aboutsummaryrefslogtreecommitdiff
path: root/drivers/gpu/drm/amd/amdgpu/amdgpu_amdkfd_gfx_v8.c
diff options
context:
space:
mode:
Diffstat (limited to 'drivers/gpu/drm/amd/amdgpu/amdgpu_amdkfd_gfx_v8.c')
-rw-r--r--drivers/gpu/drm/amd/amdgpu/amdgpu_amdkfd_gfx_v8.c124
1 files changed, 48 insertions, 76 deletions
diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_amdkfd_gfx_v8.c b/drivers/gpu/drm/amd/amdgpu/amdgpu_amdkfd_gfx_v8.c
index 5ce0ce704a21..52832cd69a93 100644
--- a/drivers/gpu/drm/amd/amdgpu/amdgpu_amdkfd_gfx_v8.c
+++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_amdkfd_gfx_v8.c
@@ -39,68 +39,54 @@ enum hqd_dequeue_request_type {
RESET_WAVES
};
-static inline struct amdgpu_device *get_amdgpu_device(struct kgd_dev *kgd)
-{
- return (struct amdgpu_device *)kgd;
-}
-
-static void lock_srbm(struct kgd_dev *kgd, uint32_t mec, uint32_t pipe,
+static void lock_srbm(struct amdgpu_device *adev, uint32_t mec, uint32_t pipe,
uint32_t queue, uint32_t vmid)
{
- struct amdgpu_device *adev = get_amdgpu_device(kgd);
uint32_t value = PIPEID(pipe) | MEID(mec) | VMID(vmid) | QUEUEID(queue);
mutex_lock(&adev->srbm_mutex);
WREG32(mmSRBM_GFX_CNTL, value);
}
-static void unlock_srbm(struct kgd_dev *kgd)
+static void unlock_srbm(struct amdgpu_device *adev)
{
- struct amdgpu_device *adev = get_amdgpu_device(kgd);
-
WREG32(mmSRBM_GFX_CNTL, 0);
mutex_unlock(&adev->srbm_mutex);
}
-static void acquire_queue(struct kgd_dev *kgd, uint32_t pipe_id,
+static void acquire_queue(struct amdgpu_device *adev, uint32_t pipe_id,
uint32_t queue_id)
{
- struct amdgpu_device *adev = get_amdgpu_device(kgd);
-
uint32_t mec = (pipe_id / adev->gfx.mec.num_pipe_per_mec) + 1;
uint32_t pipe = (pipe_id % adev->gfx.mec.num_pipe_per_mec);
- lock_srbm(kgd, mec, pipe, queue_id, 0);
+ lock_srbm(adev, mec, pipe, queue_id, 0);
}
-static void release_queue(struct kgd_dev *kgd)
+static void release_queue(struct amdgpu_device *adev)
{
- unlock_srbm(kgd);
+ unlock_srbm(adev);
}
-static void kgd_program_sh_mem_settings(struct kgd_dev *kgd, uint32_t vmid,
+static void kgd_program_sh_mem_settings(struct amdgpu_device *adev, uint32_t vmid,
uint32_t sh_mem_config,
uint32_t sh_mem_ape1_base,
uint32_t sh_mem_ape1_limit,
uint32_t sh_mem_bases)
{
- struct amdgpu_device *adev = get_amdgpu_device(kgd);
-
- lock_srbm(kgd, 0, 0, 0, vmid);
+ lock_srbm(adev, 0, 0, 0, vmid);
WREG32(mmSH_MEM_CONFIG, sh_mem_config);
WREG32(mmSH_MEM_APE1_BASE, sh_mem_ape1_base);
WREG32(mmSH_MEM_APE1_LIMIT, sh_mem_ape1_limit);
WREG32(mmSH_MEM_BASES, sh_mem_bases);
- unlock_srbm(kgd);
+ unlock_srbm(adev);
}
-static int kgd_set_pasid_vmid_mapping(struct kgd_dev *kgd, u32 pasid,
+static int kgd_set_pasid_vmid_mapping(struct amdgpu_device *adev, u32 pasid,
unsigned int vmid)
{
- struct amdgpu_device *adev = get_amdgpu_device(kgd);
-
/*
* We have to assume that there is no outstanding mapping.
