static int update_mqd_hiq(struct mqd_manager *mm, void *mqd,
struct queue_properties *q)
{
struct cik_mqd *m;
m = get_mqd(mqd);
m->cp_hqd_pq_control = DEFAULT_RPTR_BLOCK_SIZE |
DEFAULT_MIN_AVAIL_SIZE |
PRIV_STATE |
KMD_QUEUE;
/*
* Calculating queue size which is log base 2 of actual queue
* size -1 dwords
*/
m->cp_hqd_pq_control |= order_base_2(q->queue_size / 4) - 1;
m->cp_hqd_pq_base_lo = lower_32_bits((uint64_t)q->queue_address >> 8);
m->cp_hqd_pq_base_hi = upper_32_bits((uint64_t)q->queue_address >> 8);
m->cp_hqd_pq_rptr_report_addr_lo = lower_32_bits((uint64_t)q->read_ptr);
m->cp_hqd_pq_rptr_report_addr_hi = upper_32_bits((uint64_t)q->read_ptr);
m->cp_hqd_pq_doorbell_control = DOORBELL_OFFSET(q->doorbell_off);
m->cp_hqd_vmid = q->vmid;
q->is_active = (q->queue_size > 0 &&
q->queue_address != 0 &&
q->queue_percent > 0 &&
!q->is_evicted);
return 0;
}
static void update_cu_mask(struct mqd_manager *mm, void *mqd,
struct queue_properties *q)
{
struct cik_mqd *m;
struct kfd_cu_info cu_info;
uint32_t mgmt_se_mask;
uint32_t cu_sh_mask, cu_sh_shift;
uint32_t cu_mask;
int se, sh;
if (q->cu_mask == 0)
return;
m = get_mqd(mqd);
m->compute_static_thread_mgmt_se0 = 0;
m->compute_static_thread_mgmt_se1 = 0;
m->compute_static_thread_mgmt_se2 = 0;
m->compute_static_thread_mgmt_se3 = 0;
mm->dev->kfd2kgd->get_cu_info(mm->dev->kgd, &cu_info);
cu_mask = q->cu_mask;
for (se = 0; se < cu_info.num_shader_engines && cu_mask; se++) {
mgmt_se_mask = 0;
for (sh = 0; sh < 2 && cu_mask; sh++) {
cu_sh_shift = hweight32(cu_info.cu_bitmap[se][sh]);
cu_sh_mask = (1 << cu_sh_shift) - 1;
mgmt_se_mask |= (cu_mask & cu_sh_mask) << (sh * 16);
cu_mask >>= cu_sh_shift;
}
switch (se) {
case 0:
m->compute_static_thread_mgmt_se0 = mgmt_se_mask;
break;
case 1:
m->compute_static_thread_mgmt_se1 = mgmt_se_mask;
break;
case 2:
m->compute_static_thread_mgmt_se2 = mgmt_se_mask;
break;
case 3:
m->compute_static_thread_mgmt_se3 = mgmt_se_mask;
break;
default:
break;
}
}
pr_debug("kfd: update cu mask to %#x %#x %#x %#x\n",
m->compute_static_thread_mgmt_se0,
m->compute_static_thread_mgmt_se1,
m->compute_static_thread_mgmt_se2,
m->compute_static_thread_mgmt_se3);
}
static int update_mqd_hiq(struct mqd_manager *mm, void *mqd,
struct queue_properties *q)
{
struct v9_mqd *m;
int retval = update_mqd(mm, mqd, q);
if (retval != 0)
return retval;
/* TODO: what's the point? update_mqd already does this. */
m = get_mqd(mqd);
m->cp_hqd_vmid = q->vmid;
