[Intel-gfx] [PATCH] drm/i915/gtt: Introduce release_pd_entry
Chris Wilson
chris at chris-wilson.co.uk
Fri Jul 5 15:07:13 UTC 2019
From: Mika Kuoppala <mika.kuoppala at linux.intel.com>
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 at chris-wilson.co.uk>
Signed-off-by: Mika Kuoppala <mika.kuoppala at linux.intel.com>
---
drivers/gpu/drm/i915/i915_gem_gtt.c | 121 +++++++++++-----------------
1 file changed, 47 insertions(+), 74 deletions(-)
diff --git a/drivers/gpu/drm/i915/i915_gem_gtt.c b/drivers/gpu/drm/i915/i915_gem_gtt.c
index 0a55b0932c86..dd581e038fce 100644
--- a/drivers/gpu/drm/i915/i915_gem_gtt.c
+++ b/drivers/gpu/drm/i915/i915_gem_gtt.c
@@ -795,10 +795,10 @@ __set_pd_entry(struct i915_page_directory * const pd,
}
static inline void
-__clear_pd_entry(struct i915_page_directory * const pd,
- const unsigned short pde,
- struct i915_page_dma * const to,
- u64 (*encode)(const dma_addr_t, const enum i915_cache_level))
+clear_pd_entry(struct i915_page_directory * const pd,
+ const unsigned short pde,
+ struct i915_page_dma * const to,
+ u64 (*encode)(const dma_addr_t, const enum i915_cache_level))
{
GEM_BUG_ON(atomic_read(&pd->used) == 0);
@@ -810,8 +810,23 @@ __clear_pd_entry(struct i915_page_directory * const pd,
#define set_pd_entry(pd, pde, to) \
__set_pd_entry((pd), (pde), px_base(to), gen8_pde_encode)
-#define clear_pd_entry(pd, pde, to) \
- __clear_pd_entry((pd), (pde), px_base(to), gen8_pde_encode)
+static bool
+release_pd_entry(struct i915_page_directory * const pd,
+ const unsigned short pde,
+ atomic_t *counter,
+ struct i915_page_dma * const scratch)
+{
+ bool free = false;
+
+ spin_lock(&pd->lock);
+ if (atomic_dec_and_test(counter)) {
+ clear_pd_entry(pd, pde, scratch, gen8_pde_encode);
+ free = true;
+ }
+ spin_unlock(&pd->lock);
+
+ return free;
+}
/*
* PDE TLBs are a pain to invalidate on GEN8+. When we modify
@@ -827,11 +842,11 @@ 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 +854,11 @@ 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 +866,22 @@ 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_pd_entry(pd, pde, &pt->used,
+ px_base(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 +889,16 @@ 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->used,
+ px_base(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 +922,14 @@ 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->used,
+ px_base(vm->scratch_pdp)))
free_pd(vm, pdp);
}
}
@@ -1457,17 +1452,8 @@ 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->used, px_base(vm->scratch_pd)))
+ free_pd(vm, pd);
unwind:
gen8_ppgtt_clear_pdp(vm, pdp, from, start - from);
out:
@@ -1530,17 +1516,8 @@ 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->used, px_base(vm->scratch_pdp)))
+ free_pd(vm, pdp);
unwind:
gen8_ppgtt_clear_4lvl(vm, from, start - from);
out:
@@ -1570,11 +1547,7 @@ static int gen8_preallocate_top_level_pdp(struct i915_ppgtt *ppgtt)
return 0;
unwind:
- start -= from;
- gen8_for_each_pdpe(pd, pdp, from, start, pdpe) {
- clear_pd_entry(pdp, pdpe, vm->scratch_pd);
- free_pd(vm, pd);
- }
+ gen8_ppgtt_clear_pdp(vm, pdp, from, start - from);
atomic_set(&pdp->used, 0);
return -ENOMEM;
}
--
2.20.1
More information about the Intel-gfx
mailing list