[RFC v3 0/8] drm/msm: GPU crash state

[RFC v3 0/8] drm/msm: GPU crash state

[Jordan Crouse]

This is revision 3 of my stack implementing a GPU crash state for drm/msm
(https://patchwork.freedesktop.org/series/36097/).

The goal is to store and provide enough information to debug software
and hardware issues on the Adreno hardware in a semi human-readable
format that can also be parsed by scripts.

So far this is a relatively basic dump of registers and data but future patches
will add more details and target specific information.

You can see an example of the output for a simple invalid opcode error on the
db820c here: https://hastebin.com/olaruyakaz.bash

v3: Make recommended changes to ascii85 per Chris Wilson. Use devcoredump to
dump crash states as suggested by Bjorn Andersson and add a new drm_print
facility to facilitate that. Remove the now obsolete 'crash' debugfs node.
Add documentation for the crash dump output.

v2: Convert output to yaml, use ascii85 to dump ringbuffer contents.

Jordan Crouse (8):
  include: Move ascii85 functions from i915 to linux/ascii85.h
  drm: drm_printer: Add printer for devcoredump
  drm/msm/gpu: Capture the state of the GPU
  drm/msm/gpu: Convert the GPU show function to use the GPU state
  drm/msm/gpu: Capture the GPU state on a GPU hang
  drm/msm/adreno: Convert the show/crash file format
  drm/msm/adreno: Add ringbuffer data to the GPU state
  drm/msm/adreno: Add a5xx specific registers for the GPU state

 Documentation/gpu/drm-msm-crash-dump.txt |  34 +++++
 drivers/gpu/drm/drm_print.c              |  54 +++++++
 drivers/gpu/drm/i915/i915_gpu_error.c    |  38 +----
 drivers/gpu/drm/msm/Kconfig              |   1 +
 drivers/gpu/drm/msm/adreno/a3xx_gpu.c    |  28 ++--
 drivers/gpu/drm/msm/adreno/a4xx_gpu.c    |  20 ++-
 drivers/gpu/drm/msm/adreno/a5xx_gpu.c    | 239 +++++++++++++++++++++++++++++--
 drivers/gpu/drm/msm/adreno/adreno_gpu.c  | 146 ++++++++++++++++---
 drivers/gpu/drm/msm/adreno/adreno_gpu.h  |   7 +-
 drivers/gpu/drm/msm/msm_debugfs.c        |  24 +++-
 drivers/gpu/drm/msm/msm_gpu.c            | 103 +++++++++++--
 drivers/gpu/drm/msm/msm_gpu.h            |  58 +++++++-
 include/drm/drm_print.h                  |  27 ++++
 include/linux/ascii85.h                  |  39 +++++
 14 files changed, 721 insertions(+), 97 deletions(-)
 create mode 100644 Documentation/gpu/drm-msm-crash-dump.txt
 create mode 100644 include/linux/ascii85.h

-- 
1.9.1

_______________________________________________
Freedreno mailing list
Freedreno@lists.freedesktop.org
https://lists.freedesktop.org/mailman/listinfo/freedreno

[PATCH 1/8] include: Move ascii85 functions from i915 to linux/ascii85.h

[Jordan Crouse]

The i915 DRM driver very cleverly used ascii85 encoding for their
GPU state file. Move the encode functions to a general header file to
support other drivers that might be interested in the same
functionality.

[v2 - Update the API to be cleaner for the caller suggested by
Chris Wilson]

Signed-off-by: Jordan Crouse <jcrouse@codeaurora.org>
---
 drivers/gpu/drm/i915/i915_gpu_error.c | 38 ++++++----------------------------
 include/linux/ascii85.h               | 39 +++++++++++++++++++++++++++++++++++
 2 files changed, 45 insertions(+), 32 deletions(-)
 create mode 100644 include/linux/ascii85.h

diff --git a/drivers/gpu/drm/i915/i915_gpu_error.c b/drivers/gpu/drm/i915/i915_gpu_error.c
index 9440593..8bc72c7 100644
--- a/drivers/gpu/drm/i915/i915_gpu_error.c
+++ b/drivers/gpu/drm/i915/i915_gpu_error.c
@@ -31,6 +31,7 @@
 #include <linux/stop_machine.h>
 #include <linux/zlib.h>
 #include <drm/drm_print.h>
+#include <linux/ascii85.h>
 
 #include "i915_drv.h"
 
@@ -502,35 +503,12 @@ void i915_error_printf(struct drm_i915_error_state_buf *e, const char *f, ...)
 	va_end(args);
 }
 
-static int
-ascii85_encode_len(int len)
-{
-	return DIV_ROUND_UP(len, 4);
-}
-
-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)
 {
-	char out[6];
+	char out[ASCII85_BUFSZ];
 	int page;
 
 	if (!obj)
@@ -545,19 +523,15 @@ static void print_error_obj(struct drm_i915_error_state_buf *m,
 
 	err_compression_marker(m);
 	for (page = 0; page < obj->page_count; page++) {
-		int i, len;
+		int i;
+		long len = PAGE_SIZE;
 
-		len = PAGE_SIZE;
 		if (page == obj->page_count - 1)
 			len -= obj->unused;
 		len = ascii85_encode_len(len);
 
-		for (i = 0; i < len; i++) {
-			if (ascii85_encode(obj->pages[page][i], out))
-				err_puts(m, out);
-			else
-				err_puts(m, "z");
-		}
+		for (i = 0; i < len; i++)
+			error_puts(m, ascii85_encode(obj->pages[page][i], out));
 	}
 	err_puts(m, "\n");
 }
diff --git a/include/linux/ascii85.h b/include/linux/ascii85.h
new file mode 100644
index 0000000..322bbed
--- /dev/null
+++ b/include/linux/ascii85.h
@@ -0,0 +1,39 @@
+
+/*
+ * SPDX-License-Identifier: GPL-2.0
+ *
+ * Copyright (c) 2008 Intel Corporation
+ * Copyright (c) The Linux Foundation. All rights reserved.
+ */
+
+#ifndef _ASCII85_H_
+#define _ASCII85_H_
+
+#include <linux/kernel.h>
+
+#define ASCII85_BUFSZ 6
+
+static inline long
+ascii85_encode_len(long len)
+{
+	return DIV_ROUND_UP(len, 4);
+}
+
+static inline char *
+ascii85_encode(u32 in, char *out)
+{
+	int i;
+
+	if (in == 0)
+		return "z";
+
+	out[5] = '\0';
+	for (i = 5; i--; ) {
+		out[i] = '!' + in % 85;
+		in /= 85;
+	}
+
+	return out;
+}
+
+#endif
-- 
1.9.1

_______________________________________________
Freedreno mailing list
Freedreno@lists.freedesktop.org
https://lists.freedesktop.org/mailman/listinfo/freedreno

[PATCH 2/8] drm: drm_printer: Add printer for devcoredump

[Jordan Crouse]

Add a drm printer suitable for use with the read callback for
devcoredump or any other file operation read() function that
isn't otherwise covered by seq_file.

Signed-off-by: Jordan Crouse <jcrouse@codeaurora.org>
---
 drivers/gpu/drm/drm_print.c | 54 +++++++++++++++++++++++++++++++++++++++++++++
 include/drm/drm_print.h     | 27 +++++++++++++++++++++++
 2 files changed, 81 insertions(+)

diff --git a/drivers/gpu/drm/drm_print.c b/drivers/gpu/drm/drm_print.c
index 781518f..f6efde4 100644
--- a/drivers/gpu/drm/drm_print.c
+++ b/drivers/gpu/drm/drm_print.c
@@ -30,6 +30,60 @@
 #include <drm/drmP.h>
 #include <drm/drm_print.h>
 
+void __drm_printfn_coredump(struct drm_printer *p, struct va_format *vaf)
+{
+	struct drm_print_iterator *iterator = p->arg;
+
+	ssize_t len;
+
+	if (!iterator->remain)
+		return;
+
+	if (iterator->offset < iterator->start) {
+		char *buf;
+		ssize_t copy;
+
+		/* Figure out how big the string will be */
+		len = snprintf(NULL, 0, "%pV", vaf);
+
+		if (iterator->offset + len <= iterator->start) {
+			iterator->offset += len;
+			return;
+		}
+
+		/* Print the string into a temporary buffer */
+		buf = kmalloc(len + 1,
+			GFP_KERNEL | __GFP_NOWARN | __GFP_NORETRY);
+		if (!buf)
+			return;
+
+		snprintf(buf, len + 1, "%pV", vaf);
+
+		copy = len - (iterator->start - iterator->offset);
+
+		if (copy > iterator->remain)
+			copy = iterator->remain;
+
+		/* Copy out the bit of the string that we need */
+		memcpy(iterator->data,
+			buf + (iterator->start - iterator->offset), copy);
+
+		iterator->offset = iterator->start + copy;
+		iterator->remain -= copy;
+
+		kfree(buf);
+	} else {
+		ssize_t pos = iterator->offset - iterator->start;
+
+		len = scnprintf(((char *) iterator->data) + pos,
+			iterator->remain, "%pV", vaf);
+
+		iterator->offset += len;
+		iterator->remain -= len;
+	}
+}
+EXPORT_SYMBOL(__drm_printfn_coredump);
+
 void __drm_printfn_seq_file(struct drm_printer *p, struct va_format *vaf)
 {
 	seq_printf(p->arg, "%pV", vaf);
diff --git a/include/drm/drm_print.h b/include/drm/drm_print.h
index 2a4a42e..29eee51 100644
--- a/include/drm/drm_print.h
+++ b/include/drm/drm_print.h
@@ -73,6 +73,7 @@ struct drm_printer {
 	const char *prefix;
 };
 
+void __drm_printfn_coredump(struct drm_printer *p, struct va_format *vaf);
 void __drm_printfn_seq_file(struct drm_printer *p, struct va_format *vaf);
 void __drm_printfn_info(struct drm_printer *p, struct va_format *vaf);
 void __drm_printfn_debug(struct drm_printer *p, struct va_format *vaf);
@@ -104,6 +105,32 @@ struct drm_printer {
 #define drm_printf_indent(printer, indent, fmt, ...) \
 	drm_printf((printer), "%.*s" fmt, (indent), "\t\t\t\t\tX", ##__VA_ARGS__)
 
+struct drm_print_iterator {
+	void *data;
+
+	ssize_t start;
+	ssize_t offset;
+	ssize_t remain;
+};
+
+/**
+ * drm_coredump_printer - construct a &drm_printer that can output to a buffer
+ * from the read function for devcoredump
+ * @iter: A pointer to a struct drm_print_iterator for the read instance
+ *
+ * RETURNS:
+ * The &drm_printer object
+ */
+static inline struct drm_printer
+drm_coredump_printer(struct drm_print_iterator *iter)
+{
+	struct drm_printer p = {
+		.printfn = __drm_printfn_coredump,
+		.arg = iter,
+	};
+	return p;
+}
+
 /**
  * drm_seq_file_printer - construct a &drm_printer that outputs to &seq_file
  * @f:  the &struct seq_file to output to
-- 
1.9.1

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

[PATCH 3/8] drm/msm/gpu: Capture the state of the GPU

[Jordan Crouse]

Add the infrastructure to capture the state current state of the
GPU and store it in memory. This is useful for storing the state
of a hung GPU so it can be dumped later.

For now grab the same basic ringbuffer information and registers
that are provided by the debugfs 'gpu' node but obviously this can
be extended to capture a much larger set of GPU information.

