1 files changed, 182 insertions, 95 deletions

diff --git a/drivers/gpu/drm/msm/msm_rd.c b/drivers/gpu/drm/msm/msm_rd.c
index 0366b8092f97..54493a94dcb7 100644
--- a/drivers/gpu/drm/msm/msm_rd.c
+++ b/drivers/gpu/drm/msm/msm_rd.c
@@ -1,29 +1,24 @@
+// SPDX-License-Identifier: GPL-2.0-only
 /*
  * Copyright (C) 2013 Red Hat
  * Author: Rob Clark <robdclark@gmail.com>
- *
- * This program is free software; you can redistribute it and/or modify it
- * under the terms of the GNU General Public License version 2 as published by
- * the Free Software Foundation.
- *
- * This program is distributed in the hope that it will be useful, but WITHOUT
- * ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or
- * FITNESS FOR A PARTICULAR PURPOSE.  See the GNU General Public License for
- * more details.
- *
- * You should have received a copy of the GNU General Public License along with
- * this program.  If not, see <http://www.gnu.org/licenses/>.
  */
 
 /* For debugging crashes, userspace can:
  *
  *   tail -f /sys/kernel/debug/dri/<minor>/rd > logfile.rd
  *
- * To log the cmdstream in a format that is understood by freedreno/cffdump
+ * to log the cmdstream in a format that is understood by freedreno/cffdump
  * utility.  By comparing the last successfully completed fence #, to the
  * cmdstream for the next fence, you can narrow down which process and submit
  * caused the gpu crash/lockup.
  *
+ * Additionally:
+ *
+ *   tail -f /sys/kernel/debug/dri/<minor>/hangrd > logfile.rd
+ *
+ * will capture just the cmdstream from submits which triggered a GPU hang.
+ *
  * This bypasses drm_debugfs_create_files() mainly because we need to use
  * our own fops for a bit more control.  In particular, we don't want to
  * do anything if userspace doesn't have the debugfs file open.
@@ -34,21 +29,24 @@
  * or shader programs (if not emitted inline in cmdstream).
  */
 
-#ifdef CONFIG_DEBUG_FS
-
-#include <linux/kfifo.h>
-#include <linux/debugfs.h>
 #include <linux/circ_buf.h>
+#include <linux/debugfs.h>
+#include <linux/kfifo.h>
+#include <linux/uaccess.h>
 #include <linux/wait.h>
 
+#include <drm/drm_file.h>
+
 #include "msm_drv.h"
 #include "msm_gpu.h"
 #include "msm_gem.h"
 
-static bool rd_full = false;
+bool rd_full = false;
 MODULE_PARM_DESC(rd_full, "If true, $debugfs/.../rd will snapshot all buffer contents");
 module_param_named(rd_full, rd_full, bool, 0600);
 
+#ifdef CONFIG_DEBUG_FS
+
 enum rd_sect_type {
 	RD_NONE,
 	RD_TEST,       /* ascii text */
@@ -64,6 +62,7 @@ enum rd_sect_type {
 	RD_FRAG_SHADER,
 	RD_BUFFER_CONTENTS,
 	RD_GPU_ID,
+	RD_CHIP_ID,
 };
 
 #define BUF_SZ 512  /* should be power of 2 */
@@ -84,15 +83,10 @@ struct msm_rd_state {
 
 	bool open;
 
-	/* current submit to read out: */
-	struct msm_gem_submit *submit;
-
 	/* fifo access is synchronized on the producer side by
-	 * struct_mutex held by submit code (otherwise we could
-	 * end up w/ cmds logged in different order than they
-	 * were executed).  And read_lock synchronizes the reads
+	 * write_lock.  And read_lock synchronizes the reads
 	 */
-	struct mutex read_lock;
+	struct mutex read_lock, write_lock;
 
 	wait_queue_head_t fifo_event;
 	struct circ_buf fifo;
@@ -109,12 +103,18 @@ static void rd_write(struct msm_rd_state *rd, const void *buf, int sz)
 		char *fptr = &fifo->buf[fifo->head];
 		int n;
 
-		wait_event(rd->fifo_event, circ_space(&rd->fifo) > 0);
+		wait_event(rd->fifo_event, circ_space(&rd->fifo) > 0 || !rd->open);
+		if (!rd->open)
+			return;
 
