[PATCH 1/2] drm,io-mapping: Specify slot to use for atomic mappings

[PATCH 1/2] drm,io-mapping: Specify slot to use for atomic mappings

[Chris Wilson]

This is required should we ever attempt to use an io-mapping where
KM_USER0 is verboten, such as inside an IRQ context.

Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk>
Cc: Eric Anholt <eric@anholt.net>
---
 drivers/gpu/drm/i915/i915_gem.c        |    9 +++++----
 drivers/gpu/drm/i915/intel_overlay.c   |    5 +++--
 drivers/gpu/drm/nouveau/nouveau_bios.c |    8 ++++----
 include/linux/io-mapping.h             |   16 ++++++++++------
 4 files changed, 22 insertions(+), 16 deletions(-)

diff --git a/drivers/gpu/drm/i915/i915_gem.c b/drivers/gpu/drm/i915/i915_gem.c
index 862ce34..4cf2789 100644
--- a/drivers/gpu/drm/i915/i915_gem.c
+++ b/drivers/gpu/drm/i915/i915_gem.c
@@ -496,10 +496,10 @@ fast_user_write(struct io_mapping *mapping,
 	char *vaddr_atomic;
 	unsigned long unwritten;
 
-	vaddr_atomic = io_mapping_map_atomic_wc(mapping, page_base);
+	vaddr_atomic = io_mapping_map_atomic_wc(mapping, page_base, KM_USER0);
 	unwritten = __copy_from_user_inatomic_nocache(vaddr_atomic + page_offset,
 						      user_data, length);
-	io_mapping_unmap_atomic(vaddr_atomic);
+	io_mapping_unmap_atomic(vaddr_atomic, KM_USER0);
 	if (unwritten)
 		return -EFAULT;
 	return 0;
@@ -3288,7 +3288,8 @@ i915_gem_object_pin_and_relocate(struct drm_gem_object *obj,
 		reloc_offset = obj_priv->gtt_offset + reloc->offset;
 		reloc_page = io_mapping_map_atomic_wc(dev_priv->mm.gtt_mapping,
 						      (reloc_offset &
-						       ~(PAGE_SIZE - 1)));
+						       ~(PAGE_SIZE - 1)),
+						      KM_USER0);
 		reloc_entry = (uint32_t __iomem *)(reloc_page +
 						   (reloc_offset & (PAGE_SIZE - 1)));
 		reloc_val = target_obj_priv->gtt_offset + reloc->delta;
@@ -3299,7 +3300,7 @@ i915_gem_object_pin_and_relocate(struct drm_gem_object *obj,
 			  readl(reloc_entry), reloc_val);
 #endif
 		writel(reloc_val, reloc_entry);
-		io_mapping_unmap_atomic(reloc_page);
+		io_mapping_unmap_atomic(reloc_page, KM_USER0);
 
 		/* The updated presumed offset for this entry will be
 		 * copied back out to the user.
diff --git a/drivers/gpu/drm/i915/intel_overlay.c b/drivers/gpu/drm/i915/intel_overlay.c
index f26ec2f..d39aea2 100644
--- a/drivers/gpu/drm/i915/intel_overlay.c
+++ b/drivers/gpu/drm/i915/intel_overlay.c
@@ -185,7 +185,8 @@ static struct overlay_registers *intel_overlay_map_regs_atomic(struct intel_over
 
 	if (OVERLAY_NONPHYSICAL(overlay->dev)) {
 		regs = io_mapping_map_atomic_wc(dev_priv->mm.gtt_mapping,
-				overlay->reg_bo->gtt_offset);
+						overlay->reg_bo->gtt_offset,
+						KM_USER0);
 
 		if (!regs) {
 			DRM_ERROR("failed to map overlay regs in GTT\n");
@@ -200,7 +201,7 @@ static struct overlay_registers *intel_overlay_map_regs_atomic(struct intel_over
 static void intel_overlay_unmap_regs_atomic(struct intel_overlay *overlay)
 {
 	if (OVERLAY_NONPHYSICAL(overlay->dev))
-		io_mapping_unmap_atomic(overlay->virt_addr);
+		io_mapping_unmap_atomic(overlay->virt_addr, KM_USER0);
 
 	overlay->virt_addr = NULL;
 
diff --git a/drivers/gpu/drm/nouveau/nouveau_bios.c b/drivers/gpu/drm/nouveau/nouveau_bios.c
index b59f348..7369b5e 100644
--- a/drivers/gpu/drm/nouveau/nouveau_bios.c
+++ b/drivers/gpu/drm/nouveau/nouveau_bios.c
@@ -2083,11 +2083,11 @@ peek_fb(struct drm_device *dev, struct io_mapping *fb,
 	uint32_t val = 0;
 
 	if (off < pci_resource_len(dev->pdev, 1)) {
-		uint32_t __iomem *p = io_mapping_map_atomic_wc(fb, off);
+		uint32_t __iomem *p = io_mapping_map_atomic_wc(fb, off, KM_USER0);
 
 		val = ioread32(p);
 
