bybrooklyn/linux
1
0
Fork 0
mirror of https://github.com/torvalds/linux.git synced 2026-04-26 18:42:25 -04:00
Code Issues Packages Projects Releases Wiki Activity

drm/amdkfd: Introduce kfd_node struct (v5)

Browse Source 
Introduce a new structure, kfd_node, which will now represent
a compute node. kfd_node is carved out of kfd_dev structure.
kfd_dev struct now will become the parent of kfd_node, and will
store common resources such as doorbells, GTT sub-alloctor etc.
kfd_node struct will store all resources specific to a compute
node, such as device queue manager, interrupt handling etc.

This is the first step in adding compute partition support in KFD.

v2: introduce kfd_node struct to gc v11 (Hawking)
v3: make reference to kfd_dev struct through kfd_node (Morris)
v4: use kfd_node instead for kfd isr/mqd functions (Morris)
v5: rebase (Alex)

Signed-off-by: Mukul Joshi <mukul.joshi@amd.com>
Tested-by: Amber Lin <Amber.Lin@amd.com>
Reviewed-by: Felix Kuehling <Felix.Kuehling@amd.com>
Signed-off-by: Hawking Zhang <Hawking.Zhang@amd.com>
Signed-off-by: Morris Zhang <Shiwu.Zhang@amd.com>
Signed-off-by: Alex Deucher <alexander.deucher@amd.com>
This commit is contained in:
Mukul Joshi
2022-09-14 16:39:48 +08:00
committed by Alex Deucher
parent 5cf1675591
commit 8dc1db3172
38 changed files with 574 additions and 496 deletions
Download Patch File Download Diff File Expand all files Collapse all files