return retval;
}
static int init_mqd_hiq(struct mqd_manager *mm, void **mqd,
struct kfd_mem_obj **mqd_mem_obj, uint64_t *gart_addr,
struct queue_properties *q)
{
struct v9_mqd *m;
int retval = init_mqd(mm, mqd, mqd_mem_obj, gart_addr, q);
if (retval != 0)
return retval;
m = get_mqd(*mqd);
m->cp_hqd_pq_control |= 1 << CP_HQD_PQ_CONTROL__PRIV_STATE__SHIFT |
1 << CP_HQD_PQ_CONTROL__KMD_QUEUE__SHIFT;
return retval;
}
static int update_mqd(struct mqd_manager *mm, void *mqd,
struct queue_properties *q)
{
struct cik_mqd *m;
BUG_ON(!mm || !q || !mqd);
pr_debug("kfd: In func %s\n", __func__);
m = get_mqd(mqd);
m->cp_hqd_pq_control = DEFAULT_RPTR_BLOCK_SIZE |
DEFAULT_MIN_AVAIL_SIZE | PQ_ATC_EN;
/*
* Calculating queue size which is log base 2 of actual queue size -1
* dwords and another -1 for ffs
*/
m->cp_hqd_pq_control |= ffs(q->queue_size / sizeof(unsigned int))
- 1 - 1;
m->cp_hqd_pq_base_lo = lower_32_bits((uint64_t)q->queue_address >> 8);
m->cp_hqd_pq_base_hi = upper_32_bits((uint64_t)q->queue_address >> 8);
m->cp_hqd_pq_rptr_report_addr_lo = lower_32_bits((uint64_t)q->read_ptr);
m->cp_hqd_pq_rptr_report_addr_hi = upper_32_bits((uint64_t)q->read_ptr);
m->cp_hqd_pq_doorbell_control = DOORBELL_EN |
DOORBELL_OFFSET(q->doorbell_off);
m->cp_hqd_vmid = q->vmid;
if (q->format == KFD_QUEUE_FORMAT_AQL) {
m->cp_hqd_pq_control |= NO_UPDATE_RPTR;
}
update_cu_mask(mm, mqd, q);
m->cp_hqd_active = 0;
q->is_active = false;
if (q->queue_size > 0 &&
q->queue_address != 0 &&
q->queue_percent > 0) {
m->cp_hqd_active = 1;
q->is_active = true;
}
return 0;
}
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)
{
struct amdgpu_device *adev = get_amdgpu_device(kgd);
struct cik_mqd *m;
uint32_t *mqd_hqd;
uint32_t reg, wptr_val, data;
bool valid_wptr = false;
m = get_mqd(mqd);
acquire_queue(kgd, pipe_id, queue_id);
/* HQD registers extend from CP_MQD_BASE_ADDR to CP_MQD_CONTROL. */
mqd_hqd = &m->cp_mqd_base_addr_lo;
for (reg = mmCP_MQD_BASE_ADDR; reg <= mmCP_MQD_CONTROL; reg++)
WREG32(reg, mqd_hqd[reg - mmCP_MQD_BASE_ADDR]);
/* Copy userspace write pointer value to register.
* Activate doorbell logic to monitor subsequent changes.
*/
data = REG_SET_FIELD(m->cp_hqd_pq_doorbell_control,
CP_HQD_PQ_DOORBELL_CONTROL, DOORBELL_EN, 1);
WREG32(mmCP_HQD_PQ_DOORBELL_CONTROL, data);
/* read_user_ptr may take the mm->mmap_sem.
* release srbm_mutex to avoid circular dependency between
* srbm_mutex->mm_sem->reservation_ww_class_mutex->srbm_mutex.