-		io_mapping_unmap_atomic(p);
+		io_mapping_unmap_atomic(p, KM_USER0);
 	}
 
 	return val;
@@ -2098,12 +2098,12 @@ poke_fb(struct drm_device *dev, struct io_mapping *fb,
 	uint32_t off, uint32_t val)
 {
 	if (off < pci_resource_len(dev->pdev, 1)) {
-		uint32_t __iomem *p = io_mapping_map_atomic_wc(fb, off);
+		uint32_t __iomem *p = io_mapping_map_atomic_wc(fb, off, KM_USER0);
 
 		iowrite32(val, p);
 		wmb();
 
-		io_mapping_unmap_atomic(p);
+		io_mapping_unmap_atomic(p, KM_USER0);
 	}
 }
 
diff --git a/include/linux/io-mapping.h b/include/linux/io-mapping.h
index 25085dd..e0ea40f 100644
--- a/include/linux/io-mapping.h
+++ b/include/linux/io-mapping.h
@@ -79,7 +79,9 @@ io_mapping_free(struct io_mapping *mapping)
 
 /* Atomic map/unmap */
 static inline void *
-io_mapping_map_atomic_wc(struct io_mapping *mapping, unsigned long offset)
+io_mapping_map_atomic_wc(struct io_mapping *mapping,
+			 unsigned long offset,
+			 int slot)
 {
 	resource_size_t phys_addr;
 	unsigned long pfn;
@@ -87,13 +89,13 @@ io_mapping_map_atomic_wc(struct io_mapping *mapping, unsigned long offset)
 	BUG_ON(offset >= mapping->size);
 	phys_addr = mapping->base + offset;
 	pfn = (unsigned long) (phys_addr >> PAGE_SHIFT);
-	return iomap_atomic_prot_pfn(pfn, KM_USER0, mapping->prot);
+	return iomap_atomic_prot_pfn(pfn, slot, mapping->prot);
 }
 
 static inline void
-io_mapping_unmap_atomic(void *vaddr)
+io_mapping_unmap_atomic(void *vaddr, int slot)
 {
-	iounmap_atomic(vaddr, KM_USER0);
+	iounmap_atomic(vaddr, slot);
 }
 
 static inline void *
@@ -133,13 +135,15 @@ io_mapping_free(struct io_mapping *mapping)
 
 /* Atomic map/unmap */
 static inline void *
-io_mapping_map_atomic_wc(struct io_mapping *mapping, unsigned long offset)
+io_mapping_map_atomic_wc(struct io_mapping *mapping,
+			 unsigned long offset,
+			 int slot)
 {
 	return ((char *) mapping) + offset;
 }
 
 static inline void
-io_mapping_unmap_atomic(void *vaddr)
+io_mapping_unmap_atomic(void *vaddr, int slot)
 {
 }
 
-- 
1.7.1

[PATCH 2/2] drm/i915: Record error batch buffers using iomem

[Chris Wilson]

Directly read the GTT mapping for the contents of the batch buffers
rather than relying on possibly stale CPU caches. Also for completeness
scan the flushing/inactive lists for the current buffers - we are
collecting error state after all.

Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk>
---
 drivers/gpu/drm/i915/i915_irq.c |   64 ++++++++++++++++++++++++++++++++++----
 1 files changed, 57 insertions(+), 7 deletions(-)

diff --git a/drivers/gpu/drm/i915/i915_irq.c b/drivers/gpu/drm/i915/i915_irq.c
index 85785a8..8aed608 100644
--- a/drivers/gpu/drm/i915/i915_irq.c
+++ b/drivers/gpu/drm/i915/i915_irq.c
@@ -425,9 +425,11 @@ static struct drm_i915_error_object *
 i915_error_object_create(struct drm_device *dev,
 			 struct drm_gem_object *src)
 {
+	drm_i915_private_t *dev_priv = dev->dev_private;
 	struct drm_i915_error_object *dst;
 	struct drm_i915_gem_object *src_priv;
 	int page, page_count;
+	u32 reloc_offset;
 
 	if (src == NULL)
 		return NULL;
@@ -442,18 +444,27 @@ i915_error_object_create(struct drm_device *dev,
 	if (dst == NULL)
 		return NULL;
 
+	reloc_offset = src_priv->gtt_offset;
 	for (page = 0; page < page_count; page++) {
-		void *s, *d = kmalloc(PAGE_SIZE, GFP_ATOMIC);
 		unsigned long flags;
+		void __iomem *s;
+		void *d;
 
+		d = kmalloc(PAGE_SIZE, GFP_ATOMIC);
 		if (d == NULL)
 			goto unwind;
+
 		local_irq_save(flags);
-		s = kmap_atomic(src_priv->pages[page], KM_IRQ0);
-		memcpy(d, s, PAGE_SIZE);
-		kunmap_atomic(s, KM_IRQ0);
+		s = io_mapping_map_atomic_wc(dev_priv->mm.gtt_mapping,
+					     reloc_offset,
+					     KM_IRQ0);
+		memcpy_fromio(d, s, PAGE_SIZE);
+		io_mapping_unmap_atomic(s, KM_IRQ0);
 		local_irq_restore(flags);
+
 		dst->pages[page] = d;
+
+		reloc_offset += PAGE_SIZE;
 	}
 	dst->page_count = page_count;
 	dst->gtt_offset = src_priv->gtt_offset;
@@ -612,18 +623,57 @@ static void i915_capture_error_state(struct drm_device *dev)
 
 		if (batchbuffer[1] == NULL &&
 		    error->acthd >= obj_priv->gtt_offset &&
-		    error->acthd < obj_priv->gtt_offset + obj->size &&
-		    batchbuffer[0] != obj)
+		    error->acthd < obj_priv->gtt_offset + obj->size)
 			batchbuffer[1] = obj;
 
 		count++;
 	}
+	/* Scan the other lists for completeness for those bizarre errors. */
+	if (batchbuffer[0] == NULL || batchbuffer[1] == NULL) {
+		list_for_each_entry(obj_priv, &dev_priv->mm.flushing_list, list) {
+			struct drm_gem_object *obj = &obj_priv->base;
+
+			if (batchbuffer[0] == NULL &&
+			    bbaddr >= obj_priv->gtt_offset &&
+			    bbaddr < obj_priv->gtt_offset + obj->size)
+				batchbuffer[0] = obj;
+
+			if (batchbuffer[1] == NULL &&
+			    error->acthd >= obj_priv->gtt_offset &&
+			    error->acthd < obj_priv->gtt_offset + obj->size)
+				batchbuffer[1] = obj;
+
+			if (batchbuffer[0] && batchbuffer[1])
+				break;
+		}
+	}
+	if (batchbuffer[0] == NULL || batchbuffer[1] == NULL) {
+		list_for_each_entry(obj_priv, &dev_priv->mm.inactive_list, list) {
+			struct drm_gem_object *obj = &obj_priv->base;
+
+			if (batchbuffer[0] == NULL &&
+			    bbaddr >= obj_priv->gtt_offset &&
+			    bbaddr < obj_priv->gtt_offset + obj->size)
+				batchbuffer[0] = obj;
+
+			if (batchbuffer[1] == NULL &&
+			    error->acthd >= obj_priv->gtt_offset &&
+			    error->acthd < obj_priv->gtt_offset + obj->size)
+				batchbuffer[1] = obj;
+
+			if (batchbuffer[0] && batchbuffer[1])
+				break;
+		}
+	}
 
 	/* We need to copy these to an anonymous buffer as the simplest
 	 * method to avoid being overwritten by userpace.
 	 */
 	error->batchbuffer[0] = i915_error_object_create(dev, batchbuffer[0]);
-	error->batchbuffer[1] = i915_error_object_create(dev, batchbuffer[1]);
+	if (batchbuffer[1] != batchbuffer[0])
+		error->batchbuffer[1] = i915_error_object_create(dev, batchbuffer[1]);
+	else
+		error->batchbuffer[1] = NULL;
 
 	/* Record the ringbuffer */
 	error->ringbuffer = i915_error_object_create(dev,
-- 
1.7.1