Signed-off-by: Jordan Crouse <jcrouse@codeaurora.org>
---
 drivers/gpu/drm/msm/adreno/a3xx_gpu.c   | 15 +++++++++
 drivers/gpu/drm/msm/adreno/a4xx_gpu.c   | 14 +++++++++
 drivers/gpu/drm/msm/adreno/a5xx_gpu.c   | 22 ++++++++++++++
 drivers/gpu/drm/msm/adreno/adreno_gpu.c | 54 +++++++++++++++++++++++++++++++++
 drivers/gpu/drm/msm/adreno/adreno_gpu.h |  3 ++
 drivers/gpu/drm/msm/msm_gpu.h           | 19 ++++++++++++
 6 files changed, 127 insertions(+)

diff --git a/drivers/gpu/drm/msm/adreno/a3xx_gpu.c b/drivers/gpu/drm/msm/adreno/a3xx_gpu.c
index 1dd84d3..c351292 100644
--- a/drivers/gpu/drm/msm/adreno/a3xx_gpu.c
+++ b/drivers/gpu/drm/msm/adreno/a3xx_gpu.c
@@ -426,6 +426,19 @@ static void a3xx_dump(struct msm_gpu *gpu)
 			gpu_read(gpu, REG_A3XX_RBBM_STATUS));
 	adreno_dump(gpu);
 }
+
+static struct msm_gpu_state *a3xx_gpu_state_get(struct msm_gpu *gpu)
+{
+	struct msm_gpu_state *state = adreno_gpu_state_get(gpu);
+
+	if (IS_ERR(state))
+		return state;
+
+	state->rbbm_status = gpu_read(gpu, REG_A3XX_RBBM_STATUS);
+
+	return state;
+}
+
 /* Register offset defines for A3XX */
 static const unsigned int a3xx_register_offsets[REG_ADRENO_REGISTER_MAX] = {
 	REG_ADRENO_DEFINE(REG_ADRENO_CP_RB_BASE, REG_AXXX_CP_RB_BASE),
@@ -452,6 +465,8 @@ static void a3xx_dump(struct msm_gpu *gpu)
 #ifdef CONFIG_DEBUG_FS
 		.show = a3xx_show,
 #endif
+		.gpu_state_get = a3xx_gpu_state_get,
+		.gpu_state_put = adreno_gpu_state_put,
 	},
 };
 
diff --git a/drivers/gpu/drm/msm/adreno/a4xx_gpu.c b/drivers/gpu/drm/msm/adreno/a4xx_gpu.c
index 2884b1b..faf5d60 100644
--- a/drivers/gpu/drm/msm/adreno/a4xx_gpu.c
+++ b/drivers/gpu/drm/msm/adreno/a4xx_gpu.c
@@ -464,6 +464,18 @@ static void a4xx_show(struct msm_gpu *gpu, struct seq_file *m)
 }
 #endif
 
+static struct msm_gpu_state *a4xx_gpu_state_get(struct msm_gpu *gpu)
+{
+	struct msm_gpu_state *state = adreno_gpu_state_get(gpu);
+
+	if (IS_ERR(state))
+		return state;
+
+	state->rbbm_status = gpu_read(gpu, REG_A4XX_RBBM_STATUS);
+
+	return state;
+}
+
 /* Register offset defines for A4XX, in order of enum adreno_regs */
 static const unsigned int a4xx_register_offsets[REG_ADRENO_REGISTER_MAX] = {
 	REG_ADRENO_DEFINE(REG_ADRENO_CP_RB_BASE, REG_A4XX_CP_RB_BASE),
@@ -540,6 +552,8 @@ static int a4xx_get_timestamp(struct msm_gpu *gpu, uint64_t *value)
 #ifdef CONFIG_DEBUG_FS
 		.show = a4xx_show,
 #endif
+		.gpu_state_get = a4xx_gpu_state_get,
+		.gpu_state_put = adreno_gpu_state_put,
 	},
 	.get_timestamp = a4xx_get_timestamp,
 };
diff --git a/drivers/gpu/drm/msm/adreno/a5xx_gpu.c b/drivers/gpu/drm/msm/adreno/a5xx_gpu.c
index a4f68af..08f2579 100644
--- a/drivers/gpu/drm/msm/adreno/a5xx_gpu.c
+++ b/drivers/gpu/drm/msm/adreno/a5xx_gpu.c
@@ -1195,6 +1195,26 @@ static int a5xx_get_timestamp(struct msm_gpu *gpu, uint64_t *value)
 	return 0;
 }
 
+static struct msm_gpu_state *a5xx_gpu_state_get(struct msm_gpu *gpu)
+{
+	struct msm_gpu_state *state;
+
+	/*
+	 * Temporarily disable hardware clock gating before going into
+	 * adreno_show to avoid issues while reading the registers
+	 */
+	a5xx_set_hwcg(gpu, false);
+
+	state = adreno_gpu_state_get(gpu);
+
+	if (!IS_ERR(state))
+		state->rbbm_status = gpu_read(gpu, REG_A5XX_RBBM_STATUS);
+
+	a5xx_set_hwcg(gpu, true);
+
+	return state;
+}
+
 #ifdef CONFIG_DEBUG_FS
 static void a5xx_show(struct msm_gpu *gpu, struct seq_file *m)
 {
@@ -1244,6 +1264,8 @@ static int a5xx_gpu_busy(struct msm_gpu *gpu, uint64_t *value)
 		.debugfs_init = a5xx_debugfs_init,
 #endif
 		.gpu_busy = a5xx_gpu_busy,
+		.gpu_state_get = a5xx_gpu_state_get,
+		.gpu_state_put = adreno_gpu_state_put,
 	},
 	.get_timestamp = a5xx_get_timestamp,
 };
diff --git a/drivers/gpu/drm/msm/adreno/adreno_gpu.c b/drivers/gpu/drm/msm/adreno/adreno_gpu.c
index 87133c6c..35b77f0 100644
--- a/drivers/gpu/drm/msm/adreno/adreno_gpu.c
+++ b/drivers/gpu/drm/msm/adreno/adreno_gpu.c
@@ -382,6 +382,60 @@ bool adreno_idle(struct msm_gpu *gpu, struct msm_ringbuffer *ring)
 	return false;
 }
 
+struct msm_gpu_state *adreno_gpu_state_get(struct msm_gpu *gpu)
+{
+	struct adreno_gpu *adreno_gpu = to_adreno_gpu(gpu);
+	struct msm_gpu_state *state;
+	int i, count = 0;
+
+	state = kzalloc(sizeof(*state), GFP_KERNEL);
+	if (!state)
+		return ERR_PTR(-ENOMEM);
+
+	do_gettimeofday(&state->time);
+
+	for (i = 0; i < gpu->nr_rings; i++) {
+		state->ring[i].fence = gpu->rb[i]->memptrs->fence;
+		state->ring[i].seqno = gpu->rb[i]->seqno;
+		state->ring[i].rptr = get_rptr(adreno_gpu, gpu->rb[i]);
+		state->ring[i].wptr = get_wptr(gpu->rb[i]);
+	}
+
+	/* Count the number of registers */
+	for (i = 0; adreno_gpu->registers[i] != ~0; i += 2)
+		count += adreno_gpu->registers[i + 1] -
+			adreno_gpu->registers[i] + 1;
+
+	state->registers = kcalloc(count * 2, sizeof(u32), GFP_KERNEL);
+	if (state->registers) {
+		int pos = 0;
+
+		for (i = 0; adreno_gpu->registers[i] != ~0; i += 2) {
+			uint32_t start = adreno_gpu->registers[i];
+			uint32_t end   = adreno_gpu->registers[i+1];
+			uint32_t addr;
+
+			for (addr = start; addr <= end; addr++) {
+				state->registers[pos++] = addr;
+				state->registers[pos++] = gpu_read(gpu, addr);
+			}
+		}
+
+		state->nr_registers = count;
+	}
+
+	return state;
+}
+
+void adreno_gpu_state_put(struct msm_gpu_state *state)
+{
+	if (IS_ERR_OR_NULL(state))
+		return;
+
+	kfree(state->registers);
+	kfree(state);
+}
+
 #ifdef CONFIG_DEBUG_FS
 void adreno_show(struct msm_gpu *gpu, struct seq_file *m)
 {
diff --git a/drivers/gpu/drm/msm/adreno/adreno_gpu.h b/drivers/gpu/drm/msm/adreno/adreno_gpu.h
index d6b0e7b..0beb078 100644
--- a/drivers/gpu/drm/msm/adreno/adreno_gpu.h
+++ b/drivers/gpu/drm/msm/adreno/adreno_gpu.h
@@ -228,6 +228,9 @@ int adreno_gpu_init(struct drm_device *drm, struct platform_device *pdev,
 void adreno_gpu_cleanup(struct adreno_gpu *gpu);
 
 
+struct msm_gpu_state *adreno_gpu_state_get(struct msm_gpu *gpu);
+void adreno_gpu_state_put(struct msm_gpu_state *state);
+
 /* ringbuffer helpers (the parts that are adreno specific) */
 
 static inline void
diff --git a/drivers/gpu/drm/msm/msm_gpu.h b/drivers/gpu/drm/msm/msm_gpu.h
index b824117..4be72a6 100644
--- a/drivers/gpu/drm/msm/msm_gpu.h
+++ b/drivers/gpu/drm/msm/msm_gpu.h
@@ -27,6 +27,7 @@
 
 struct msm_gem_submit;
 struct msm_gpu_perfcntr;
+struct msm_gpu_state;
 
 struct msm_gpu_config {
 	const char *ioname;
@@ -69,6 +70,8 @@ struct msm_gpu_funcs {
 	int (*debugfs_init)(struct msm_gpu *gpu, struct drm_minor *minor);
 #endif
 	int (*gpu_busy)(struct msm_gpu *gpu, uint64_t *value);
+	struct msm_gpu_state *(*gpu_state_get)(struct msm_gpu *gpu);
+	void (*gpu_state_put)(struct msm_gpu_state *state);
 };
 
 struct msm_gpu {
@@ -175,6 +178,22 @@ struct msm_gpu_submitqueue {
 	struct kref ref;
 };
 
+struct msm_gpu_state {
+	struct timeval time;
+
+	struct {
+		u32 fence;
+		u32 seqno;
+		u32 rptr;
+		u32 wptr;
+	} ring[MSM_GPU_MAX_RINGS];
+
+	int nr_registers;
+	u32 *registers;
+
+	u32 rbbm_status;
+};
+
 static inline void gpu_write(struct msm_gpu *gpu, u32 reg, u32 data)
 {
 	msm_writel(data, gpu->mmio + (reg << 2));
-- 
1.9.1

_______________________________________________
Freedreno mailing list
Freedreno@lists.freedesktop.org
https://lists.freedesktop.org/mailman/listinfo/freedreno

[PATCH 4/8] drm/msm/gpu: Convert the GPU show function to use the GPU state

[Jordan Crouse]

Convert the existing GPU show function to use the GPU state to
dump the information rather than reading it directly from the hardware.
This will require an additional step to capture the state before
dumping it for the existing nodes but it will greatly facilitate reusing
the same code for dumping a previously captured state from a GPU hang.

