diff options
author | Linus Torvalds <torvalds@linux-foundation.org> | 2015-02-16 15:48:00 -0800 |
---|---|---|
committer | Linus Torvalds <torvalds@linux-foundation.org> | 2015-02-16 15:48:00 -0800 |
commit | 796e1c55717e9a6ff5c81b12289ffa1ffd919b6f (patch) | |
tree | 27ce45cb1227156b72c641dbcbf2b399d23ba63d /drivers/gpu/drm/amd/amdkfd/kfd_device_queue_manager.c | |
parent | 8c334ce8f0fec7122fc3059c52a697b669a01b41 (diff) | |
parent | 45ee2dbc65cbf6910892c480e6f428be342fa733 (diff) | |
download | op-kernel-dev-796e1c55717e9a6ff5c81b12289ffa1ffd919b6f.zip op-kernel-dev-796e1c55717e9a6ff5c81b12289ffa1ffd919b6f.tar.gz |
Merge branch 'drm-next' of git://people.freedesktop.org/~airlied/linux
Pull drm updates from Dave Airlie:
"This is the main drm pull, it has a shared branch with some alsa
crossover but everything should be acked by relevant people.
New drivers:
- ATMEL HLCDC driver
- designware HDMI core support (used in multiple SoCs).
core:
- lots more atomic modesetting work, properties and atomic ioctl
(hidden under option)
- bridge rework allows support for Samsung exynos chromebooks to
work finally.
- some more panels supported
i915:
- atomic plane update support
- DSI uses shared DSI infrastructure
- Skylake basic support is all merged now
- component framework used for i915/snd-hda interactions
- write-combine cpu memory mappings
- engine init code refactored
- full ppgtt enabled where execlists are enabled.
- cherryview rps/gpu turbo and pipe CRC support.
radeon:
- indirect draw support for evergreen/cayman
- SMC and manual fan control for SI/CI
- Displayport audio support
amdkfd:
- SDMA usermode queue support
- replace suballocator usage with more suitable one
- rework for allowing interfacing to more than radeon
nouveau:
- major renaming in prep for later splitting work
- merge arm platform driver into nouveau
- GK20A reclocking support
msm:
- conversion to atomic modesetting
- YUV support for mdp4/5
- eDP support
- hw cursor for mdp5
tegra:
- conversion to atomic modesetting
- better suspend/resume support for child devices
rcar-du:
- interlaced support
imx:
- move to using dw_hdmi shared support
- mode_fixup support
sti:
- DVO support
- HDMI infoframe support
exynos:
- refactoring and cleanup, removed lots of internal unnecessary
abstraction
- exynos7 DECON display controller support
Along with the usual bunch of fixes, cleanups etc"
* 'drm-next' of git://people.freedesktop.org/~airlied/linux: (724 commits)
drm/radeon: fix voltage setup on hawaii
drm/radeon/dp: Set EDP_CONFIGURATION_SET for bridge chips if necessary
drm/radeon: only enable kv/kb dpm interrupts once v3
drm/radeon: workaround for CP HW bug on CIK
drm/radeon: Don't try to enable write-combining without PAT
drm/radeon: use 0-255 rather than 0-100 for pwm fan range
drm/i915: Clamp efficient frequency to valid range
drm/i915: Really ignore long HPD pulses on eDP
drm/exynos: Add DECON driver
drm/i915: Correct the base value while updating LP_OUTPUT_HOLD in MIPI_PORT_CTRL
drm/i915: Insert a command barrier on BLT/BSD cache flushes
drm/i915: Drop vblank wait from intel_dp_link_down
drm/exynos: fix NULL pointer reference
drm/exynos: remove exynos_plane_dpms
drm/exynos: remove mode property of exynos crtc
drm/exynos: Remove exynos_plane_dpms() call with no effect
drm/i915: Squelch overzealous uncore reset WARN_ON
drm/i915: Take runtime pm reference on hangcheck_info
drm/i915: Correct the IOSF Dev_FN field for IOSF transfers
drm/exynos: fix DMA_ATTR_NO_KERNEL_MAPPING usage
...
