Skip to contentSkip to conversation

Use drm_clflush* instead of clflush

PATCH Superseded Michael Cheng 2 participants 8 messages Feb 10, 2022

Subsystems (3)

v8 of 13 Superseded Latest is v13 →

lore
Share & tools
lore
#0 lore raw

v8 cover Use drm_clflush* instead of clflush

This patch series re-work a few i915 functions to use drm_clflush_virt_range instead of calling clflush or clflushopt directly. This will prevent errors when building for non-x86 architectures.

v2: s/PAGE_SIZE/sizeof(value) for Re-work intel_write_status_page and added more patches to convert additional clflush/clflushopt to use drm_clflush*. (Michael Cheng)

v3: Drop invalidate_csb_entries and directly invoke drm_clflush_virt_ran

v4: Remove extra memory barriers

v5: s/cache_clflush_range/drm_clflush_virt_range

v6: Fix up "Drop invalidate_csb_entries" to use correct parameters. Also added in arm64 support for drm_clflush_virt_range.

v7: Re-order patches, and use correct macro for dcache flush for arm64.

v8: Remove ifdef for asm/cacheflush.

Michael Cheng (6):
drm: Add arch arm64 for drm_clflush_virt_range
drm/i915/gt: Re-work intel_write_status_page
drm/i915/gt: Drop invalidate_csb_entries
drm/i915/gt: Re-work reset_csb
drm/i915/: Re-work clflush_write32
drm/i915/gt: replace cache_clflush_range

 drivers/gpu/drm/drm_cache.c                   |  5 +++++
 .../gpu/drm/i915/gem/i915_gem_execbuffer.c    |  8 +++-----
 drivers/gpu/drm/i915/gt/gen8_ppgtt.c          | 12 ++++++------
 drivers/gpu/drm/i915/gt/intel_engine.h        | 13 ++++---------
 .../drm/i915/gt/intel_execlists_submission.c  | 19 ++++++-------------
 drivers/gpu/drm/i915/gt/intel_gtt.c           |  2 +-
 drivers/gpu/drm/i915/gt/intel_ppgtt.c         |  2 +-
 .../gpu/drm/i915/gt/uc/intel_guc_submission.c |  2 +-
 8 files changed, 27 insertions(+), 36 deletions(-)
Signature · 1 line
2.25.1
#1 lore raw

PATCH v8 1/6 drm: Add arch arm64 for drm_clflush_virt_range

Add arm64 support for drm_clflush_virt_range. dcache_clean_inval_poc performs a flush by first performing a clean, follow by an invalidation operation.

v2 (Michael Cheng): Use correct macro for cleaning and invalidation the
dcache.

v3 (Michael Cheng): Remove ifdef for asm/cacheflush.h

---
 drivers/gpu/drm/drm_cache.c | 5 +++++
 1 file changed, 5 insertions(+)
drivers/gpu/drm/drm_cache.c +5 -0
index f19d9acbe959..f167bc19afb9 100644
@@ -28,6 +28,7 @@
  * Authors: Thomas Hellström <thomas-at-tungstengraphics-dot-com>
  */
 
+#include <asm/cacheflush.h>
 #include <linux/dma-buf-map.h>
 #include <linux/export.h>
 #include <linux/highmem.h>
@@ -176,6 +177,10 @@ drm_clflush_virt_range(void *addr, unsigned long length)
 
 	if (wbinvd_on_all_cpus())
 		pr_err("Timed out waiting for cache flush\n");
+
+#elif defined(CONFIG_ARM64)
+	void *end = addr + length;
+	dcache_clean_inval_poc((unsigned long)addr, (unsigned long)end);
 #else
 	pr_err("Architecture has no drm_cache.c support\n");
 	WARN_ON_ONCE(1);
Signature · 1 line
2.25.1
#2 lore raw

PATCH v8 2/6 drm/i915/gt: Re-work intel_write_status_page

In reply to Michael Cheng

Re-work intel_write_status_page to use drm_clflush_virt_range. This will prevent compiler errors when building for non-x86 architectures.

