On 2018-05-11 03:06 PM, Oded Gabbay wrote:
> On Fri, May 11, 2018 at 9:15 PM, Felix Kuehling <felix.kuehling at amd.com> wrote:
>> This patch series was meant to be applied after the userptr changes. I
>> haven't tested this without the userptr changes.
>>
>> I think your main concern about userptr is the use of GFP_NOIO. I
>> remember considering memalloc_noio_save/restore when I worked on this
>> over a year ago. I found an old email thread I had with Christian about
>> this (subject: MMU notifier deadlock on kernel 4.9):
>>
>>> memalloc_noio_save doesn't affect kmalloc directly. It sets
>>> current->flags, which is used deep inside the page allocator, after
>>> lockdep_trace_alloc(gfp) flags a lock as being used with IO enabled. The
>>> slob allocator looks at gfp_allowed_mask, but I'm not sure how safe it
>>> is to change that in a driver and it doesn't seem to be an exported
>>> symbol anyway.
>> Maybe this has changed in the mean time, but a year ago, using
>> memalloc_noio_save/restore may have prevented real deadlocks, but would
>> not shut up the lockdep warnings. Maybe this has changed in the mean
>> time. FWIW, I don't find lockdep_trace_alloc in the current kernel.
>>
>> Regards,
>> Felix
> I'm not familiar with this API, but from reading about it, it seems a
> more robust solution then to change the GFP flags directly in each
> kmalloc from the reasons I mentioned in the original email I sent.
I agree.
> Having said that, if no one else objects and we say we will look at
> moving to that API in the future, I don't object to taking your
> patch-set as is now.
Yeah. I'll look into this.
Regards,
 Felix
>
> Oded
>
>>
>> On 2018-05-11 05:10 AM, Oded Gabbay wrote:
>>> On Wed, Apr 11, 2018 at 12:33 AM, Felix Kuehling <Felix.Kuehling at amd.com> wrote:
>>>> Signed-off-by: John Bridgman <john.bridgman at amd.com>
>>>> Signed-off-by: Jay Cornwall <Jay.Cornwall at amd.com>
>>>> Signed-off-by: Felix Kuehling <Felix.Kuehling at amd.com>
>>>> ---
>>>> drivers/gpu/drm/amd/amdkfd/Makefile | 1 +
>>>> drivers/gpu/drm/amd/amdkfd/kfd_device.c | 2 +-
>>>> drivers/gpu/drm/amd/amdkfd/kfd_mqd_manager.c | 3 +
>>>> drivers/gpu/drm/amd/amdkfd/kfd_mqd_manager_v9.c | 443 ++++++++++++++++++++++++
>>>> drivers/gpu/drm/amd/amdkfd/kfd_priv.h | 3 +
>>>> 5 files changed, 451 insertions(+), 1 deletion(-)
>>>> create mode 100644 drivers/gpu/drm/amd/amdkfd/kfd_mqd_manager_v9.c
>>>>
>>>> diff --git a/drivers/gpu/drm/amd/amdkfd/Makefile b/drivers/gpu/drm/amd/amdkfd/Makefile
>>>> index 52b3c1b..094b591 100644
>>>> --- a/drivers/gpu/drm/amd/amdkfd/Makefile
>>>> +++ b/drivers/gpu/drm/amd/amdkfd/Makefile
>>>> @@ -30,6 +30,7 @@ amdkfd-y := kfd_module.o kfd_device.o kfd_chardev.o kfd_topology.o \
>>>> kfd_pasid.o kfd_doorbell.o kfd_flat_memory.o \
>>>> kfd_process.o kfd_queue.o kfd_mqd_manager.o \
>>>> kfd_mqd_manager_cik.o kfd_mqd_manager_vi.o \
>>>> + kfd_mqd_manager_v9.o \
>>>> kfd_kernel_queue.o kfd_kernel_queue_cik.o \
>>>> kfd_kernel_queue_vi.o kfd_kernel_queue_v9.o \
>>>> kfd_packet_manager.o kfd_process_queue_manager.o \
>>>> diff --git a/drivers/gpu/drm/amd/amdkfd/kfd_device.c b/drivers/gpu/drm/amd/amdkfd/kfd_device.c
>>>> index f563acb..c368ce3 100644
>>>> --- a/drivers/gpu/drm/amd/amdkfd/kfd_device.c
>>>> +++ b/drivers/gpu/drm/amd/amdkfd/kfd_device.c
>>>> @@ -700,7 +700,7 @@ int kfd_gtt_sa_allocate(struct kfd_dev *kfd, unsigned int size,
>>>> if (size > kfd->gtt_sa_num_of_chunks * kfd->gtt_sa_chunk_size)
>>>> return -ENOMEM;
>>>>
>>>> - *mem_obj = kmalloc(sizeof(struct kfd_mem_obj), GFP_NOIO);
>>>> + *mem_obj = kzalloc(sizeof(struct kfd_mem_obj), GFP_NOIO);
>>> This assumes the patch in the userptr patch-set is applied. I changed
>>> it to GFP_KERNEL for now.