* The ATC_VMID_PASID_MAPPING_UPDATE_STATUS bit could be 0 because
@@ -123,21 +109,20 @@ static int kgd_set_pasid_vmid_mapping(struct kgd_dev *kgd, u32 pasid,
return 0;
}
-static int kgd_init_interrupts(struct kgd_dev *kgd, uint32_t pipe_id)
+static int kgd_init_interrupts(struct amdgpu_device *adev, uint32_t pipe_id)
{
- struct amdgpu_device *adev = get_amdgpu_device(kgd);
uint32_t mec;
uint32_t pipe;
mec = (pipe_id / adev->gfx.mec.num_pipe_per_mec) + 1;
pipe = (pipe_id % adev->gfx.mec.num_pipe_per_mec);
- lock_srbm(kgd, mec, pipe, 0, 0);
+ lock_srbm(adev, mec, pipe, 0, 0);
WREG32(mmCPC_INT_CNTL, CP_INT_CNTL_RING0__TIME_STAMP_INT_ENABLE_MASK |
CP_INT_CNTL_RING0__OPCODE_ERROR_INT_ENABLE_MASK);
- unlock_srbm(kgd);
+ unlock_srbm(adev);
return 0;
}
@@ -165,12 +150,11 @@ static inline struct vi_sdma_mqd *get_sdma_mqd(void *mqd)
return (struct vi_sdma_mqd *)mqd;
}
-static int kgd_hqd_load(struct kgd_dev *kgd, void *mqd, uint32_t pipe_id,
- uint32_t queue_id, uint32_t __user *wptr,
- uint32_t wptr_shift, uint32_t wptr_mask,
- struct mm_struct *mm)
+static int kgd_hqd_load(struct amdgpu_device *adev, void *mqd,
+ uint32_t pipe_id, uint32_t queue_id,
+ uint32_t __user *wptr, uint32_t wptr_shift,
+ uint32_t wptr_mask, struct mm_struct *mm)
{
- struct amdgpu_device *adev = get_amdgpu_device(kgd);
struct vi_mqd *m;
uint32_t *mqd_hqd;
uint32_t reg, wptr_val, data;
@@ -178,7 +162,7 @@ static int kgd_hqd_load(struct kgd_dev *kgd, void *mqd, uint32_t pipe_id,
m = get_mqd(mqd);
- acquire_queue(kgd, pipe_id, queue_id);
+ acquire_queue(adev, pipe_id, queue_id);
/* HIQ is set during driver init period with vmid set to 0*/
if (m->cp_hqd_vmid == 0) {
@@ -206,7 +190,7 @@ static int kgd_hqd_load(struct kgd_dev *kgd, void *mqd, uint32_t pipe_id,
* on ASICs that do not support context-save.
* EOP writes/reads can start anywhere in the ring.
*/
- if (get_amdgpu_device(kgd)->asic_type != CHIP_TONGA) {
+ if (adev->asic_type != CHIP_TONGA) {
WREG32(mmCP_HQD_EOP_RPTR, m->cp_hqd_eop_rptr);
WREG32(mmCP_HQD_EOP_WPTR, m->cp_hqd_eop_wptr);
WREG32(mmCP_HQD_EOP_WPTR_MEM, m->cp_hqd_eop_wptr_mem);
@@ -226,25 +210,24 @@ static int kgd_hqd_load(struct kgd_dev *kgd, void *mqd, uint32_t pipe_id,
* release srbm_mutex to avoid circular dependency between
* srbm_mutex->mm_sem->reservation_ww_class_mutex->srbm_mutex.
