[Intel-gfx] [PATCH 33/33] drm/i915: Compress GPU objects in error state

Chris Wilson chris at chris-wilson.co.uk
Sun Aug 7 14:45:41 UTC 2016


Our error states are quickly growing, pinning kernel memory with them.
The majority of the space is taken up by the error objects. These
compress well using zlib and without decode are mostly meaningless, so
encoding them does not hinder quickly parsing the error state for
familiarity.

Signed-off-by: Chris Wilson <chris at chris-wilson.co.uk>
---
 drivers/gpu/drm/i915/Kconfig          |  1 +
 drivers/gpu/drm/i915/i915_drv.h       |  3 +-
 drivers/gpu/drm/i915/i915_gpu_error.c | 96 ++++++++++++++++++++++++++++++-----
 3 files changed, 85 insertions(+), 15 deletions(-)

diff --git a/drivers/gpu/drm/i915/Kconfig b/drivers/gpu/drm/i915/Kconfig
index 7badcee88ebf..c8ea20526aef 100644
--- a/drivers/gpu/drm/i915/Kconfig
+++ b/drivers/gpu/drm/i915/Kconfig
@@ -5,6 +5,7 @@ config DRM_I915
 	select INTEL_GTT
 	select INTERVAL_TREE
 	select STOP_MACHINE
+	select ZLIB_DEFLATE
 	# we need shmfs for the swappable backing store, and in particular
 	# the shmem_readpage() which depends upon tmpfs
 	select SHMEM
diff --git a/drivers/gpu/drm/i915/i915_drv.h b/drivers/gpu/drm/i915/i915_drv.h
index b5abf88e908e..7d4038a2525a 100644
--- a/drivers/gpu/drm/i915/i915_drv.h
+++ b/drivers/gpu/drm/i915/i915_drv.h
@@ -551,9 +551,10 @@ struct drm_i915_error_state {
 		u32 semaphore_mboxes[I915_NUM_ENGINES - 1];
 
 		struct drm_i915_error_object {
-			int page_count;
 			u64 gtt_offset;
 			u64 gtt_size;
+			int page_count;
+			int unused;
 			u32 *pages[0];
 		} *ringbuffer, *batchbuffer, *wa_batchbuffer, *ctx, *hws_page;
 
diff --git a/drivers/gpu/drm/i915/i915_gpu_error.c b/drivers/gpu/drm/i915/i915_gpu_error.c
index a0e906bc64dc..80c4ca3d4db9 100644
--- a/drivers/gpu/drm/i915/i915_gpu_error.c
+++ b/drivers/gpu/drm/i915/i915_gpu_error.c
@@ -29,6 +29,7 @@
 
 #include <generated/utsrelease.h>
 #include <linux/stop_machine.h>
+#include <linux/zlib.h>
 #include "i915_drv.h"
 
 static const char *engine_str(int engine)
@@ -309,12 +310,30 @@ void i915_error_printf(struct drm_i915_error_state_buf *e, const char *f, ...)
 	va_end(args);
 }
 
+static bool
+ascii85_encode(u32 in, char *out)
+{
+	int i;
+
+	if (in == 0)
+		return false;
+
+	out[5] = '\0';
+	for (i = 5; i--; ) {
+		out[i] = '!' + in % 85;
+		in /= 85;
+	}
+
+	return true;
+}
+
 static void print_error_obj(struct drm_i915_error_state_buf *m,
 			    struct intel_engine_cs *engine,
 			    const char *name,
 			    struct drm_i915_error_object *obj)
 {
-	int page, offset, elt;
+	char out[6];
+	int page;
 
 	if (!obj)
 		return;
@@ -326,13 +345,23 @@ static void print_error_obj(struct drm_i915_error_state_buf *m,
 			   lower_32_bits(obj->gtt_offset));
 	}
 
-	for (page = offset = 0; page < obj->page_count; page++) {
-		for (elt = 0; elt < PAGE_SIZE/4; elt++) {
-			err_printf(m, "%08x :  %08x\n", offset,
-				   obj->pages[page][elt]);
-			offset += 4;
+	err_puts(m, ":"); /* indicate compressed data */
+	for (page = 0; page < obj->page_count; page++) {
+		int i, len;
+
+		len = PAGE_SIZE;
+		if (page == obj->page_count - 1)
+			len -= obj->unused;
+		len = (len + 3) / 4;
+
+		for (i = 0; i < len; i++) {
+			if (ascii85_encode(obj->pages[page][i], out))
+				err_puts(m, out);
+			else
+				err_puts(m, "z");
 		}
 	}
+	err_puts(m, "\n");
 }
 
 int i915_error_state_to_str(struct drm_i915_error_state_buf *m,
@@ -593,17 +622,37 @@ static void i915_error_state_free(struct kref *error_ref)
 	kfree(error);
 }
 