>>>
>>>> if ((*mem_obj) == NULL)
>>>> return -ENOMEM;
>>>>
>>>> diff --git a/drivers/gpu/drm/amd/amdkfd/kfd_mqd_manager.c b/drivers/gpu/drm/amd/amdkfd/kfd_mqd_manager.c
>>>> index ee7061e..4b8eb50 100644
>>>> --- a/drivers/gpu/drm/amd/amdkfd/kfd_mqd_manager.c
>>>> +++ b/drivers/gpu/drm/amd/amdkfd/kfd_mqd_manager.c
>>>> @@ -38,6 +38,9 @@ struct mqd_manager *mqd_manager_init(enum KFD_MQD_TYPE type,
>>>> case CHIP_POLARIS10:
>>>> case CHIP_POLARIS11:
>>>> return mqd_manager_init_vi_tonga(type, dev);
>>>> + case CHIP_VEGA10:
>>>> + case CHIP_RAVEN:
>>>> + return mqd_manager_init_v9(type, dev);
>>>> default:
>>>> WARN(1, "Unexpected ASIC family %u",
>>>> dev->device_info->asic_family);
>>>> diff --git a/drivers/gpu/drm/amd/amdkfd/kfd_mqd_manager_v9.c b/drivers/gpu/drm/amd/amdkfd/kfd_mqd_manager_v9.c
>>>> new file mode 100644
>>>> index 0000000..684054f
>>>> --- /dev/null
>>>> +++ b/drivers/gpu/drm/amd/amdkfd/kfd_mqd_manager_v9.c
>>>> @@ -0,0 +1,443 @@
>>>> +/*
>>>> + * Copyright 2016-2018 Advanced Micro Devices, Inc.
>>>> + *
>>>> + * Permission is hereby granted, free of charge, to any person obtaining a
>>>> + * copy of this software and associated documentation files (the "Software"),
>>>> + * to deal in the Software without restriction, including without limitation
>>>> + * the rights to use, copy, modify, merge, publish, distribute, sublicense,
>>>> + * and/or sell copies of the Software, and to permit persons to whom the
>>>> + * Software is furnished to do so, subject to the following conditions:
>>>> + *
>>>> + * The above copyright notice and this permission notice shall be included in
>>>> + * all copies or substantial portions of the Software.
>>>> + *
>>>> + * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
>>>> + * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
>>>> + * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL
>>>> + * THE COPYRIGHT HOLDER(S) OR AUTHOR(S) BE LIABLE FOR ANY CLAIM, DAMAGES OR
>>>> + * OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE,
>>>> + * ARISING FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR
>>>> + * OTHER DEALINGS IN THE SOFTWARE.
