drm/amdgpu: clean up init sequence for failures
If we fail during device init, record what state each
block is in so that we can tear down clearly.
Fixes various problems on device init failure.
Reviewed-by: Christian König <christian.koenig@amd.com>
Signed-off-by: Alex Deucher <alexander.deucher@amd.com>
diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu.h b/drivers/gpu/drm/amd/amdgpu/amdgpu.h
index e9fde72..31b00f9 100644
--- a/drivers/gpu/drm/amd/amdgpu/amdgpu.h
+++ b/drivers/gpu/drm/amd/amdgpu/amdgpu.h
@@ -1866,6 +1866,12 @@
typedef uint32_t (*amdgpu_block_rreg_t)(struct amdgpu_device*, uint32_t, uint32_t);
typedef void (*amdgpu_block_wreg_t)(struct amdgpu_device*, uint32_t, uint32_t, uint32_t);
+struct amdgpu_ip_block_status {
+ bool valid;
+ bool sw;
+ bool hw;
+};
+
struct amdgpu_device {
struct device *dev;
struct drm_device *ddev;
@@ -2008,7 +2014,7 @@
const struct amdgpu_ip_block_version *ip_blocks;
int num_ip_blocks;
- bool *ip_block_enabled;
+ struct amdgpu_ip_block_status *ip_block_status;
struct mutex mn_lock;
DECLARE_HASHTABLE(mn_hash, 7);
diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_device.c b/drivers/gpu/drm/amd/amdgpu/amdgpu_device.c
index d79009b..99f158e 100644
--- a/drivers/gpu/drm/amd/amdgpu/amdgpu_device.c
+++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_device.c
@@ -1191,8 +1191,9 @@
return -EINVAL;
}
- adev->ip_block_enabled = kcalloc(adev->num_ip_blocks, sizeof(bool), GFP_KERNEL);
- if (adev->ip_block_enabled == NULL)
+ adev->ip_block_status = kcalloc(adev->num_ip_blocks,
+ sizeof(struct amdgpu_ip_block_status), GFP_KERNEL);
+ if (adev->ip_block_status == NULL)
return -ENOMEM;
if (adev->ip_blocks == NULL) {
@@ -1203,18 +1204,18 @@
for (i = 0; i < adev->num_ip_blocks; i++) {
if ((amdgpu_ip_block_mask & (1 << i)) == 0) {
DRM_ERROR("disabled ip block: %d\n", i);
- adev->ip_block_enabled[i] = false;
+ adev->ip_block_status[i].valid = false;
} else {
if (adev->ip_blocks[i].funcs->early_init) {
r = adev->ip_blocks[i].funcs->early_init((void *)adev);
if (r == -ENOENT)
- adev->ip_block_enabled[i] = false;
+ adev->ip_block_status[i].valid = false;
else if (r)
return r;
else
- adev->ip_block_enabled[i] = true;
+ adev->ip_block_status[i].valid = true;
} else {
- adev->ip_block_enabled[i] = true;
+ adev->ip_block_status[i].valid = true;
}
}
}
@@ -1227,11 +1228,12 @@
int i, r;
for (i = 0; i < adev->num_ip_blocks; i++) {
- if (!adev->ip_block_enabled[i])
+ if (!adev->ip_block_status[i].valid)
continue;
r = adev->ip_blocks[i].funcs->sw_init((void *)adev);
if (r)
return r;
+ adev->ip_block_status[i].sw = true;
/* need to do gmc hw init early so we can allocate gpu mem */
if (adev->ip_blocks[i].type == AMD_IP_BLOCK_TYPE_GMC) {
r = amdgpu_vram_scratch_init(adev);
@@ -1243,11 +1245,12 @@
r = amdgpu_wb_init(adev);
if (r)
return r;
+ adev->ip_block_status[i].hw = true;
}
}
for (i = 0; i < adev->num_ip_blocks; i++) {
