AI/rocm-systems
2
0
Fork 0
Code Issues Pull Requests Actions Packages Projects Releases Wiki Activity

gfx900: Allow doorbell allocation independent of queue ID

Browse Source 
On SOC15 chips, the ABI for the create_queue ioctl is changed to
allow doorbell allocation independent of the queue ID. This is
necessary to accommodate doorbell routing to specific engines in
the BIF.

Change-Id: Ie98d0a758758149dd5fc09ae088afccc29904124
Signed-off-by: Felix Kuehling <Felix.Kuehling@amd.com>
This commit is contained in:
Felix Kuehling
2016-11-25 19:29:33 -05:00
parent d7063dd102
commit 7de66d149b
1 changed files with 22 additions and 3 deletions
Download Patch File Download Diff File Expand all files Collapse all files
src/queues.c
+22 -3
View File
@@ -55,6 +55,7 @@ enum asic_family_type {
#define IS_VI(chip) ((chip) >= CHIP_CARRIZO && (chip) <= CHIP_POLARIS11) #define IS_VI(chip) ((chip) >= CHIP_CARRIZO && (chip) <= CHIP_POLARIS11)
#define IS_DGPU(chip) (((chip) >= CHIP_TONGA && (chip) <= CHIP_VEGA10) || \ #define IS_DGPU(chip) (((chip) >= CHIP_TONGA && (chip) <= CHIP_VEGA10) || \
(chip) == CHIP_HAWAII) (chip) == CHIP_HAWAII)
#define IS_SOC15(chip) ((chip) >= CHIP_VEGA10)
#define WG_CONTEXT_DATA_SIZE_PER_CU_VI 344576 #define WG_CONTEXT_DATA_SIZE_PER_CU_VI 344576
#define WAVES_PER_CU_VI 32 #define WAVES_PER_CU_VI 32
@@ -563,6 +564,8 @@ hsaKmtCreateQueue(
HSAKMT_STATUS result; HSAKMT_STATUS result;
uint32_t gpu_id; uint32_t gpu_id;
uint16_t dev_id; uint16_t dev_id;
uint64_t doorbell_mmap_offset;
unsigned doorbell_offset;
struct device_info *dev_info; struct device_info *dev_info;
int err; int err;
CHECK_KFD_OPEN(); CHECK_KFD_OPEN();
@@ -631,7 +634,24 @@ hsaKmtCreateQueue(
q->queue_id = args.queue_id; q->queue_id = args.queue_id;
err = map_doorbell(NodeId, gpu_id, args.doorbell_offset); if (IS_SOC15(dev_info->asic_family)) {
/* On SOC15 chips, the doorbell offset within the
* doorbell page is included in the doorbell offset
* returned by KFD. This allows doorbells to be
* allocated per-device, independent of the
* per-process queue ID. */
doorbell_mmap_offset = args.doorbell_offset &
~(HSAuint64)(doorbells[NodeId].size - 1);
doorbell_offset = args.doorbell_offset &
(doorbells[NodeId].size - 1);
} else {
/* On older chips, the doorbell offset within the
* doorbell page is based on the queue ID. */
doorbell_mmap_offset = args.doorbell_offset;
doorbell_offset = q->queue_id * dev_info->doorbell_size;
}
err = map_doorbell(NodeId, gpu_id, doorbell_mmap_offset);
if (err != HSAKMT_STATUS_SUCCESS) { if (err != HSAKMT_STATUS_SUCCESS) {
hsaKmtDestroyQueue(q->queue_id); hsaKmtDestroyQueue(q->queue_id);
free_queue(q); free_queue(q);
@@ -640,8 +660,7 @@ hsaKmtCreateQueue(
QueueResource->QueueId = PORT_VPTR_TO_UINT64(q); QueueResource->QueueId = PORT_VPTR_TO_UINT64(q);
QueueResource->Queue_DoorBell = VOID_PTR_ADD(doorbells[NodeId].doorbells, QueueResource->Queue_DoorBell = VOID_PTR_ADD(doorbells[NodeId].doorbells,
(q->queue_id * doorbell_offset);
dev_info->doorbell_size));
return HSAKMT_STATUS_SUCCESS; return HSAKMT_STATUS_SUCCESS;
} }