Signed-off-by: Jordan Crouse <jcrouse@codeaurora.org>
---
 drivers/gpu/drm/msm/adreno/a3xx_gpu.c   | 11 +----------
 drivers/gpu/drm/msm/adreno/a4xx_gpu.c   | 12 +-----------
 drivers/gpu/drm/msm/adreno/a5xx_gpu.c   | 18 +-----------------
 drivers/gpu/drm/msm/adreno/adreno_gpu.c | 30 +++++++++++++-----------------
 drivers/gpu/drm/msm/adreno/adreno_gpu.h |  4 ++--
 drivers/gpu/drm/msm/msm_debugfs.c       | 21 +++++++++++++++------
 drivers/gpu/drm/msm/msm_gpu.h           |  3 ++-
 7 files changed, 35 insertions(+), 64 deletions(-)

diff --git a/drivers/gpu/drm/msm/adreno/a3xx_gpu.c b/drivers/gpu/drm/msm/adreno/a3xx_gpu.c
index c351292..4b9284a 100644
--- a/drivers/gpu/drm/msm/adreno/a3xx_gpu.c
+++ b/drivers/gpu/drm/msm/adreno/a3xx_gpu.c
@@ -410,15 +410,6 @@ static irqreturn_t a3xx_irq(struct msm_gpu *gpu)
 	~0   /* sentinel */
 };
 
-#ifdef CONFIG_DEBUG_FS
-static void a3xx_show(struct msm_gpu *gpu, struct seq_file *m)
-{
-	seq_printf(m, "status:   %08x\n",
-			gpu_read(gpu, REG_A3XX_RBBM_STATUS));
-	adreno_show(gpu, m);
-}
-#endif
-
 /* would be nice to not have to duplicate the _show() stuff with printk(): */
 static void a3xx_dump(struct msm_gpu *gpu)
 {
@@ -463,7 +454,7 @@ static struct msm_gpu_state *a3xx_gpu_state_get(struct msm_gpu *gpu)
 		.irq = a3xx_irq,
 		.destroy = a3xx_destroy,
 #ifdef CONFIG_DEBUG_FS
-		.show = a3xx_show,
+		.show = adreno_show,
 #endif
 		.gpu_state_get = a3xx_gpu_state_get,
 		.gpu_state_put = adreno_gpu_state_put,
diff --git a/drivers/gpu/drm/msm/adreno/a4xx_gpu.c b/drivers/gpu/drm/msm/adreno/a4xx_gpu.c
index faf5d60..b8dcbb1 100644
--- a/drivers/gpu/drm/msm/adreno/a4xx_gpu.c
+++ b/drivers/gpu/drm/msm/adreno/a4xx_gpu.c
@@ -454,16 +454,6 @@ static irqreturn_t a4xx_irq(struct msm_gpu *gpu)
 	~0 /* sentinel */
 };
 
-#ifdef CONFIG_DEBUG_FS
-static void a4xx_show(struct msm_gpu *gpu, struct seq_file *m)
-{
-	seq_printf(m, "status:   %08x\n",
-			gpu_read(gpu, REG_A4XX_RBBM_STATUS));
-	adreno_show(gpu, m);
-
-}
-#endif
-
 static struct msm_gpu_state *a4xx_gpu_state_get(struct msm_gpu *gpu)
 {
 	struct msm_gpu_state *state = adreno_gpu_state_get(gpu);
@@ -550,7 +540,7 @@ static int a4xx_get_timestamp(struct msm_gpu *gpu, uint64_t *value)
 		.irq = a4xx_irq,
 		.destroy = a4xx_destroy,
 #ifdef CONFIG_DEBUG_FS
-		.show = a4xx_show,
+		.show = adreno_show,
 #endif
 		.gpu_state_get = a4xx_gpu_state_get,
 		.gpu_state_put = adreno_gpu_state_put,
diff --git a/drivers/gpu/drm/msm/adreno/a5xx_gpu.c b/drivers/gpu/drm/msm/adreno/a5xx_gpu.c
index 08f2579..b0910bb 100644
--- a/drivers/gpu/drm/msm/adreno/a5xx_gpu.c
+++ b/drivers/gpu/drm/msm/adreno/a5xx_gpu.c
@@ -1215,22 +1215,6 @@ static struct msm_gpu_state *a5xx_gpu_state_get(struct msm_gpu *gpu)
 	return state;
 }
 
-#ifdef CONFIG_DEBUG_FS
-static void a5xx_show(struct msm_gpu *gpu, struct seq_file *m)
-{
-	seq_printf(m, "status:   %08x\n",
-			gpu_read(gpu, REG_A5XX_RBBM_STATUS));
-
-	/*
-	 * Temporarily disable hardware clock gating before going into
-	 * adreno_show to avoid issues while reading the registers
-	 */
-	a5xx_set_hwcg(gpu, false);
-	adreno_show(gpu, m);
-	a5xx_set_hwcg(gpu, true);
-}
-#endif
-
 static struct msm_ringbuffer *a5xx_active_ring(struct msm_gpu *gpu)
 {
 	struct adreno_gpu *adreno_gpu = to_adreno_gpu(gpu);
@@ -1260,7 +1244,7 @@ static int a5xx_gpu_busy(struct msm_gpu *gpu, uint64_t *value)
 		.irq = a5xx_irq,
 		.destroy = a5xx_destroy,
 #ifdef CONFIG_DEBUG_FS
-		.show = a5xx_show,
+		.show = adreno_show,
 		.debugfs_init = a5xx_debugfs_init,
 #endif
 		.gpu_busy = a5xx_gpu_busy,
diff --git a/drivers/gpu/drm/msm/adreno/adreno_gpu.c b/drivers/gpu/drm/msm/adreno/adreno_gpu.c
index 35b77f0..af776f1 100644
--- a/drivers/gpu/drm/msm/adreno/adreno_gpu.c
+++ b/drivers/gpu/drm/msm/adreno/adreno_gpu.c
@@ -437,38 +437,34 @@ void adreno_gpu_state_put(struct msm_gpu_state *state)
 }
 
 #ifdef CONFIG_DEBUG_FS
-void adreno_show(struct msm_gpu *gpu, struct seq_file *m)
+void adreno_show(struct msm_gpu *gpu, struct msm_gpu_state *state,
+		struct seq_file *m)
 {
 	struct adreno_gpu *adreno_gpu = to_adreno_gpu(gpu);
 	int i;
 
+	if (IS_ERR_OR_NULL(state))
+		return;
+
+	seq_printf(m, "status:   %08x\n", state->rbbm_status);
 	seq_printf(m, "revision: %d (%d.%d.%d.%d)\n",
 			adreno_gpu->info->revn, adreno_gpu->rev.core,
 			adreno_gpu->rev.major, adreno_gpu->rev.minor,
 			adreno_gpu->rev.patchid);
 
 	for (i = 0; i < gpu->nr_rings; i++) {
-		struct msm_ringbuffer *ring = gpu->rb[i];
-
 		seq_printf(m, "rb %d: fence:    %d/%d\n", i,
-			ring->memptrs->fence, ring->seqno);
+			state->ring[i].fence, state->ring[i].seqno);
 
-		seq_printf(m, "      rptr:     %d\n",
-			get_rptr(adreno_gpu, ring));
-		seq_printf(m, "rb wptr:  %d\n", get_wptr(ring));
+		seq_printf(m, "      rptr:     %d\n", state->ring[i].rptr);
+		seq_printf(m, "rb wptr:  %d\n", state->ring[i].wptr);
 	}
 
-	/* dump these out in a form that can be parsed by demsm: */
 	seq_printf(m, "IO:region %s 00000000 00020000\n", gpu->name);
-	for (i = 0; adreno_gpu->registers[i] != ~0; i += 2) {
-		uint32_t start = adreno_gpu->registers[i];
-		uint32_t end   = adreno_gpu->registers[i+1];
-		uint32_t addr;
-
-		for (addr = start; addr <= end; addr++) {
-			uint32_t val = gpu_read(gpu, addr);
-			seq_printf(m, "IO:R %08x %08x\n", addr<<2, val);
-		}
+	for (i = 0; i < state->nr_registers; i++) {
+		seq_printf(m, "IO:R %08x %08x\n",
+			state->registers[i * 2] << 2,
+			state->registers[(i * 2) + 1]);
 	}
 }
 #endif
diff --git a/drivers/gpu/drm/msm/adreno/adreno_gpu.h b/drivers/gpu/drm/msm/adreno/adreno_gpu.h
index 0beb078..815ae98 100644
--- a/drivers/gpu/drm/msm/adreno/adreno_gpu.h
+++ b/drivers/gpu/drm/msm/adreno/adreno_gpu.h
@@ -215,7 +215,8 @@ void adreno_submit(struct msm_gpu *gpu, struct msm_gem_submit *submit,
 void adreno_flush(struct msm_gpu *gpu, struct msm_ringbuffer *ring);
 bool adreno_idle(struct msm_gpu *gpu, struct msm_ringbuffer *ring);
 #ifdef CONFIG_DEBUG_FS
-void adreno_show(struct msm_gpu *gpu, struct seq_file *m);
+void adreno_show(struct msm_gpu *gpu, struct msm_gpu_state *state,
+		struct seq_file *m);
 #endif
 void adreno_dump_info(struct msm_gpu *gpu);
 void adreno_dump(struct msm_gpu *gpu);
@@ -227,7 +228,6 @@ int adreno_gpu_init(struct drm_device *drm, struct platform_device *pdev,
 		int nr_rings);
 void adreno_gpu_cleanup(struct adreno_gpu *gpu);
 
-
 struct msm_gpu_state *adreno_gpu_state_get(struct msm_gpu *gpu);
 void adreno_gpu_state_put(struct msm_gpu_state *state);
 
diff --git a/drivers/gpu/drm/msm/msm_debugfs.c b/drivers/gpu/drm/msm/msm_debugfs.c
index ba74cb4..fd535da 100644
--- a/drivers/gpu/drm/msm/msm_debugfs.c
+++ b/drivers/gpu/drm/msm/msm_debugfs.c
@@ -25,13 +25,22 @@ static int msm_gpu_show(struct drm_device *dev, struct seq_file *m)
 {
 	struct msm_drm_private *priv = dev->dev_private;
 	struct msm_gpu *gpu = priv->gpu;
+	struct msm_gpu_state *state;
 
-	if (gpu) {
-		seq_printf(m, "%s Status:\n", gpu->name);
-		pm_runtime_get_sync(&gpu->pdev->dev);
-		gpu->funcs->show(gpu, m);
-		pm_runtime_put_sync(&gpu->pdev->dev);
-	}
+	if (!gpu)
+		return 0;
+
+	pm_runtime_get_sync(&gpu->pdev->dev);
+	state = gpu->funcs->gpu_state_get(gpu);
+	pm_runtime_put_sync(&gpu->pdev->dev);
+
+	if (IS_ERR(state))
+		return PTR_ERR(state);
+
+	seq_printf(m, "%s Status:\n", gpu->name);
+	gpu->funcs->show(gpu, state, m);
+
+	gpu->funcs->gpu_state_put(state);
 
 	return 0;
 }
diff --git a/drivers/gpu/drm/msm/msm_gpu.h b/drivers/gpu/drm/msm/msm_gpu.h
index 4be72a6..470f3bb 100644
--- a/drivers/gpu/drm/msm/msm_gpu.h
+++ b/drivers/gpu/drm/msm/msm_gpu.h
@@ -65,7 +65,8 @@ struct msm_gpu_funcs {
 	void (*destroy)(struct msm_gpu *gpu);
 #ifdef CONFIG_DEBUG_FS
 	/* show GPU status in debugfs: */
-	void (*show)(struct msm_gpu *gpu, struct seq_file *m);
+	void (*show)(struct msm_gpu *gpu, struct msm_gpu_state *state,
+			struct seq_file *m);
 	/* for generation specific debugfs: */
 	int (*debugfs_init)(struct msm_gpu *gpu, struct drm_minor *minor);
 #endif
-- 
1.9.1

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

[PATCH 5/8] drm/msm/gpu: Capture the GPU state on a GPU hang

[Jordan Crouse]

Capture the GPU state on a GPU hang and store it for later playback
via the devcoredump facility. Only one crash state is stored at a
time on the assumption that the first hang is usually the most
interesting. The existing crash state can be cleared after capturing
it and then a new one will be captured on the next hang.