*/
- release_queue(kgd);
+ release_queue(adev);
valid_wptr = read_user_wptr(mm, wptr, wptr_val);
- acquire_queue(kgd, pipe_id, queue_id);
+ acquire_queue(adev, pipe_id, queue_id);
if (valid_wptr)
WREG32(mmCP_HQD_PQ_WPTR, (wptr_val << wptr_shift) & wptr_mask);
data = REG_SET_FIELD(m->cp_hqd_active, CP_HQD_ACTIVE, ACTIVE, 1);
WREG32(mmCP_HQD_ACTIVE, data);
- release_queue(kgd);
+ release_queue(adev);
return 0;
}
-static int kgd_hqd_dump(struct kgd_dev *kgd,
+static int kgd_hqd_dump(struct amdgpu_device *adev,
uint32_t pipe_id, uint32_t queue_id,
uint32_t (**dump)[2], uint32_t *n_regs)
{
- struct amdgpu_device *adev = get_amdgpu_device(kgd);
uint32_t i = 0, reg;
#define HQD_N_REGS (54+4)
#define DUMP_REG(addr) do { \
@@ -258,7 +241,7 @@ static int kgd_hqd_dump(struct kgd_dev *kgd,
if (*dump == NULL)
return -ENOMEM;
- acquire_queue(kgd, pipe_id, queue_id);
+ acquire_queue(adev, pipe_id, queue_id);
DUMP_REG(mmCOMPUTE_STATIC_THREAD_MGMT_SE0);
DUMP_REG(mmCOMPUTE_STATIC_THREAD_MGMT_SE1);
@@ -268,7 +251,7 @@ static int kgd_hqd_dump(struct kgd_dev *kgd,
for (reg = mmCP_MQD_BASE_ADDR; reg <= mmCP_HQD_EOP_DONES; reg++)
DUMP_REG(reg);
- release_queue(kgd);
+ release_queue(adev);
WARN_ON_ONCE(i != HQD_N_REGS);
*n_regs = i;
@@ -276,10 +259,9 @@ static int kgd_hqd_dump(struct kgd_dev *kgd,
return 0;
}
-static int kgd_hqd_sdma_load(struct kgd_dev *kgd, void *mqd,
+static int kgd_hqd_sdma_load(struct amdgpu_device *adev, void *mqd,
uint32_t __user *wptr, struct mm_struct *mm)
{
- struct amdgpu_device *adev = get_amdgpu_device(kgd);
struct vi_sdma_mqd *m;
unsigned long end_jiffies;
uint32_t sdma_rlc_reg_offset;
@@ -331,11 +313,10 @@ static int kgd_hqd_sdma_load(struct kgd_dev *kgd, void *mqd,
return 0;
}
-static int kgd_hqd_sdma_dump(struct kgd_dev *kgd,
+static int kgd_hqd_sdma_dump(struct amdgpu_device *adev,
uint32_t engine_id, uint32_t queue_id,
uint32_t (**dump)[2], uint32_t *n_regs)
{
- struct amdgpu_device *adev = get_amdgpu_device(kgd);
uint32_t sdma_offset = engine_id * SDMA1_REGISTER_OFFSET +
queue_id * KFD_VI_SDMA_QUEUE_OFFSET;
uint32_t i = 0, reg;
@@ -367,15 +348,15 @@ static int kgd_hqd_sdma_dump(struct kgd_dev *kgd,
return 0;
}
-static bool kgd_hqd_is_occupied(struct kgd_dev *kgd, uint64_t queue_address,
- uint32_t pipe_id, uint32_t queue_id)
+static bool kgd_hqd_is_occupied(struct amdgpu_device *adev,
+ uint64_t queue_address, uint32_t pipe_id,
+ uint32_t queue_id)
{
- struct amdgpu_device *adev = get_amdgpu_device(kgd);
uint32_t act;
bool retval = false;
uint32_t low, high;
- acquire_queue(kgd, pipe_id, queue_id);
+ acquire_queue(adev, pipe_id, queue_id);
act = RREG32(mmCP_HQD_ACTIVE);
if (act) {
low = lower_32_bits(queue_address >> 8);
@@ -385,13 +366,12 @@ static bool kgd_hqd_is_occupied(struct kgd_dev *kgd, uint64_t queue_address,
high == RREG32(mmCP_HQD_PQ_BASE_HI))
retval = true;
}
- release_queue(kgd);
+ release_queue(adev);
return retval;
}
-static bool kgd_hqd_sdma_is_occupied(struct kgd_dev *kgd, void *mqd)
+static bool kgd_hqd_sdma_is_occupied(struct amdgpu_device *adev, void *mqd)
{
- struct amdgpu_device *adev = get_amdgpu_device(kgd);
struct vi_sdma_mqd *m;
uint32_t sdma_rlc_reg_offset;
uint32_t sdma_rlc_rb_cntl;
@@ -407,12 +387,11 @@ static bool kgd_hqd_sdma_is_occupied(struct kgd_dev *kgd, void *mqd)
return false;
}
-static int kgd_hqd_destroy(struct kgd_dev *kgd, void *mqd,
+static int kgd_hqd_destroy(struct amdgpu_device *adev, void *mqd,
enum kfd_preempt_type reset_type,
