When task fail, we can find its process with this information.

Signed-off-by: Qiang Yu yuq825@gmail.com --- drivers/gpu/drm/lima/lima_ctx.c | 3 +++ drivers/gpu/drm/lima/lima_ctx.h | 5 +++++ 2 files changed, 8 insertions(+)

diff --git a/drivers/gpu/drm/lima/lima_ctx.c b/drivers/gpu/drm/lima/lima_ctx.c index 22fff6caa961..891d5cd5019a 100644 --- a/drivers/gpu/drm/lima/lima_ctx.c +++ b/drivers/gpu/drm/lima/lima_ctx.c @@ -27,6 +27,9 @@ int lima_ctx_create(struct lima_device *dev, struct lima_ctx_mgr *mgr, u32 *id) if (err < 0) goto err_out0;

+ ctx->pid = task_pid_nr(current); + get_task_comm(ctx->pname, current); + return 0;

err_out0: diff --git a/drivers/gpu/drm/lima/lima_ctx.h b/drivers/gpu/drm/lima/lima_ctx.h index 6154e5c9bfe4..74e2be09090f 100644 --- a/drivers/gpu/drm/lima/lima_ctx.h +++ b/drivers/gpu/drm/lima/lima_ctx.h @@ -5,6 +5,7 @@ #define __LIMA_CTX_H__

#include <linux/xarray.h> +#include <linux/sched.h>

#include "lima_device.h"

@@ -13,6 +14,10 @@ struct lima_ctx { struct lima_device *dev; struct lima_sched_context context[lima_pipe_num]; atomic_t guilty; + + /* debug info */ + char pname[TASK_COMM_LEN]; + pid_t pid; };

struct lima_ctx_mgr {

Save all information to start a task which can be exported to user for debug usage. Dump file data format is specified in lima_dump.h

Signed-off-by: Qiang Yu yuq825@gmail.com --- drivers/gpu/drm/lima/lima_device.c | 13 +++ drivers/gpu/drm/lima/lima_device.h | 8 ++ drivers/gpu/drm/lima/lima_dump.h | 77 +++++++++++++++++ drivers/gpu/drm/lima/lima_sched.c | 128 +++++++++++++++++++++++++++++ drivers/gpu/drm/lima/lima_sched.h | 7 ++ 5 files changed, 233 insertions(+) create mode 100644 drivers/gpu/drm/lima/lima_dump.h

diff --git a/drivers/gpu/drm/lima/lima_device.c b/drivers/gpu/drm/lima/lima_device.c index 19829b543024..42a00171fea5 100644 --- a/drivers/gpu/drm/lima/lima_device.c +++ b/drivers/gpu/drm/lima/lima_device.c @@ -344,6 +344,12 @@ int lima_device_init(struct lima_device *ldev) if (err) goto err_out5;

+ ldev->dump.magic = LIMA_DUMP_MAGIC; + ldev->dump.version_major = LIMA_DUMP_MAJOR; + ldev->dump.version_minor = LIMA_DUMP_MINOR; + INIT_LIST_HEAD(&ldev->error_task_list); + mutex_init(&ldev->error_task_list_lock); + dev_info(ldev->dev, "bus rate = %lu\n", clk_get_rate(ldev->clk_bus)); dev_info(ldev->dev, "mod rate = %lu", clk_get_rate(ldev->clk_gpu));

@@ -370,6 +376,13 @@ int lima_device_init(struct lima_device *ldev) void lima_device_fini(struct lima_device *ldev) { int i; + struct lima_sched_error_task *et, *tmp; + + list_for_each_entry_safe(et, tmp, &ldev->error_task_list, list) { + list_del(&et->list); + kvfree(et); + } + mutex_destroy(&ldev->error_task_list_lock);

lima_fini_pp_pipe(ldev); lima_fini_gp_pipe(ldev); diff --git a/drivers/gpu/drm/lima/lima_device.h b/drivers/gpu/drm/lima/lima_device.h index 31158d86271c..f17173f47f26 100644 --- a/drivers/gpu/drm/lima/lima_device.h +++ b/drivers/gpu/drm/lima/lima_device.h @@ -6,8 +6,11 @@

#include <drm/drm_device.h> #include <linux/delay.h> +#include <linux/list.h> +#include <linux/mutex.h>

