#include "kfd_mqd_manager.h"
#include "amdgpu_amdkfd.h"
#include "kfd_device_queue_manager.h"
int pipe_priority_map[] = {
KFD_PIPE_PRIORITY_CS_LOW,
KFD_PIPE_PRIORITY_CS_LOW,
KFD_PIPE_PRIORITY_CS_LOW,
KFD_PIPE_PRIORITY_CS_LOW,
KFD_PIPE_PRIORITY_CS_LOW,
KFD_PIPE_PRIORITY_CS_LOW,
KFD_PIPE_PRIORITY_CS_LOW,
KFD_PIPE_PRIORITY_CS_MEDIUM,
KFD_PIPE_PRIORITY_CS_MEDIUM,
KFD_PIPE_PRIORITY_CS_MEDIUM,
KFD_PIPE_PRIORITY_CS_MEDIUM,
KFD_PIPE_PRIORITY_CS_HIGH,
KFD_PIPE_PRIORITY_CS_HIGH,
KFD_PIPE_PRIORITY_CS_HIGH,
KFD_PIPE_PRIORITY_CS_HIGH,
KFD_PIPE_PRIORITY_CS_HIGH
};
struct kfd_mem_obj *allocate_hiq_mqd(struct kfd_node *dev, struct queue_properties *q)
{
struct kfd_mem_obj *mqd_mem_obj;
mqd_mem_obj = kzalloc(sizeof(struct kfd_mem_obj), GFP_KERNEL);
if (!mqd_mem_obj)
return NULL;
mqd_mem_obj->gtt_mem = dev->dqm->hiq_sdma_mqd.gtt_mem;
mqd_mem_obj->gpu_addr = dev->dqm->hiq_sdma_mqd.gpu_addr;
mqd_mem_obj->cpu_ptr = dev->dqm->hiq_sdma_mqd.cpu_ptr;
return mqd_mem_obj;
}
struct kfd_mem_obj *allocate_sdma_mqd(struct kfd_node *dev,
struct queue_properties *q)
{
struct kfd_mem_obj *mqd_mem_obj;
uint64_t offset;
mqd_mem_obj = kzalloc(sizeof(struct kfd_mem_obj), GFP_KERNEL);
if (!mqd_mem_obj)
return NULL;
offset = (q->sdma_engine_id *
dev->kfd->device_info.num_sdma_queues_per_engine +
q->sdma_queue_id) *
dev->dqm->mqd_mgrs[KFD_MQD_TYPE_SDMA]->mqd_size;
offset += dev->dqm->mqd_mgrs[KFD_MQD_TYPE_HIQ]->mqd_size *
NUM_XCC(dev->xcc_mask);
mqd_mem_obj->gtt_mem = (void *)((uint64_t)dev->dqm->hiq_sdma_mqd.gtt_mem
+ offset);
mqd_mem_obj->gpu_addr = dev->dqm->hiq_sdma_mqd.gpu_addr + offset;
mqd_mem_obj->cpu_ptr = (uint32_t *)((uint64_t)
dev->dqm->hiq_sdma_mqd.cpu_ptr + offset);
return mqd_mem_obj;
}
void free_mqd_hiq_sdma(struct mqd_manager *mm, void *mqd,
struct kfd_mem_obj *mqd_mem_obj)
{
WARN_ON(!mqd_mem_obj->gtt_mem);
kfree(mqd_mem_obj);
}
void mqd_symmetrically_map_cu_mask(struct mqd_manager *mm,
const uint32_t *cu_mask, uint32_t cu_mask_count,
uint32_t *se_mask, uint32_t inst)
{
struct kfd_cu_info cu_info;
uint32_t cu_per_sh[KFD_MAX_NUM_SE][KFD_MAX_NUM_SH_PER_SE] = {0};
bool wgp_mode_req = KFD_GC_VERSION(mm->dev) >= IP_VERSION(10, 0, 0);
uint32_t en_mask = wgp_mode_req ? 0x3 : 0x1;
int i, se, sh, cu, cu_bitmap_sh_mul, cu_inc = wgp_mode_req ? 2 : 1;
uint32_t cu_active_per_node;
int inc = cu_inc * NUM_XCC(mm->dev->xcc_mask);
int xcc_inst = inst + ffs(mm->dev->xcc_mask) - 1;
amdgpu_amdkfd_get_cu_info(mm->dev->adev, &cu_info);
cu_active_per_node = cu_info.cu_active_number / mm->dev->kfd->num_nodes;
if (cu_mask_count > cu_active_per_node)
cu_mask_count = cu_active_per_node;
if (cu_info.num_shader_engines > KFD_MAX_NUM_SE) {
pr_err("Exceeded KFD_MAX_NUM_SE, chip reports %d\n", cu_info.num_shader_engines);
return;
}
if (cu_info.num_shader_arrays_per_engine > KFD_MAX_NUM_SH_PER_SE) {
pr_err("Exceeded KFD_MAX_NUM_SH, chip reports %d\n",
cu_info.num_shader_arrays_per_engine * cu_info.num_shader_engines);
return;
}
cu_bitmap_sh_mul = (KFD_GC_VERSION(mm->dev) >= IP_VERSION(11, 0, 0) &&