unsigned int utimeout, uint32_t pipe_id,
uint32_t queue_id)
{
- struct amdgpu_device *adev = get_amdgpu_device(kgd);
uint32_t temp;
enum hqd_dequeue_request_type type;
unsigned long flags, end_jiffies;
@@ -422,7 +401,7 @@ static int kgd_hqd_destroy(struct kgd_dev *kgd, void *mqd,
if (amdgpu_in_reset(adev))
return -EIO;
- acquire_queue(kgd, pipe_id, queue_id);
+ acquire_queue(adev, pipe_id, queue_id);
if (m->cp_hqd_vmid == 0)
WREG32_FIELD(RLC_CP_SCHEDULERS, scheduler1, 0);
@@ -502,20 +481,19 @@ loop:
break;
if (time_after(jiffies, end_jiffies)) {
pr_err("cp queue preemption time out.\n");
- release_queue(kgd);
+ release_queue(adev);
return -ETIME;
}
usleep_range(500, 1000);
}
- release_queue(kgd);
+ release_queue(adev);
return 0;
}
-static int kgd_hqd_sdma_destroy(struct kgd_dev *kgd, void *mqd,
+static int kgd_hqd_sdma_destroy(struct amdgpu_device *adev, void *mqd,
unsigned int utimeout)
{
- struct amdgpu_device *adev = get_amdgpu_device(kgd);
struct vi_sdma_mqd *m;
uint32_t sdma_rlc_reg_offset;
uint32_t temp;
@@ -549,11 +527,10 @@ static int kgd_hqd_sdma_destroy(struct kgd_dev *kgd, void *mqd,
return 0;
}
-static bool get_atc_vmid_pasid_mapping_info(struct kgd_dev *kgd,
+static bool get_atc_vmid_pasid_mapping_info(struct amdgpu_device *adev,
uint8_t vmid, uint16_t *p_pasid)
{
uint32_t value;
- struct amdgpu_device *adev = (struct amdgpu_device *) kgd;
value = RREG32(mmATC_VMID0_PASID_MAPPING + vmid);
*p_pasid = value & ATC_VMID0_PASID_MAPPING__PASID_MASK;
@@ -561,12 +538,12 @@ static bool get_atc_vmid_pasid_mapping_info(struct kgd_dev *kgd,
return !!(value & ATC_VMID0_PASID_MAPPING__VALID_MASK);
}
-static int kgd_address_watch_disable(struct kgd_dev *kgd)
+static int kgd_address_watch_disable(struct amdgpu_device *adev)
{
return 0;
}
-static int kgd_address_watch_execute(struct kgd_dev *kgd,
+static int kgd_address_watch_execute(struct amdgpu_device *adev,
unsigned int watch_point_id,
uint32_t cntl_val,
uint32_t addr_hi,
@@ -575,11 +552,10 @@ static int kgd_address_watch_execute(struct kgd_dev *kgd,
return 0;
}
-static int kgd_wave_control_execute(struct kgd_dev *kgd,
+static int kgd_wave_control_execute(struct amdgpu_device *adev,
uint32_t gfx_index_val,
uint32_t sq_cmd)
{
- struct amdgpu_device *adev = get_amdgpu_device(kgd);
uint32_t data = 0;
mutex_lock(&adev->grbm_idx_mutex);
@@ -600,28 +576,24 @@ static int kgd_wave_control_execute(struct kgd_dev *kgd,
return 0;
}
-static uint32_t kgd_address_watch_get_offset(struct kgd_dev *kgd,
+static uint32_t kgd_address_watch_get_offset(struct amdgpu_device *adev,
unsigned int watch_point_id,
unsigned int reg_offset)
{
return 0;
}
-static void set_scratch_backing_va(struct kgd_dev *kgd,
+static void set_scratch_backing_va(struct amdgpu_device *adev,
uint64_t va, uint32_t vmid)
{
- struct amdgpu_device *adev = (struct amdgpu_device *) kgd;
-
- lock_srbm(kgd, 0, 0, 0, vmid);
+ lock_srbm(adev, 0, 0, 0, vmid);
WREG32(mmSH_HIDDEN_PRIVATE_BASE_VMID, va);
- unlock_srbm(kgd);
+ unlock_srbm(adev);
}
-static void set_vm_context_page_table_base(struct kgd_dev *kgd, uint32_t vmid,
- uint64_t page_table_base)
+static void set_vm_context_page_table_base(struct amdgpu_device *adev,
+ uint32_t vmid, uint64_t page_table_base)
{
- struct amdgpu_device *adev = get_amdgpu_device(kgd);
-
if (!amdgpu_amdkfd_is_kfd_vmid(adev, vmid)) {
pr_err("trying to set page table base for wrong VMID\n");
return;
Powered by cgit, Gruvboxed by Blaster4385.