>>>> + *
>>>> + */
>>>> +
>>>> +#include <linux/printk.h>
>>>> +#include <linux/slab.h>
>>>> +#include <linux/uaccess.h>
>>>> +#include "kfd_priv.h"
>>>> +#include "kfd_mqd_manager.h"
>>>> +#include "v9_structs.h"
>>>> +#include "gc/gc_9_0_offset.h"
>>>> +#include "gc/gc_9_0_sh_mask.h"
>>>> +#include "sdma0/sdma0_4_0_sh_mask.h"
>>>> +
>>>> +static inline struct v9_mqd *get_mqd(void *mqd)
>>>> +{
>>>> + return (struct v9_mqd *)mqd;
>>>> +}
>>>> +
>>>> +static inline struct v9_sdma_mqd *get_sdma_mqd(void *mqd)
>>>> +{
>>>> + return (struct v9_sdma_mqd *)mqd;
>>>> +}
>>>> +
>>>> +static int init_mqd(struct mqd_manager *mm, void **mqd,
>>>> + struct kfd_mem_obj **mqd_mem_obj, uint64_t *gart_addr,
>>>> + struct queue_properties *q)
>>>> +{
>>>> + int retval;
>>>> + uint64_t addr;
>>>> + struct v9_mqd *m;
>>>> + struct kfd_dev *kfd = mm->dev;
>>>> +
>>>> + /* From V9, for CWSR, the control stack is located on the next page
>>>> + * boundary after the mqd, we will use the gtt allocation function
>>>> + * instead of sub-allocation function.
>>>> + */
>>>> + if (kfd->cwsr_enabled && (q->type == KFD_QUEUE_TYPE_COMPUTE)) {
>>>> + *mqd_mem_obj = kzalloc(sizeof(struct kfd_mem_obj), GFP_NOIO);
>>> Using GFP_NOIO directly is not recommended. Can we use the scope
>>> functions instead ?
>>>
>>>> + if (!*mqd_mem_obj)
>>>> + return -ENOMEM;
>>>> + retval = kfd->kfd2kgd->init_gtt_mem_allocation(kfd->kgd,
>>>> + ALIGN(q->ctl_stack_size, PAGE_SIZE) +
>>>> + ALIGN(sizeof(struct v9_mqd), PAGE_SIZE),
>>>> + &((*mqd_mem_obj)->gtt_mem),
>>>> + &((*mqd_mem_obj)->gpu_addr),
>>>> + (void *)&((*mqd_mem_obj)->cpu_ptr));
>>>> + } else
>>>> + retval = kfd_gtt_sa_allocate(mm->dev, sizeof(struct v9_mqd),
>>>> + mqd_mem_obj);
>>>> + if (retval != 0)
>>>> + return -ENOMEM;
>>>> +
>>>> + m = (struct v9_mqd *) (*mqd_mem_obj)->cpu_ptr;
>>>> + addr = (*mqd_mem_obj)->gpu_addr;
>>>> +
>>>> + memset(m, 0, sizeof(struct v9_mqd));
>>>> +
>>>> + m->header = 0xC0310800;
>>>> + m->compute_pipelinestat_enable = 1;
>>>> + m->compute_static_thread_mgmt_se0 = 0xFFFFFFFF;
>>>> + m->compute_static_thread_mgmt_se1 = 0xFFFFFFFF;
>>>> + m->compute_static_thread_mgmt_se2 = 0xFFFFFFFF;
>>>> + m->compute_static_thread_mgmt_se3 = 0xFFFFFFFF;
>>>> +
>>>> + m->cp_hqd_persistent_state = CP_HQD_PERSISTENT_STATE__PRELOAD_REQ_MASK |
>>>> + 0x53 << CP_HQD_PERSISTENT_STATE__PRELOAD_SIZE__SHIFT;
>>>> +
>>>> + m->cp_mqd_control = 1 << CP_MQD_CONTROL__PRIV_STATE__SHIFT;
>>>> +
>>>> + m->cp_mqd_base_addr_lo = lower_32_bits(addr);
>>>> + m->cp_mqd_base_addr_hi = upper_32_bits(addr);
>>>> +
>>>> + m->cp_hqd_quantum = 1 << CP_HQD_QUANTUM__QUANTUM_EN__SHIFT |
>>>> + 1 << CP_HQD_QUANTUM__QUANTUM_SCALE__SHIFT |
>>>> + 10 << CP_HQD_QUANTUM__QUANTUM_DURATION__SHIFT;
>>>> +
>>>> + m->cp_hqd_pipe_priority = 1;
>>>> + m->cp_hqd_queue_priority = 15;
>>>> +
>>>> + if (q->format == KFD_QUEUE_FORMAT_AQL) {
>>>> + m->cp_hqd_aql_control =
>>>> + 1 << CP_HQD_AQL_CONTROL__CONTROL0__SHIFT;