- if (!adev->ip_block_enabled[i])
+ if (!adev->ip_block_status[i].sw)
continue;
/* gmc hw init is done early */
if (adev->ip_blocks[i].type == AMD_IP_BLOCK_TYPE_GMC)
@@ -1255,6 +1258,7 @@
r = adev->ip_blocks[i].funcs->hw_init((void *)adev);
if (r)
return r;
+ adev->ip_block_status[i].hw = true;
}
return 0;
@@ -1265,7 +1269,7 @@
int i = 0, r;
for (i = 0; i < adev->num_ip_blocks; i++) {
- if (!adev->ip_block_enabled[i])
+ if (!adev->ip_block_status[i].valid)
continue;
/* enable clockgating to save power */
r = adev->ip_blocks[i].funcs->set_clockgating_state((void *)adev,
@@ -1287,7 +1291,7 @@
int i, r;
for (i = adev->num_ip_blocks - 1; i >= 0; i--) {
- if (!adev->ip_block_enabled[i])
+ if (!adev->ip_block_status[i].hw)
continue;
if (adev->ip_blocks[i].type == AMD_IP_BLOCK_TYPE_GMC) {
amdgpu_wb_fini(adev);
@@ -1300,14 +1304,16 @@
return r;
r = adev->ip_blocks[i].funcs->hw_fini((void *)adev);
/* XXX handle errors */
+ adev->ip_block_status[i].hw = false;
}
for (i = adev->num_ip_blocks - 1; i >= 0; i--) {
- if (!adev->ip_block_enabled[i])
+ if (!adev->ip_block_status[i].sw)
continue;
r = adev->ip_blocks[i].funcs->sw_fini((void *)adev);
/* XXX handle errors */
- adev->ip_block_enabled[i] = false;
+ adev->ip_block_status[i].sw = false;
+ adev->ip_block_status[i].valid = false;
}
return 0;
@@ -1318,7 +1324,7 @@
int i, r;
for (i = adev->num_ip_blocks - 1; i >= 0; i--) {
- if (!adev->ip_block_enabled[i])
+ if (!adev->ip_block_status[i].valid)
continue;
/* ungate blocks so that suspend can properly shut them down */
r = adev->ip_blocks[i].funcs->set_clockgating_state((void *)adev,
@@ -1336,7 +1342,7 @@
int i, r;
for (i = 0; i < adev->num_ip_blocks; i++) {
- if (!adev->ip_block_enabled[i])
+ if (!adev->ip_block_status[i].valid)
continue;
r = adev->ip_blocks[i].funcs->resume(adev);
if (r)
@@ -1582,8 +1588,8 @@
amdgpu_fence_driver_fini(adev);
amdgpu_fbdev_fini(adev);
r = amdgpu_fini(adev);
- kfree(adev->ip_block_enabled);
- adev->ip_block_enabled = NULL;
+ kfree(adev->ip_block_status);
+ adev->ip_block_status = NULL;
adev->accel_working = false;
/* free i2c buses */
amdgpu_i2c_fini(adev);
diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_kms.c b/drivers/gpu/drm/amd/amdgpu/amdgpu_kms.c
index 31ad444..d316bda 100644
--- a/drivers/gpu/drm/amd/amdgpu/amdgpu_kms.c
+++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_kms.c
@@ -235,7 +235,7 @@
for (i = 0; i < adev->num_ip_blocks; i++) {
if (adev->ip_blocks[i].type == type &&
- adev->ip_block_enabled[i]) {
+ adev->ip_block_status[i].valid) {
ip.hw_ip_version_major = adev->ip_blocks[i].major;
ip.hw_ip_version_minor = adev->ip_blocks[i].minor;
ip.capabilities_flags = 0;
@@ -274,7 +274,7 @@
for (i = 0; i < adev->num_ip_blocks; i++)
if (adev->ip_blocks[i].type == type &&
- adev->ip_block_enabled[i] &&
+ adev->ip_block_status[i].valid &&
count < AMDGPU_HW_IP_INSTANCE_MAX_COUNT)
count++;