Signed-off-by: Jordan Crouse <jcrouse@codeaurora.org>
---
 drivers/gpu/drm/msm/Kconfig             |   1 +
 drivers/gpu/drm/msm/adreno/adreno_gpu.c |  34 +++++++----
 drivers/gpu/drm/msm/adreno/adreno_gpu.h |   4 +-
 drivers/gpu/drm/msm/msm_debugfs.c       |   7 ++-
 drivers/gpu/drm/msm/msm_gpu.c           | 103 +++++++++++++++++++++++++++++---
 drivers/gpu/drm/msm/msm_gpu.h           |  38 +++++++++++-
 6 files changed, 161 insertions(+), 26 deletions(-)

diff --git a/drivers/gpu/drm/msm/Kconfig b/drivers/gpu/drm/msm/Kconfig
index 38cbde9..843a9d4 100644
--- a/drivers/gpu/drm/msm/Kconfig
+++ b/drivers/gpu/drm/msm/Kconfig
@@ -12,6 +12,7 @@ config DRM_MSM
 	select SHMEM
 	select TMPFS
 	select QCOM_SCM
+	select WANT_DEV_COREDUMP
 	select SND_SOC_HDMI_CODEC if SND_SOC
 	select SYNC_FILE
 	select PM_OPP
diff --git a/drivers/gpu/drm/msm/adreno/adreno_gpu.c b/drivers/gpu/drm/msm/adreno/adreno_gpu.c
index af776f1..e86238ce 100644
--- a/drivers/gpu/drm/msm/adreno/adreno_gpu.c
+++ b/drivers/gpu/drm/msm/adreno/adreno_gpu.c
@@ -392,6 +392,8 @@ struct msm_gpu_state *adreno_gpu_state_get(struct msm_gpu *gpu)
 	if (!state)
 		return ERR_PTR(-ENOMEM);
 
+	kref_init(&state->ref);
+
 	do_gettimeofday(&state->time);
 
 	for (i = 0; i < gpu->nr_rings; i++) {
@@ -427,18 +429,28 @@ struct msm_gpu_state *adreno_gpu_state_get(struct msm_gpu *gpu)
 	return state;
 }
 
-void adreno_gpu_state_put(struct msm_gpu_state *state)
+static void adreno_gpu_state_destroy(struct kref *kref)
 {
-	if (IS_ERR_OR_NULL(state))
-		return;
+	struct msm_gpu_state *state = container_of(kref,
+		struct msm_gpu_state, ref);
 
+	kfree(state->comm);
+	kfree(state->cmd);
 	kfree(state->registers);
 	kfree(state);
 }
 
+int adreno_gpu_state_put(struct msm_gpu_state *state)
+{
+	if (IS_ERR_OR_NULL(state))
+		return 1;
+
+	return kref_put(&state->ref, adreno_gpu_state_destroy);
+}
+
 #ifdef CONFIG_DEBUG_FS
 void adreno_show(struct msm_gpu *gpu, struct msm_gpu_state *state,
-		struct seq_file *m)
+		struct drm_printer *p)
 {
 	struct adreno_gpu *adreno_gpu = to_adreno_gpu(gpu);
 	int i;
@@ -446,23 +458,23 @@ void adreno_show(struct msm_gpu *gpu, struct msm_gpu_state *state,
 	if (IS_ERR_OR_NULL(state))
 		return;
 
-	seq_printf(m, "status:   %08x\n", state->rbbm_status);
-	seq_printf(m, "revision: %d (%d.%d.%d.%d)\n",
+	drm_printf(p, "status:   %08x\n", state->rbbm_status);
+	drm_printf(p, "revision: %d (%d.%d.%d.%d)\n",
 			adreno_gpu->info->revn, adreno_gpu->rev.core,
 			adreno_gpu->rev.major, adreno_gpu->rev.minor,
 			adreno_gpu->rev.patchid);
 
 	for (i = 0; i < gpu->nr_rings; i++) {
-		seq_printf(m, "rb %d: fence:    %d/%d\n", i,
+		drm_printf(p, "rb %d: fence:    %d/%d\n", i,
 			state->ring[i].fence, state->ring[i].seqno);
 
-		seq_printf(m, "      rptr:     %d\n", state->ring[i].rptr);
-		seq_printf(m, "rb wptr:  %d\n", state->ring[i].wptr);
+		drm_printf(p, "      rptr:     %d\n", state->ring[i].rptr);
+		drm_printf(p, "rb wptr:  %d\n", state->ring[i].wptr);
 	}
 
-	seq_printf(m, "IO:region %s 00000000 00020000\n", gpu->name);
+	drm_printf(p, "IO:region %s 00000000 00020000\n", gpu->name);
 	for (i = 0; i < state->nr_registers; i++) {
-		seq_printf(m, "IO:R %08x %08x\n",
+		drm_printf(p, "IO:R %08x %08x\n",
 			state->registers[i * 2] << 2,
 			state->registers[(i * 2) + 1]);
 	}
diff --git a/drivers/gpu/drm/msm/adreno/adreno_gpu.h b/drivers/gpu/drm/msm/adreno/adreno_gpu.h
index 815ae98..077bf11 100644
--- a/drivers/gpu/drm/msm/adreno/adreno_gpu.h
+++ b/drivers/gpu/drm/msm/adreno/adreno_gpu.h
@@ -216,7 +216,7 @@ void adreno_submit(struct msm_gpu *gpu, struct msm_gem_submit *submit,
 bool adreno_idle(struct msm_gpu *gpu, struct msm_ringbuffer *ring);
 #ifdef CONFIG_DEBUG_FS
 void adreno_show(struct msm_gpu *gpu, struct msm_gpu_state *state,
-		struct seq_file *m);
+		struct drm_printer *p);
 #endif
 void adreno_dump_info(struct msm_gpu *gpu);
 void adreno_dump(struct msm_gpu *gpu);
@@ -229,7 +229,7 @@ int adreno_gpu_init(struct drm_device *drm, struct platform_device *pdev,
 void adreno_gpu_cleanup(struct adreno_gpu *gpu);
 
 struct msm_gpu_state *adreno_gpu_state_get(struct msm_gpu *gpu);
-void adreno_gpu_state_put(struct msm_gpu_state *state);
+int adreno_gpu_state_put(struct msm_gpu_state *state);
 
 /* ringbuffer helpers (the parts that are adreno specific) */
 
diff --git a/drivers/gpu/drm/msm/msm_debugfs.c b/drivers/gpu/drm/msm/msm_debugfs.c
index fd535da..165d94b 100644
--- a/drivers/gpu/drm/msm/msm_debugfs.c
+++ b/drivers/gpu/drm/msm/msm_debugfs.c
@@ -16,6 +16,8 @@
  */
 
 #ifdef CONFIG_DEBUG_FS
+
+#include <linux/debugfs.h>
 #include "msm_drv.h"
 #include "msm_gpu.h"
 #include "msm_kms.h"
@@ -26,6 +28,7 @@ static int msm_gpu_show(struct drm_device *dev, struct seq_file *m)
 	struct msm_drm_private *priv = dev->dev_private;
 	struct msm_gpu *gpu = priv->gpu;
 	struct msm_gpu_state *state;
+	struct drm_printer p = drm_seq_file_printer(m);
 
 	if (!gpu)
 		return 0;
@@ -37,8 +40,8 @@ static int msm_gpu_show(struct drm_device *dev, struct seq_file *m)
 	if (IS_ERR(state))
 		return PTR_ERR(state);
 
-	seq_printf(m, "%s Status:\n", gpu->name);
-	gpu->funcs->show(gpu, state, m);
+	drm_printf(&p, "%s Status:\n", gpu->name);
+	gpu->funcs->show(gpu, state, &p);
 
 	gpu->funcs->gpu_state_put(state);
 
diff --git a/drivers/gpu/drm/msm/msm_gpu.c b/drivers/gpu/drm/msm/msm_gpu.c
index 1c09acf..f19b452 100644
--- a/drivers/gpu/drm/msm/msm_gpu.c
+++ b/drivers/gpu/drm/msm/msm_gpu.c
@@ -20,10 +20,11 @@
 #include "msm_mmu.h"
 #include "msm_fence.h"
 
+#include <generated/utsrelease.h>
 #include <linux/string_helpers.h>
 #include <linux/pm_opp.h>
 #include <linux/devfreq.h>
-
+#include <linux/devcoredump.h>
 
 /*
  * Power Management:
@@ -273,6 +274,83 @@ int msm_gpu_hw_init(struct msm_gpu *gpu)
 	return ret;
 }
 
+#ifdef CONFIG_DEV_COREDUMP
+static ssize_t msm_gpu_devcoredump_read(char *buffer, loff_t offset,
+		size_t count, void *data, size_t datalen)
+{
+	struct msm_gpu *gpu = data;
+	struct drm_print_iterator iter;
+	struct drm_printer p;
+	struct msm_gpu_state *state;
+
+	state = msm_gpu_crashstate_get(gpu);
+	if (!state)
+		return 0;
+
+	iter.data = buffer;
+	iter.offset = 0;
+	iter.start = offset;
+	iter.remain = count;
+
+	p = drm_coredump_printer(&iter);
+
+	drm_printf(&p, "---\n");
+	drm_printf(&p, "kernel: " UTS_RELEASE "\n");
+	drm_printf(&p, "module: " KBUILD_MODNAME "\n");
+	drm_printf(&p, "time: %ld.%ld\n",
+		state->time.tv_sec, state->time.tv_usec);
+	if (state->comm)
+		drm_printf(&p, "comm: %s\n", state->comm);
+	if (state->cmd)
+		drm_printf(&p, "cmdline: %s\n", state->cmd);
+
+	gpu->funcs->show(gpu, state, &p);
+
+	msm_gpu_crashstate_put(gpu);
+
+	return count - iter.remain;
+}
+
+static void msm_gpu_devcoredump_free(void *data)
+{
+	struct msm_gpu *gpu = data;
+
+	msm_gpu_crashstate_put(gpu);
+}
+
+static void msm_gpu_crashstate_capture(struct msm_gpu *gpu, char *comm,
+		char *cmd)
+{
+	struct msm_gpu_state *state;
+
+	/* Only save one crash state at a a time */
+	if (gpu->crashstate)
+		return;
+
+	state = gpu->funcs->gpu_state_get(gpu);
+	if (IS_ERR_OR_NULL(state))
+		return;
+
+	/* Fill in the additional crash state information */
+	state->comm = kstrdup(comm, GFP_KERNEL);
+	state->cmd = kstrdup(cmd, GFP_KERNEL);
+
+	kref_get(&state->ref);
+
+	/* Set the active crash state to be dumped on failure */
+	gpu->crashstate = state;
+
+	/* FIXME: Release the crashstate if this errors out? */
+	dev_coredumpm(gpu->dev->dev, THIS_MODULE, gpu, 0, GFP_KERNEL,
+		msm_gpu_devcoredump_read, msm_gpu_devcoredump_free);
+}
+#else
+static void msm_gpu_crashstate_capture(struct msm_gpu *gpu, char *comm,
+		char *cmd)
+{
+}
+#endif
+
 /*
  * Hangcheck detection for locked gpu:
  */
@@ -314,6 +392,7 @@ static void recover_worker(struct work_struct *work)
 	struct msm_drm_private *priv = dev->dev_private;
 	struct msm_gem_submit *submit;
 	struct msm_ringbuffer *cur_ring = gpu->funcs->active_ring(gpu);
+	char *comm = NULL, *cmd = NULL;
 	int i;
 
 	mutex_lock(&dev->struct_mutex);
@@ -326,8 +405,9 @@ static void recover_worker(struct work_struct *work)
 
 		rcu_read_lock();
 		task = pid_task(submit->pid, PIDTYPE_PID);
+
 		if (task) {
-			char *cmd;
+			comm = kstrdup(task->comm, GFP_KERNEL);
 
 			/*
 			 * So slightly annoying, in other paths like
@@ -342,20 +422,25 @@ static void recover_worker(struct work_struct *work)
 			mutex_unlock(&dev->struct_mutex);
 			cmd = kstrdup_quotable_cmdline(task, GFP_KERNEL);
 			mutex_lock(&dev->struct_mutex);
+		}
+
+		rcu_read_unlock();
 
+		if (comm && cmd) {
 			dev_err(dev->dev, "%s: offending task: %s (%s)\n",
-				gpu->name, task->comm, cmd);
+				gpu->name, comm, cmd);
 
 			msm_rd_dump_submit(priv->hangrd, submit,
-				"offending task: %s (%s)", task->comm, cmd);
-
-			kfree(cmd);
-		} else {
+				"offending task: %s (%s)", comm, cmd);
+		} else
 			msm_rd_dump_submit(priv->hangrd, submit, NULL);
-		}
-		rcu_read_unlock();
 	}
 
+	/* Record the crash state */
+	msm_gpu_crashstate_capture(gpu, comm, cmd);
+
+	kfree(cmd);
+	kfree(comm);
 