>>>> + }
>>>> +
>>>> + if (q->tba_addr) {
>>>> + m->compute_pgm_rsrc2 |=
>>>> + (1 << COMPUTE_PGM_RSRC2__TRAP_PRESENT__SHIFT);
>>>> + }
>>>> +
>>>> + if (mm->dev->cwsr_enabled && q->ctx_save_restore_area_address) {
>>>> + m->cp_hqd_persistent_state |=
>>>> + (1 << CP_HQD_PERSISTENT_STATE__QSWITCH_MODE__SHIFT);
>>>> + m->cp_hqd_ctx_save_base_addr_lo =
>>>> + lower_32_bits(q->ctx_save_restore_area_address);
>>>> + m->cp_hqd_ctx_save_base_addr_hi =
>>>> + upper_32_bits(q->ctx_save_restore_area_address);
>>>> + m->cp_hqd_ctx_save_size = q->ctx_save_restore_area_size;
>>>> + m->cp_hqd_cntl_stack_size = q->ctl_stack_size;
>>>> + m->cp_hqd_cntl_stack_offset = q->ctl_stack_size;
>>>> + m->cp_hqd_wg_state_offset = q->ctl_stack_size;
>>>> + }
>>>> +
>>>> + *mqd = m;
>>>> + if (gart_addr)
>>>> + *gart_addr = addr;
>>>> + retval = mm->update_mqd(mm, m, q);
>>>> +
>>>> + return retval;
>>>> +}
>>>> +
>>>> +static int load_mqd(struct mqd_manager *mm, void *mqd,
>>>> + uint32_t pipe_id, uint32_t queue_id,
>>>> + struct queue_properties *p, struct mm_struct *mms)
>>>> +{
>>>> + /* AQL write pointer counts in 64B packets, PM4/CP counts in dwords. */
>>>> + uint32_t wptr_shift = (p->format == KFD_QUEUE_FORMAT_AQL ? 4 : 0);
>>>> +
>>>> + return mm->dev->kfd2kgd->hqd_load(mm->dev->kgd, mqd, pipe_id, queue_id,
>>>> + (uint32_t __user *)p->write_ptr,
>>>> + wptr_shift, 0, mms);
>>>> +}
>>>> +
>>>> +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;
>>>> +
>>>> + q->is_active = (q->queue_size > 0 &&
>>>> + q->queue_address != 0 &&
>>>> + q->queue_percent > 0 &&
>>>> + !q->is_evicted);
>>>> +
>>>> + return 0;
>>>> +}
>>>> +
>>>> +
>>>> +static int destroy_mqd(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->kgd, mqd, type, timeout,
>>>> + pipe_id, queue_id);
>>>> +}
>>>> +
>>>> +static void uninit_mqd(struct mqd_manager *mm, void *mqd,
>>>> + struct kfd_mem_obj *mqd_mem_obj)
>>>> +{
>>>> + struct kfd_dev *kfd = mm->dev;
>>>> +
>>>> + if (mqd_mem_obj->gtt_mem) {
>>>> + kfd->kfd2kgd->free_gtt_mem(kfd->kgd, mqd_mem_obj->gtt_mem);
>>>> + kfree(mqd_mem_obj);
>>>> + } else {
>>>> + kfd_gtt_sa_free(mm->dev, mqd_mem_obj);
>>>> + }
>>>> +}
>>>> +
>>>> +static bool is_occupied(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->kgd, queue_address,
>>>> + pipe_id, queue_id);
>>>> +}
>>>> +
>>>> +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_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_sdma(struct mqd_manager *mm, void **mqd,
>>>> + struct kfd_mem_obj **mqd_mem_obj, uint64_t *gart_addr,
>>>> + struct queue_properties *q)
>>>> +{
>>>> + int retval;
>>>> + struct v9_sdma_mqd *m;
>>>> +
>>>> +
>>>> + retval = kfd_gtt_sa_allocate(mm->dev,
>>>> + sizeof(struct v9_sdma_mqd),
>>>> + mqd_mem_obj);
>>>> +
>>>> + if (retval != 0)
>>>> + return -ENOMEM;
>>>> +
>>>> + m = (struct v9_sdma_mqd *) (*mqd_mem_obj)->cpu_ptr;
>>>> +
>>>> + memset(m, 0, sizeof(struct v9_sdma_mqd));
>>>> +