*/
release_queue(kgd);
valid_wptr = read_user_wptr(mm, wptr, wptr_val);
acquire_queue(kgd, 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);
return 0;
}
static int get_wave_state(struct mqd_manager *mm, void *mqd,
void __user *ctl_stack,
u32 *ctl_stack_used_size,
u32 *save_area_used_size)
{
struct v9_mqd *m;
/* Control stack is located one page after MQD. */
void *mqd_ctl_stack = (void *)((uintptr_t)mqd + PAGE_SIZE);
m = get_mqd(mqd);
*ctl_stack_used_size = m->cp_hqd_cntl_stack_size -
m->cp_hqd_cntl_stack_offset;
*save_area_used_size = m->cp_hqd_wg_state_offset;
if (copy_to_user(ctl_stack, mqd_ctl_stack, m->cp_hqd_cntl_stack_size))
return -EFAULT;
return 0;
}
static int __update_mqd(struct mqd_manager *mm, void *mqd,
struct queue_properties *q, unsigned int atc_bit)
{
struct cik_mqd *m;
m = get_mqd(mqd);
m->cp_hqd_pq_control = DEFAULT_RPTR_BLOCK_SIZE |
DEFAULT_MIN_AVAIL_SIZE;
m->cp_hqd_ib_control = DEFAULT_MIN_IB_AVAIL_SIZE;
if (atc_bit) {
m->cp_hqd_pq_control |= PQ_ATC_EN;
m->cp_hqd_ib_control |= IB_ATC_EN;
}
/*
* Calculating queue size which is log base 2 of actual queue size -1
* dwords and another -1 for ffs
*/
m->cp_hqd_pq_control |= order_base_2(q->queue_size / 4) - 1;
m->cp_hqd_pq_base_lo = lower_32_bits((uint64_t)q->queue_address >> 8);
m->cp_hqd_pq_base_hi = upper_32_bits((uint64_t)q->queue_address >> 8);
m->cp_hqd_pq_rptr_report_addr_lo = lower_32_bits((uint64_t)q->read_ptr);
m->cp_hqd_pq_rptr_report_addr_hi = upper_32_bits((uint64_t)q->read_ptr);
m->cp_hqd_pq_doorbell_control = DOORBELL_OFFSET(q->doorbell_off);
m->cp_hqd_vmid = q->vmid;
if (q->format == KFD_QUEUE_FORMAT_AQL)
m->cp_hqd_pq_control |= NO_UPDATE_RPTR;
update_cu_mask(mm, mqd, q);
q->is_active = (q->queue_size > 0 &&
q->queue_address != 0 &&
q->queue_percent > 0 &&
!q->is_evicted);
return 0;
}
static void update_cu_mask(struct mqd_manager *mm, void *mqd,
struct queue_properties *q)
{
struct cik_mqd *m;
uint32_t se_mask[4] = {0}; /* 4 is the max # of SEs */
if (q->cu_mask_count == 0)
return;
mqd_symmetrically_map_cu_mask(mm,
q->cu_mask, q->cu_mask_count, se_mask);
m = get_mqd(mqd);
m->compute_static_thread_mgmt_se0 = se_mask[0];
m->compute_static_thread_mgmt_se1 = se_mask[1];
m->compute_static_thread_mgmt_se2 = se_mask[2];
m->compute_static_thread_mgmt_se3 = se_mask[3];
pr_debug("Update cu mask to %#x %#x %#x %#x\n",
m->compute_static_thread_mgmt_se0,
m->compute_static_thread_mgmt_se1,
m->compute_static_thread_mgmt_se2,
m->compute_static_thread_mgmt_se3);
}
static int update_mqd(struct mqd_manager *mm, void *mqd,
struct queue_properties *q)
{
struct v9_mqd *m;
m = get_mqd(mqd);
m->cp_hqd_pq_control = 5 << CP_HQD_PQ_CONTROL__RPTR_BLOCK_SIZE__SHIFT;
m->cp_hqd_pq_control |= order_base_2(q->queue_size / 4) - 1;
pr_debug("cp_hqd_pq_control 0x%x\n", m->cp_hqd_pq_control);
m->cp_hqd_pq_base_lo = lower_32_bits((uint64_t)q->queue_address >> 8);
m->cp_hqd_pq_base_hi = upper_32_bits((uint64_t)q->queue_address >> 8);
m->cp_hqd_pq_rptr_report_addr_lo = lower_32_bits((uint64_t)q->read_ptr);
m->cp_hqd_pq_rptr_report_addr_hi = upper_32_bits((uint64_t)q->read_ptr);
m->cp_hqd_pq_wptr_poll_addr_lo = lower_32_bits((uint64_t)q->write_ptr);
m->cp_hqd_pq_wptr_poll_addr_hi = upper_32_bits((uint64_t)q->write_ptr);
m->cp_hqd_pq_doorbell_control =
q->doorbell_off <<
CP_HQD_PQ_DOORBELL_CONTROL__DOORBELL_OFFSET__SHIFT;
pr_debug("cp_hqd_pq_doorbell_control 0x%x\n",
m->cp_hqd_pq_doorbell_control);
m->cp_hqd_ib_control =
3 << CP_HQD_IB_CONTROL__MIN_IB_AVAIL_SIZE__SHIFT |
1 << CP_HQD_IB_CONTROL__IB_EXE_DISABLE__SHIFT;
/*
* HW does not clamp this field correctly. Maximum EOP queue size
* is constrained by per-SE EOP done signal count, which is 8-bit.
