From 093c7d8cfd2a26ff05e3bdff63d371147df1951c Mon Sep 17 00:00:00 2001 From: Alexey Skidanov Date: Tue, 18 Nov 2014 14:00:04 +0200 Subject: drm/amdkfd: Process-device data creation and lookup split This patch splits the current kfd_get_process_device_data() to two functions, one that specifically creates a pdd and another one which just do lookup. This is done to enhance the readability and maintainability of the code. Signed-off-by: Alexey Skidanov Signed-off-by: Oded Gabbay --- drivers/gpu/drm/amd/amdkfd/kfd_device_queue_manager.c | 1 - 1 file changed, 1 deletion(-) (limited to 'drivers/gpu/drm/amd/amdkfd/kfd_device_queue_manager.c') 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 924e90c072e5..fb94f1a2b911 100644 --- a/drivers/gpu/drm/amd/amdkfd/kfd_device_queue_manager.c +++ b/drivers/gpu/drm/amd/amdkfd/kfd_device_queue_manager.c @@ -75,7 +75,6 @@ get_sh_mem_bases_nybble_64(struct kfd_process_device *pdd) nybble = (pdd->lds_base >> 60) & 0x0E; return nybble; - } static inline unsigned int get_sh_mem_bases_32(struct kfd_process_device *pdd) -- cgit v1.2.3 From bcea308175748339b872cc50972e0a31c1999c64 Mon Sep 17 00:00:00 2001 From: Ben Goz Date: Sat, 3 Jan 2015 22:12:32 +0200 Subject: drm/amdkfd: Add SDMA user-mode queues support to QCM This patch adds support for SDMA user-mode queues to the QCM - the Queue management system that manages queues-per-device and queues-per-process. v2: Remove calls to interface function that initializes sdma engines. v3: Use the new names of some of the defines. Signed-off-by: Ben Goz Signed-off-by: Oded Gabbay Reviewed-by: Alex Deucher --- .../gpu/drm/amd/amdkfd/kfd_device_queue_manager.c | 159 +++++++++++++++++++-- 1 file changed, 148 insertions(+), 11 deletions(-) (limited to 'drivers/gpu/drm/amd/amdkfd/kfd_device_queue_manager.c') 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 fb94f1a2b911..7ead0802883d 100644 --- a/drivers/gpu/drm/amd/amdkfd/kfd_device_queue_manager.c +++ b/drivers/gpu/drm/amd/amdkfd/kfd_device_queue_manager.c @@ -46,9 +46,24 @@ static int set_pasid_vmid_mapping(struct device_queue_manager *dqm, 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 +enum KFD_MQD_TYPE get_mqd_type_from_queue_type(enum kfd_queue_type type) +{ + if (type == KFD_QUEUE_TYPE_SDMA) + return KFD_MQD_TYPE_CIK_SDMA; + return KFD_MQD_TYPE_CIK_CP; +} static inline unsigned int get_pipes_num(struct device_queue_manager *dqm) { @@ -189,7 +204,10 @@ 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)) { @@ -202,7 +220,8 @@ static int create_queue_nocpsch(struct device_queue_manager *dqm, list_add(&q->list, &qpd->queues_list); dqm->queue_count++; - + if (q->properties.type == KFD_QUEUE_TYPE_SDMA) + dqm->sdma_queue_count++; mutex_unlock(&dqm->lock); return 0; } @@ -279,8 +298,7 @@ static int destroy_queue_nocpsch(struct device_queue_manager *dqm, struct queue *q) { int retval; - struct mqd_manager *mqd; - + struct mqd_manager *mqd, *mqd_sdma; BUG_ON(!dqm || !q || !q->mqd || !qpd); retval = 0; @@ -294,6 +312,12 @@ static int destroy_queue_nocpsch(struct device_queue_manager *dqm, goto out; } + mqd_sdma = dqm->get_mqd_manager(dqm, KFD_MQD_TYPE_CIK_SDMA); + if (mqd_sdma == NULL) { + mutex_unlock(&dqm->lock); + return -ENOMEM; + } + retval = mqd->destroy_mqd(mqd, q->mqd, KFD_PREEMPT_TYPE_WAVEFRONT, QUEUE_PREEMPT_DEFAULT_TIMEOUT_MS, @@ -302,7 +326,12 @@ static int destroy_queue_nocpsch(struct device_queue_manager *dqm, if (retval != 0) goto out; - deallocate_hqd(dqm, q); + if (q->properties.type == KFD_QUEUE_TYPE_COMPUTE) + deallocate_hqd(dqm, q); + else if (q->properties.type == KFD_QUEUE_TYPE_SDMA) { + dqm->sdma_queue_count--; + deallocate_sdma_queue(dqm, q->sdma_id); + } mqd->uninit_mqd(mqd, q->mqd, q->mqd_mem_obj); @@ -323,7 +352,7 @@ 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->get_mqd_manager(dqm, q->properties.type); if (mqd == NULL) { mutex_unlock(&dqm->lock); return -ENOMEM; @@ -526,7 +555,6 @@ static int init_pipelines(struct device_queue_manager *dqm, return 0; } - static int init_scheduler(struct device_queue_manager *dqm) { int retval; @@ -556,6 +584,7 @@ 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) { @@ -567,6 +596,7 @@ 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; @@ -598,6 +628,77 @@ 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 < 0 || 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->get_mqd_manager(dqm, KFD_MQD_TYPE_CIK_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 */ @@ -639,6 +740,7 @@ 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); if (retval != 0) @@ -682,7 +784,6 @@ static int start_cpsch(struct device_queue_manager *dqm) 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, @@ -753,6 +854,14 @@ 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) { @@ -768,7 +877,12 @@ static int create_queue_cpsch(struct device_queue_manager *dqm, struct queue *q, mutex_lock(&dqm->lock); - 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->get_mqd_manager(dqm, + get_mqd_type_from_queue_type(q->properties.type)); + if (mqd == NULL) { mutex_unlock(&dqm->lock); return -ENOMEM; @@ -785,6 +899,9 @@ 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++; + out: mutex_unlock(&dqm->lock); return retval; @@ -808,6 +925,14 @@ static int fence_wait_timeout(unsigned int *fence_addr, 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; @@ -820,6 +945,15 @@ 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) @@ -891,13 +1025,16 @@ 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->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--; -- cgit v1.2.3 From a86aa3ca5a2f16772653782c078f62a7d76dd57e Mon Sep 17 00:00:00 2001 From: Oded Gabbay Date: Sun, 26 Oct 2014 22:00:31 +0200 Subject: drm/amdkfd: Using new gtt sa in amdkfd This patch change the calls throughout the amdkfd driver from the old kfd-->kgd interface to the new kfd gtt sa inside amdkfd v2: change the new call in sdma code that appeared because of the sdma feature Signed-off-by: Oded Gabbay Reviewed-by: Alexey Skidanov Reviewed-by: Alex Deucher --- .../gpu/drm/amd/amdkfd/kfd_device_queue_manager.c | 23 +++++++--------------- 1 file changed, 7 insertions(+), 16 deletions(-) (limited to 'drivers/gpu/drm/amd/amdkfd/kfd_device_queue_manager.c') 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 7ead0802883d..6806e64c5ffd 100644 --- a/drivers/gpu/drm/amd/amdkfd/kfd_device_queue_manager.c +++ b/drivers/gpu/drm/amd/amdkfd/kfd_device_queue_manager.c @@ -519,11 +519,8 @@ 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", @@ -538,8 +535,7 @@ static int init_pipelines(struct device_queue_manager *dqm, mqd = dqm->get_mqd_manager(dqm, KFD_MQD_TYPE_CIK_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; } @@ -614,8 +610,7 @@ 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) @@ -773,11 +768,8 @@ 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; @@ -812,8 +804,7 @@ 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; -- cgit v1.2.3 From fe502804205e4103bdff4854bb41cd78fa82b099 Mon Sep 17 00:00:00 2001 From: Ben Goz Date: Sun, 26 Oct 2014 18:07:34 +0200 Subject: drm/amdkfd: Remove call to deprecated init_memory interface This patch removes a call to kfd-->kgd interface function that is doing H/W initialization. That function is moved into radeon to be part of the common