+		/* Note that smp_load_acquire() is not strictly required
+		 * as CIRC_SPACE_TO_END() does not access the tail more
+		 * than once.
+		 */
 		n = min(sz, circ_space_to_end(&rd->fifo));
 		memcpy(fptr, ptr, n);
 
-		fifo->head = (fifo->head + n) & (BUF_SZ - 1);
+		smp_store_release(&fifo->head, (fifo->head + n) & (BUF_SZ - 1));
 		sz  -= n;
 		ptr += n;
 
@@ -145,13 +145,17 @@ static ssize_t rd_read(struct file *file, char __user *buf,
 	if (ret)
 		goto out;
 
+	/* Note that smp_load_acquire() is not strictly required
+	 * as CIRC_CNT_TO_END() does not access the head more than
+	 * once.
+	 */
 	n = min_t(int, sz, circ_count_to_end(&rd->fifo));
 	if (copy_to_user(buf, fptr, n)) {
 		ret = -EFAULT;
 		goto out;
 	}
 
-	fifo->tail = (fifo->tail + n) & (BUF_SZ - 1);
+	smp_store_release(&fifo->tail, (fifo->tail + n) & (BUF_SZ - 1));
 	*ppos += n;
 
 	wake_up_all(&rd->fifo_event);
@@ -171,11 +175,15 @@ static int rd_open(struct inode *inode, struct file *file)
 	struct msm_gpu *gpu = priv->gpu;
 	uint64_t val;
 	uint32_t gpu_id;
+	uint32_t zero = 0;
 	int ret = 0;
 
-	mutex_lock(&dev->struct_mutex);
+	if (!gpu)
+		return -ENODEV;
 
-	if (rd->open || !gpu) {
+	mutex_lock(&gpu->lock);
+
+	if (rd->open) {
 		ret = -EBUSY;
 		goto out;
 	}
@@ -183,23 +191,34 @@ static int rd_open(struct inode *inode, struct file *file)
 	file->private_data = rd;
 	rd->open = true;
 
+	/* Reset fifo to clear any previously unread data: */
+	rd->fifo.head = rd->fifo.tail = 0;
+
 	/* the parsing tools need to know gpu-id to know which
 	 * register database to load.
+	 *
+	 * Note: These particular params do not require a context
 	 */
-	gpu->funcs->get_param(gpu, MSM_PARAM_GPU_ID, &val);
+	gpu->funcs->get_param(gpu, NULL, MSM_PARAM_GPU_ID, &val, &zero);
 	gpu_id = val;
 
 	rd_write_section(rd, RD_GPU_ID, &gpu_id, sizeof(gpu_id));
 
+	gpu->funcs->get_param(gpu, NULL, MSM_PARAM_CHIP_ID, &val, &zero);
+	rd_write_section(rd, RD_CHIP_ID, &val, sizeof(val));
+
 out:
-	mutex_unlock(&dev->struct_mutex);
+	mutex_unlock(&gpu->lock);
 	return ret;
 }
 
 static int rd_release(struct inode *inode, struct file *file)
 {
 	struct msm_rd_state *rd = inode->i_private;
+
 	rd->open = false;
+	wake_up_all(&rd->fifo_event);
+
 	return 0;
 }
 
@@ -208,116 +227,188 @@ static const struct file_operations rd_debugfs_fops = {
 	.owner = THIS_MODULE,
 	.open = rd_open,
 	.read = rd_read,
-	.llseek = no_llseek,
 	.release = rd_release,
 };
 
-int msm_rd_debugfs_init(struct drm_minor *minor)
+
+static void rd_cleanup(struct msm_rd_state *rd)
 {
-	struct msm_drm_private *priv = minor->dev->dev_private;
-	struct msm_rd_state *rd;
-	struct dentry *ent;
+	if (!rd)
+		return;
 
-	/* only create on first minor: */
-	if (priv->rd)
-		return 0;
+	mutex_destroy(&rd->read_lock);
+	mutex_destroy(&rd->write_lock);
+	kfree(rd);
+}
+
+static struct msm_rd_state *rd_init(struct drm_minor *minor, const char *name)
+{
+	struct msm_rd_state *rd;
 
 	rd = kzalloc(sizeof(*rd), GFP_KERNEL);
 	if (!rd)
-		return -ENOMEM;
+		return ERR_PTR(-ENOMEM);
 
 	rd->dev = minor->dev;
 	rd->fifo.buf = rd->buf;
 
 	mutex_init(&rd->read_lock);
-	priv->rd = rd;
+	mutex_init(&rd->write_lock);
 
 	init_waitqueue_head(&rd->fifo_event);
 