Diffstat (limited to 'drivers/gpu/drm/amd/amdkfd/kfd_device_queue_manager.c')
-rw-r--r-- | drivers/gpu/drm/amd/amdkfd/kfd_device_queue_manager.c | 420 |
1 files changed, 228 insertions, 192 deletions
diff --git a/drivers/gpu/drm/amd/amdkfd/kfd_device_queue_manager.c b/drivers/gpu/drm/amd/amdkfd/kfd_device_queue_manager.c index 0fd5927..b3589d0 100644 --- a/drivers/gpu/drm/amd/amdkfd/kfd_device_queue_manager.c +++ b/drivers/gpu/drm/amd/amdkfd/kfd_device_queue_manager.c @@ -26,34 +26,40 @@ #include <linux/types.h> #include <linux/printk.h> #include <linux/bitops.h> +#include <linux/sched.h> #include "kfd_priv.h" #include "kfd_device_queue_manager.h" #include "kfd_mqd_manager.h" #include "cik_regs.h" #include "kfd_kernel_queue.h" -#include "../../radeon/cik_reg.h" /* Size of the per-pipe EOP queue */ #define CIK_HPD_EOP_BYTES_LOG2 11 #define CIK_HPD_EOP_BYTES (1U << CIK_HPD_EOP_BYTES_LOG2) -static bool is_mem_initialized; - -static int init_memory(struct device_queue_manager *dqm); static int set_pasid_vmid_mapping(struct device_queue_manager *dqm, unsigned int pasid, unsigned int vmid); static int create_compute_queue_nocpsch(struct device_queue_manager *dqm, struct queue *q, struct qcm_process_device *qpd); + static int execute_queues_cpsch(struct device_queue_manager *dqm, bool lock); static int destroy_queues_cpsch(struct device_queue_manager *dqm, bool lock); +static int create_sdma_queue_nocpsch(struct device_queue_manager *dqm, + struct queue *q, + struct qcm_process_device *qpd); + +static void deallocate_sdma_queue(struct device_queue_manager *dqm, + unsigned int sdma_queue_id); -static inline unsigned int get_pipes_num(struct device_queue_manager *dqm) +static inline +enum KFD_MQD_TYPE get_mqd_type_from_queue_type(enum kfd_queue_type type) { - BUG_ON(!dqm || !dqm->dev); - return dqm->dev->shared_resources.compute_pipe_count; + if (type == KFD_QUEUE_TYPE_SDMA) + return KFD_MQD_TYPE_SDMA; + return KFD_MQD_TYPE_CP; } static inline unsigned int get_first_pipe(struct device_queue_manager *dqm) @@ -67,61 +73,7 @@ static inline unsigned int get_pipes_num_cpsch(void) return PIPE_PER_ME_CP_SCHEDULING; } -static inline unsigned int -get_sh_mem_bases_nybble_64(struct kfd_process_device *pdd) -{ - uint32_t nybble; - - nybble = (pdd->lds_base >> 60) & 0x0E; - - return nybble; - -} - -static inline unsigned int get_sh_mem_bases_32(struct kfd_process_device *pdd) -{ - unsigned int shared_base; - - shared_base = (pdd->lds_base >> 16) & 0xFF; - - return shared_base; -} - -static uint32_t compute_sh_mem_bases_64bit(unsigned int top_address_nybble); -static void init_process_memory(struct device_queue_manager *dqm, - struct qcm_process_device *qpd) -{ - struct kfd_process_device *pdd; - unsigned int temp; - - BUG_ON(!dqm || !qpd); - - pdd = qpd_to_pdd(qpd); - - /* check if sh_mem_config register already configured */ - if (qpd->sh_mem_config == 0) { - qpd->sh_mem_config = - ALIGNMENT_MODE(SH_MEM_ALIGNMENT_MODE_UNALIGNED) | - DEFAULT_MTYPE(MTYPE_NONCACHED) | - APE1_MTYPE(MTYPE_NONCACHED); - qpd->sh_mem_ape1_limit = 0; - qpd->sh_mem_ape1_base = 0; - } - - if (qpd->pqm->process->is_32bit_user_mode) { - temp = get_sh_mem_bases_32(pdd); - qpd->sh_mem_bases = SHARED_BASE(temp); - qpd->sh_mem_config |= PTR32; - } else { - temp = get_sh_mem_bases_nybble_64(pdd); - qpd->sh_mem_bases = compute_sh_mem_bases_64bit(temp); - } - - pr_debug("kfd: is32bit process: %d sh_mem_bases nybble: 0x%X and register 0x%X\n", - qpd->pqm->process->is_32bit_user_mode, temp, qpd->sh_mem_bases); -} - -static void program_sh_mem_settings(struct device_queue_manager *dqm, +void program_sh_mem_settings(struct device_queue_manager *dqm, struct qcm_process_device *qpd) { return kfd2kgd->program_sh_mem_settings(dqm->dev->kgd, qpd->vmid, @@ -200,7 +152,10 @@ static int create_queue_nocpsch(struct device_queue_manager *dqm, *allocated_vmid = qpd->vmid; q->properties.vmid = qpd->vmid; - retval = create_compute_queue_nocpsch(dqm, q, qpd); + if (q->properties.type == KFD_QUEUE_TYPE_COMPUTE) + retval = create_compute_queue_nocpsch(dqm, q, qpd); + if (q->properties.type == KFD_QUEUE_TYPE_SDMA) + retval = create_sdma_queue_nocpsch(dqm, q, qpd); if (retval != 0) { if (list_empty(&qpd->queues_list)) { @@ -212,7 +167,11 @@ static int create_queue_nocpsch(struct device_queue_manager *dqm, } list_add(&q->list, &qpd->queues_list); - dqm->queue_count++; + if (q->properties.is_active) + dqm->queue_count++; + + if (q->properties.type == KFD_QUEUE_TYPE_SDMA) + dqm->sdma_queue_count++; /* * Unconditionally increment this counter, regardless of the queue's @@ -229,12 +188,12 @@ static int create_queue_nocpsch(struct device_queue_manager *dqm, static int allocate_hqd(struct device_queue_manager *dqm, struct queue *q) { bool set; - int pipe, bit; + int pipe, bit, i; set = false; - for (pipe = dqm->next_pipe_to_allocate; pipe < get_pipes_num(dqm); - pipe = (pipe + 1) % get_pipes_num(dqm)) { + for (pipe = dqm->next_pipe_to_allocate, i = 0; i < get_pipes_num(dqm); + pipe = ((pipe + 1) % get_pipes_num(dqm)), ++i) { if (dqm->allocated_queues[pipe] != 0) { bit = find_first_bit( (unsigned long *)&dqm->allocated_queues[pipe], @@ -275,7 +234,7 @@ static int create_compute_queue_nocpsch(struct device_queue_manager *dqm, BUG_ON(!dqm || !q || !qpd); - mqd = dqm->get_mqd_manager(dqm, KFD_MQD_TYPE_CIK_COMPUTE); + mqd = dqm->ops.get_mqd_manager(dqm, KFD_MQD_TYPE_COMPUTE); if (mqd == NULL) return -ENOMEM; @@ -319,28 +278,44 @@ static int destroy_queue_nocpsch(struct device_queue_manager *dqm, pr_debug("kfd: In Func %s\n", __func__); mutex_lock(&dqm->lock); - mqd = dqm->get_mqd_manager(dqm, KFD_MQD_TYPE_CIK_COMPUTE); - if (mqd == NULL) { - retval = -ENOMEM; + + if (q->properties.type == KFD_QUEUE_TYPE_COMPUTE) { + mqd = dqm->ops.get_mqd_manager(dqm, KFD_MQD_TYPE_COMPUTE); + if (mqd == NULL) { + retval = -ENOMEM; + goto out; + } + deallocate_hqd(dqm, q); + } else if (q->properties.type == KFD_QUEUE_TYPE_SDMA) { + mqd = dqm->ops.get_mqd_manager(dqm, KFD_MQD_TYPE_SDMA); + if (mqd == NULL) { + retval = -ENOMEM; + goto out; + } + dqm->sdma_queue_count--; + deallocate_sdma_queue(dqm, q->sdma_id); + } else { + pr_debug("q->properties.type is invalid (%d)\n", + q->properties.type); + retval = -EINVAL; goto out; } retval = mqd->destroy_mqd(mqd, q->mqd, - KFD_PREEMPT_TYPE_WAVEFRONT, + KFD_PREEMPT_TYPE_WAVEFRONT_RESET, QUEUE_PREEMPT_DEFAULT_TIMEOUT_MS, q->pipe, q->queue); if (retval != 0) goto out; - deallocate_hqd(dqm, q); - mqd->uninit_mqd(mqd, q->mqd, q->mqd_mem_obj); list_del(&q->list); if (list_empty(&qpd->queues_list)) deallocate_vmid(dqm, qpd, q); - dqm->queue_count--; + if (q->properties.is_active) + dqm->queue_count--; /* * Unconditionally decrement this counter, regardless of the queue's @@ -364,7 +339,8 @@ static int update_queue(struct device_queue_manager *dqm, struct queue *q) BUG_ON(!dqm || !q || !q->mqd); mutex_lock(&dqm->lock); - mqd = dqm->get_mqd_manager(dqm, KFD_MQD_TYPE_CIK_COMPUTE); + mqd = dqm->ops.get_mqd_manager(dqm, + get_mqd_type_from_queue_type(q->properties.type)); if (mqd == NULL) { mutex_unlock(&dqm->lock); return -ENOMEM; @@ -415,6 +391,7 @@ static int register_process_nocpsch(struct device_queue_manager *dqm, struct qcm_process_device *qpd) { struct device_process_node *n; + int retval; BUG_ON(!dqm || !qpd); @@ -429,12 +406,13 @@ static int register_process_nocpsch(struct device_queue_manager *dqm, mutex_lock(&dqm->lock); list_add(&n->list, &dqm->queues); - init_process_memory(dqm, qpd); + retval = dqm->ops_asic_specific.register_process(dqm, qpd); + dqm->processes_count++; mutex_unlock(&dqm->lock); - return 0; + return retval; } static int unregister_process_nocpsch(struct device_queue_manager *dqm, @@ -479,48 +457,7 @@ set_pasid_vmid_mapping(struct device_queue_manager *dqm, unsigned int pasid, vmid); } -static uint32_t compute_sh_mem_bases_64bit(unsigned int top_address_nybble) -{ - /* In 64-bit mode, we can only control the top 3 bits of the LDS, - * scratch and GPUVM apertures. - * The hardware fills in the remaining 59 bits according to the - * following pattern: - * LDS: X0000000'00000000 - X0000001'00000000 (4GB) - * Scratch: X0000001'00000000 - X0000002'00000000 (4GB) - * GPUVM: Y0010000'00000000 - Y0020000'00000000 (1TB) - * - * (where X/Y is the configurable nybble with the low-bit 0) - * - * LDS and scratch will have the same top nybble programmed in the - * top 3 bits of SH_MEM_BASES.PRIVATE_BASE. - * GPUVM can have a different top nybble programmed in the - * top 3 bits of SH_MEM_BASES.SHARED_BASE. - * We don't bother to support different top nybbles - * for LDS/Scratch and GPUVM. - */ - - BUG_ON((top_address_nybble & 1) || top_address_nybble > 0xE || - top_address_nybble == 0); - - return PRIVATE_BASE(top_address_nybble << 12) | - SHARED_BASE(top_address_nybble << 12); -} - -static int init_memory(struct device_queue_manager *dqm) -{ - int i, retval; - - for (i = 8; i < 16; i++) - set_pasid_vmid_mapping(dqm, 0, i); - - retval = kfd2kgd->init_memory(dqm->dev->kgd); - if (retval == 0) - is_mem_initialized = true; - return retval; -} - - -static int init_pipelines(struct device_queue_manager *dqm, +int init_pipelines(struct device_queue_manager *dqm, unsigned int pipes_num, unsigned int first_pipe) { void *hpdptr; @@ -539,11 +476,8 @@ static int init_pipelines(struct device_queue_manager *dqm, * because it contains no data when there are no active queues. */ - err = kfd2kgd->allocate_mem(dqm->dev->kgd, - CIK_HPD_EOP_BYTES * pipes_num, - PAGE_SIZE, - KFD_MEMPOOL_SYSTEM_WRITECOMBINE, - (struct