H/W initialization sequence. The interface function will be deleted. Signed-off-by: Ben Goz Signed-off-by: Oded Gabbay Reviewed-by: Alex Deucher --- .../gpu/drm/amd/amdkfd/kfd_device_queue_manager.c | 21 --------------------- 1 file changed, 21 deletions(-) (limited to 'drivers/gpu/drm/amd/amdkfd/kfd_device_queue_manager.c') 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 6806e64c5ffd..60f4ccbbb301 100644 --- a/drivers/gpu/drm/amd/amdkfd/kfd_device_queue_manager.c +++ b/drivers/gpu/drm/amd/amdkfd/kfd_device_queue_manager.c @@ -37,9 +37,6 @@ #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); @@ -486,20 +483,6 @@ static uint32_t compute_sh_mem_bases_64bit(unsigned int top_address_nybble) 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, unsigned int pipes_num, unsigned int first_pipe) { @@ -560,10 +543,6 @@ static int init_scheduler(struct device_queue_manager *dqm) pr_debug("kfd: In %s\n", __func__); retval = init_pipelines(dqm, get_pipes_num(dqm), KFD_DQM_FIRST_PIPE); - if (retval != 0) - return retval; - - retval = init_memory(dqm); return retval; } -- cgit v1.2.3 From 71273adc52fafcda181e62cef64ddffd76d91944 Mon Sep 17 00:00:00 2001 From: Oded Gabbay Date: Fri, 2 Jan 2015 23:18:54 +0200 Subject: drm/amdkfd: Don't include header files from radeon Because amdkfd will need to work both with radeon and amdgpu, don't include header files that are in radeon's folder. Instead, use the common amd include folder and move amdkfd specific defines to amdkfd header files. Signed-off-by: Oded Gabbay Reviewed-by: Alex Deucher --- drivers/gpu/drm/amd/amdkfd/kfd_device_queue_manager.c | 1 - 1 file changed, 1 deletion(-) (limited to 'drivers/gpu/drm/amd/amdkfd/kfd_device_queue_manager.c') 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 60f4ccbbb301..fead2d7c0877 100644 --- a/drivers/gpu/drm/amd/amdkfd/kfd_device_queue_manager.c +++ b/drivers/gpu/drm/amd/amdkfd/kfd_device_queue_manager.c @@ -31,7 +31,6 @@ #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 -- cgit v1.2.3 From 85d258f9a7e827dc321d54d15ce1c6d5b0048a17 Mon Sep 17 00:00:00 2001 From: Ben Goz Date: Sun, 4 Jan 2015 10:36:30 +0200 Subject: drm/amdkfd: Make KFD_MQD_TYPE enum types H/W agnostic As the MQD types are common across all AMD GPUs/APUs, let's remove the CIK part from the name. Signed-off-by: Ben Goz Signed-off-by: Oded Gabbay Reviewed-by: Alex Deucher --- drivers/gpu/drm/amd/amdkfd/kfd_device_queue_manager.c | 14 +++++++------- 1 file changed, 7 insertions(+), 7 deletions(-) (limited to 'drivers/gpu/drm/amd/amdkfd/kfd_device_queue_manager.c') 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 fead2d7c0877..3d5f71a6e147 100644 --- a/drivers/gpu/drm/amd/amdkfd/kfd_device_queue_manager.c +++ b/drivers/gpu/drm/amd/amdkfd/kfd_device_queue_manager.c @@ -57,8 +57,8 @@ static inline enum KFD_MQD_TYPE get_mqd_type_from_queue_type(enum kfd_queue_type type) { if (type == KFD_QUEUE_TYPE_SDMA) - return KFD_MQD_TYPE_CIK_SDMA; - return KFD_MQD_TYPE_CIK_CP; + return KFD_MQD_TYPE_SDMA; + return KFD_MQD_TYPE_CP; } static inline unsigned int get_pipes_num(struct device_queue_manager *dqm) @@ -271,7 +271,7 @@ 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->get_mqd_manager(dqm, KFD_MQD_TYPE_COMPUTE); if (mqd == NULL) return -ENOMEM; @@ -302,13 +302,13 @@ 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); + mqd = dqm->get_mqd_manager(dqm, KFD_MQD_TYPE_COMPUTE); if (mqd == NULL) { retval = -ENOMEM; goto out; } - mqd_sdma = dqm->get_mqd_manager(dqm, KFD_MQD_TYPE_CIK_SDMA); + mqd_sdma = dqm->get_mqd_manager(dqm, KFD_MQD_TYPE_SDMA); if (mqd_sdma == NULL) { mutex_unlock(&dqm->lock); return -ENOMEM; @@ -515,7 +515,7 @@ 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->get_mqd_manager(dqm, KFD_MQD_TYPE_COMPUTE); if (mqd == NULL) { kfd_gtt_sa_free(dqm->dev, dqm->pipeline_mem); return -ENOMEM; @@ -646,7 +646,7 @@ static int create_sdma_queue_nocpsch(struct device_queue_manager *dqm, struct mqd_manager *mqd; int retval; - mqd = dqm->get_mqd_manager(dqm, KFD_MQD_TYPE_CIK_SDMA); + mqd = dqm->get_mqd_manager(dqm, KFD_MQD_TYPE_SDMA); if (!mqd) return -ENOMEM; -- cgit v1.2.3 From c2e1b3a496332b90d073bc9dddd3324d660fca24 Mon Sep 17 00:00:00 2001 From: Ben Goz Date: Mon, 18 Aug 2014 14:55:59 +0300 Subject: drm/amdkfd: Fix logic of destroy_queue_nocpsch() This patch rewrites destroy_queue_nocpsch() as the current logic that is implemented in the function is completely flawed. This function is used only in non-HWS mode. Signed-off-by: Ben Goz Signed-off-by: Oded Gabbay Acked-by: Alex Deucher --- .../gpu/drm/amd/amdkfd/kfd_device_queue_manager.c | 36 +++++++++++----------- 1 file changed, 18 insertions(+), 18 deletions(-) (limited to 'drivers/gpu/drm/amd/amdkfd/kfd_device_queue_manager.c') 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 3d5f71a6e147..c83f01153440 100644 --- a/drivers/gpu/drm/amd/amdkfd/kfd_device_queue_manager.c +++ b/drivers/gpu/drm/amd/amdkfd/kfd_device_queue_manager.c @@ -294,7 +294,8 @@ static int destroy_queue_nocpsch(struct device_queue_manager *dqm, struct queue *q) { int retval; - struct mqd_manager *mqd, *mqd_sdma; + struct mqd_manager *mqd; + BUG_ON(!dqm || !q || !q->mqd || !qpd); retval = 0; @@ -302,33 +303,32 @@ 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_COMPUTE); - if (mqd == NULL) { - retval = -ENOMEM; - goto out; - } - mqd_sdma = dqm->get_mqd_manager(dqm, KFD_MQD_TYPE_SDMA); - if (mqd_sdma == NULL) { - mutex_unlock(&dqm->lock); - return -ENOMEM; + if (q->properties.type == KFD_QUEUE_TYPE_COMPUTE) { + mqd = dqm->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->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); } 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; - if (q->properties.type == KFD_QUEUE_TYPE_COMPUTE) - deallocate_hqd(dqm, q); - else if (q->properties.type == KFD_QUEUE_TYPE_SDMA) { - dqm->sdma_queue_count--; - deallocate_sdma_queue(dqm, q->sdma_id); - } - mqd->uninit_mqd(mqd, q->mqd, q->mqd_mem_obj); list_del(&q->list); -- cgit v1.2.3 From 45c9a5e4297b9a07d94ff8195ff6f21ba3581ad6 Mon Sep 17 00:00:00 2001 From: Oded Gabbay Date: Mon, 12 Jan 2015 14:26:10 +0200 Subject: drm/amdkfd: Encapsulate DQM functions in ops structure This patch does some re-org on the device_queue_manager structure. It takes out all the function pointers from the structure and puts them in a new structure, called device_queue_manager_ops. Then, it puts an instance of that structure inside device_queue_manager. This re-org is done to prepare the DQM module to support more than one AMD APU (Kaveri). Signed-off-by: Oded Gabbay Reviewed-by: Alex Deucher --- .../gpu/drm/amd/amdkfd/kfd_device_queue_manager.c | 68 +++++++++++----------- 1 file changed, 34 insertions(+), 34 deletions(-) (limited to 'drivers/gpu/drm/amd/amdkfd/kfd_device_queue_manager.c') 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 c83f01153440..12c84488551e 100644 --- a/drivers/gpu/drm/amd/amdkfd/kfd_device_queue_manager.c +++ b/drivers/gpu/drm/amd/amdkfd/kfd_device_queue_manager.c @@ -271,7 +271,7 @@ 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_COMPUTE); + mqd = dqm->ops.get_mqd_manager(dqm, KFD_MQD_TYPE_COMPUTE); if (mqd == NULL) return -ENOMEM; @@ -305,14 +305,14 @@ static int destroy_queue_nocpsch(struct device_queue_manager *dqm, mutex_lock(&dqm->lock); if (q->properties.type == KFD_QUEUE_TYPE_COMPUTE) { - mqd = dqm->get_mqd_manager(dqm, KFD_MQD_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->get_mqd_manager(dqm, KFD_MQD_TYPE_SDMA); + mqd = dqm->ops.get_mqd_manager(dqm, KFD_MQD_TYPE_SDMA); if (mqd == NULL) { retval = -ENOMEM; goto out; @@ -348,7 +348,7 @@ 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, q->properties.type); + mqd = dqm->ops.get_mqd_manager(dqm, q->properties.type); if (mqd == NULL) { mutex_unlock(&dqm->lock); return -ENOMEM; @@ -515,7 +515,7 @@ 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_COMPUTE); + mqd = dqm->ops.get_mqd_manager(dqm, KFD_MQD_TYPE_COMPUTE); if (mqd == NULL) { kfd_gtt_sa_free(dqm->dev, dqm->pipeline_mem); return -ENOMEM; @@ -646,7 +646,7 @@ static int create_sdma_queue_nocpsch(struct device_queue_manager *dqm, struct mqd_manager *mqd; int retval; - mqd = dqm->get_mqd_manager(dqm, KFD_MQD_TYPE_SDMA); + mqd = dqm->ops.get_mqd_manager(dqm, KFD_MQD_TYPE_SDMA); if (!mqd) return -ENOMEM; @@ -849,7 +849,7 @@ static int create_queue_cpsch(struct device_queue_manager *dqm, struct queue *q, if (q->properties.type == KFD_QUEUE_TYPE_SDMA) select_sdma_engine_id(q); - mqd = dqm->get_mqd_manager(dqm, + mqd = dqm->ops.get_mqd_manager(dqm, get_mqd_type_from_queue_type(q->properties.type)); if (mqd == NULL) { @@ -994,7 +994,7 @@ 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, + mqd = dqm->ops.get_mqd_manager(dqm, get_mqd_type_from_queue_type(q->properties.type)); if (!mqd) { retval = -ENOMEM; @@ -1116,40 +1116,40 @@ 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) { + if (dqm->ops.initialize(dqm) != 0) { kfree(dqm); return NULL; } @@ -1161,7 +1161,7 @@ void device_queue_manager_uninit(struct device_queue_manager *dqm) { BUG_ON(!dqm); - dqm->uninitialize(dqm); + dqm->ops.uninitialize(dqm); kfree(dqm); } -- cgit v1.2.3 From a22fc85495575d81c36db24b12f66fd314b7ced1 Mon Sep 17 00:00:00 2001 From: Ben Goz Date: Mon, 12 Jan 2015 14:28:46 +0200 Subject: drm/amdkfd: Add initial VI support for DQM This patch starts to add support for the VI APU in the DQM module. Because most (more than 90%) of the DQM code is shared among AMD's APUs, we chose a design that performs most/all the code in the shared DQM file (kfd_device_queue_manager.c). If there is H/W specific code to be executed, than it is written in an asic-specific extension function for that H/W. That asic-specific extension function is called from the shared function at the appropriate time. This requires that for every asic-specific extension function that is implemented in a specific ASIC, there will be an equivalent implementation in ALL ASICs, even if those implementations are just stubs. That way we achieve: - Maintainability: by having one copy of most of the code, we only need to fix bugs at one locations - Readability: very clear what is the shared code and what is done per ASIC - Extensibility: very easy to add new H/W specific files/functions Signed-off-by: Ben Goz Signed-off-by: Oded Gabbay Reviewed-by: Alex Deucher --- .../gpu/drm/amd/amdkfd/kfd_device_queue_manager.c | 113 +++++---------------- 1 file changed, 28 insertions(+), 85 deletions(-) (limited to 'drivers/gpu/drm/amd/amdkfd/kfd_device_queue_manager.c') 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 12c84488551e..b2016245073e 100644 --- a/drivers/gpu/drm/amd/amdkfd/kfd_device_queue_manager.c +++ b/drivers/gpu/drm/amd/amdkfd/kfd_device_queue_manager.c @@ -61,7 +61,7 @@ enum KFD_MQD_TYPE get_mqd_type_from_queue_type(enum kfd_queue_type type) return KFD_MQD_TYPE_CP; } -static inline unsigned int get_pipes_num(struct device_queue_manager *dqm) +inline