[PATCH] drm/msm: Optimize adreno_show_object()

[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index]

 



When the userspace tries to read the crashstate dump, the read side
implementation in the driver currently ascii85 encodes all the binary
buffers and it does this each time the read system call is called.
A userspace tool like cat typically does a page by page read and the
number of read calls depends on the size of the data captured by the
driver. This is certainly not desirable and does not scale well with
large captures.

This patch encodes the buffer only once in the read path. With this there
is an immediate >10X speed improvement in crashstate save time.

Signed-off-by: Sharat Masetty <smasetty@xxxxxxxxxxxxxx>
---
This is a different and a simpler version of [1]. We move the encode back to
read time and make sure each buffer is encoded just once. This is showing
similar performance numbers as [1] for similar capture sizes.

1) https://patchwork.freedesktop.org/patch/259627/

 drivers/gpu/drm/msm/adreno/adreno_gpu.c | 82 ++++++++++++++++++++++++---------
 drivers/gpu/drm/msm/msm_gpu.h           |  2 +
 2 files changed, 63 insertions(+), 21 deletions(-)

diff --git a/drivers/gpu/drm/msm/adreno/adreno_gpu.c b/drivers/gpu/drm/msm/adreno/adreno_gpu.c
index 141062f..6251e76 100644
--- a/drivers/gpu/drm/msm/adreno/adreno_gpu.c
+++ b/drivers/gpu/drm/msm/adreno/adreno_gpu.c
@@ -406,7 +406,7 @@ int adreno_gpu_state_get(struct msm_gpu *gpu, struct msm_gpu_state *state)
 				size = j + 1;

 		if (size) {
-			state->ring[i].data = kmalloc(size << 2, GFP_KERNEL);
+			state->ring[i].data = kvmalloc(size << 2, GFP_KERNEL);
 			if (state->ring[i].data) {
 				memcpy(state->ring[i].data, gpu->rb[i]->start, size << 2);
 				state->ring[i].data_size = size << 2;
@@ -445,7 +445,7 @@ void adreno_gpu_state_destroy(struct msm_gpu_state *state)
 	int i;

 	for (i = 0; i < ARRAY_SIZE(state->ring); i++)
-		kfree(state->ring[i].data);
+		kvfree(state->ring[i].data);

 	for (i = 0; state->bos && i < state->nr_bos; i++)
 		kvfree(state->bos[i].data);
@@ -475,34 +475,74 @@ int adreno_gpu_state_put(struct msm_gpu_state *state)

 #if defined(CONFIG_DEBUG_FS) || defined(CONFIG_DEV_COREDUMP)

-static void adreno_show_object(struct drm_printer *p, u32 *ptr, int len)
+static char *adreno_gpu_ascii85_encode(u32 *src, size_t len)
 {
+	void *buf;
+	size_t buf_itr = 0, buffer_size;
 	char out[ASCII85_BUFSZ];
-	long l, datalen, i;
+	long l;
+	int i;

-	if (!ptr || !len)
-		return;
+	if (!src || !len)
+		return NULL;
+
+	l = ascii85_encode_len(len);

 	/*
-	 * Only dump the non-zero part of the buffer - rarely will any data
-	 * completely fill the entire allocated size of the buffer
+	 * Ascii85 outputs either a 5 byte string or a 1 byte string. So we
+	 * account for the worst case of 5 bytes per dword plus the 1 for '\0'
 	 */
-	for (datalen = 0, i = 0; i < len >> 2; i++) {
-		if (ptr[i])
-			datalen = (i << 2) + 1;
-	}
+	buffer_size = (l * 5) + 1;
+
+	buf = kvmalloc(buffer_size, GFP_KERNEL);
+	if (!buf)
+		return NULL;

-	/* Skip printing the object if it is empty */
-	if (datalen == 0)
+	for (i = 0; i < l; i++)
+		buf_itr += snprintf(buf + buf_itr, buffer_size - buf_itr, "%s",
+				ascii85_encode(src[i], out));
+
+	return buf;
+}
+
+/* len is expected to be in bytes */
+static void adreno_show_object(struct drm_printer *p, void **ptr, int len,
+		bool *encoded)
+{
+	if (!*ptr || !len)
 		return;

-	l = ascii85_encode_len(datalen);
+	if (!*encoded) {
+		long datalen, i;
+		u32 *buf = *ptr;
+
+		/*
+		 * Only dump the non-zero part of the buffer - rarely will
+		 * any data completely fill the entire allocated size of
+		 * the buffer.
+		 */
+		for (datalen = 0, i = 0; i < len >> 2; i++)
+			if (buf[i])
+				datalen = ((i + 1) << 2);
+
+		/*
+		 * If we reach here, then the originally captured binary buffer
+		 * will be replaced with the ascii85 encoded string
+		 */
+		*ptr = adreno_gpu_ascii85_encode(buf, datalen);
+
+		kvfree(buf);
+
+		*encoded = true;
+	}
+
+	if (!*ptr)
+		return;

 	drm_puts(p, "    data: !!ascii85 |\n");
 	drm_puts(p, "     ");

-	for (i = 0; i < l; i++)
-		drm_puts(p, ascii85_encode(ptr[i], out));
+	drm_puts(p, *ptr);

 	drm_puts(p, "\n");
 }
@@ -534,8 +574,8 @@ void adreno_show(struct msm_gpu *gpu, struct msm_gpu_state *state,
 		drm_printf(p, "    wptr: %d\n", state->ring[i].wptr);
 		drm_printf(p, "    size: %d\n", MSM_GPU_RINGBUFFER_SZ);

-		adreno_show_object(p, state->ring[i].data,
-			state->ring[i].data_size);
+		adreno_show_object(p, &state->ring[i].data,
+			state->ring[i].data_size, &state->ring[i].encoded);
 	}

 	if (state->bos) {
@@ -546,8 +586,8 @@ void adreno_show(struct msm_gpu *gpu, struct msm_gpu_state *state,
 				state->bos[i].iova);
 			drm_printf(p, "    size: %zd\n", state->bos[i].size);

-			adreno_show_object(p, state->bos[i].data,
-				state->bos[i].size);
+			adreno_show_object(p, &state->bos[i].data,
+				state->bos[i].size, &state->bos[i].encoded);
 		}
 	}

diff --git a/drivers/gpu/drm/msm/msm_gpu.h b/drivers/gpu/drm/msm/msm_gpu.h
index f82bac0..efb49bb 100644
--- a/drivers/gpu/drm/msm/msm_gpu.h
+++ b/drivers/gpu/drm/msm/msm_gpu.h
@@ -187,6 +187,7 @@ struct msm_gpu_state_bo {
 	u64 iova;
 	size_t size;
 	void *data;
+	bool encoded;
 };

 struct msm_gpu_state {
@@ -201,6 +202,7 @@ struct msm_gpu_state {
 		u32 wptr;
 		void *data;
 		int data_size;
+		bool encoded;
 	} ring[MSM_GPU_MAX_RINGS];

 	int nr_registers;
--
1.9.1

_______________________________________________
dri-devel mailing list
dri-devel@xxxxxxxxxxxxxxxxxxxxx
https://lists.freedesktop.org/mailman/listinfo/dri-devel




[Index of Archives]     [Linux DRI Users]     [Linux Intel Graphics]     [Linux USB Devel]     [Video for Linux]     [Linux Audio Users]     [Yosemite News]     [Linux Kernel]     [Linux SCSI]     [XFree86]     [Linux USB Devel]     [Video for Linux]     [Linux Audio Users]     [Linux Kernel]     [Linux SCSI]     [XFree86]
  Powered by Linux