#include "lima_sched.h" +#include "lima_dump.h"

enum lima_gpu_id { lima_gpu_mali400 = 0, @@ -94,6 +97,11 @@ struct lima_device {

u32 *dlbu_cpu; dma_addr_t dlbu_dma; + + /* debug info */ + struct lima_dump_head dump; + struct list_head error_task_list; + struct mutex error_task_list_lock; };

static inline struct lima_device * diff --git a/drivers/gpu/drm/lima/lima_dump.h b/drivers/gpu/drm/lima/lima_dump.h new file mode 100644 index 000000000000..ca243d99c51b --- /dev/null +++ b/drivers/gpu/drm/lima/lima_dump.h @@ -0,0 +1,77 @@ +/* SPDX-License-Identifier: GPL-2.0 OR MIT */ +/* Copyright 2020 Qiang Yu yuq825@gmail.com */ + +#ifndef __LIMA_DUMP_H__ +#define __LIMA_DUMP_H__ + +#include <linux/types.h> + +/** + * dump file format for all the information to start a lima task + * + * top level format + * | magic code "LIMA" | format version | num tasks | data size | + * | reserved | reserved | reserved | reserved | + * | task 1 ID | task 1 size | num chunks | reserved | task 1 data | + * | task 2 ID | task 2 size | num chunks | reserved | task 2 data | + * ... + * + * task data format + * | chunk 1 ID | chunk 1 size | reserved | reserved | chunk 1 data | + * | chunk 2 ID | chunk 2 size | reserved | reserved | chunk 2 data | + * ... + * + */ + +#define LIMA_DUMP_MAJOR 1 +#define LIMA_DUMP_MINOR 0 + +#define LIMA_DUMP_MAGIC 0x414d494c + +struct lima_dump_head { + __u32 magic; + __u16 version_major; + __u16 version_minor; + __u32 num_tasks; + __u32 size; + __u32 reserved[4]; +}; + +#define LIMA_DUMP_TASK_GP 0 +#define LIMA_DUMP_TASK_PP 1 +#define LIMA_DUMP_TASK_NUM 2 + +struct lima_dump_task { + __u32 id; + __u32 size; + __u32 num_chunks; + __u32 reserved; +}; + +#define LIMA_DUMP_CHUNK_FRAME 0 +#define LIMA_DUMP_CHUNK_BUFFER 1 +#define LIMA_DUMP_CHUNK_PROCESS_NAME 2 +#define LIMA_DUMP_CHUNK_PROCESS_ID 3 +#define LIMA_DUMP_CHUNK_NUM 4 + +struct lima_dump_chunk { + __u32 id; + __u32 size; + __u32 reserved[2]; +}; + +struct lima_dump_chunk_buffer { + __u32 id; + __u32 size; + __u32 va; + __u32 reserved; +}; + +struct lima_dump_chunk_pid { + __u32 id; + __u32 size; + __u32 pid; + __u32 reserved; +}; + +#endif diff --git a/drivers/gpu/drm/lima/lima_sched.c b/drivers/gpu/drm/lima/lima_sched.c index 3886999b4533..cd1bf3ad9bb5 100644 --- a/drivers/gpu/drm/lima/lima_sched.c +++ b/drivers/gpu/drm/lima/lima_sched.c @@ -256,6 +256,132 @@ static struct dma_fence *lima_sched_run_job(struct drm_sched_job *job) return task->fence; }