54
drivers/gpu/drm/amd/amdkfd/kfd_process.c
View File

@@ -269,7 +269,7 @@ static int kfd_get_cu_occupancy(struct attribute *attr, char *buffer)
int cu_cnt;
int wave_cnt;
int max_waves_per_cu;
struct kfd_dev *dev = NULL;
struct kfd_node *dev = NULL;
struct kfd_process *proc = NULL;
struct kfd_process_device *pdd = NULL;
@@ -691,7 +691,7 @@ void kfd_process_destroy_wq(void)
static void kfd_process_free_gpuvm(struct kgd_mem *mem,
struct kfd_process_device *pdd, void **kptr)
{
struct kfd_dev *dev = pdd->dev;
struct kfd_node *dev = pdd->dev;
if (kptr && *kptr) {
amdgpu_amdkfd_gpuvm_unmap_gtt_bo_from_kernel(mem);
@@ -713,7 +713,7 @@ static int kfd_process_alloc_gpuvm(struct kfd_process_device *pdd,
uint64_t gpu_va, uint32_t size,
uint32_t flags, struct kgd_mem **mem, void **kptr)
{
struct kfd_dev *kdev = pdd->dev;
struct kfd_node *kdev = pdd->dev;
int err;
err = amdgpu_amdkfd_gpuvm_alloc_memory_of_gpu(kdev->adev, gpu_va, size,
@@ -982,7 +982,7 @@ static void kfd_process_device_free_bos(struct kfd_process_device *pdd)
static void kfd_process_kunmap_signal_bo(struct kfd_process *p)
{
struct kfd_process_device *pdd;
struct kfd_dev *kdev;
struct kfd_node *kdev;
void *mem;
kdev = kfd_device_by_id(GET_GPU_ID(p->signal_handle));
@@ -1040,9 +1040,9 @@ static void kfd_process_destroy_pdds(struct kfd_process *p)
bitmap_free(pdd->qpd.doorbell_bitmap);
idr_destroy(&pdd->alloc_idr);
kfd_free_process_doorbells(pdd->dev, pdd->doorbell_index);
kfd_free_process_doorbells(pdd->dev->kfd, pdd->doorbell_index);
if (pdd->dev->shared_resources.enable_mes)
if (pdd->dev->kfd->shared_resources.enable_mes)
amdgpu_amdkfd_free_gtt_mem(pdd->dev->adev,
pdd->proc_ctx_bo);
/*
@@ -1259,10 +1259,10 @@ static int kfd_process_init_cwsr_apu(struct kfd_process *p, struct file *filep)
int i;
for (i = 0; i < p->n_pdds; i++) {
struct kfd_dev *dev = p->pdds[i]->dev;
struct kfd_node *dev = p->pdds[i]->dev;
struct qcm_process_device *qpd = &p->pdds[i]->qpd;
if (!dev->cwsr_enabled || qpd->cwsr_kaddr || qpd->cwsr_base)
if (!dev->kfd->cwsr_enabled || qpd->cwsr_kaddr || qpd->cwsr_base)
continue;
offset = KFD_MMAP_TYPE_RESERVED_MEM | KFD_MMAP_GPU_ID(dev->id);
@@ -1279,7 +1279,7 @@ static int kfd_process_init_cwsr_apu(struct kfd_process *p, struct file *filep)
return err;
}
memcpy(qpd->cwsr_kaddr, dev->cwsr_isa, dev->cwsr_isa_size);
memcpy(qpd->cwsr_kaddr, dev->kfd->cwsr_isa, dev->kfd->cwsr_isa_size);
qpd->tma_addr = qpd->tba_addr + KFD_CWSR_TMA_OFFSET;
pr_debug("set tba :0x%llx, tma:0x%llx, cwsr_kaddr:%p for pqm.\n",
@@ -1291,7 +1291,7 @@ static int kfd_process_init_cwsr_apu(struct kfd_process *p, struct file *filep)
static int kfd_process_device_init_cwsr_dgpu(struct kfd_process_device *pdd)
{
struct kfd_dev *dev = pdd->dev;
struct kfd_node *dev = pdd->dev;
struct qcm_process_device *qpd = &pdd->qpd;
uint32_t flags = KFD_IOC_ALLOC_MEM_FLAGS_GTT
| KFD_IOC_ALLOC_MEM_FLAGS_NO_SUBSTITUTE
@@ -1300,7 +1300,7 @@ static int kfd_process_device_init_cwsr_dgpu(struct kfd_process_device *pdd)
void *kaddr;
int ret;
if (!dev->cwsr_enabled || qpd->cwsr_kaddr || !qpd->cwsr_base)
if (!dev->kfd->cwsr_enabled || qpd->cwsr_kaddr || !qpd->cwsr_base)
return 0;
/* cwsr_base is only set for dGPU */
@@ -1313,7 +1313,7 @@ static int kfd_process_device_init_cwsr_dgpu(struct kfd_process_device *pdd)
qpd->cwsr_kaddr = kaddr;
qpd->tba_addr = qpd->cwsr_base;
memcpy(qpd->cwsr_kaddr, dev->cwsr_isa, dev->cwsr_isa_size);
memcpy(qpd->cwsr_kaddr, dev->kfd->cwsr_isa, dev->kfd->cwsr_isa_size);
qpd->tma_addr = qpd->tba_addr + KFD_CWSR_TMA_OFFSET;
pr_debug("set tba :0x%llx, tma:0x%llx, cwsr_kaddr:%p for pqm.\n",
@@ -1324,10 +1324,10 @@ static int kfd_process_device_init_cwsr_dgpu(struct kfd_process_device *pdd)
static void kfd_process_device_destroy_cwsr_dgpu(struct kfd_process_device *pdd)
{
struct kfd_dev *dev = pdd->dev;
struct kfd_node *dev = pdd->dev;
struct qcm_process_device *qpd = &pdd->qpd;
if (!dev->cwsr_enabled || !qpd->cwsr_kaddr || !qpd->cwsr_base)
if (!dev->kfd->cwsr_enabled || !qpd->cwsr_kaddr || !qpd->cwsr_base)