>>>> + *mqd = m;
>>>> + if (gart_addr)
>>>> + *gart_addr = (*mqd_mem_obj)->gpu_addr;
>>>> +
>>>> + retval = mm->update_mqd(mm, m, q);
>>>> +
>>>> + return retval;
>>>> +}
>>>> +
>>>> +static void uninit_mqd_sdma(struct mqd_manager *mm, void *mqd,
>>>> + struct kfd_mem_obj *mqd_mem_obj)
>>>> +{
>>>> + kfd_gtt_sa_free(mm->dev, mqd_mem_obj);
>>>> +}
>>>> +
>>>> +static int 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->kgd, mqd,
>>>> + (uint32_t __user *)p->write_ptr,
>>>> + mms);
>>>> +}
>>>> +
>>>> +#define SDMA_RLC_DUMMY_DEFAULT 0xf
>>>> +
>>>> +static int update_mqd_sdma(struct mqd_manager *mm, void *mqd,
>>>> + struct queue_properties *q)
>>>> +{
>>>> + struct v9_sdma_mqd *m;
>>>> +
>>>> + m = get_sdma_mqd(mqd);
>>>> + m->sdmax_rlcx_rb_cntl = order_base_2(q->queue_size / 4)
>>>> + << SDMA0_RLC0_RB_CNTL__RB_SIZE__SHIFT |
>>>> + q->vmid << SDMA0_RLC0_RB_CNTL__RB_VMID__SHIFT |
>>>> + 1 << SDMA0_RLC0_RB_CNTL__RPTR_WRITEBACK_ENABLE__SHIFT |
>>>> + 6 << SDMA0_RLC0_RB_CNTL__RPTR_WRITEBACK_TIMER__SHIFT;
>>>> +
>>>> + m->sdmax_rlcx_rb_base = lower_32_bits(q->queue_address >> 8);
>>>> + m->sdmax_rlcx_rb_base_hi = upper_32_bits(q->queue_address >> 8);
>>>> + m->sdmax_rlcx_rb_rptr_addr_lo = lower_32_bits((uint64_t)q->read_ptr);
>>>> + m->sdmax_rlcx_rb_rptr_addr_hi = upper_32_bits((uint64_t)q->read_ptr);
>>>> + m->sdmax_rlcx_doorbell_offset =
>>>> + q->doorbell_off << SDMA0_RLC0_DOORBELL_OFFSET__OFFSET__SHIFT;
>>>> +
>>>> + m->sdma_engine_id = q->sdma_engine_id;
>>>> + m->sdma_queue_id = q->sdma_queue_id;
>>>> + m->sdmax_rlcx_dummy_reg = SDMA_RLC_DUMMY_DEFAULT;
>>>> +
>>>> + q->is_active = (q->queue_size > 0 &&
>>>> + q->queue_address != 0 &&
>>>> + q->queue_percent > 0 &&
>>>> + !q->is_evicted);
>>>> +
>>>> + return 0;
>>>> +}
>>>> +
>>>> +/*
>>>> + * * preempt type here is ignored because there is only one way
>>>> + * * to preempt sdma queue
>>>> + */
>>>> +static int 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->kgd, mqd, timeout);
>>>> +}
>>>> +
>>>> +static bool 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->kgd, mqd);
>>>> +}
>>>> +
>>>> +#if defined(CONFIG_DEBUG_FS)
>>>> +
>>>> +static int debugfs_show_mqd(struct seq_file *m, void *data)
>>>> +{
>>>> + seq_hex_dump(m, " ", DUMP_PREFIX_OFFSET, 32, 4,
>>>> + data, sizeof(struct v9_mqd), false);
>>>> + return 0;
>>>> +}
>>>> +
>>>> +static int debugfs_show_mqd_sdma(struct seq_file *m, void *data)
>>>> +{
>>>> + seq_hex_dump(m, " ", DUMP_PREFIX_OFFSET, 32, 4,
>>>> + data, sizeof(struct v9_sdma_mqd), false);
>>>> + return 0;
>>>> +}
>>>> +
>>>> +#endif
>>>> +
>>>> +struct mqd_manager *mqd_manager_init_v9(enum KFD_MQD_TYPE type,
>>>> + struct kfd_dev *dev)
>>>> +{
>>>> + struct mqd_manager *mqd;
>>>> +
>>>> + if (WARN_ON(type >= KFD_MQD_TYPE_MAX))
>>>> + return NULL;
>>>> +
>>>> + mqd = kzalloc(sizeof(*mqd), GFP_NOIO);
>>> Using GFP_NOIO directly is not recommended. Can we use the scope
>>> functions instead ?