kgd_mem **) &dqm->pipeline_mem); + err = kfd_gtt_sa_allocate(dqm->dev, CIK_HPD_EOP_BYTES * pipes_num, + &dqm->pipeline_mem); if (err) { pr_err("kfd: error allocate vidmem num pipes: %d\n", @@ -556,10 +490,9 @@ static int init_pipelines(struct device_queue_manager *dqm, memset(hpdptr, 0, CIK_HPD_EOP_BYTES * pipes_num); - mqd = dqm->get_mqd_manager(dqm, KFD_MQD_TYPE_CIK_COMPUTE); + mqd = dqm->ops.get_mqd_manager(dqm, KFD_MQD_TYPE_COMPUTE); if (mqd == NULL) { - kfd2kgd->free_mem(dqm->dev->kgd, - (struct kgd_mem *) dqm->pipeline_mem); + kfd_gtt_sa_free(dqm->dev, dqm->pipeline_mem); return -ENOMEM; } @@ -579,7 +512,6 @@ static int init_pipelines(struct device_queue_manager *dqm, return 0; } - static int init_scheduler(struct device_queue_manager *dqm) { int retval; @@ -589,11 +521,6 @@ static int init_scheduler(struct device_queue_manager *dqm) pr_debug("kfd: In %s\n", __func__); retval = init_pipelines(dqm, get_pipes_num(dqm), get_first_pipe(dqm)); - if (retval != 0) - return retval; - - retval = init_memory(dqm); - return retval; } @@ -609,6 +536,7 @@ static int initialize_nocpsch(struct device_queue_manager *dqm) mutex_init(&dqm->lock); INIT_LIST_HEAD(&dqm->queues); dqm->queue_count = dqm->next_pipe_to_allocate = 0; + dqm->sdma_queue_count = 0; dqm->allocated_queues = kcalloc(get_pipes_num(dqm), sizeof(unsigned int), GFP_KERNEL); if (!dqm->allocated_queues) { @@ -620,6 +548,7 @@ static int initialize_nocpsch(struct device_queue_manager *dqm) dqm->allocated_queues[i] = (1 << QUEUES_PER_PIPE) - 1; dqm->vmid_bitmap = (1 << VMID_PER_DEVICE) - 1; + dqm->sdma_bitmap = (1 << CIK_SDMA_QUEUES) - 1; init_scheduler(dqm); return 0; @@ -637,8 +566,7 @@ static void uninitialize_nocpsch(struct device_queue_manager *dqm) for (i = 0 ; i < KFD_MQD_TYPE_MAX ; i++) kfree(dqm->mqds[i]); mutex_destroy(&dqm->lock); - kfd2kgd->free_mem(dqm->dev->kgd, - (struct kgd_mem *) dqm->pipeline_mem); + kfd_gtt_sa_free(dqm->dev, dqm->pipeline_mem); } static int start_nocpsch(struct device_queue_manager *dqm) @@ -651,6 +579,77 @@ static int stop_nocpsch(struct device_queue_manager *dqm) return 0; } +static int allocate_sdma_queue(struct device_queue_manager *dqm, + unsigned int *sdma_queue_id) +{ + int bit; + + if (dqm->sdma_bitmap == 0) + return -ENOMEM; + + bit = find_first_bit((unsigned long *)&dqm->sdma_bitmap, + CIK_SDMA_QUEUES); + + clear_bit(bit, (unsigned long *)&dqm->sdma_bitmap); + *sdma_queue_id = bit; + + return 0; +} + +static void deallocate_sdma_queue(struct device_queue_manager *dqm, + unsigned int sdma_queue_id) +{ + if (sdma_queue_id >= CIK_SDMA_QUEUES) + return; + set_bit(sdma_queue_id, (unsigned long *)&dqm->sdma_bitmap); +} + +static void init_sdma_vm(struct device_queue_manager *dqm, struct queue *q, + struct qcm_process_device *qpd) +{ + uint32_t value = SDMA_ATC; + + if (q->process->is_32bit_user_mode) + value |= SDMA_VA_PTR32 | get_sh_mem_bases_32(qpd_to_pdd(qpd)); + else + value |= SDMA_VA_SHARED_BASE(get_sh_mem_bases_nybble_64( + qpd_to_pdd(qpd))); + q->properties.sdma_vm_addr = value; +} + +static int create_sdma_queue_nocpsch(struct device_queue_manager *dqm, + struct queue *q, + struct