+static void lima_sched_build_error_task_list(struct lima_sched_task *task) +{ + struct lima_sched_error_task *et; + struct lima_sched_pipe *pipe = to_lima_pipe(task->base.sched); + struct lima_ip *ip = pipe->processor[0]; + int pipe_id = ip->id == lima_ip_gp ? lima_pipe_gp : lima_pipe_pp; + struct lima_device *dev = ip->dev; + struct lima_sched_context *sched_ctx = + container_of(task->base.entity, struct lima_sched_context, base); + struct lima_ctx *ctx = + container_of(sched_ctx, struct lima_ctx, context[pipe_id]); + struct lima_dump_task *dt; + struct lima_dump_chunk *chunk; + struct lima_dump_chunk_pid *pid_chunk; + struct lima_dump_chunk_buffer *buffer_chunk; + uint32_t size, task_size, mem_size; + int i; + + mutex_lock(&dev->error_task_list_lock); + + if (dev->dump.num_tasks >= lima_max_error_tasks) { + dev_info(dev->dev, "fail to save task state: error task list is full\n"); + goto out; + } + + /* frame chunk */ + size = sizeof(struct lima_dump_chunk) + pipe->frame_size; + /* process name chunk */ + size += sizeof(struct lima_dump_chunk) + sizeof(ctx->pname); + /* pid chunk */ + size += sizeof(struct lima_dump_chunk); + /* buffer chunks */ + for (i = 0; i < task->num_bos; i++) { + struct lima_bo *bo = task->bos[i]; + + size += sizeof(struct lima_dump_chunk); + size += bo->heap_size ? bo->heap_size : lima_bo_size(bo); + } + + task_size = size + sizeof(struct lima_dump_task); + mem_size = task_size + sizeof(*et); + et = kvmalloc(mem_size, GFP_KERNEL); + if (!et) { + dev_err(dev->dev, "fail to alloc task dump buffer of size %x\n", + mem_size); + goto out; + } + + et->data = et + 1; + et->size = task_size; + + dt = et->data; + memset(dt, 0, sizeof(*dt)); + dt->id = pipe_id; + dt->size = size; + + chunk = (struct lima_dump_chunk *)(dt + 1); + memset(chunk, 0, sizeof(*chunk)); + chunk->id = LIMA_DUMP_CHUNK_FRAME; + chunk->size = pipe->frame_size; + memcpy(chunk + 1, task->frame, pipe->frame_size); + dt->num_chunks++; + + chunk = (void *)(chunk + 1) + chunk->size; + memset(chunk, 0, sizeof(*chunk)); + chunk->id = LIMA_DUMP_CHUNK_PROCESS_NAME; + chunk->size = sizeof(ctx->pname); + memcpy(chunk + 1, ctx->pname, sizeof(ctx->pname)); + dt->num_chunks++; + + pid_chunk = (void *)(chunk + 1) + chunk->size; + memset(pid_chunk, 0, sizeof(*pid_chunk)); + pid_chunk->id = LIMA_DUMP_CHUNK_PROCESS_ID; + pid_chunk->pid = ctx->pid; + dt->num_chunks++; + + buffer_chunk = (void *)(pid_chunk + 1) + pid_chunk->size; + for (i = 0; i < task->num_bos; i++) { + struct lima_bo *bo = task->bos[i]; + void *data; + + memset(buffer_chunk, 0, sizeof(*buffer_chunk)); + buffer_chunk->id = LIMA_DUMP_CHUNK_BUFFER; + buffer_chunk->va = lima_vm_get_va(task->vm, bo); + + if (bo->heap_size) { + buffer_chunk->size = bo->heap_size; + + data = vmap(bo->base.pages, bo->heap_size >> PAGE_SHIFT, + VM_MAP, pgprot_writecombine(PAGE_KERNEL)); + if (!data) { + kvfree(et); + goto out; + } + + memcpy(buffer_chunk + 1, data, buffer_chunk->size); + + vunmap(data); + } else { + buffer_chunk->size = lima_bo_size(bo); + + data = drm_gem_shmem_vmap(&bo->base.base); + if (IS_ERR_OR_NULL(data)) { + kvfree(et); + goto out; + } + + memcpy(buffer_chunk + 1, data, buffer_chunk->size); + + drm_gem_shmem_vunmap(&bo->base.base, data); + } + + buffer_chunk = (void *)(buffer_chunk + 1) + buffer_chunk->size; + dt->num_chunks++; + } + + list_add(&et->list, &dev->error_task_list); + dev->dump.size += et->size; + dev->dump.num_tasks++; + + dev_info(dev->dev, "save error task state success\n"); + +out: + mutex_unlock(&dev->error_task_list_lock); +} + static void lima_sched_timedout_job(struct drm_sched_job *job) { struct lima_sched_pipe *pipe = to_lima_pipe(job->sched); @@ -268,6 +394,8 @@ static void lima_sched_timedout_job(struct drm_sched_job *job)

drm_sched_increase_karma(&task->base);

+ lima_sched_build_error_task_list(task); + pipe->task_error(pipe);

if (pipe->bcast_mmu) diff --git a/drivers/gpu/drm/lima/lima_sched.h b/drivers/gpu/drm/lima/lima_sched.h index d64393fb50a9..e29f5e3b675b 100644 --- a/drivers/gpu/drm/lima/lima_sched.h +++ b/drivers/gpu/drm/lima/lima_sched.h @@ -5,9 +5,16 @@ #define __LIMA_SCHED_H__

#include <drm/gpu_scheduler.h> +#include <linux/list.h>

struct lima_vm;

+struct lima_sched_error_task { + struct list_head list; + void *data; + uint32_t size; +}; + struct lima_sched_task { struct drm_sched_job base;

Export /sys/class/drm/cardX/device/error sysfs for user read out error task dump file.