>>>
>>>> + if (!mqd)
>>>> + return NULL;
>>>> +
>>>> + mqd->dev = dev;
>>>> +
>>>> + switch (type) {
>>>> + case KFD_MQD_TYPE_CP:
>>>> + case KFD_MQD_TYPE_COMPUTE:
>>>> + mqd->init_mqd = init_mqd;
>>>> + mqd->uninit_mqd = uninit_mqd;
>>>> + mqd->load_mqd = load_mqd;
>>>> + mqd->update_mqd = update_mqd;
>>>> + mqd->destroy_mqd = destroy_mqd;
>>>> + mqd->is_occupied = is_occupied;
>>>> +#if defined(CONFIG_DEBUG_FS)
>>>> + mqd->debugfs_show_mqd = debugfs_show_mqd;
>>>> +#endif
>>>> + break;
>>>> + case KFD_MQD_TYPE_HIQ:
>>>> + mqd->init_mqd = init_mqd_hiq;
>>>> + mqd->uninit_mqd = uninit_mqd;
>>>> + mqd->load_mqd = load_mqd;
>>>> + mqd->update_mqd = update_mqd_hiq;
>>>> + mqd->destroy_mqd = destroy_mqd;
>>>> + mqd->is_occupied = is_occupied;
>>>> +#if defined(CONFIG_DEBUG_FS)
>>>> + mqd->debugfs_show_mqd = debugfs_show_mqd;
>>>> +#endif
>>>> + break;
>>>> + case KFD_MQD_TYPE_SDMA:
>>>> + mqd->init_mqd = init_mqd_sdma;
>>>> + mqd->uninit_mqd = uninit_mqd_sdma;
>>>> + mqd->load_mqd = load_mqd_sdma;
>>>> + mqd->update_mqd = update_mqd_sdma;
>>>> + mqd->destroy_mqd = destroy_mqd_sdma;
>>>> + mqd->is_occupied = is_occupied_sdma;
>>>> +#if defined(CONFIG_DEBUG_FS)
>>>> + mqd->debugfs_show_mqd = debugfs_show_mqd_sdma;
>>>> +#endif
>>>> + break;
>>>> + default:
>>>> + kfree(mqd);
>>>> + return NULL;
>>>> + }
>>>> +
>>>> + return mqd;
>>>> +}
>>>> diff --git a/drivers/gpu/drm/amd/amdkfd/kfd_priv.h b/drivers/gpu/drm/amd/amdkfd/kfd_priv.h
>>>> index b68299a..fac2882 100644
>>>> --- a/drivers/gpu/drm/amd/amdkfd/kfd_priv.h
>>>> +++ b/drivers/gpu/drm/amd/amdkfd/kfd_priv.h
>>>> @@ -197,6 +197,7 @@ struct kfd_mem_obj {
>>>> uint32_t range_end;
>>>> uint64_t gpu_addr;
>>>> uint32_t *cpu_ptr;
>>>> + void *gtt_mem;
>>>> };
>>>>
>>>> struct kfd_vmid_info {
>>>> @@ -822,6 +823,8 @@ struct mqd_manager *mqd_manager_init_vi(enum KFD_MQD_TYPE type,
>>>> struct kfd_dev *dev);
>>>> struct mqd_manager *mqd_manager_init_vi_tonga(enum KFD_MQD_TYPE type,
>>>> struct kfd_dev *dev);
>>>> +struct mqd_manager *mqd_manager_init_v9(enum KFD_MQD_TYPE type,
>>>> + struct kfd_dev *dev);
>>>> struct device_queue_manager *device_queue_manager_init(struct kfd_dev *dev);
>>>> void device_queue_manager_uninit(struct device_queue_manager *dqm);
>>>> struct kernel_queue *kernel_queue_init(struct kfd_dev *dev,
>>>> --
>>>> 2.7.4
>>>>