return;
kfd_process_free_gpuvm(qpd->cwsr_mem, pdd, &qpd->cwsr_kaddr);
@@ -1371,7 +1371,7 @@ bool kfd_process_xnack_mode(struct kfd_process *p, bool supported)
* support retry.
*/
for (i = 0; i < p->n_pdds; i++) {
struct kfd_dev *dev = p->pdds[i]->dev;
struct kfd_node *dev = p->pdds[i]->dev;
/* Only consider GFXv9 and higher GPUs. Older GPUs don't
* support the SVM APIs and don't need to be considered
@@ -1394,7 +1394,7 @@ bool kfd_process_xnack_mode(struct kfd_process *p, bool supported)
if (KFD_GC_VERSION(dev) >= IP_VERSION(10, 1, 1))
return false;
if (dev->noretry)
if (dev->kfd->noretry)
return false;
}
@@ -1528,7 +1528,7 @@ static int init_doorbell_bitmap(struct qcm_process_device *qpd,
return 0;
}
struct kfd_process_device *kfd_get_process_device_data(struct kfd_dev *dev,
struct kfd_process_device *kfd_get_process_device_data(struct kfd_node *dev,
struct kfd_process *p)
{
int i;
@@ -1540,7 +1540,7 @@ struct kfd_process_device *kfd_get_process_device_data(struct kfd_dev *dev,
return NULL;
}
struct kfd_process_device *kfd_create_process_device_data(struct kfd_dev *dev,
struct kfd_process_device *kfd_create_process_device_data(struct kfd_node *dev,
struct kfd_process *p)
{
struct kfd_process_device *pdd = NULL;
@@ -1552,7 +1552,7 @@ struct kfd_process_device *kfd_create_process_device_data(struct kfd_dev *dev,
if (!pdd)
return NULL;
if (init_doorbell_bitmap(&pdd->qpd, dev)) {
if (init_doorbell_bitmap(&pdd->qpd, dev->kfd)) {
pr_err("Failed to init doorbell for process\n");
goto err_free_pdd;
}
@@ -1573,7 +1573,7 @@ struct kfd_process_device *kfd_create_process_device_data(struct kfd_dev *dev,
pdd->user_gpu_id = dev->id;
atomic64_set(&pdd->evict_duration_counter, 0);
if (dev->shared_resources.enable_mes) {
if (dev->kfd->shared_resources.enable_mes) {
retval = amdgpu_amdkfd_alloc_gtt_mem(dev->adev,
AMDGPU_MES_PROC_CTX_SIZE,
&pdd->proc_ctx_bo,
@@ -1619,7 +1619,7 @@ int kfd_process_device_init_vm(struct kfd_process_device *pdd,
struct amdgpu_fpriv *drv_priv;
struct amdgpu_vm *avm;
struct kfd_process *p;
struct kfd_dev *dev;
struct kfd_node *dev;
int ret;
if (!drm_file)
@@ -1679,7 +1679,7 @@ err_reserve_ib_mem:
*
* Assumes that the process lock is held.
*/
struct kfd_process_device *kfd_bind_process_to_device(struct kfd_dev *dev,
struct kfd_process_device *kfd_bind_process_to_device(struct kfd_node *dev,
struct kfd_process *p)
{
struct kfd_process_device *pdd;
@@ -1811,7 +1811,7 @@ int kfd_process_evict_queues(struct kfd_process *p, uint32_t trigger)
for (i = 0; i < p->n_pdds; i++) {
struct kfd_process_device *pdd = p->pdds[i];
kfd_smi_event_queue_eviction(pdd->dev, p->lead_thread->pid,
kfd_smi_event_queue_eviction(pdd->dev->kfd, p->lead_thread->pid,
trigger);
r = pdd->dev->dqm->ops.evict_process_queues(pdd->dev->dqm,
@@ -1839,7 +1839,7 @@ fail:
if (n_evicted == 0)
break;
kfd_smi_event_queue_restore(pdd->dev, p->lead_thread->pid);
kfd_smi_event_queue_restore(pdd->dev->kfd, p->lead_thread->pid);
if (pdd->dev->dqm->ops.restore_process_queues(pdd->dev->dqm,
&pdd->qpd))
@@ -1860,7 +1860,7 @@ int kfd_process_restore_queues(struct kfd_process *p)
for (i = 0; i < p->n_pdds; i++) {
struct kfd_process_device *pdd = p->pdds[i];
kfd_smi_event_queue_restore(pdd->dev, p->lead_thread->pid);
kfd_smi_event_queue_restore(pdd->dev->kfd, p->lead_thread->pid);
r = pdd->dev->dqm->ops.restore_process_queues(pdd->dev->dqm,
&pdd->qpd);
@@ -2016,7 +2016,7 @@ int kfd_resume_all_processes(void)
return ret;
}
int kfd_reserved_mem_mmap(struct kfd_dev *dev, struct kfd_process *process,
int kfd_reserved_mem_mmap(struct kfd_node *dev, struct kfd_process *process,
struct vm_area_struct *vma)
{
struct kfd_process_device *pdd;
@@ -2051,7 +2051,7 @@ void kfd_flush_tlb(struct kfd_process_device *pdd, enum TLB_FLUSH_TYPE type)
{
struct amdgpu_vm *vm = drm_priv_to_vm(pdd->drm_priv);
uint64_t tlb_seq = amdgpu_vm_tlb_seq(vm);
struct kfd_dev *dev = pdd->dev;
struct kfd_node *dev = pdd->dev;
/*
* It can be that we race and lose here, but that is extremely unlikely