Signed-off-by: Qiang Yu yuq825@gmail.com --- drivers/gpu/drm/lima/lima_drv.c | 93 +++++++++++++++++++++++++++++++++ 1 file changed, 93 insertions(+)

diff --git a/drivers/gpu/drm/lima/lima_drv.c b/drivers/gpu/drm/lima/lima_drv.c index e235d4545b6c..8c5adc025902 100644 --- a/drivers/gpu/drm/lima/lima_drv.c +++ b/drivers/gpu/drm/lima/lima_drv.c @@ -276,6 +276,92 @@ static struct drm_driver lima_drm_driver = { .gem_prime_mmap = drm_gem_prime_mmap, };

+struct lima_block_reader { + void *dst; + size_t base; + size_t count; + size_t off; + ssize_t read; +}; + +static bool lima_read_block(struct lima_block_reader *reader, void *src, size_t src_size) +{ + size_t max_off = reader->base + src_size; + + if (reader->off < max_off) { + size_t size = min_t(size_t, max_off - reader->off, reader->count); + + memcpy(reader->dst, src + (reader->off - reader->base), size); + + reader->dst += size; + reader->off += size; + reader->read += size; + reader->count -= size; + } + + reader->base = max_off; + + return !!reader->count; +} + +static ssize_t lima_error_state_read(struct file *filp, struct kobject *kobj, + struct bin_attribute *attr, char *buf, + loff_t off, size_t count) +{ + + struct device *dev = kobj_to_dev(kobj); + struct lima_device *ldev = dev_get_drvdata(dev); + struct lima_sched_error_task *et; + struct lima_block_reader reader = { + .dst = buf, + .count = count, + .off = off, + }; + + mutex_lock(&ldev->error_task_list_lock); + + if (lima_read_block(&reader, &ldev->dump, sizeof(ldev->dump))) { + list_for_each_entry(et, &ldev->error_task_list, list) { + if (!lima_read_block(&reader, et->data, et->size)) + break; + } + } + + mutex_unlock(&ldev->error_task_list_lock); + return reader.read; +} + +static ssize_t lima_error_state_write(struct file *file, struct kobject *kobj, + struct bin_attribute *attr, char *buf, + loff_t off, size_t count) +{ + struct device *dev = kobj_to_dev(kobj); + struct lima_device *ldev = dev_get_drvdata(dev); + struct lima_sched_error_task *et, *tmp; + + mutex_lock(&ldev->error_task_list_lock); + + list_for_each_entry_safe(et, tmp, &ldev->error_task_list, list) { + list_del(&et->list); + kvfree(et); + } + + ldev->dump.size = 0; + ldev->dump.num_tasks = 0; + + mutex_unlock(&ldev->error_task_list_lock); + + return count; +} + +static const struct bin_attribute lima_error_state_attr = { + .attr.name = "error", + .attr.mode = S_IRUSR | S_IWUSR, + .size = 0, + .read = lima_error_state_read, + .write = lima_error_state_write, +}; + static int lima_pdev_probe(struct platform_device *pdev) { struct lima_device *ldev; @@ -318,6 +404,11 @@ static int lima_pdev_probe(struct platform_device *pdev) if (err < 0) goto err_out2;

+ platform_set_drvdata(pdev, ldev); + + if (sysfs_create_bin_file(&ldev->dev->kobj, &lima_error_state_attr)) + dev_warn(ldev->dev, "fail to create error state sysfs\n"); + return 0;

err_out2: @@ -334,6 +425,8 @@ static int lima_pdev_remove(struct platform_device *pdev) struct lima_device *ldev = platform_get_drvdata(pdev); struct drm_device *ddev = ldev->ddev;

+ sysfs_remove_bin_file(&ldev->dev->kobj, &lima_error_state_attr); + platform_set_drvdata(pdev, NULL); drm_dev_unregister(ddev); lima_device_fini(ldev); drm_dev_put(ddev);