* Limit is 0xFF EOP entries (= 0x7F8 dwords). CP will not submit
* more than (EOP entry count - 1) so a queue size of 0x800 dwords
* is safe, giving a maximum field value of 0xA.
*/
m->cp_hqd_eop_control = min(0xA,
order_base_2(q->eop_ring_buffer_size / 4) - 1);
m->cp_hqd_eop_base_addr_lo =
lower_32_bits(q->eop_ring_buffer_address >> 8);
m->cp_hqd_eop_base_addr_hi =
upper_32_bits(q->eop_ring_buffer_address >> 8);
m->cp_hqd_iq_timer = 0;
m->cp_hqd_vmid = q->vmid;
if (q->format == KFD_QUEUE_FORMAT_AQL) {
m->cp_hqd_pq_control |= CP_HQD_PQ_CONTROL__NO_UPDATE_RPTR_MASK |
2 << CP_HQD_PQ_CONTROL__SLOT_BASED_WPTR__SHIFT |
1 << CP_HQD_PQ_CONTROL__QUEUE_FULL_EN__SHIFT |
1 << CP_HQD_PQ_CONTROL__WPP_CLAMP_EN__SHIFT;
m->cp_hqd_pq_doorbell_control |= 1 <<
CP_HQD_PQ_DOORBELL_CONTROL__DOORBELL_BIF_DROP__SHIFT;
}
if (mm->dev->cwsr_enabled && q->ctx_save_restore_area_address)
m->cp_hqd_ctx_save_control = 0;
update_cu_mask(mm, mqd, q);
q->is_active = (q->queue_size > 0 &&
q->queue_address != 0 &&
q->queue_percent > 0 &&
!q->is_evicted);
return 0;
}
static int kgd_hqd_destroy(struct kgd_dev *kgd, 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;
int retry;
struct vi_mqd *m = get_mqd(mqd);
if (adev->in_gpu_reset)
return -EIO;
acquire_queue(kgd, pipe_id, queue_id);
if (m->cp_hqd_vmid == 0)
WREG32_FIELD(RLC_CP_SCHEDULERS, scheduler1, 0);
switch (reset_type) {
case KFD_PREEMPT_TYPE_WAVEFRONT_DRAIN:
type = DRAIN_PIPE;
break;
case KFD_PREEMPT_TYPE_WAVEFRONT_RESET:
type = RESET_WAVES;
break;
default:
type = DRAIN_PIPE;
break;
}
/* Workaround: If IQ timer is active and the wait time is close to or
* equal to 0, dequeueing is not safe. Wait until either the wait time
* is larger or timer is cleared. Also, ensure that IQ_REQ_PEND is
* cleared before continuing. Also, ensure wait times are set to at
* least 0x3.
