Linux kernel mirror (for testing)
git.kernel.org/pub/scm/linux/kernel/git/torvalds/linux.git
kernel
os
linux
drm/amdgpu: disable ras query and iject during gpu reset
added flag to ras context to indicate if ras query functionality is ready
Reviewed-by: Hawking Zhang <Hawking.Zhang@amd.com>
Signed-off-by: John Clements <john.clements@amd.com>
Signed-off-by: Alex Deucher <alexander.deucher@amd.com>
authored by
John Clements
and committed by
Alex Deucher
61380faa
66399248
+28
-3
+3
drivers/gpu/drm/amd/amdgpu/amdgpu_device.c
+3
drivers/gpu/drm/amd/amdgpu/amdgpu_device.c
···
4168
4168
need_full_reset = job_signaled = false;
4169
4169
INIT_LIST_HEAD(&device_list);
4170
4170
4171
+
amdgpu_ras_set_error_query_ready(adev, false);
4172
+
4171
4173
dev_info(adev->dev, "GPU %s begin!\n",
4172
4174
(in_ras_intr && !use_baco) ? "jobs stop":"reset");
4173
4175
···
4226
4224
/* block all schedulers and reset given job's ring */
4227
4225
list_for_each_entry(tmp_adev, device_list_handle, gmc.xgmi.head) {
4228
4226
if (tmp_adev != adev) {
4227
+
amdgpu_ras_set_error_query_ready(tmp_adev, false);
4229
4228
amdgpu_device_lock_adev(tmp_adev, false);
4230
4229
if (!amdgpu_sriov_vf(tmp_adev))
4231
4230
amdgpu_amdkfd_pre_reset(tmp_adev);
+21
-3
drivers/gpu/drm/amd/amdgpu/amdgpu_ras.c
+21
-3
drivers/gpu/drm/amd/amdgpu/amdgpu_ras.c
···
80
80
static bool amdgpu_ras_check_bad_page(struct amdgpu_device *adev,
81
81
uint64_t addr);
82
82
83
+
void amdgpu_ras_set_error_query_ready(struct amdgpu_device *adev, bool ready)
84
+
{
85
+
if (adev)
86
+
amdgpu_ras_get_context(adev)->error_query_ready = ready;
87
+
}
88
+
89
+
bool amdgpu_ras_get_error_query_ready(struct amdgpu_device *adev)
90
+
{
91
+
if (adev)
92
+
return amdgpu_ras_get_context(adev)->error_query_ready;
93
+
94
+
return false;
95
+
}
96
+
83
97
static ssize_t amdgpu_ras_debugfs_read(struct file *f, char __user *buf,
84
98
size_t size, loff_t *pos)
85
99
{
···
295
281
struct ras_debug_if data;
296
282
int ret = 0;
297
283
298
-
if (amdgpu_ras_intr_triggered()) {
284
+
if (!amdgpu_ras_get_error_query_ready(adev)) {
299
285
DRM_WARN("RAS WARN: error injection currently inaccessible\n");
300
286
return size;
301
287
}
···
413
399
.head = obj->head,
414
400
};
415
401
416
-
if (amdgpu_ras_intr_triggered())
402
+
if (!amdgpu_ras_get_error_query_ready(obj->adev))
417
403
return snprintf(buf, PAGE_SIZE,
418
404
"Query currently inaccessible\n");
419
405
···
1900
1886
}
1901
1887
1902
1888
/* in resume phase, no need to create ras fs node */
1903
-
if (adev->in_suspend || adev->in_gpu_reset)
1889
+
if (adev->in_suspend || adev->in_gpu_reset) {
1890
+
amdgpu_ras_set_error_query_ready(adev, true);
1904
1891
return 0;
1892
+
}
1905
1893
1906
1894
if (ih_info->cb) {
1907
1895
r = amdgpu_ras_interrupt_add_handler(adev, ih_info);
···
1914
1898
r = amdgpu_ras_sysfs_create(adev, fs_info);
1915
1899
if (r)
1916
1900
goto sysfs;
1901
+
1902
+
amdgpu_ras_set_error_query_ready(adev, true);
1917
1903
1918
1904
return 0;
1919
1905
cleanup:
+4
drivers/gpu/drm/amd/amdgpu/amdgpu_ras.h
+4
drivers/gpu/drm/amd/amdgpu/amdgpu_ras.h
···
334
334
uint32_t flags;
335
335
bool reboot;
336
336
struct amdgpu_ras_eeprom_control eeprom_control;
337
+
338
+
bool error_query_ready;
337
339
};
338
340
339
341
struct ras_fs_data {
···
630
628
}
631
629
632
630
void amdgpu_ras_global_ras_isr(struct amdgpu_device *adev);
631
+
632
+
void amdgpu_ras_set_error_query_ready(struct amdgpu_device *adev, bool ready);
633
633
634
634
#endif