-static int compress_page(void *src, struct drm_i915_error_object *dst)
+static int compress_page(struct z_stream_s *zstream,
+			 void *src,
+			 struct drm_i915_error_object *dst)
 {
-	unsigned long page;
+	zstream->next_in = src;
+	zstream->avail_in = PAGE_SIZE;
 
-	page = __get_free_page(GFP_ATOMIC | __GFP_NOWARN);
-	if (!page)
-		return -ENOMEM;
+	do {
+		if (zstream->avail_out == 0) {
+			unsigned long page;
+
+			page = __get_free_page(GFP_ATOMIC | __GFP_NOWARN);
+			if (!page)
+				return -ENOMEM;
+
+			dst->pages[dst->page_count++] = (void *)page;
+
+			zstream->next_out = (void *)page;
+			zstream->avail_out = PAGE_SIZE;
+		}
 
-	dst->pages[dst->page_count++] = (void *)page;
+		if (zlib_deflate(zstream, Z_SYNC_FLUSH) != Z_OK)
+			return -EIO;
+
+#if 0
+		/* XXX fallback to uncompressed if we increases size? */
+		if (zstream->total_out > zstream->total_in)
+			return -E2BIG;
+#endif
+	} while (zstream->avail_in);
 
-	memcpy((void *)page, src, PAGE_SIZE);
 	return 0;
 }
 
@@ -614,6 +663,7 @@ i915_error_object_create(struct drm_i915_private *i915,
 	struct i915_ggtt *ggtt = &i915->ggtt;
 	const u64 slot = ggtt->gpu_error.start;
 	struct drm_i915_error_object *dst;
+	struct z_stream_s zstream;
 	unsigned long num_pages;
 	struct sgt_iter iter;
 	dma_addr_t dma;
@@ -622,6 +672,7 @@ i915_error_object_create(struct drm_i915_private *i915,
 		return NULL;
 
 	num_pages = min_t(u64, vma->size, vma->obj->base.size) >> PAGE_SHIFT;
+	num_pages = DIV_ROUND_UP(10 * num_pages, 8); /* worstcase zlib growth */
 	dst = kmalloc(sizeof(*dst) + num_pages * sizeof(u32 *),
 		      GFP_ATOMIC | __GFP_NOWARN);
 	if (!dst)
@@ -629,6 +680,18 @@ i915_error_object_create(struct drm_i915_private *i915,
 
 	dst->gtt_offset = vma->node.start;
 	dst->page_count = 0;
+	dst->unused = 0;
+
+	memset(&zstream, 0, sizeof(zstream));
+	zstream.workspace = kmalloc(zlib_deflate_workspacesize(MAX_WBITS,
+							       MAX_MEM_LEVEL),
+				    GFP_ATOMIC | __GFP_NOWARN);
+	if (!zstream.workspace ||
+	    zlib_deflateInit(&zstream, Z_DEFAULT_COMPRESSION) != Z_OK) {
+		kfree(zstream.workspace);
+		kfree(dst);
+		return NULL;
+	}
 
 	for_each_sgt_dma(dma, iter,
 			 vma->ggtt_view.pages ?: vma->obj->pages) {
@@ -640,13 +703,18 @@ i915_error_object_create(struct drm_i915_private *i915,
 
 		s = (void *__force)
 			io_mapping_map_atomic_wc(ggtt->mappable, slot);
-		ret = compress_page(s, dst);
+		ret = compress_page(&zstream, s, dst);
 		io_mapping_unmap_atomic(s);
 
 		if (ret)
 			goto unwind;
 	}
+	zlib_deflate(&zstream, Z_FINISH);
+	dst->unused = zstream.avail_out;
 out:
+	zlib_deflateEnd(&zstream);
+	kfree(zstream.workspace);
+
 	ggtt->base.clear_range(&ggtt->base, slot, PAGE_SIZE, true);
 	return dst;
 
-- 
2.8.1



More information about the Intel-gfx mailing list