qcm_process_device *qpd) +{ + struct mqd_manager *mqd; + int retval; + + mqd = dqm->ops.get_mqd_manager(dqm, KFD_MQD_TYPE_SDMA); + if (!mqd) + return -ENOMEM; + + retval = allocate_sdma_queue(dqm, &q->sdma_id); + if (retval != 0) + return retval; + + q->properties.sdma_queue_id = q->sdma_id % CIK_SDMA_QUEUES_PER_ENGINE; + q->properties.sdma_engine_id = q->sdma_id / CIK_SDMA_ENGINE_NUM; + + pr_debug("kfd: sdma id is: %d\n", q->sdma_id); + pr_debug(" sdma queue id: %d\n", q->properties.sdma_queue_id); + pr_debug(" sdma engine id: %d\n", q->properties.sdma_engine_id); + + retval = mqd->init_mqd(mqd, &q->mqd, &q->mqd_mem_obj, + &q->gart_mqd_addr, &q->properties); + if (retval != 0) { + deallocate_sdma_queue(dqm, q->sdma_id); + return retval; + } + + init_sdma_vm(dqm, q, qpd); + return 0; +} + /* * Device Queue Manager implementation for cp scheduler */ @@ -692,8 +691,9 @@ static int initialize_cpsch(struct device_queue_manager *dqm) mutex_init(&dqm->lock); INIT_LIST_HEAD(&dqm->queues); dqm->queue_count = dqm->processes_count = 0; + dqm->sdma_queue_count = 0; dqm->active_runlist = false; - retval = init_pipelines(dqm, get_pipes_num(dqm), 0); + retval = dqm->ops_asic_specific.initialize(dqm); if (retval != 0) goto fail_init_pipelines; @@ -724,18 +724,14 @@ static int start_cpsch(struct device_queue_manager *dqm) pr_debug("kfd: allocating fence memory\n"); /* allocate fence memory on the gart */ - retval = kfd2kgd->allocate_mem(dqm->dev->kgd, - sizeof(*dqm->fence_addr), - 32, - KFD_MEMPOOL_SYSTEM_WRITECOMBINE, - (struct kgd_mem **) &dqm->fence_mem); + retval = kfd_gtt_sa_allocate(dqm->dev, sizeof(*dqm->fence_addr), + &dqm->fence_mem); if (retval != 0) goto fail_allocate_vidmem; dqm->fence_addr = dqm->fence_mem->cpu_ptr; dqm->fence_gpu_addr = dqm->fence_mem->gpu_addr; - list_for_each_entry(node, &dqm->queues, list) if (node->qpd->pqm->process && dqm->dev) kfd_bind_process_to_device(dqm->dev, @@ -764,8 +760,7 @@ static int stop_cpsch(struct device_queue_manager *dqm) pdd = qpd_to_pdd(node->qpd); pdd->bound = false; } - kfd2kgd->free_mem(dqm->dev->kgd, - (struct kgd_mem *) dqm->fence_mem); + kfd_gtt_sa_free(dqm->dev, dqm->fence_mem); pm_uninit(&dqm->packets); return 0; @@ -828,6 +823,14 @@ static void destroy_kernel_queue_cpsch(struct device_queue_manager *dqm, mutex_unlock(&dqm->lock); } +static void select_sdma_engine_id(struct queue *q) +{ + static int sdma_id; + + q->sdma_id = sdma_id; + sdma_id = (sdma_id + 1) % 2; +} + static int create_queue_cpsch(struct device_queue_manager *dqm, struct queue *q, struct qcm_process_device *qpd, int *allocate_vmid) { @@ -850,7 +853,12 @@ static int create_queue_cpsch(struct device_queue_manager *dqm, struct queue *q, goto out; } - mqd = dqm->get_mqd_manager(dqm, KFD_MQD_TYPE_CIK_CP); + if (q->properties.type == KFD_QUEUE_TYPE_SDMA) + select_sdma_engine_id(q); + + mqd = dqm->ops.get_mqd_manager(dqm, + get_mqd_type_from_queue_type(q->properties.type)); + if (mqd == NULL) { mutex_unlock(&dqm->lock); return -ENOMEM; @@ -867,6 +875,8 @@ static int create_queue_cpsch(struct device_queue_manager *dqm, struct queue *q, retval = execute_queues_cpsch(dqm, false); } + if (q->properties.type == KFD_QUEUE_TYPE_SDMA) + dqm->sdma_queue_count++; /* * Unconditionally increment this counter, regardless of the queue's * type or whether the queue is active. @@ -893,12 +903,20 @@ static int fence_wait_timeout(unsigned int *fence_addr, pr_err("kfd: qcm fence wait loop timeout expired\n"); return -ETIME; } - cpu_relax(); + schedule(); } return 0; } +static int destroy_sdma_queues(struct device_queue_manager *dqm, + unsigned int sdma_engine) +{ + return pm_send_unmap_queue(&dqm->packets, KFD_QUEUE_TYPE_SDMA, + KFD_PREEMPT_TYPE_FILTER_ALL_QUEUES, 0, false, + sdma_engine); +} + static int destroy_queues_cpsch(struct device_queue_manager *dqm, bool lock) { int retval; @@ -911,6 +929,15 @@ static int destroy_queues_cpsch(struct device_queue_manager *dqm, bool lock) mutex_lock(&dqm->lock); if (dqm->active_runlist == false) goto out; + + pr_debug("kfd: Before destroying queues, sdma queue count is : %u\n", + dqm->sdma_queue_count); + + if (dqm->sdma_queue_count > 0) { + destroy_sdma_queues(dqm, 0); + destroy_sdma_queues(dqm, 1); + } + retval = pm_send_unmap_queue(&dqm->packets, KFD_QUEUE_TYPE_COMPUTE, KFD_PREEMPT_TYPE_FILTER_ALL_QUEUES, 0, false, 0); if (retval != 0) @@ -982,15 +1009,19 @@ static int destroy_queue_cpsch(struct device_queue_manager *dqm, /* remove queue from list to prevent rescheduling after preemption */ mutex_lock(&dqm->lock); - - mqd = dqm->get_mqd_manager(dqm, KFD_MQD_TYPE_CIK_CP); + mqd = dqm->ops.get_mqd_manager(dqm, + get_mqd_type_from_queue_type(q->properties.type)); if (!mqd) { retval = -ENOMEM; goto failed; } + if (q->properties.type == KFD_QUEUE_TYPE_SDMA) + dqm->sdma_queue_count--; + list_del(&q->list); - dqm->queue_count--; + if (q->properties.is_active) + dqm->queue_count--; execute_queues_cpsch(dqm, false); @@ -1028,8 +1059,7 @@ static bool set_cache_memory_policy(struct device_queue_manager *dqm, void __user *alternate_aperture_base, uint64_t alternate_aperture_size) { - uint32_t default_mtype; - uint32_t ape1_mtype; + bool retval; pr_debug("kfd: In func %s\n", __func__); @@ -1066,18 +1096,13 @@ static bool set_cache_memory_policy(struct device_queue_manager *dqm, qpd->sh_mem_ape1_limit = limit >> 16; } - default_mtype = (default_policy == cache_policy_coherent) ? - MTYPE_NONCACHED : - MTYPE_CACHED; - - ape1_mtype = (alternate_policy == cache_policy_coherent) ? - MTYPE_NONCACHED : - MTYPE_CACHED; - - qpd->sh_mem_config = (qpd->sh_mem_config & PTR32) - | ALIGNMENT_MODE(SH_MEM_ALIGNMENT_MODE_UNALIGNED) - | DEFAULT_MTYPE(default_mtype) - | APE1_MTYPE(ape1_mtype); + retval = dqm->ops_asic_specific.set_cache_memory_policy( + dqm, + qpd, + default_policy, + alternate_policy, + alternate_aperture_base, + alternate_aperture_size); if ((sched_policy == KFD_SCHED_POLICY_NO_HWS) && (qpd->vmid != 0)) program_sh_mem_settings(dqm, qpd); @@ -1087,7 +1112,7 @@ static bool set_cache_memory_policy(struct device_queue_manager *dqm, qpd->sh_mem_ape1_limit); mutex_unlock(&dqm->lock); - return true; + return retval; out: mutex_unlock(&dqm->lock); @@ -1100,6 +1125,8 @@ struct device_queue_manager *device_queue_manager_init(struct kfd_dev *dev) BUG_ON(!dev); + pr_debug("kfd: loading device queue