*/
local_irq_save(flags);
preempt_disable();
retry = 5000; /* wait for 500 usecs at maximum */
while (true) {
temp = RREG32(mmCP_HQD_IQ_TIMER);
if (REG_GET_FIELD(temp, CP_HQD_IQ_TIMER, PROCESSING_IQ)) {
pr_debug("HW is processing IQ\n");
goto loop;
}
if (REG_GET_FIELD(temp, CP_HQD_IQ_TIMER, ACTIVE)) {
if (REG_GET_FIELD(temp, CP_HQD_IQ_TIMER, RETRY_TYPE)
== 3) /* SEM-rearm is safe */
break;
/* Wait time 3 is safe for CP, but our MMIO read/write
* time is close to 1 microsecond, so check for 10 to
* leave more buffer room
*/
if (REG_GET_FIELD(temp, CP_HQD_IQ_TIMER, WAIT_TIME)
>= 10)
break;
pr_debug("IQ timer is active\n");
} else
break;
loop:
if (!retry) {
pr_err("CP HQD IQ timer status time out\n");
break;
}
ndelay(100);
--retry;
}
retry = 1000;
while (true) {
temp = RREG32(mmCP_HQD_DEQUEUE_REQUEST);
if (!(temp & CP_HQD_DEQUEUE_REQUEST__IQ_REQ_PEND_MASK))
break;
pr_debug("Dequeue request is pending\n");
if (!retry) {
pr_err("CP HQD dequeue request time out\n");
break;
}
ndelay(100);
--retry;
}
local_irq_restore(flags);
preempt_enable();
WREG32(mmCP_HQD_DEQUEUE_REQUEST, type);
end_jiffies = (utimeout * HZ / 1000) + jiffies;
while (true) {
temp = RREG32(mmCP_HQD_ACTIVE);
if (!(temp & CP_HQD_ACTIVE__ACTIVE_MASK))
break;
if (time_after(jiffies, end_jiffies)) {
pr_err("cp queue preemption time out.\n");
release_queue(kgd);
return -ETIME;
}
usleep_range(500, 1000);
}
release_queue(kgd);
return 0;
}
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)
{
struct amdgpu_device *adev = get_amdgpu_device(kgd);
struct vi_mqd *m;
uint32_t *mqd_hqd;
uint32_t reg, wptr_val, data;
bool valid_wptr = false;
m = get_mqd(mqd);
acquire_queue(kgd, pipe_id, queue_id);
/* HIQ is set during driver init period with vmid set to 0*/
if (m->cp_hqd_vmid == 0) {
uint32_t value, mec, pipe;
mec = (pipe_id / adev->gfx.mec.num_pipe_per_mec) + 1;
pipe = (pipe_id % adev->gfx.mec.num_pipe_per_mec);
pr_debug("kfd: set HIQ, mec:%d, pipe:%d, queue:%d.\n",
mec, pipe, queue_id);
value = RREG32(mmRLC_CP_SCHEDULERS);
value = REG_SET_FIELD(value, RLC_CP_SCHEDULERS, scheduler1,
((mec << 5) | (pipe << 3) | queue_id | 0x80));
WREG32(mmRLC_CP_SCHEDULERS, value);
}
/* HQD registers extend from CP_MQD_BASE_ADDR to CP_HQD_EOP_WPTR_MEM. */
mqd_hqd = &m->cp_mqd_base_addr_lo;
for (reg = mmCP_MQD_BASE_ADDR; reg <= mmCP_HQD_EOP_CONTROL; reg++)
WREG32(reg, mqd_hqd[reg - mmCP_MQD_BASE_ADDR]);
/* Tonga errata: EOP RPTR/WPTR should be left unmodified.
* This is safe since EOP RPTR==WPTR for any inactive HQD
* 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) {
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);
}
for (reg = mmCP_HQD_EOP_EVENTS; reg <= mmCP_HQD_ERROR; reg++)
WREG32(reg, mqd_hqd[reg - mmCP_MQD_BASE_ADDR]);
/* Copy userspace write pointer value to register.
* Activate doorbell logic to monitor subsequent changes.
*/
data = REG_SET_FIELD(m->cp_hqd_pq_doorbell_control,
CP_HQD_PQ_DOORBELL_CONTROL, DOORBELL_EN, 1);
WREG32(mmCP_HQD_PQ_DOORBELL_CONTROL, data);
/* read_user_ptr may take the mm->mmap_sem.
* release srbm_mutex to avoid circular dependency between
* srbm_mutex->mm_sem->reservation_ww_class_mutex->srbm_mutex.