 	/*
 	 * Update all the rings with the latest and greatest fence.. this
diff --git a/drivers/gpu/drm/msm/msm_gpu.h b/drivers/gpu/drm/msm/msm_gpu.h
index 470f3bb..e65f507 100644
--- a/drivers/gpu/drm/msm/msm_gpu.h
+++ b/drivers/gpu/drm/msm/msm_gpu.h
@@ -66,13 +66,13 @@ struct msm_gpu_funcs {
 #ifdef CONFIG_DEBUG_FS
 	/* show GPU status in debugfs: */
 	void (*show)(struct msm_gpu *gpu, struct msm_gpu_state *state,
-			struct seq_file *m);
+			struct drm_printer *p);
 	/* for generation specific debugfs: */
 	int (*debugfs_init)(struct msm_gpu *gpu, struct drm_minor *minor);
 #endif
 	int (*gpu_busy)(struct msm_gpu *gpu, uint64_t *value);
 	struct msm_gpu_state *(*gpu_state_get)(struct msm_gpu *gpu);
-	void (*gpu_state_put)(struct msm_gpu_state *state);
+	int (*gpu_state_put)(struct msm_gpu_state *state);
 };
 
 struct msm_gpu {
@@ -133,6 +133,8 @@ struct msm_gpu {
 		u64 busy_cycles;
 		ktime_t time;
 	} devfreq;
+
+	struct msm_gpu_state *crashstate;
 };
 
 /* It turns out that all targets use the same ringbuffer size */
@@ -180,6 +182,7 @@ struct msm_gpu_submitqueue {
 };
 
 struct msm_gpu_state {
+	struct kref ref;
 	struct timeval time;
 
 	struct {
@@ -193,6 +196,9 @@ struct msm_gpu_state {
 	u32 *registers;
 
 	u32 rbbm_status;
+
+	char *comm;
+	char *cmd;
 };
 
 static inline void gpu_write(struct msm_gpu *gpu, u32 reg, u32 data)
@@ -274,4 +280,32 @@ static inline void msm_submitqueue_put(struct msm_gpu_submitqueue *queue)
 		kref_put(&queue->ref, msm_submitqueue_destroy);
 }
 
+static inline struct msm_gpu_state *msm_gpu_crashstate_get(struct msm_gpu *gpu)
+{
+	struct msm_gpu_state *state = NULL;
+
+	mutex_lock(&gpu->dev->struct_mutex);
+
+	if (gpu->crashstate) {
+		kref_get(&gpu->crashstate->ref);
+		state = gpu->crashstate;
+	}
+
+	mutex_unlock(&gpu->dev->struct_mutex);
+
+	return state;
+}
+
+static inline void msm_gpu_crashstate_put(struct msm_gpu *gpu)
+{
+	mutex_lock(&gpu->dev->struct_mutex);
+
+	if (gpu->crashstate) {
+		if (gpu->funcs->gpu_state_put(gpu->crashstate))
+			gpu->crashstate = NULL;
+	}
+
+	mutex_unlock(&gpu->dev->struct_mutex);
+}
+
 #endif /* __MSM_GPU_H__ */
-- 
1.9.1

_______________________________________________
Freedreno mailing list
Freedreno@lists.freedesktop.org
https://lists.freedesktop.org/mailman/listinfo/freedreno

[PATCH 6/8] drm/msm/adreno: Convert the show/crash file format

[Jordan Crouse]

Convert the format of the 'show' debugfs file and the crash
dump to a  format resembling YAML. This should be easier to
parse and be more flexible for future changes and expansions.

Signed-off-by: Jordan Crouse <jcrouse@codeaurora.org>
---
 Documentation/gpu/drm-msm-crash-dump.txt | 29 +++++++++++++++++++++++++++++
 drivers/gpu/drm/msm/adreno/adreno_gpu.c  | 23 ++++++++++++++---------
 2 files changed, 43 insertions(+), 9 deletions(-)
 create mode 100644 Documentation/gpu/drm-msm-crash-dump.txt

diff --git a/Documentation/gpu/drm-msm-crash-dump.txt b/Documentation/gpu/drm-msm-crash-dump.txt
new file mode 100644
index 0000000..ec56640
--- /dev/null
+++ b/Documentation/gpu/drm-msm-crash-dump.txt
@@ -0,0 +1,29 @@
+# drm/msm GPU crash dump format
+#
+# This is a description of the format of the drm/msm GPU crash dump format that
+# can be read from /sys/kernel/dri/X/show or from devcoredump following a GPU
+# hang or fault
+
+---
+kernel:       # [string] The kernel version as printed by UTS_RELEASE
+module:       # [string] The module that generated the crash dump
+time:         # [seconds.microseconds] The kernel time at crash
+comm:         # [string] comm string for the binary that generated the fault
+              # (if known)
+cmdline:      # [string] the cmdline for the binary that generated the fault
+              # (if known)
+revision:     # [ id core.major.minor.patchlevel] The GPU id followed by the
+              # individual components of the id separated by dots
+rbbm-status:  # [hex] The current value of RBBM_STATUS which shows what GPU
+              # components were in use at the time of the crash
+ringbuffer:   # Ringbuffer data. There will be a sequence for each ringbuffer
+  -id:             # [decimal] Ringbuffer identifier (0 based index)
+   last-fence:     # [decimal] The last fence issued on the ring
+   retired-fence:  # [decimal] THe last fence retired on the ring
+   rptr:           # [decimal] The current read pointer (rptr) for the ring
+   wptr:           # [decimal] The current write pointer (wptr) for the ring
+registers:    # Sets of register values. This section can be used multiple
+              # times for different ranges of registers. Each register will be
+              # on its own line.
+  - [offset, value] # offset: [hex] byte offset of the register
+                    # value: [hex] value of the register
diff --git a/drivers/gpu/drm/msm/adreno/adreno_gpu.c b/drivers/gpu/drm/msm/adreno/adreno_gpu.c
index e86238ce..920db2e 100644
--- a/drivers/gpu/drm/msm/adreno/adreno_gpu.c
+++ b/drivers/gpu/drm/msm/adreno/adreno_gpu.c
@@ -458,23 +458,28 @@ void adreno_show(struct msm_gpu *gpu, struct msm_gpu_state *state,
 	if (IS_ERR_OR_NULL(state))
 		return;
 
-	drm_printf(p, "status:   %08x\n", state->rbbm_status);
 	drm_printf(p, "revision: %d (%d.%d.%d.%d)\n",
 			adreno_gpu->info->revn, adreno_gpu->rev.core,
 			adreno_gpu->rev.major, adreno_gpu->rev.minor,
 			adreno_gpu->rev.patchid);
 
-	for (i = 0; i < gpu->nr_rings; i++) {
-		drm_printf(p, "rb %d: fence:    %d/%d\n", i,
-			state->ring[i].fence, state->ring[i].seqno);
+	drm_printf(p, "rbbm-status: 0x%08x\n", state->rbbm_status);
+
+	drm_printf(p, "ringbuffer:\n");
 
-		drm_printf(p, "      rptr:     %d\n", state->ring[i].rptr);
-		drm_printf(p, "rb wptr:  %d\n", state->ring[i].wptr);
+	for (i = 0; i < gpu->nr_rings; i++) {
+		drm_printf(p, "  - id: %d\n", i);
+		drm_printf(p, "    last-fence: %d\n", state->ring[i].seqno);
+		drm_printf(p, "    retired-fence: %d\n", state->ring[i].fence);
+		drm_printf(p, "    rptr: %d\n", state->ring[i].rptr);
+		drm_printf(p, "    wptr: %d\n", state->ring[i].wptr);
 	}
 
-	drm_printf(p, "IO:region %s 00000000 00020000\n", gpu->name);
-	for (i = 0; i < state->nr_registers; i++) {
-		drm_printf(p, "IO:R %08x %08x\n",
+	drm_printf(p, "registers:\n");
+	drm_printf(p, "  - [offset, value]\n");
+
+	for(i = 0; i < state->nr_registers; i++) {
+		drm_printf(p, "  - [0x%04x, 0x%08x]\n",
 			state->registers[i * 2] << 2,
 			state->registers[(i * 2) + 1]);
 	}
-- 
1.9.1

_______________________________________________
Freedreno mailing list
Freedreno@lists.freedesktop.org
https://lists.freedesktop.org/mailman/listinfo/freedreno

[PATCH 7/8] drm/msm/adreno: Add ringbuffer data to the GPU state

[Jordan Crouse]

Add the contents of each ringbuffer to the GPU state and dump the
data in the crash file encoded with ascii85. To save space only
the used portions of the ringbuffer are dumped.

Signed-off-by: Jordan Crouse <jcrouse@codeaurora.org>
---
 Documentation/gpu/drm-msm-crash-dump.txt |  5 ++++
 drivers/gpu/drm/msm/adreno/adreno_gpu.c  | 40 ++++++++++++++++++++++++++++++++
 drivers/gpu/drm/msm/msm_gpu.h            |  2 ++
 3 files changed, 47 insertions(+)

diff --git a/Documentation/gpu/drm-msm-crash-dump.txt b/Documentation/gpu/drm-msm-crash-dump.txt
index ec56640..3f54b44 100644
--- a/Documentation/gpu/drm-msm-crash-dump.txt
+++ b/Documentation/gpu/drm-msm-crash-dump.txt
@@ -22,6 +22,11 @@ ringbuffer:   # Ringbuffer data. There will be a sequence for each ringbuffer
    retired-fence:  # [decimal] THe last fence retired on the ring
    rptr:           # [decimal] The current read pointer (rptr) for the ring
    wptr:           # [decimal] The current write pointer (wptr) for the ring
+   size:           # [decimal] The maximum size of the ring programmed in the
+                   # hardware
+   data:           # [ascii85] The contents of the ring encoded as ascii85.
+		   # Only the unused portions of the ring will be printed (up
+                   # to a maximum of 'size' bytes
 registers:    # Sets of register values. This section can be used multiple
               # times for different ranges of registers. Each register will be
               # on its own line.
diff --git a/drivers/gpu/drm/msm/adreno/adreno_gpu.c b/drivers/gpu/drm/msm/adreno/adreno_gpu.c
index 920db2e..e9b7b36 100644
--- a/drivers/gpu/drm/msm/adreno/adreno_gpu.c
+++ b/drivers/gpu/drm/msm/adreno/adreno_gpu.c
@@ -17,6 +17,7 @@
  * this program.  If not, see <http://www.gnu.org/licenses/>.
  */
 
+#include <linux/ascii85.h>
 #include <linux/pm_opp.h>
 #include "adreno_gpu.h"
 #include "msm_gem.h"
@@ -397,10 +398,29 @@ struct msm_gpu_state *adreno_gpu_state_get(struct msm_gpu *gpu)
 	do_gettimeofday(&state->time);
 
 	for (i = 0; i < gpu->nr_rings; i++) {
+		int size = 0, j;
+
 		state->ring[i].fence = gpu->rb[i]->memptrs->fence;
 		state->ring[i].seqno = gpu->rb[i]->seqno;
 		state->ring[i].rptr = get_rptr(adreno_gpu, gpu->rb[i]);
 		state->ring[i].wptr = get_wptr(gpu->rb[i]);
+
+		/*
+		 * Only copy used parts of the ring buffers (this should save
+		 * data size for lightly used rings)
+		 */
+		for(j = 0; j < MSM_GPU_RINGBUFFER_SZ >> 2; j++)
+			if (gpu->rb[i]->start[j])
+				size = j;
+
+		if (size) {
+			state->ring[i].data = kmalloc((size + 1) << 2, GFP_KERNEL);
+			if (state->ring[i].data) {
+				memcpy(state->ring[i].data, gpu->rb[i]->start,
+				(size + 1) << 2);
+				state->ring[i].data_size = (size + 1) << 2;
+			}
+		}
 	}
 
 	/* Count the number of registers */
@@ -431,9 +451,13 @@ struct msm_gpu_state *adreno_gpu_state_get(struct msm_gpu *gpu)
 
 static void adreno_gpu_state_destroy(struct kref *kref)
 {
+	int i;
 	struct msm_gpu_state *state = container_of(kref,
 		struct msm_gpu_state, ref);
 