---
 drivers/gpu/drm/i915/gt/intel_engine.h | 13 ++++---------
 1 file changed, 4 insertions(+), 9 deletions(-)
drivers/gpu/drm/i915/gt/intel_engine.h +4 -9
index 0e353d8c2bc8..986777c2430d 100644
@@ -4,6 +4,7 @@
 
 #include <asm/cacheflush.h>
 #include <drm/drm_util.h>
+#include <drm/drm_cache.h>
 
 #include <linux/hashtable.h>
 #include <linux/irq_work.h>
@@ -143,15 +144,9 @@ intel_write_status_page(struct intel_engine_cs *engine, int reg, u32 value)
 	 * of extra paranoia to try and ensure that the HWS takes the value
 	 * we give and that it doesn't end up trapped inside the CPU!
 	 */
-	if (static_cpu_has(X86_FEATURE_CLFLUSH)) {
-		mb();
-		clflush(&engine->status_page.addr[reg]);
-		engine->status_page.addr[reg] = value;
-		clflush(&engine->status_page.addr[reg]);
-		mb();
-	} else {
-		WRITE_ONCE(engine->status_page.addr[reg], value);
-	}
+	drm_clflush_virt_range(&engine->status_page.addr[reg], sizeof(value));
+	WRITE_ONCE(engine->status_page.addr[reg], value);
+	drm_clflush_virt_range(&engine->status_page.addr[reg], sizeof(value));
 }
 
 /*
Signature · 1 line
2.25.1
#3 lore raw

PATCH v8 3/6 drm/i915/gt: Drop invalidate_csb_entries

In reply to Michael Cheng

Drop invalidate_csb_entries and directly call drm_clflush_virt_range. This allows for one less function call, and prevent complier errors when building for non-x86 architectures.

v2(Michael Cheng): Drop invalidate_csb_entries function and directly
invoke drm_clflush_virt_range. Thanks to Tvrtko for the sugguestion.

v3(Michael Cheng): Use correct parameters for drm_clflush_virt_range.
Thanks to Tvrtko for pointing this out.

---
 drivers/gpu/drm/i915/gt/intel_execlists_submission.c | 12 +++---------
 1 file changed, 3 insertions(+), 9 deletions(-)
drivers/gpu/drm/i915/gt/intel_execlists_submission.c +3 -9
index 9bb7c863172f..28f2581d3046 100644
@@ -1646,12 +1646,6 @@ cancel_port_requests(struct intel_engine_execlists * const execlists,
 	return inactive;
 }
 
-static void invalidate_csb_entries(const u64 *first, const u64 *last)
-{
-	clflush((void *)first);
-	clflush((void *)last);
-}
-
 /*
  * Starting with Gen12, the status has a new format:
  *
@@ -1999,7 +1993,7 @@ process_csb(struct intel_engine_cs *engine, struct i915_request **inactive)
 	 * the wash as hardware, working or not, will need to do the
 	 * invalidation before.
 	 */
-	invalidate_csb_entries(&buf[0], &buf[num_entries - 1]);
+	drm_clflush_virt_range(&buf[0], num_entries * sizeof(buf[0]));
 
 	/*
 	 * We assume that any event reflects a change in context flow
@@ -2783,8 +2777,8 @@ static void reset_csb_pointers(struct intel_engine_cs *engine)
 
 	/* Check that the GPU does indeed update the CSB entries! */
 	memset(execlists->csb_status, -1, (reset_value + 1) * sizeof(u64));
-	invalidate_csb_entries(&execlists->csb_status[0],
-			       &execlists->csb_status[reset_value]);
+	drm_clflush_virt_range(&execlists->csb_status[0],
+		execlists->csb_size * sizeof(execlists->csb_status[0]));
 
 	/* Once more for luck and our trusty paranoia */
 	ENGINE_WRITE(engine, RING_CONTEXT_STATUS_PTR,
Signature · 1 line
2.25.1
#4 lore raw

PATCH v8 4/6 drm/i915/gt: Re-work reset_csb

In reply to Michael Cheng

Use drm_clflush_virt_range instead of directly invoking clflush. This will prevent compiler errors when building for non-x86 architectures.

v2(Michael Cheng): Remove extra clflush

v3(Michael Cheng): Remove memory barrier since drm_clflush_virt_range
takes care of it.

---
 drivers/gpu/drm/i915/gt/intel_execlists_submission.c | 5 ++---
 1 file changed, 2 insertions(+), 3 deletions(-)
drivers/gpu/drm/i915/gt/intel_execlists_submission.c +2 -3
index 28f2581d3046..cc561cfae808 100644
@@ -2944,9 +2944,8 @@ reset_csb(struct intel_engine_cs *engine, struct i915_request **inactive)
 {
 	struct intel_engine_execlists * const execlists = &engine->execlists;
 
-	mb(); /* paranoia: read the CSB pointers from after the reset */
-	clflush(execlists->csb_write);
-	mb();
+	drm_clflush_virt_range(execlists->csb_write,
+			sizeof(execlists->csb_write));
 
 	inactive = process_csb(engine, inactive); /* drain preemption events */
 
Signature · 1 line
2.25.1
#5 lore raw

PATCH v8 5/6 drm/i915/: Re-work clflush_write32

In reply to Michael Cheng

Use drm_clflush_virt_range instead of clflushopt and remove the memory barrier, since drm_clflush_virt_range takes care of that.

