_BUG_ON(num_entries > atomic_read(&pt->used));
- return !atomic_sub_return(num_entries, &pt->used);
+
+ atomic_sub(num_entries, &pt->used);
}
-static bool gen8_ppgtt_clear_pd(struct i915_address_space *vm,
+static void gen8_ppgtt_clear_pd(struct i915_address_space *vm,
struct i915_page_directory *pd,
u64 start, u64 length)
{
@@ -850,30 +887,21 @@ static bool gen8_ppgtt_clear_pd(struct i915_address_space *vm,
u32 pde;
gen8_for_each_pde(pt, pd, start, length, pde) {
- bool free = false;
-
GEM_BUG_ON(pt == vm->scratch_pt);
- if (!gen8_ppgtt_clear_pt(vm, pt, start, length))
- continue;
+ atomic_inc(&pt->used);
- spin_lock(&pd->lock);
- if (!atomic_read(&pt->used)) {
- clear_pd_entry(pd, pde, vm->scratch_pt);
- free = true;
- }
- spin_unlock(&pd->lock);
- if (free)
+ gen8_ppgtt_clear_pt(vm, pt, start, length);
+
+ if (release_pt_entry(pd, pde, pt, vm->scratch_pt))
free_pt(vm, pt);
}
-
- return !atomic_read(&pd->used);
}
/* Removes entries from a single page dir pointer, releasing it if it's empty.
* Caller can use the return value to update higher-level entries
*/
-static bool gen8_ppgtt_clear_pdp(struct i915_address_space *vm,
+static void gen8_ppgtt_clear_pdp(struct i915_address_space *vm,
struct i915_page_directory * const pdp,
u64 start, u64 length)
{
@@ -881,24 +909,15 @@ static bool gen8_ppgtt_clear_pdp(struct i915_address_space *vm,
unsigned int pdpe;
gen8_for_each_pdpe(pd, pdp, start, length, pdpe) {
- bool free = false;
-
GEM_BUG_ON(pd == vm->scratch_pd);
- if (!gen8_ppgtt_clear_pd(vm, pd, start, length))
- continue;
+ atomic_inc(&pd->used);
- spin_lock(&pdp->lock);
- if (!atomic_read(&pd->used)) {
- clear_pd_entry(pdp, pdpe, vm->scratch_pd);
- free = true;
- }
- spin_unlock(&pdp->lock);
- if (free)
+ gen8_ppgtt_clear_pd(vm, pd, start, length);
+
+ if (release_pd_entry(pdp, pdpe, pd, vm->scratch_pd))
free_pd(vm, pd);
}
-
- return !atomic_read(&pdp->used);
}
static void gen8_ppgtt_clear_3lvl(struct i915_address_space *vm,
@@ -922,19 +941,13 @@ static void gen8_ppgtt_clear_4lvl(struct i915_address_space *vm,
GEM_BUG_ON(!i915_vm_is_4lvl(vm));
gen8_for_each_pml4e(pdp, pml4, start, length, pml4e) {
- bool free = false;
GEM_BUG_ON(pdp == vm->scratch_pdp);
- if (!gen8_ppgtt_clear_pdp(vm, pdp, start, length))
- continue;
+ atomic_inc(&pdp->used);
- spin_lock(&pml4->lock);
- if (!atomic_read(&pdp->used)) {
- clear_pd_entry(pml4, pml4e, vm->scratch_pdp);
- free = true;
- }
- spin_unlock(&pml4->lock);
- if (free)
+ gen8_ppgtt_clear_pdp(vm, pdp, start, length);
+
+ if (release_pd_entry(pml4, pml4e, pdp, vm->scratch_pdp))
free_pd(vm, pdp);
}
}
@@ -1457,17 +1470,8 @@ static int gen8_ppgtt_alloc_pdp(struct i915_address_space *vm,
goto out;
unwind_pd:
- if (alloc) {
- free_pd(vm, alloc);
- alloc = NULL;
- }
- spin_lock(&pdp->lock);
- if (atomic_dec_and_test(&pd->used)) {
- GEM_BUG_ON(alloc);
- alloc = pd; /* defer the free to after the lock */
- clear_pd_entry(pdp, pdpe, vm->scratch_pd);
- }
- spin_unlock(&pdp->lock);
+ if (release_pd_entry(pdp, pdpe, pd, vm->scratch_pd))
+ free_pd(vm, pd);
unwind:
gen8_ppgtt_clear_pdp(vm, pdp, from, start - from);
out:
@@ -1530,17 +1534,8 @@ static int gen8_ppgtt_alloc_4lvl(struct i915_address_space *vm,
goto out;
unwind_pdp:
- if (alloc) {
- free_pd(vm, alloc);
- alloc = NULL;
- }
- spin_lock(&pml4->lock);
- if (atomic_dec_and_test(&pdp->used)) {
- GEM_BUG_ON(alloc);
- alloc = pdp; /* defer the free until after the lock */
- clear_pd_entry(pml4, pml4e, vm->scratch_pdp);
- }
- spin_unlock(&pml4->lock);
+ if (release_pd_entry(pml4, pml4e, pdp, vm->scratch_pdp))
+ free_pd(vm, pdp);
unwind:
gen8_ppgtt_clear_4lvl(vm, from, start - from);
out:
@@ -1572,8 +1567,8 @@ static int gen8_preallocate_top_level_pdp(struct i915_ppgtt *ppgtt)
unwind:
start -= from;
gen8_for_each_pdpe(pd, pdp, from, start, pdpe) {
- clear_pd_entry(pdp, pdpe, vm->scratch_pd);
- free_pd(vm, pd);
+ if (release_pd_entry(pdp, pdpe, pd, vm->scratch_pd))
+ free_pd(vm, pd);
}
atomic_set(&pdp->used, 0);
return -ENOMEM;
By encapsulating the locking upper level and used check for entry into a helper function, we can use it in all callsites. v2: get rid of atomic_reads on lower level clears (Chris) Cc: Chris Wilson <chris@chris-wilson.co.uk> Signed-off-by: Mika Kuoppala <mika.kuoppala@linux.intel.com> --- drivers/gpu/drm/i915/i915_gem_gtt.c | 125 +++++++++++++--------------- 1 file changed, 60 insertions(+), 65 deletions(-)