+	for(i = 0; i < ARRAY_SIZE(state->ring); i++)
+		kfree(state->ring[i].data);
+
 	kfree(state->comm);
 	kfree(state->cmd);
 	kfree(state->registers);
@@ -473,6 +497,22 @@ void adreno_show(struct msm_gpu *gpu, struct msm_gpu_state *state,
 		drm_printf(p, "    retired-fence: %d\n", state->ring[i].fence);
 		drm_printf(p, "    rptr: %d\n", state->ring[i].rptr);
 		drm_printf(p, "    wptr: %d\n", state->ring[i].wptr);
+		drm_printf(p, "    size: %d\n", MSM_GPU_RINGBUFFER_SZ);
+
+		if (state->ring[i].data && state->ring[i].data_size) {
+			u32 *ptr = (u32 *) state->ring[i].data;
+			char out[ASCII85_BUFSZ];
+			long len = ascii85_encode_len(state->ring[i].data_size);
+			int j;
+
+			drm_printf(p, "    data: !!ascii85 |\n");
+			drm_printf(p, "     ");
+
+			for(j = 0; j < len; j++)
+				drm_printf(p, ascii85_encode(ptr[j], out));
+
+			drm_printf(p, "\n");
+		}
 	}
 
 	drm_printf(p, "registers:\n");
diff --git a/drivers/gpu/drm/msm/msm_gpu.h b/drivers/gpu/drm/msm/msm_gpu.h
index e65f507..48f7b21 100644
--- a/drivers/gpu/drm/msm/msm_gpu.h
+++ b/drivers/gpu/drm/msm/msm_gpu.h
@@ -190,6 +190,8 @@ struct msm_gpu_state {
 		u32 seqno;
 		u32 rptr;
 		u32 wptr;
+		void *data;
+		int data_size;
 	} ring[MSM_GPU_MAX_RINGS];
 
 	int nr_registers;
-- 
1.9.1

[PATCH 8/8] drm/msm/adreno: Add a5xx specific registers for the GPU state

[Jordan Crouse]

HLSQ, SP and TP registers are only accessible from a special
aperture and to make matters worse the aperture is blocked from
the CPU on targets that can support secure rendering. Luckily the
GPU hardware has its own purpose built register dumper that can
access the registers from the aperture.  Add a5xx specific code
to program the crashdumper and retrieve the wayward registers
and dump them for the crash state.

Also, remove a block of registers the regular CPU accessible
list that aren't useful for debug which helps reduce the size
of the crash state file by a goodly amount.

Signed-off-by: Jordan Crouse <jcrouse@codeaurora.org>
---
 drivers/gpu/drm/msm/adreno/a3xx_gpu.c   |   8 +-
 drivers/gpu/drm/msm/adreno/a4xx_gpu.c   |   8 +-
 drivers/gpu/drm/msm/adreno/a5xx_gpu.c   | 235 ++++++++++++++++++++++++++++++--
 drivers/gpu/drm/msm/adreno/adreno_gpu.c |  23 ++--
 drivers/gpu/drm/msm/adreno/adreno_gpu.h |   4 +-
 5 files changed, 247 insertions(+), 31 deletions(-)

diff --git a/drivers/gpu/drm/msm/adreno/a3xx_gpu.c b/drivers/gpu/drm/msm/adreno/a3xx_gpu.c
index 4b9284a..6dd2b13 100644
--- a/drivers/gpu/drm/msm/adreno/a3xx_gpu.c
+++ b/drivers/gpu/drm/msm/adreno/a3xx_gpu.c
@@ -420,10 +420,12 @@ static void a3xx_dump(struct msm_gpu *gpu)
 
 static struct msm_gpu_state *a3xx_gpu_state_get(struct msm_gpu *gpu)
 {
-	struct msm_gpu_state *state = adreno_gpu_state_get(gpu);
+	struct msm_gpu_state *state = kzalloc(sizeof(*state), GFP_KERNEL);
 
-	if (IS_ERR(state))
-		return state;
+	if (!state)
+		return ERR_PTR(-ENOMEM);
+
+	adreno_gpu_state_get(gpu, state);
 
 	state->rbbm_status = gpu_read(gpu, REG_A3XX_RBBM_STATUS);
 
diff --git a/drivers/gpu/drm/msm/adreno/a4xx_gpu.c b/drivers/gpu/drm/msm/adreno/a4xx_gpu.c
index b8dcbb1..093ca30 100644
--- a/drivers/gpu/drm/msm/adreno/a4xx_gpu.c
+++ b/drivers/gpu/drm/msm/adreno/a4xx_gpu.c
@@ -456,10 +456,12 @@ static irqreturn_t a4xx_irq(struct msm_gpu *gpu)
 
 static struct msm_gpu_state *a4xx_gpu_state_get(struct msm_gpu *gpu)
 {
-	struct msm_gpu_state *state = adreno_gpu_state_get(gpu);
+	struct msm_gpu_state *state = kzalloc(sizeof(*state), GFP_KERNEL);
 
-	if (IS_ERR(state))
-		return state;
+	if (!state)
+		return ERR_PTR(-ENOMEM);
+
+	adreno_gpu_state_get(gpu, state);
 
 	state->rbbm_status = gpu_read(gpu, REG_A4XX_RBBM_STATUS);
 
diff --git a/drivers/gpu/drm/msm/adreno/a5xx_gpu.c b/drivers/gpu/drm/msm/adreno/a5xx_gpu.c
index b0910bb..764407c 100644
--- a/drivers/gpu/drm/msm/adreno/a5xx_gpu.c
+++ b/drivers/gpu/drm/msm/adreno/a5xx_gpu.c
@@ -19,6 +19,7 @@
 #include <linux/soc/qcom/mdt_loader.h>
 #include <linux/pm_opp.h>
 #include <linux/nvmem-consumer.h>
+#include <linux/iopoll.h>
 #include "msm_gem.h"
 #include "msm_mmu.h"
 #include "a5xx_gpu.h"
@@ -1123,8 +1124,9 @@ static irqreturn_t a5xx_irq(struct msm_gpu *gpu)
 	0xE800, 0xE806, 0xE810, 0xE89A, 0xE8A0, 0xE8A4, 0xE8AA, 0xE8EB,
 	0xE900, 0xE905, 0xEB80, 0xEB8F, 0xEBB0, 0xEBB0, 0xEC00, 0xEC05,
 	0xEC08, 0xECE9, 0xECF0, 0xECF0, 0xEA80, 0xEA80, 0xEA82, 0xEAA3,
-	0xEAA5, 0xEAC2, 0xA800, 0xA8FF, 0xAC60, 0xAC60, 0xB000, 0xB97F,
-	0xB9A0, 0xB9BF, ~0
+	0xEAA5, 0xEAC2, 0xA800, 0xA800, 0xA820, 0xA828, 0xA840, 0xA87D,
+	0XA880, 0xA88D, 0xA890, 0xA8A3, 0xA8D0, 0xA8D8, 0xA8E0, 0xA8F5,
+	0xAC60, 0xAC60, ~0,
 };
 
 static void a5xx_dump(struct msm_gpu *gpu)
@@ -1195,24 +1197,231 @@ static int a5xx_get_timestamp(struct msm_gpu *gpu, uint64_t *value)
 	return 0;
 }
 
+struct a5xx_crashdumper {
+	void *ptr;
+	struct drm_gem_object *bo;
+	u64 iova;
+};
+
+struct a5xx_gpu_state {
+	struct msm_gpu_state base;
+	u32 *hlsqregs;
+};
+
+#define gpu_poll_timeout(gpu, addr, val, cond, interval, timeout) \
+	readl_poll_timeout((gpu)->mmio + ((addr) << 2), val, cond, \
+		interval, timeout)
+
+static int a5xx_crashdumper_init(struct msm_gpu *gpu,
+		struct a5xx_crashdumper *dumper)
+{
+	dumper->ptr = msm_gem_kernel_new_locked(gpu->dev,
+		SZ_1M, MSM_BO_UNCACHED, gpu->aspace,
+		&dumper->bo, &dumper->iova);
+
+	if (IS_ERR(dumper->ptr))
+		return PTR_ERR(dumper->ptr);
+
+	return 0;
+}
+
+static void a5xx_crashdumper_free(struct msm_gpu *gpu,
+		struct a5xx_crashdumper *dumper)
+{
+	msm_gem_put_iova(dumper->bo, gpu->aspace);
+	msm_gem_put_vaddr(dumper->bo);
+
+	drm_gem_object_unreference(dumper->bo);
+}
+
+static int a5xx_crashdumper_run(struct msm_gpu *gpu,
+		struct a5xx_crashdumper *dumper)
+{
+	u32 val;
+
+	if (IS_ERR_OR_NULL(dumper->ptr))
+		return -EINVAL;
+
+	gpu_write64(gpu, REG_A5XX_CP_CRASH_SCRIPT_BASE_LO,
+		REG_A5XX_CP_CRASH_SCRIPT_BASE_HI, dumper->iova);
+
+	gpu_write(gpu, REG_A5XX_CP_CRASH_DUMP_CNTL, 1);
+
+	return gpu_poll_timeout(gpu, REG_A5XX_CP_CRASH_DUMP_CNTL, val,
+		val & 0x04, 100, 10000);
+}
+
+/*
+ * These are a list of the registers that need to be read through the HLSQ
+ * aperture through the crashdumper.  These are not nominally accessible from
+ * the CPU on a secure platform.
+ */
+static const struct {
+	u32 type;
+	u32 regoffset;
+	u32 count;
+} a5xx_hlsq_aperture_regs[] = {
+	{ 0x35, 0xe00, 0x32 },   /* HSLQ non-context */
+	{ 0x31, 0x2080, 0x1 },   /* HLSQ 2D context 0 */
+	{ 0x33, 0x2480, 0x1 },   /* HLSQ 2D context 1 */
+	{ 0x32, 0xe780, 0x62 },  /* HLSQ 3D context 0 */
+	{ 0x34, 0xef80, 0x62 },  /* HLSQ 3D context 1 */
+	{ 0x3f, 0x0ec0, 0x40 },  /* SP non-context */
+	{ 0x3d, 0x2040, 0x1 },   /* SP 2D context 0 */
+	{ 0x3b, 0x2440, 0x1 },   /* SP 2D context 1 */
+	{ 0x3e, 0xe580, 0x170 }, /* SP 3D context 0 */
+	{ 0x3c, 0xed80, 0x170 }, /* SP 3D context 1 */
+	{ 0x3a, 0x0f00, 0x1c },  /* TP non-context */
+	{ 0x38, 0x2000, 0xa },   /* TP 2D context 0 */
+	{ 0x36, 0x2400, 0xa },   /* TP 2D context 1 */
+	{ 0x39, 0xe700, 0x80 },  /* TP 3D context 0 */
+	{ 0x37, 0xef00, 0x80 },  /* TP 3D context 1 */
+};
+
+static void a5xx_gpu_state_get_hlsq_regs(struct msm_gpu *gpu,
+		struct a5xx_gpu_state *a5xx_state)
+{
+	struct a5xx_crashdumper dumper = { 0 };
+	u32 offset, count = 0;
+	u64 *ptr;
+	int i;
+
+	if (a5xx_crashdumper_init(gpu, &dumper))
+		return;
+
+	/* The script will be written at offset 0 */
+	ptr = dumper.ptr;
+
+	/* Start writing the data at offset 256k */
+	offset = dumper.iova + (256 * SZ_1K);
+
+	/* Count how many additional registers to get from the HLSQ aperture */
+	for (i = 0; i < ARRAY_SIZE(a5xx_hlsq_aperture_regs); i++)
+		count += a5xx_hlsq_aperture_regs[i].count;
+
+	a5xx_state->hlsqregs = kcalloc(count, sizeof(u32), GFP_KERNEL);
+	if (!a5xx_state->hlsqregs)
+		return;
+
+	/* Build the crashdump script */
+	for (i = 0; i < ARRAY_SIZE(a5xx_hlsq_aperture_regs); i++) {
+		u32 type = a5xx_hlsq_aperture_regs[i].type;
+		u32 c = a5xx_hlsq_aperture_regs[i].count;
+
+		/* Write the register to select the desired bank */
+		*ptr++ = type << 8;
+		*ptr++ = (((u64) REG_A5XX_HLSQ_DBG_READ_SEL) << 44) |
+			(1 << 21) | 1;
+
+		*ptr++ = offset;
+		*ptr++ = (((u64) REG_A5XX_HLSQ_DBG_AHB_READ_APERTURE) << 44)
+			| c;
+
+		offset += c * sizeof(u32);
+	}
+
+	/* Write two zeros to close off the script */
+	*ptr++ = 0;
+	*ptr++ = 0;
+
+	if (a5xx_crashdumper_run(gpu, &dumper)) {
+		kfree(a5xx_state->hlsqregs);
+		a5xx_crashdumper_free(gpu, &dumper);
+		return;
+	}
+
+	/* Copy the data from the crashdumper to the state */
+	memcpy(a5xx_state->hlsqregs, dumper.ptr + (256 * SZ_1K),
+		count * sizeof(u32));
+
+	a5xx_crashdumper_free(gpu, &dumper);
+}
+
 static struct msm_gpu_state *a5xx_gpu_state_get(struct msm_gpu *gpu)
 {
-	struct msm_gpu_state *state;
+	struct a5xx_gpu_state *a5xx_state = kzalloc(sizeof(*a5xx_state),
+			GFP_KERNEL);
 