---
 drivers/gpu/drm/i915/gem/i915_gem_execbuffer.c | 8 +++-----
 1 file changed, 3 insertions(+), 5 deletions(-)
drivers/gpu/drm/i915/gem/i915_gem_execbuffer.c +3 -5
index 498b458fd784..0854276ff7ba 100644
@@ -1332,10 +1332,8 @@ static void *reloc_vaddr(struct i915_vma *vma,
 static void clflush_write32(u32 *addr, u32 value, unsigned int flushes)
 {
 	if (unlikely(flushes & (CLFLUSH_BEFORE | CLFLUSH_AFTER))) {
-		if (flushes & CLFLUSH_BEFORE) {
-			clflushopt(addr);
-			mb();
-		}
+		if (flushes & CLFLUSH_BEFORE)
+			drm_clflush_virt_range(addr, sizeof(addr));
 
 		*addr = value;
 
@@ -1347,7 +1345,7 @@ static void clflush_write32(u32 *addr, u32 value, unsigned int flushes)
 		 * to ensure ordering of clflush wrt to the system.
 		 */
 		if (flushes & CLFLUSH_AFTER)
-			clflushopt(addr);
+			drm_clflush_virt_range(addr, sizeof(addr));
 	} else
 		*addr = value;
 }
Signature · 1 line
2.25.1
#6 lore raw

PATCH v8 6/6 drm/i915/gt: replace cache_clflush_range

In reply to Michael Cheng

Replace all occurance of cache_clflush_range with drm_clflush_virt_range. This will prevent compile errors on non-x86 platforms.

---
 drivers/gpu/drm/i915/gt/gen8_ppgtt.c                 | 12 ++++++------
 drivers/gpu/drm/i915/gt/intel_execlists_submission.c |  2 +-
 drivers/gpu/drm/i915/gt/intel_gtt.c                  |  2 +-
 drivers/gpu/drm/i915/gt/intel_ppgtt.c                |  2 +-
 drivers/gpu/drm/i915/gt/uc/intel_guc_submission.c    |  2 +-
 5 files changed, 10 insertions(+), 10 deletions(-)
drivers/gpu/drm/i915/gt/gen8_ppgtt.c +6 -6
index c43e724afa9f..d0999e92621b 100644
@@ -444,11 +444,11 @@ gen8_ppgtt_insert_pte(struct i915_ppgtt *ppgtt,
 				pd = pdp->entry[gen8_pd_index(idx, 2)];
 			}
 
-			clflush_cache_range(vaddr, PAGE_SIZE);
+			drm_clflush_virt_range(vaddr, PAGE_SIZE);
 			vaddr = px_vaddr(i915_pt_entry(pd, gen8_pd_index(idx, 1)));
 		}
 	} while (1);
-	clflush_cache_range(vaddr, PAGE_SIZE);
+	drm_clflush_virt_range(vaddr, PAGE_SIZE);
 
 	return idx;
 }
@@ -532,7 +532,7 @@ static void gen8_ppgtt_insert_huge(struct i915_address_space *vm,
 			}
 		} while (rem >= page_size && index < I915_PDES);
 
-		clflush_cache_range(vaddr, PAGE_SIZE);
+		drm_clflush_virt_range(vaddr, PAGE_SIZE);
 
 		/*
 		 * Is it safe to mark the 2M block as 64K? -- Either we have
@@ -548,7 +548,7 @@ static void gen8_ppgtt_insert_huge(struct i915_address_space *vm,
 					      I915_GTT_PAGE_SIZE_2M)))) {
 			vaddr = px_vaddr(pd);
 			vaddr[maybe_64K] |= GEN8_PDE_IPS_64K;
-			clflush_cache_range(vaddr, PAGE_SIZE);
+			drm_clflush_virt_range(vaddr, PAGE_SIZE);
 			page_size = I915_GTT_PAGE_SIZE_64K;
 