*/
release_queue(kgd);
valid_wptr = read_user_wptr(mm, wptr, wptr_val);
acquire_queue(kgd, 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);
return 0;
}
static int kgd_hqd_load(struct kgd_dev *kgd, void *mqd, uint32_t pipe_id,
uint32_t queue_id, uint32_t __user *wptr)
{
struct vi_mqd *m;
uint32_t shadow_wptr, valid_wptr;
struct amdgpu_device *adev = get_amdgpu_device(kgd);
m = get_mqd(mqd);
valid_wptr = copy_from_user(&shadow_wptr, wptr, sizeof(shadow_wptr));
acquire_queue(kgd, pipe_id, queue_id);
WREG32(mmCP_MQD_CONTROL, m->cp_mqd_control);
WREG32(mmCP_MQD_BASE_ADDR, m->cp_mqd_base_addr_lo);
WREG32(mmCP_MQD_BASE_ADDR_HI, m->cp_mqd_base_addr_hi);
WREG32(mmCP_HQD_VMID, m->cp_hqd_vmid);
WREG32(mmCP_HQD_PERSISTENT_STATE, m->cp_hqd_persistent_state);
WREG32(mmCP_HQD_PIPE_PRIORITY, m->cp_hqd_pipe_priority);
WREG32(mmCP_HQD_QUEUE_PRIORITY, m->cp_hqd_queue_priority);
WREG32(mmCP_HQD_QUANTUM, m->cp_hqd_quantum);
WREG32(mmCP_HQD_PQ_BASE, m->cp_hqd_pq_base_lo);
WREG32(mmCP_HQD_PQ_BASE_HI, m->cp_hqd_pq_base_hi);
WREG32(mmCP_HQD_PQ_RPTR_REPORT_ADDR, m->cp_hqd_pq_rptr_report_addr_lo);
WREG32(mmCP_HQD_PQ_RPTR_REPORT_ADDR_HI,
m->cp_hqd_pq_rptr_report_addr_hi);
if (valid_wptr > 0)
WREG32(mmCP_HQD_PQ_WPTR, shadow_wptr);
WREG32(mmCP_HQD_PQ_CONTROL, m->cp_hqd_pq_control);
WREG32(mmCP_HQD_PQ_DOORBELL_CONTROL, m->cp_hqd_pq_doorbell_control);
WREG32(mmCP_HQD_EOP_BASE_ADDR, m->cp_hqd_eop_base_addr_lo);
WREG32(mmCP_HQD_EOP_BASE_ADDR_HI, m->cp_hqd_eop_base_addr_hi);
WREG32(mmCP_HQD_EOP_CONTROL, m->cp_hqd_eop_control);
WREG32(mmCP_HQD_EOP_RPTR, m->cp_hqd_eop_rptr);
WREG32(mmCP_HQD_EOP_WPTR, m->cp_hqd_eop_wptr);
WREG32(mmCP_HQD_EOP_EVENTS, m->cp_hqd_eop_done_events);
WREG32(mmCP_HQD_CTX_SAVE_BASE_ADDR_LO, m->cp_hqd_ctx_save_base_addr_lo);
WREG32(mmCP_HQD_CTX_SAVE_BASE_ADDR_HI, m->cp_hqd_ctx_save_base_addr_hi);
WREG32(mmCP_HQD_CTX_SAVE_CONTROL, m->cp_hqd_ctx_save_control);
WREG32(mmCP_HQD_CNTL_STACK_OFFSET, m->cp_hqd_cntl_stack_offset);
WREG32(mmCP_HQD_CNTL_STACK_SIZE, m->cp_hqd_cntl_stack_size);
WREG32(mmCP_HQD_WG_STATE_OFFSET, m->cp_hqd_wg_state_offset);
WREG32(mmCP_HQD_CTX_SAVE_SIZE, m->cp_hqd_ctx_save_size);
WREG32(mmCP_HQD_IB_CONTROL, m->cp_hqd_ib_control);
WREG32(mmCP_HQD_DEQUEUE_REQUEST, m->cp_hqd_dequeue_request);
WREG32(mmCP_HQD_ERROR, m->cp_hqd_error);
WREG32(mmCP_HQD_EOP_WPTR_MEM, m->cp_hqd_eop_wptr_mem);
WREG32(mmCP_HQD_EOP_DONES, m->cp_hqd_eop_dones);
WREG32(mmCP_HQD_ACTIVE, m->cp_hqd_active);
release_queue(kgd);
return 0;
}