-	/*
-	 * Temporarily disable hardware clock gating before going into
-	 * adreno_show to avoid issues while reading the registers
-	 */
+	if (!a5xx_state)
+		return ERR_PTR(-ENOMEM);
+
+	/* Temporarily disable hardware clock gating before reading the hw */
 	a5xx_set_hwcg(gpu, false);
 
-	state = adreno_gpu_state_get(gpu);
+	/* First get the generic state from the adreno core */
+	adreno_gpu_state_get(gpu, &(a5xx_state->base));
 
-	if (!IS_ERR(state))
-		state->rbbm_status = gpu_read(gpu, REG_A5XX_RBBM_STATUS);
+	a5xx_state->base.rbbm_status = gpu_read(gpu, REG_A5XX_RBBM_STATUS);
+
+	/* Get the HLSQ regs with the help of the crashdumper */
+	a5xx_gpu_state_get_hlsq_regs(gpu, a5xx_state);
 
 	a5xx_set_hwcg(gpu, true);
 
-	return state;
+	return &a5xx_state->base;
+}
+
+static void a5xx_gpu_state_destroy(struct kref *kref)
+{
+	struct msm_gpu_state *state = container_of(kref,
+		struct msm_gpu_state, ref);
+	struct a5xx_gpu_state *a5xx_state = container_of(state,
+		struct a5xx_gpu_state, base);
+
+	kfree(a5xx_state->hlsqregs);
+
+	adreno_gpu_state_destroy(state);
+	kfree(a5xx_state);
+}
+
+int a5xx_gpu_state_put(struct msm_gpu_state *state)
+{
+	if (IS_ERR_OR_NULL(state))
+		return 1;
+
+	return kref_put(&state->ref, a5xx_gpu_state_destroy);
+}
+
+
+void a5xx_show(struct msm_gpu *gpu, struct msm_gpu_state *state,
+		struct drm_printer *p)
+{
+	int i, j;
+	u32 pos = 0;
+	struct a5xx_gpu_state *a5xx_state = container_of(state,
+		struct a5xx_gpu_state, base);
+
+	if (IS_ERR_OR_NULL(state))
+		return;
+
+	adreno_show(gpu, state, p);
+
+	/* Dump the additional a5xx HLSQ registers */
+	if (!a5xx_state->hlsqregs)
+		return;
+
+	drm_printf(p, "registers:\n");
+	drm_printf(p, "  - [offset, value]\n");
+
+	for (i = 0; i < ARRAY_SIZE(a5xx_hlsq_aperture_regs); i++) {
+		u32 o = a5xx_hlsq_aperture_regs[i].regoffset;
+		u32 c = a5xx_hlsq_aperture_regs[i].count;
+
+		for (j = 0; j < c; j++, pos++, o++) {
+			/*
+			 * To keep the crashdump simple we pull the entire range
+			 * for each register type but not all of the registers
+			 * in the range are valid. Fortunately invalid registers
+			 * stick out like a sore thumb with a value of
+			 * 0xdeadbeef
+			 */
+			if (a5xx_state->hlsqregs[pos] == 0xdeadbeef)
+				continue;
+
+			drm_printf(p, "  - [0x%04x, 0x%08x]\n",
+				o << 2, a5xx_state->hlsqregs[pos]);
+		}
+	}
 }
 
 static struct msm_ringbuffer *a5xx_active_ring(struct msm_gpu *gpu)
@@ -1244,12 +1453,12 @@ static int a5xx_gpu_busy(struct msm_gpu *gpu, uint64_t *value)
 		.irq = a5xx_irq,
 		.destroy = a5xx_destroy,
 #ifdef CONFIG_DEBUG_FS
-		.show = adreno_show,
+		.show = a5xx_show,
 		.debugfs_init = a5xx_debugfs_init,
 #endif
 		.gpu_busy = a5xx_gpu_busy,
 		.gpu_state_get = a5xx_gpu_state_get,
-		.gpu_state_put = adreno_gpu_state_put,
+		.gpu_state_put = a5xx_gpu_state_put,
 	},
 	.get_timestamp = a5xx_get_timestamp,
 };
diff --git a/drivers/gpu/drm/msm/adreno/adreno_gpu.c b/drivers/gpu/drm/msm/adreno/adreno_gpu.c
index e9b7b36..d85d106 100644
--- a/drivers/gpu/drm/msm/adreno/adreno_gpu.c
+++ b/drivers/gpu/drm/msm/adreno/adreno_gpu.c
@@ -383,16 +383,11 @@ bool adreno_idle(struct msm_gpu *gpu, struct msm_ringbuffer *ring)
 	return false;
 }
 
-struct msm_gpu_state *adreno_gpu_state_get(struct msm_gpu *gpu)
+int adreno_gpu_state_get(struct msm_gpu *gpu, struct msm_gpu_state *state)
 {
 	struct adreno_gpu *adreno_gpu = to_adreno_gpu(gpu);
-	struct msm_gpu_state *state;
 	int i, count = 0;
 
-	state = kzalloc(sizeof(*state), GFP_KERNEL);
-	if (!state)
-		return ERR_PTR(-ENOMEM);
-
 	kref_init(&state->ref);
 
 	do_gettimeofday(&state->time);
@@ -446,14 +441,12 @@ struct msm_gpu_state *adreno_gpu_state_get(struct msm_gpu *gpu)
 		state->nr_registers = count;
 	}
 
-	return state;
+	return 0;
 }
 
-static void adreno_gpu_state_destroy(struct kref *kref)
+void adreno_gpu_state_destroy(struct msm_gpu_state *state)
 {
 	int i;
-	struct msm_gpu_state *state = container_of(kref,
-		struct msm_gpu_state, ref);
 
 	for(i = 0; i < ARRAY_SIZE(state->ring); i++)
 		kfree(state->ring[i].data);
@@ -461,6 +454,14 @@ static void adreno_gpu_state_destroy(struct kref *kref)
 	kfree(state->comm);
 	kfree(state->cmd);
 	kfree(state->registers);
+}
+
+static void adreno_gpu_state_kref_destroy(struct kref *kref)
+{
+	struct msm_gpu_state *state = container_of(kref,
+		struct msm_gpu_state, ref);
+
+	adreno_gpu_state_destroy(state);
 	kfree(state);
 }
 
@@ -469,7 +470,7 @@ int adreno_gpu_state_put(struct msm_gpu_state *state)
 	if (IS_ERR_OR_NULL(state))
 		return 1;
 
-	return kref_put(&state->ref, adreno_gpu_state_destroy);
+	return kref_put(&state->ref, adreno_gpu_state_kref_destroy);
 }
 
 #ifdef CONFIG_DEBUG_FS
diff --git a/drivers/gpu/drm/msm/adreno/adreno_gpu.h b/drivers/gpu/drm/msm/adreno/adreno_gpu.h
index 077bf11..942d695 100644
--- a/drivers/gpu/drm/msm/adreno/adreno_gpu.h
+++ b/drivers/gpu/drm/msm/adreno/adreno_gpu.h
@@ -228,7 +228,9 @@ int adreno_gpu_init(struct drm_device *drm, struct platform_device *pdev,
 		int nr_rings);
 void adreno_gpu_cleanup(struct adreno_gpu *gpu);
 
-struct msm_gpu_state *adreno_gpu_state_get(struct msm_gpu *gpu);
+void adreno_gpu_state_destroy(struct msm_gpu_state *state);
+
+int adreno_gpu_state_get(struct msm_gpu *gpu, struct msm_gpu_state *state);
 int adreno_gpu_state_put(struct msm_gpu_state *state);
 
 /* ringbuffer helpers (the parts that are adreno specific) */
-- 
1.9.1

_______________________________________________
Freedreno mailing list
Freedreno@lists.freedesktop.org
https://lists.freedesktop.org/mailman/listinfo/freedreno

Re: [PATCH 1/8] include: Move ascii85 functions from i915 to linux/ascii85.h

[Chris Wilson]

Quoting Jordan Crouse (2018-02-08 17:31:50)
> The i915 DRM driver very cleverly used ascii85 encoding for their
> GPU state file. Move the encode functions to a general header file to
> support other drivers that might be interested in the same
> functionality.
> 
> [v2 - Update the API to be cleaner for the caller suggested by
> Chris Wilson]
> 
> Signed-off-by: Jordan Crouse <jcrouse@codeaurora.org>
> ---
> @@ -545,19 +523,15 @@ static void print_error_obj(struct drm_i915_error_state_buf *m,
>  
>         err_compression_marker(m);
>         for (page = 0; page < obj->page_count; page++) {
> -               int i, len;
> +               int i;
> +               long len = PAGE_SIZE;

Please keep len = PAGE_SIZE below, so that the computation of len is
grouped. In fact, just ignore the s/int/long/ here as we know it is
bounded to int.