manager\n"); + dqm = kzalloc(sizeof(struct device_queue_manager), GFP_KERNEL); if (!dqm) return NULL; @@ -1109,40 +1136,50 @@ struct device_queue_manager *device_queue_manager_init(struct kfd_dev *dev) case KFD_SCHED_POLICY_HWS: case KFD_SCHED_POLICY_HWS_NO_OVERSUBSCRIPTION: /* initialize dqm for cp scheduling */ - dqm->create_queue = create_queue_cpsch; - dqm->initialize = initialize_cpsch; - dqm->start = start_cpsch; - dqm->stop = stop_cpsch; - dqm->destroy_queue = destroy_queue_cpsch; - dqm->update_queue = update_queue; - dqm->get_mqd_manager = get_mqd_manager_nocpsch; - dqm->register_process = register_process_nocpsch; - dqm->unregister_process = unregister_process_nocpsch; - dqm->uninitialize = uninitialize_nocpsch; - dqm->create_kernel_queue = create_kernel_queue_cpsch; - dqm->destroy_kernel_queue = destroy_kernel_queue_cpsch; - dqm->set_cache_memory_policy = set_cache_memory_policy; + dqm->ops.create_queue = create_queue_cpsch; + dqm->ops.initialize = initialize_cpsch; + dqm->ops.start = start_cpsch; + dqm->ops.stop = stop_cpsch; + dqm->ops.destroy_queue = destroy_queue_cpsch; + dqm->ops.update_queue = update_queue; + dqm->ops.get_mqd_manager = get_mqd_manager_nocpsch; + dqm->ops.register_process = register_process_nocpsch; + dqm->ops.unregister_process = unregister_process_nocpsch; + dqm->ops.uninitialize = uninitialize_nocpsch; + dqm->ops.create_kernel_queue = create_kernel_queue_cpsch; + dqm->ops.destroy_kernel_queue = destroy_kernel_queue_cpsch; + dqm->ops.set_cache_memory_policy = set_cache_memory_policy; break; case KFD_SCHED_POLICY_NO_HWS: /* initialize dqm for no cp scheduling */ - dqm->start = start_nocpsch; - dqm->stop = stop_nocpsch; - dqm->create_queue = create_queue_nocpsch; - dqm->destroy_queue = destroy_queue_nocpsch; - dqm->update_queue = update_queue; - dqm->get_mqd_manager = get_mqd_manager_nocpsch; - dqm->register_process = register_process_nocpsch; - dqm->unregister_process = unregister_process_nocpsch; - dqm->initialize = initialize_nocpsch; - dqm->uninitialize = uninitialize_nocpsch; - dqm->set_cache_memory_policy = set_cache_memory_policy; + dqm->ops.start = start_nocpsch; + dqm->ops.stop = stop_nocpsch; + dqm->ops.create_queue = create_queue_nocpsch; + dqm->ops.destroy_queue = destroy_queue_nocpsch; + dqm->ops.update_queue = update_queue; + dqm->ops.get_mqd_manager = get_mqd_manager_nocpsch; + dqm->ops.register_process = register_process_nocpsch; + dqm->ops.unregister_process = unregister_process_nocpsch; + dqm->ops.initialize = initialize_nocpsch; + dqm->ops.uninitialize = uninitialize_nocpsch; + dqm->ops.set_cache_memory_policy = set_cache_memory_policy; break; default: BUG(); break; } - if (dqm->initialize(dqm) != 0) { + switch (dev->device_info->asic_family) { + case CHIP_CARRIZO: + device_queue_manager_init_vi(&dqm->ops_asic_specific); + break; + + case CHIP_KAVERI: + device_queue_manager_init_cik(&dqm->ops_asic_specific); + break; + } + + if (dqm->ops.initialize(dqm) != 0) { kfree(dqm); return NULL; } @@ -1154,7 +1191,6 @@ void device_queue_manager_uninit(struct device_queue_manager *dqm) { BUG_ON(!dqm); - dqm->uninitialize(dqm); + dqm->ops.uninitialize(dqm); kfree(dqm); } - |