tjh.dev
Linux kernel mirror (for testing)
git.kernel.org/pub/scm/linux/kernel/git/torvalds/linux.git
kernel
os
linux
drm/amdkfd: Enable processes isolation on gfx9
When amdgpu enable enforce_isolation, KFD enables single-process mode in
HWS and sets exec_cleaner_shader bit in MAP_PROCESS.
Signed-off-by: Amber Lin <Amber.Lin@amd.com>
Signed-off-by: Alex Deucher <alexander.deucher@amd.com>
authored by
Amber Lin
and committed by
Alex Deucher
87758a0e
f846250b
+17
-4
+13
-1
drivers/gpu/drm/amd/amdkfd/kfd_packet_manager_v9.c
+13
-1
drivers/gpu/drm/amd/amdkfd/kfd_packet_manager_v9.c
···
37
37
struct kfd_node *kfd = pm->dqm->dev;
38
38
struct kfd_process_device *pdd =
39
39
container_of(qpd, struct kfd_process_device, qpd);
40
+
struct amdgpu_device *adev = kfd->adev;
40
41
41
42
packet = (struct pm4_mes_map_process *)buffer;
42
43
memset(buffer, 0, sizeof(struct pm4_mes_map_process));
43
44
packet->header.u32All = pm_build_pm4_header(IT_MAP_PROCESS,
44
45
sizeof(struct pm4_mes_map_process));
46
+
if (adev->enforce_isolation[kfd->node_id])
47
+
packet->bitfields2.exec_cleaner_shader = 1;
45
48
packet->bitfields2.diq_enable = (qpd->is_debug) ? 1 : 0;
46
49
packet->bitfields2.process_quantum = 10;
47
50
packet->bitfields2.pasid = qpd->pqm->process->pasid;
···
92
89
struct pm4_mes_map_process_aldebaran *packet;
93
90
uint64_t vm_page_table_base_addr = qpd->page_table_base;
94
91
struct kfd_dev *kfd = pm->dqm->dev->kfd;
92
+
struct kfd_node *knode = pm->dqm->dev;
95
93
struct kfd_process_device *pdd =
96
94
container_of(qpd, struct kfd_process_device, qpd);
97
95
int i;
96
+
struct amdgpu_device *adev = kfd->adev;
98
97
99
98
packet = (struct pm4_mes_map_process_aldebaran *)buffer;
100
99
memset(buffer, 0, sizeof(struct pm4_mes_map_process_aldebaran));
101
100
packet->header.u32All = pm_build_pm4_header(IT_MAP_PROCESS,
102
101
sizeof(struct pm4_mes_map_process_aldebaran));
102
+
if (adev->enforce_isolation[knode->node_id])
103
+
packet->bitfields2.exec_cleaner_shader = 1;
103
104
packet->bitfields2.diq_enable = (qpd->is_debug) ? 1 : 0;
104
105
packet->bitfields2.process_quantum = 10;
105
106
packet->bitfields2.pasid = qpd->pqm->process->pasid;
···
151
144
152
145
int concurrent_proc_cnt = 0;
153
146
struct kfd_node *kfd = pm->dqm->dev;
147
+
struct amdgpu_device *adev = kfd->adev;
154
148
155
149
/* Determine the number of processes to map together to HW:
156
150
* it can not exceed the number of VMIDs available to the
157
151
* scheduler, and it is determined by the smaller of the number
158
152
* of processes in the runlist and kfd module parameter
159
153
* hws_max_conc_proc.
154
+
* However, if enforce_isolation is set (toggle LDS/VGPRs/SGPRs
155
+
* cleaner between process switch), enable single-process mode
156
+
* in HWS.
160
157
* Note: the arbitration between the number of VMIDs and
161
158
* hws_max_conc_proc has been done in
162
159
* kgd2kfd_device_init().
163
160
*/
164
-
concurrent_proc_cnt = min(pm->dqm->processes_count,
161
+
concurrent_proc_cnt = adev->enforce_isolation[kfd->node_id] ?
162
+
1 : min(pm->dqm->processes_count,
165
163
kfd->max_proc_per_quantum);
166
164
167
165
packet = (struct pm4_mes_runlist *)buffer;
+3
-2
drivers/gpu/drm/amd/amdkfd/kfd_pm4_headers_ai.h
+3
-2
drivers/gpu/drm/amd/amdkfd/kfd_pm4_headers_ai.h
···
145
145
146
146
union {
147
147
struct {
148
-
uint32_t pasid:16;
149
-
uint32_t reserved1:2;
148
+
uint32_t pasid:16; /* 0 - 15 */
149
+
uint32_t reserved1:1; /* 16 */
150
+
uint32_t exec_cleaner_shader:1; /* 17 */
150
151
uint32_t debug_vmid:4;
151
152
uint32_t new_debug:1;
152
153
uint32_t reserved2:1;
+1
-1
drivers/gpu/drm/amd/amdkfd/kfd_pm4_headers_aldebaran.h
+1
-1
drivers/gpu/drm/amd/amdkfd/kfd_pm4_headers_aldebaran.h