With that,
Reviewed-by: Chris Wilson <chris@chris-wilson.co.uk>

Pick a tree where you want this applied...
-Chris
_______________________________________________
Intel-gfx mailing list
Intel-gfx@lists.freedesktop.org
https://lists.freedesktop.org/mailman/listinfo/intel-gfx

Re: [Freedreno] [PATCH 4/8] drm/msm/gpu: Convert the GPU show function to use the GPU state

[Rob Clark]

On Thu, Feb 8, 2018 at 12:31 PM, Jordan Crouse <jcrouse@codeaurora.org> wrote:
> Convert the existing GPU show function to use the GPU state to
> dump the information rather than reading it directly from the hardware.
> This will require an additional step to capture the state before
> dumping it for the existing nodes but it will greatly facilitate reusing
> the same code for dumping a previously captured state from a GPU hang.
>
> Signed-off-by: Jordan Crouse <jcrouse@codeaurora.org>
> ---
>  drivers/gpu/drm/msm/adreno/a3xx_gpu.c   | 11 +----------
>  drivers/gpu/drm/msm/adreno/a4xx_gpu.c   | 12 +-----------
>  drivers/gpu/drm/msm/adreno/a5xx_gpu.c   | 18 +-----------------
>  drivers/gpu/drm/msm/adreno/adreno_gpu.c | 30 +++++++++++++-----------------
>  drivers/gpu/drm/msm/adreno/adreno_gpu.h |  4 ++--
>  drivers/gpu/drm/msm/msm_debugfs.c       | 21 +++++++++++++++------
>  drivers/gpu/drm/msm/msm_gpu.h           |  3 ++-
>  7 files changed, 35 insertions(+), 64 deletions(-)
>
> diff --git a/drivers/gpu/drm/msm/adreno/a3xx_gpu.c b/drivers/gpu/drm/msm/adreno/a3xx_gpu.c
> index c351292..4b9284a 100644
> --- a/drivers/gpu/drm/msm/adreno/a3xx_gpu.c
> +++ b/drivers/gpu/drm/msm/adreno/a3xx_gpu.c
> @@ -410,15 +410,6 @@ static irqreturn_t a3xx_irq(struct msm_gpu *gpu)
>         ~0   /* sentinel */
>  };
>
> -#ifdef CONFIG_DEBUG_FS
> -static void a3xx_show(struct msm_gpu *gpu, struct seq_file *m)
> -{
> -       seq_printf(m, "status:   %08x\n",
> -                       gpu_read(gpu, REG_A3XX_RBBM_STATUS));
> -       adreno_show(gpu, m);
> -}
> -#endif
> -
>  /* would be nice to not have to duplicate the _show() stuff with printk(): */
>  static void a3xx_dump(struct msm_gpu *gpu)
>  {
> @@ -463,7 +454,7 @@ static struct msm_gpu_state *a3xx_gpu_state_get(struct msm_gpu *gpu)
>                 .irq = a3xx_irq,
>                 .destroy = a3xx_destroy,
>  #ifdef CONFIG_DEBUG_FS
> -               .show = a3xx_show,
> +               .show = adreno_show,
>  #endif
>                 .gpu_state_get = a3xx_gpu_state_get,
>                 .gpu_state_put = adreno_gpu_state_put,
> diff --git a/drivers/gpu/drm/msm/adreno/a4xx_gpu.c b/drivers/gpu/drm/msm/adreno/a4xx_gpu.c
> index faf5d60..b8dcbb1 100644
> --- a/drivers/gpu/drm/msm/adreno/a4xx_gpu.c
> +++ b/drivers/gpu/drm/msm/adreno/a4xx_gpu.c
> @@ -454,16 +454,6 @@ static irqreturn_t a4xx_irq(struct msm_gpu *gpu)
>         ~0 /* sentinel */
>  };
>
> -#ifdef CONFIG_DEBUG_FS
> -static void a4xx_show(struct msm_gpu *gpu, struct seq_file *m)
> -{
> -       seq_printf(m, "status:   %08x\n",
> -                       gpu_read(gpu, REG_A4XX_RBBM_STATUS));
> -       adreno_show(gpu, m);
> -
> -}
> -#endif
> -
>  static struct msm_gpu_state *a4xx_gpu_state_get(struct msm_gpu *gpu)
>  {
>         struct msm_gpu_state *state = adreno_gpu_state_get(gpu);
> @@ -550,7 +540,7 @@ static int a4xx_get_timestamp(struct msm_gpu *gpu, uint64_t *value)
>                 .irq = a4xx_irq,
>                 .destroy = a4xx_destroy,
>  #ifdef CONFIG_DEBUG_FS
> -               .show = a4xx_show,
> +               .show = adreno_show,
>  #endif
>                 .gpu_state_get = a4xx_gpu_state_get,
>                 .gpu_state_put = adreno_gpu_state_put,
> diff --git a/drivers/gpu/drm/msm/adreno/a5xx_gpu.c b/drivers/gpu/drm/msm/adreno/a5xx_gpu.c
> index 08f2579..b0910bb 100644
> --- a/drivers/gpu/drm/msm/adreno/a5xx_gpu.c
> +++ b/drivers/gpu/drm/msm/adreno/a5xx_gpu.c
> @@ -1215,22 +1215,6 @@ static struct msm_gpu_state *a5xx_gpu_state_get(struct msm_gpu *gpu)
>         return state;
>  }
>
> -#ifdef CONFIG_DEBUG_FS
> -static void a5xx_show(struct msm_gpu *gpu, struct seq_file *m)
> -{
> -       seq_printf(m, "status:   %08x\n",
> -                       gpu_read(gpu, REG_A5XX_RBBM_STATUS));
> -
> -       /*
> -        * Temporarily disable hardware clock gating before going into
> -        * adreno_show to avoid issues while reading the registers
> -        */
> -       a5xx_set_hwcg(gpu, false);
> -       adreno_show(gpu, m);
> -       a5xx_set_hwcg(gpu, true);
> -}
> -#endif
> -
>  static struct msm_ringbuffer *a5xx_active_ring(struct msm_gpu *gpu)
>  {
>         struct adreno_gpu *adreno_gpu = to_adreno_gpu(gpu);
> @@ -1260,7 +1244,7 @@ static int a5xx_gpu_busy(struct msm_gpu *gpu, uint64_t *value)
>                 .irq = a5xx_irq,
>                 .destroy = a5xx_destroy,
>  #ifdef CONFIG_DEBUG_FS
> -               .show = a5xx_show,
> +               .show = adreno_show,
>                 .debugfs_init = a5xx_debugfs_init,
>  #endif
>                 .gpu_busy = a5xx_gpu_busy,
> diff --git a/drivers/gpu/drm/msm/adreno/adreno_gpu.c b/drivers/gpu/drm/msm/adreno/adreno_gpu.c
> index 35b77f0..af776f1 100644
> --- a/drivers/gpu/drm/msm/adreno/adreno_gpu.c
> +++ b/drivers/gpu/drm/msm/adreno/adreno_gpu.c
> @@ -437,38 +437,34 @@ void adreno_gpu_state_put(struct msm_gpu_state *state)
>  }
>
>  #ifdef CONFIG_DEBUG_FS
> -void adreno_show(struct msm_gpu *gpu, struct seq_file *m)
> +void adreno_show(struct msm_gpu *gpu, struct msm_gpu_state *state,
> +               struct seq_file *m)
>  {
>         struct adreno_gpu *adreno_gpu = to_adreno_gpu(gpu);
>         int i;
>
> +       if (IS_ERR_OR_NULL(state))
> +               return;
> +
> +       seq_printf(m, "status:   %08x\n", state->rbbm_status);
>         seq_printf(m, "revision: %d (%d.%d.%d.%d)\n",
>                         adreno_gpu->info->revn, adreno_gpu->rev.core,
>                         adreno_gpu->rev.major, adreno_gpu->rev.minor,
>                         adreno_gpu->rev.patchid);
>
>         for (i = 0; i < gpu->nr_rings; i++) {
> -               struct msm_ringbuffer *ring = gpu->rb[i];
> -
>                 seq_printf(m, "rb %d: fence:    %d/%d\n", i,
> -                       ring->memptrs->fence, ring->seqno);
> +                       state->ring[i].fence, state->ring[i].seqno);
>
> -               seq_printf(m, "      rptr:     %d\n",
> -                       get_rptr(adreno_gpu, ring));
> -               seq_printf(m, "rb wptr:  %d\n", get_wptr(ring));
> +               seq_printf(m, "      rptr:     %d\n", state->ring[i].rptr);
> +               seq_printf(m, "rb wptr:  %d\n", state->ring[i].wptr);
>         }
>
> -       /* dump these out in a form that can be parsed by demsm: */
>         seq_printf(m, "IO:region %s 00000000 00020000\n", gpu->name);
> -       for (i = 0; adreno_gpu->registers[i] != ~0; i += 2) {
> -               uint32_t start = adreno_gpu->registers[i];
> -               uint32_t end   = adreno_gpu->registers[i+1];
> -               uint32_t addr;
> -
> -               for (addr = start; addr <= end; addr++) {
> -                       uint32_t val = gpu_read(gpu, addr);
> -                       seq_printf(m, "IO:R %08x %08x\n", addr<<2, val);
> -               }
> +       for (i = 0; i < state->nr_registers; i++) {
> +               seq_printf(m, "IO:R %08x %08x\n",
> +                       state->registers[i * 2] << 2,
> +                       state->registers[(i * 2) + 1]);
>         }
>  }
>  #endif
> diff --git a/drivers/gpu/drm/msm/adreno/adreno_gpu.h b/drivers/gpu/drm/msm/adreno/adreno_gpu.h
> index 0beb078..815ae98 100644
> --- a/drivers/gpu/drm/msm/adreno/adreno_gpu.h
> +++ b/drivers/gpu/drm/msm/adreno/adreno_gpu.h
> @@ -215,7 +215,8 @@ void adreno_submit(struct msm_gpu *gpu, struct msm_gem_submit *submit,
>  void adreno_flush(struct msm_gpu *gpu, struct msm_ringbuffer *ring);
>  bool adreno_idle(struct msm_gpu *gpu, struct msm_ringbuffer *ring);
>  #ifdef CONFIG_DEBUG_FS
> -void adreno_show(struct msm_gpu *gpu, struct seq_file *m);
> +void adreno_show(struct msm_gpu *gpu, struct msm_gpu_state *state,
> +               struct seq_file *m);
>  #endif
>  void adreno_dump_info(struct msm_gpu *gpu);
>  void adreno_dump(struct msm_gpu *gpu);
> @@ -227,7 +228,6 @@ int adreno_gpu_init(struct drm_device *drm, struct platform_device *pdev,
>                 int nr_rings);
>  void adreno_gpu_cleanup(struct adreno_gpu *gpu);
>
> -
>  struct msm_gpu_state *adreno_gpu_state_get(struct msm_gpu *gpu);
>  void adreno_gpu_state_put(struct msm_gpu_state *state);
>
> diff --git a/drivers/gpu/drm/msm/msm_debugfs.c b/drivers/gpu/drm/msm/msm_debugfs.c
> index ba74cb4..fd535da 100644
> --- a/drivers/gpu/drm/msm/msm_debugfs.c
> +++ b/drivers/gpu/drm/msm/msm_debugfs.c
> @@ -25,13 +25,22 @@ static int msm_gpu_show(struct drm_device *dev, struct seq_file *m)
>  {
>         struct msm_drm_private *priv = dev->dev_private;
>         struct msm_gpu *gpu = priv->gpu;
> +       struct msm_gpu_state *state;
>
> -       if (gpu) {
> -               seq_printf(m, "%s Status:\n", gpu->name);
> -               pm_runtime_get_sync(&gpu->pdev->dev);
> -               gpu->funcs->show(gpu, m);
> -               pm_runtime_put_sync(&gpu->pdev->dev);
> -       }
> +       if (!gpu)
> +               return 0;
> +
> +       pm_runtime_get_sync(&gpu->pdev->dev);
> +       state = gpu->funcs->gpu_state_get(gpu);
> +       pm_runtime_put_sync(&gpu->pdev->dev);
> +
> +       if (IS_ERR(state))
> +               return PTR_ERR(state);
> +
> +       seq_printf(m, "%s Status:\n", gpu->name);
> +       gpu->funcs->show(gpu, state, m);
> +
> +       gpu->funcs->gpu_state_put(state);
>
>         return 0;
>  }
> diff --git a/drivers/gpu/drm/msm/msm_gpu.h b/drivers/gpu/drm/msm/msm_gpu.h
> index 4be72a6..470f3bb 100644
> --- a/drivers/gpu/drm/msm/msm_gpu.h
> +++ b/drivers/gpu/drm/msm/msm_gpu.h
> @@ -65,7 +65,8 @@ struct msm_gpu_funcs {
>         void (*destroy)(struct msm_gpu *gpu);
>  #ifdef CONFIG_DEBUG_FS
>         /* show GPU status in debugfs: */
> -       void (*show)(struct msm_gpu *gpu, struct seq_file *m);
> +       void (*show)(struct msm_gpu *gpu, struct msm_gpu_state *state,
> +                       struct seq_file *m);

I wonder, now that 'state msm_gpu_state' captures all the state, is
there really much point for this to be a vfunc still?

BR,
-R

>         /* for generation specific debugfs: */
>         int (*debugfs_init)(struct msm_gpu *gpu, struct drm_minor *minor);
>  #endif
> --
> 1.9.1
>
> _______________________________________________
> Freedreno mailing list
> Freedreno@lists.freedesktop.org
> https://lists.freedesktop.org/mailman/listinfo/freedreno