 			/*
@@ -569,7 +569,7 @@ static void gen8_ppgtt_insert_huge(struct i915_address_space *vm,
 				for (i = 1; i < index; i += 16)
 					memset64(vaddr + i, encode, 15);
 
-				clflush_cache_range(vaddr, PAGE_SIZE);
+				drm_clflush_virt_range(vaddr, PAGE_SIZE);
 			}
 		}
 
@@ -617,7 +617,7 @@ static void gen8_ppgtt_insert_entry(struct i915_address_space *vm,
 
 	vaddr = px_vaddr(i915_pt_entry(pd, gen8_pd_index(idx, 1)));
 	vaddr[gen8_pd_index(idx, 0)] = gen8_pte_encode(addr, level, flags);
-	clflush_cache_range(&vaddr[gen8_pd_index(idx, 0)], sizeof(*vaddr));
+	drm_clflush_virt_range(&vaddr[gen8_pd_index(idx, 0)], sizeof(*vaddr));
 }
 
 static int gen8_init_scratch(struct i915_address_space *vm)
drivers/gpu/drm/i915/gt/intel_execlists_submission.c +1 -1
index cc561cfae808..bbe33794b34d 100644
@@ -2822,7 +2822,7 @@ static void execlists_sanitize(struct intel_engine_cs *engine)
 	sanitize_hwsp(engine);
 
 	/* And scrub the dirty cachelines for the HWSP */
-	clflush_cache_range(engine->status_page.addr, PAGE_SIZE);
+	drm_clflush_virt_range(engine->status_page.addr, PAGE_SIZE);
 
 	intel_engine_reset_pinned_contexts(engine);
 }
drivers/gpu/drm/i915/gt/intel_gtt.c +1 -1
index 0d6bbc8c57f2..9b594be9102f 100644
@@ -255,7 +255,7 @@ fill_page_dma(struct drm_i915_gem_object *p, const u64 val, unsigned int count)
 	void *vaddr = __px_vaddr(p);
 
 	memset64(vaddr, val, count);
-	clflush_cache_range(vaddr, PAGE_SIZE);
+	drm_clflush_virt_range(vaddr, PAGE_SIZE);
 }
 
 static void poison_scratch_page(struct drm_i915_gem_object *scratch)
drivers/gpu/drm/i915/gt/intel_ppgtt.c +1 -1
index 48e6e2f87700..bd474a5123cb 100644
@@ -90,7 +90,7 @@ write_dma_entry(struct drm_i915_gem_object * const pdma,
 	u64 * const vaddr = __px_vaddr(pdma);
 
 	vaddr[idx] = encoded_entry;
-	clflush_cache_range(&vaddr[idx], sizeof(u64));
+	drm_clflush_virt_range(&vaddr[idx], sizeof(u64));
 }
 
 void
drivers/gpu/drm/i915/gt/uc/intel_guc_submission.c +1 -1
index b3a429a92c0d..89020706adc4 100644
@@ -3573,7 +3573,7 @@ static void guc_sanitize(struct intel_engine_cs *engine)
 	sanitize_hwsp(engine);
 
 	/* And scrub the dirty cachelines for the HWSP */
-	clflush_cache_range(engine->status_page.addr, PAGE_SIZE);
+	drm_clflush_virt_range(engine->status_page.addr, PAGE_SIZE);
 
 	intel_engine_reset_pinned_contexts(engine);
 }
Signature · 1 line
2.25.1
#7 lore raw

[Intel-gfx] ✗ Fi.CI.BUILD: failure for Use drm_clflush* instead of clflush (rev7)

In reply to Michael Cheng
== Series Details ==

== Series Details ==

Series: Use drm_clflush* instead of clflush (rev7)
URL : https://patchwork.freedesktop.org/series/99450/
State : failure

== Summary ==

Applying: drm: Add arch arm64 for drm_clflush_virt_range Using index info to reconstruct a base tree...
M drivers/gpu/drm/drm_cache.c
Falling back to patching base and 3-way merge... Auto-merging drivers/gpu/drm/drm_cache.c
CONFLICT (content): Merge conflict in drivers/gpu/drm/drm_cache.c
error: Failed to merge in the changes.
hint: Use 'git am --show-current-patch=diff' to see the failed patch Patch failed at 0001 drm: Add arch arm64 for drm_clflush_virt_range When you have resolved this problem, run "git am --continue". If you prefer to skip this patch, run "git am --skip" instead. To restore the original branch and stop patching, run "git am --abort".