Re: [PATCH v2 2/2] drm/amdgpu: adding device coredump support

["Somalapuram, Amaranath" <asomalap@xxxxxxx>]

On 5/26/2022 3:56 PM, Wang, Yang(Kevin) wrote:

[AMD Official Use Only - General]

---

From: amd-gfx <amd-gfx-bounces@xxxxxxxxxxxxxxxxxxxxx> on behalf of Somalapuram Amaranath <Amaranath.Somalapuram@xxxxxxx>
Sent: Thursday, May 26, 2022 5:48 PM
To: amd-gfx@xxxxxxxxxxxxxxxxxxxxx <amd-gfx@xxxxxxxxxxxxxxxxxxxxx>
Cc: Deucher, Alexander <Alexander.Deucher@xxxxxxx>; Somalapuram, Amaranath <Amaranath.Somalapuram@xxxxxxx>; Koenig, Christian <Christian.Koenig@xxxxxxx>; Sharma, Shashank <Shashank.Sharma@xxxxxxx>
Subject: [PATCH v2 2/2] drm/amdgpu: adding device coredump support

 

Added device coredump information:
- Kernel version
- Module
- Time
- VRAM status
- Guilty process name and PID
- GPU register dumps
v1 -> v2: Variable name change
v1 -> v2: NULL check
v1 -> v2: Code alignment
v1 -> v2: Adding dummy amdgpu_devcoredump_free
v1 -> v2: memset reset_task_info to zero

Signed-off-by: Somalapuram Amaranath <Amaranath.Somalapuram@xxxxxxx>
---
 drivers/gpu/drm/amd/amdgpu/amdgpu.h        |  3 +
 drivers/gpu/drm/amd/amdgpu/amdgpu_device.c | 67 ++++++++++++++++++++++
 2 files changed, 70 insertions(+)

diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu.h b/drivers/gpu/drm/amd/amdgpu/amdgpu.h
index c79d9992b113..25a7b2c74928 100644
--- a/drivers/gpu/drm/amd/amdgpu/amdgpu.h
+++ b/drivers/gpu/drm/amd/amdgpu/amdgpu.h
@@ -1044,6 +1044,9 @@ struct amdgpu_device {
         uint32_t                        *reset_dump_reg_list;
         uint32_t                        *reset_dump_reg_value;
         int                             num_regs;
+       struct amdgpu_task_info         reset_task_info;
+       bool                            reset_vram_lost;
+       struct timespec64               reset_time;

[kevin]:

the CONFIG_DEV_COREDUMP check is needed for above variable to avoid compiler warning when coredump feautre is not enabled.

 

    Agreed.

         struct amdgpu_reset_domain      *reset_domain;
 
diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_device.c b/drivers/gpu/drm/amd/amdgpu/amdgpu_device.c
index 866b4980a6fa..ca97afe5be63 100644
--- a/drivers/gpu/drm/amd/amdgpu/amdgpu_device.c
+++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_device.c
@@ -32,6 +32,8 @@
 #include <linux/slab.h>
 #include <linux/iommu.h>
 #include <linux/pci.h>
+#include <linux/devcoredump.h>
+#include <generated/utsrelease.h>
 
 #include <drm/drm_atomic_helper.h>
 #include <drm/drm_probe_helper.h>
@@ -4734,6 +4736,62 @@ static int amdgpu_reset_reg_dumps(struct amdgpu_device *adev)
         return 0;
 }
 
+#ifdef CONFIG_DEV_COREDUMP
+static ssize_t amdgpu_devcoredump_read(char *buffer, loff_t offset,
+               size_t count, void *data, size_t datalen)
+{
+       struct drm_printer p;
+       struct amdgpu_device *adev = data;
+       struct drm_print_iterator iter;
+       int i;
+
+       if (adev == NULL)
+               return 0;

[kevin]:

 this check is not needed, because this private data is passed by our driver as below:

In my testing if the reset is unsuccessful amdgpu_devcoredump_read will not be called.

Shashank: Any inputs on this.

Regards,

S.Amarnath

       dev_coredumpm(dev->dev, THIS_MODULE, adev, 0, GFP_KERNEL,

                       amdgpu_devcoredump_read, amdgpu_devcoredump_free);

+
+       iter.data = ""> +       iter.offset = 0;
+       iter.start = offset;
+       iter.remain = count;
+
+       p = drm_coredump_printer(&iter);
+
+       drm_printf(&p, "**** AMDGPU Device Coredump ****\n");
+       drm_printf(&p, "kernel: " UTS_RELEASE "\n");
+       drm_printf(&p, "module: " KBUILD_MODNAME "\n");
+       drm_printf(&p, "time: %lld.%09ld\n", adev->reset_time.tv_sec, adev->reset_time.tv_nsec);
+       if (adev->reset_task_info.pid)
+               drm_printf(&p, "process_name: %s PID: %d\n",
+                          adev->reset_task_info.process_name,
+                          adev->reset_task_info.pid);
+
+       if (adev->reset_vram_lost)
+               drm_printf(&p, "VRAM is lost due to GPU reset!\n");
+       if (adev->num_regs) {
+               drm_printf(&p, "AMDGPU register dumps:\nOffset:     Value:\n");
+
+               for (i = 0; i < adev->num_regs; i++)
+                       drm_printf(&p, "0x%08x: 0x%08x\n",
+                                  adev->reset_dump_reg_list[i],
+                                  adev->reset_dump_reg_value[i]);
+       }
+
+       return count - iter.remain;
+}
+
+static void amdgpu_devcoredump_free(void *data)
+{
+}
+
+static void amdgpu_reset_capture_coredumpm(struct amdgpu_device *adev)
+{
+       struct drm_device *dev = adev_to_drm(adev);
+
+       ktime_get_ts64(&adev->reset_time);
+       dev_coredumpm(dev->dev, THIS_MODULE, adev, 0, GFP_KERNEL,
+                       amdgpu_devcoredump_read, amdgpu_devcoredump_free);
+}
+#endif
+
 int amdgpu_do_asic_reset(struct list_head *device_list_handle,
                          struct amdgpu_reset_context *reset_context)
 {
@@ -4818,6 +4876,15 @@ int amdgpu_do_asic_reset(struct list_head *device_list_handle,
                                         goto out;
 
                                 vram_lost = amdgpu_device_check_vram_lost(tmp_adev);
+#ifdef CONFIG_DEV_COREDUMP
+                               tmp_adev->reset_vram_lost = vram_lost;
+                               memset(&tmp_adev->reset_task_info, 0,
+                                               sizeof(tmp_adev->reset_task_info));
+                               if (reset_context->job && reset_context->job->vm)
+                                       tmp_adev->reset_task_info =
+                                               reset_context->job->vm->task_info;
+                               amdgpu_reset_capture_coredumpm(tmp_adev);
+#endif
                                 if (vram_lost) {
                                         DRM_INFO("VRAM is lost due to GPU reset!\n");
                                         amdgpu_inc_vram_lost(tmp_adev);
--
2.32.0