KFD_GC_VERSION(mm->dev) < IP_VERSION(12, 0, 0)) ? 2 : 1;
for (se = 0; se < cu_info.num_shader_engines; se++)
for (sh = 0; sh < cu_info.num_shader_arrays_per_engine; sh++)
cu_per_sh[se][sh] = hweight32(
cu_info.cu_bitmap[xcc_inst][se % 4][sh + (se / 4) *
cu_bitmap_sh_mul]);
for (i = 0; i < cu_info.num_shader_engines; i++)
se_mask[i] = 0;
i = inst;
for (cu = 0; cu < 16; cu += cu_inc) {
for (sh = 0; sh < cu_info.num_shader_arrays_per_engine; sh++) {
for (se = 0; se < cu_info.num_shader_engines; se++) {
if (cu_per_sh[se][sh] > cu) {
if (cu_mask[i / 32] & (en_mask << (i % 32)))
se_mask[se] |= en_mask << (cu + sh * 16);
i += inc;
if (i >= cu_mask_count)
return;
}
}
}
}
}
int kfd_hiq_load_mqd_kiq(struct mqd_manager *mm, void *mqd,
uint32_t pipe_id, uint32_t queue_id,
struct queue_properties *p, struct mm_struct *mms)
{
return mm->dev->kfd2kgd->hiq_mqd_load(mm->dev->adev, mqd, pipe_id,
queue_id, p->doorbell_off, 0);
}
int kfd_destroy_mqd_cp(struct mqd_manager *mm, void *mqd,
enum kfd_preempt_type type, unsigned int timeout,
uint32_t pipe_id, uint32_t queue_id)
{
return mm->dev->kfd2kgd->hqd_destroy(mm->dev->adev, mqd, type, timeout,
pipe_id, queue_id, 0);
}
void kfd_free_mqd_cp(struct mqd_manager *mm, void *mqd,
struct kfd_mem_obj *mqd_mem_obj)
{
if (mqd_mem_obj->gtt_mem) {
amdgpu_amdkfd_free_gtt_mem(mm->dev->adev, mqd_mem_obj->gtt_mem);
kfree(mqd_mem_obj);
} else {
kfd_gtt_sa_free(mm->dev, mqd_mem_obj);
}
}
bool kfd_is_occupied_cp(struct mqd_manager *mm, void *mqd,
uint64_t queue_address, uint32_t pipe_id,
uint32_t queue_id)
{
return mm->dev->kfd2kgd->hqd_is_occupied(mm->dev->adev, queue_address,
pipe_id, queue_id, 0);
}
int kfd_load_mqd_sdma(struct mqd_manager *mm, void *mqd,
uint32_t pipe_id, uint32_t queue_id,
struct queue_properties *p, struct mm_struct *mms)
{
return mm->dev->kfd2kgd->hqd_sdma_load(mm->dev->adev, mqd,
(uint32_t __user *)p->write_ptr,
mms);
}
int kfd_destroy_mqd_sdma(struct mqd_manager *mm, void *mqd,
enum kfd_preempt_type type,
unsigned int timeout, uint32_t pipe_id,
uint32_t queue_id)
{
return mm->dev->kfd2kgd->hqd_sdma_destroy(mm->dev->adev, mqd, timeout);
}
bool kfd_is_occupied_sdma(struct mqd_manager *mm, void *mqd,
uint64_t queue_address, uint32_t pipe_id,
uint32_t queue_id)
{
return mm->dev->kfd2kgd->hqd_sdma_is_occupied(mm->dev->adev, mqd);
}
uint64_t kfd_hiq_mqd_stride(struct kfd_node *dev)
{
return dev->dqm->mqd_mgrs[KFD_MQD_TYPE_HIQ]->mqd_size;
}
void kfd_get_hiq_xcc_mqd(struct kfd_node *dev, struct kfd_mem_obj *mqd_mem_obj,
uint32_t virtual_xcc_id)
{
uint64_t offset;
offset = kfd_hiq_mqd_stride(dev) * virtual_xcc_id;
mqd_mem_obj->gtt_mem = (virtual_xcc_id == 0) ?
dev->dqm->hiq_sdma_mqd.gtt_mem : NULL;
mqd_mem_obj->gpu_addr = dev->dqm->hiq_sdma_mqd.gpu_addr + offset;
mqd_mem_obj->cpu_ptr = (uint32_t *)((uintptr_t)
dev->dqm->hiq_sdma_mqd.cpu_ptr + offset);
}
uint64_t kfd_mqd_stride(struct mqd_manager *mm,
struct queue_properties *q)
{
return mm->mqd_size;
}