-	ent = debugfs_create_file("rd", S_IFREG | S_IRUGO,
-			minor->debugfs_root, rd, &rd_debugfs_fops);
-	if (!ent) {
-		DRM_ERROR("Cannot create /sys/kernel/debug/dri/%pd/rd\n",
-				minor->debugfs_root);
+	debugfs_create_file(name, S_IFREG | S_IRUGO, minor->debugfs_root, rd,
+			    &rd_debugfs_fops);
+
+	return rd;
+}
+
+int msm_rd_debugfs_init(struct drm_minor *minor)
+{
+	struct msm_drm_private *priv = minor->dev->dev_private;
+	struct msm_rd_state *rd;
+	int ret;
+
+	if (!priv->gpu_pdev)
+		return 0;
+
+	/* only create on first minor: */
+	if (priv->rd)
+		return 0;
+
+	rd = rd_init(minor, "rd");
+	if (IS_ERR(rd)) {
+		ret = PTR_ERR(rd);
+		goto fail;
+	}
+
+	priv->rd = rd;
+
+	rd = rd_init(minor, "hangrd");
+	if (IS_ERR(rd)) {
+		ret = PTR_ERR(rd);
 		goto fail;
 	}
 
+	priv->hangrd = rd;
+
 	return 0;
 
 fail:
 	msm_rd_debugfs_cleanup(priv);
-	return -1;
+	return ret;
 }
 
 void msm_rd_debugfs_cleanup(struct msm_drm_private *priv)
 {
-	struct msm_rd_state *rd = priv->rd;
-
-	if (!rd)
-		return;
-
+	rd_cleanup(priv->rd);
 	priv->rd = NULL;
-	mutex_destroy(&rd->read_lock);
-	kfree(rd);
+
+	rd_cleanup(priv->hangrd);
+	priv->hangrd = NULL;
 }
 
-static void snapshot_buf(struct msm_rd_state *rd,
-		struct msm_gem_submit *submit, int idx,
-		uint64_t iova, uint32_t size)
+static void snapshot_buf(struct msm_rd_state *rd, struct drm_gem_object *obj,
+			 uint64_t iova, bool full, size_t offset, size_t size)
 {
-	struct msm_gem_object *obj = submit->bos[idx].obj;
 	const char *buf;
 
-	buf = msm_gem_get_vaddr(&obj->base);
+	/*
+	 * Always write the GPUADDR header so can get a complete list of all the
+	 * buffers in the cmd
+	 */
+	rd_write_section(rd, RD_GPUADDR,
+			(uint32_t[3]){ iova, size, iova >> 32 }, 12);
+
+	if (!full)
+		return;
+
+	buf = msm_gem_get_vaddr_active(obj);
 	if (IS_ERR(buf))
 		return;
 
-	if (iova) {
-		buf += iova - submit->bos[idx].iova;
-	} else {
-		iova = submit->bos[idx].iova;
-		size = obj->base.size;
-	}
+	buf += offset;
 
-	rd_write_section(rd, RD_GPUADDR,
-			(uint32_t[3]){ iova, size, iova >> 32 }, 12);
 	rd_write_section(rd, RD_BUFFER_CONTENTS, buf, size);
 
-	msm_gem_put_vaddr(&obj->base);
+	msm_gem_put_vaddr_locked(obj);
 }
 
-/* called under struct_mutex */
-void msm_rd_dump_submit(struct msm_gem_submit *submit)
+/* called under gpu->lock */
+void msm_rd_dump_submit(struct msm_rd_state *rd, struct msm_gem_submit *submit,
+		const char *fmt, ...)
 {
-	struct drm_device *dev = submit->dev;
-	struct msm_drm_private *priv = dev->dev_private;
-	struct msm_rd_state *rd = priv->rd;
-	char msg[128];
+	extern bool rd_full;
+	struct task_struct *task;
+	char msg[256];
 	int i, n;
 
 	if (!rd->open)
 		return;
 
-	/* writing into fifo is serialized by caller, and
-	 * rd->read_lock is used to serialize the reads
-	 */
-	WARN_ON(!mutex_is_locked(&dev->struct_mutex));
+	mutex_lock(&rd->write_lock);
 
-	n = snprintf(msg, sizeof(msg), "%.*s/%d: fence=%u",
-			TASK_COMM_LEN, current->comm, task_pid_nr(current),
-			submit->fence->seqno);
+	if (fmt) {
+		va_list args;
+
+		va_start(args, fmt);
+		n = vscnprintf(msg, sizeof(msg), fmt, args);
+		va_end(args);
+
+		rd_write_section(rd, RD_CMD, msg, ALIGN(n, 4));
+	}
+
+	rcu_read_lock();
+	task = pid_task(submit->pid, PIDTYPE_PID);
+	if (task) {
+		n = scnprintf(msg, sizeof(msg), "%.*s/%d: fence=%u",
+				TASK_COMM_LEN, task->comm,
+				pid_nr(submit->pid), submit->seqno);
+	} else {
+		n = scnprintf(msg, sizeof(msg), "???/%d: fence=%u",
+				pid_nr(submit->pid), submit->seqno);
+	}
+	rcu_read_unlock();
 
 	rd_write_section(rd, RD_CMD, msg, ALIGN(n, 4));
 
-	if (rd_full) {
-		for (i = 0; i < submit->nr_bos; i++) {
-			/* buffers that are written to probably don't start out
-			 * with anything interesting:
-			 */
-			if (submit->bos[i].flags & MSM_SUBMIT_BO_WRITE)
+	if (msm_context_is_vmbind(submit->queue->ctx)) {
+		struct drm_gpuva *vma;
+
+		drm_gpuvm_resv_assert_held(submit->vm);
+
+		drm_gpuvm_for_each_va (vma, submit->vm) {
+			bool dump = rd_full || (vma->flags & MSM_VMA_DUMP);
+
+			/* Skip MAP_NULL/PRR VMAs: */
+			if (!vma->gem.obj)
 				continue;
 
-			snapshot_buf(rd, submit, i, 0, 0);
+			snapshot_buf(rd, vma->gem.obj, vma->va.addr, dump,
+				     vma->gem.offset, vma->va.range);
+		}
+
+	} else {
+		for (i = 0; i < submit->nr_bos; i++) {
+			struct drm_gem_object *obj = submit->bos[i].obj;
+			bool dump = rd_full || (submit->bos[i].flags & MSM_SUBMIT_BO_DUMP);
+
+			snapshot_buf(rd, obj, submit->bos[i].iova, dump, 0, obj->size);
+		}
+
+		for (i = 0; i < submit->nr_cmds; i++) {
+			uint32_t szd  = submit->cmd[i].size; /* in dwords */
+			int idx = submit->cmd[i].idx;
+			bool dump = rd_full || (submit->bos[idx].flags & MSM_SUBMIT_BO_DUMP);
+
+			/* snapshot cmdstream bo's (if we haven't already): */
+			if (!dump) {
+				struct drm_gem_object *obj = submit->bos[idx].obj;
+				size_t offset = submit->cmd[i].iova - submit->bos[idx].iova;
+
+				snapshot_buf(rd, obj, submit->cmd[i].iova, true,
+					offset, szd * 4);
+			}
 		}
 	}
 
@@ -325,12 +416,6 @@ void msm_rd_dump_submit(struct msm_gem_submit *submit)
 		uint64_t iova = submit->cmd[i].iova;
 		uint32_t szd  = submit->cmd[i].size; /* in dwords */
 
-		/* snapshot cmdstream bo's (if we haven't already): */
-		if (!rd_full) {
-			snapshot_buf(rd, submit, submit->cmd[i].idx,
-					submit->cmd[i].iova, szd * 4);
-		}
-
 		switch (submit->cmd[i].type) {
 		case MSM_SUBMIT_CMD_IB_TARGET_BUF:
 			/* ignore IB-targets, we've logged the buffer, the
@@ -345,5 +430,7 @@ void msm_rd_dump_submit(struct msm_gem_submit *submit)
 			break;
 		}
 	}
+
+	mutex_unlock(&rd->write_lock);
 }
 #endif