unsigned int get_pipes_num(struct device_queue_manager *dqm) { BUG_ON(!dqm || !dqm->dev); return dqm->dev->shared_resources.compute_pipe_count; @@ -78,7 +78,7 @@ static inline unsigned int get_pipes_num_cpsch(void) return PIPE_PER_ME_CP_SCHEDULING; } -static inline unsigned int +inline unsigned int get_sh_mem_bases_nybble_64(struct kfd_process_device *pdd) { uint32_t nybble; @@ -88,7 +88,7 @@ get_sh_mem_bases_nybble_64(struct kfd_process_device *pdd) return nybble; } -static inline unsigned int get_sh_mem_bases_32(struct kfd_process_device *pdd) +inline unsigned int get_sh_mem_bases_32(struct kfd_process_device *pdd) { unsigned int shared_base; @@ -97,41 +97,7 @@ static inline unsigned int get_sh_mem_bases_32(struct kfd_process_device *pdd) 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, @@ -391,6 +357,7 @@ 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); @@ -405,12 +372,13 @@ 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, @@ -455,34 +423,7 @@ 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_pipelines(struct device_queue_manager *dqm, +int init_pipelines(struct device_queue_manager *dqm, unsigned int pipes_num, unsigned int first_pipe) { void *hpdptr; @@ -715,7 +656,7 @@ static int initialize_cpsch(struct device_queue_manager *dqm) 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; @@ -1035,8 +976,7 @@ 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__); @@ -1073,18 +1013,13 @@ 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); @@ -1094,7 +1029,7 @@ 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); @@ -1107,6 +1042,8 @@ 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; @@ -1149,6 +1086,13 @@ struct device_queue_manager *device_queue_manager_init(struct kfd_dev *dev) break; } + switch (dev->device_info->asic_family) { + case CHIP_CARRIZO: + device_queue_manager_init_vi(&dqm->ops_asic_specific); + case CHIP_KAVERI: + device_queue_manager_init_cik(&dqm->ops_asic_specific); + } + if (dqm->ops.initialize(dqm) != 0) { kfree(dqm); return NULL; @@ -1164,4 +1108,3 @@ void device_queue_manager_uninit(struct device_queue_manager *dqm) dqm->ops.uninitialize(dqm); kfree(dqm); } - -- cgit v1.2.3 From f0ec5b9905ca8dfc44c6419dc068a335208398a7 Mon Sep 17 00:00:00 2001 From: Ben Goz Date: Tue, 13 Jan 2015 11:18:06 +0200 Subject: drm/amdkfd: Fix for-loop when allocating HQD (non-HWS) This patch fixes a minor bug in allocate_hqd(), where the loop run from the next-to-allocate pipe until the number of pipes. This is wrong because we need to consider the possibility where next-to-allocate pipe is not 0, and thus, the for-loop only checks part of the pipes and doesn't wrap-around, as it supposed to do. Therefore, we add another counting variable to make sure we go over all the pipes, regardless of where we start to look at the first iteration of the loop. This bug only affected non-HWS mode. In HWS mode, the CP fw is responsible for allocating the HQD. Signed-off-by: Ben Goz Signed-off-by: Oded Gabbay Acked-by: Alex Deucher --- drivers/gpu/drm/amd/amdkfd/kfd_device_queue_manager.c | 6 +++--- 1 file changed, 3 insertions(+), 3 deletions(-) (limited to 'drivers/gpu/drm/amd/amdkfd/kfd_device_queue_manager.c') 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 b2016245073e..fad6877d33ad 100644 --- a/drivers/gpu/drm/amd/amdkfd/kfd_device_queue_manager.c +++ b/drivers/gpu/drm/amd/amdkfd/kfd_device_queue_manager.c @@ -191,12 +191,12 @@ 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], -- cgit v1.2.3 From 99331a51cc788cb60780278cf2d6b1105cf23828 Mon Sep 17 00:00:00 2001 From: Oded Gabbay Date: Thu, 15 Jan 2015 12:01:10 +0200 Subject: drm/amdkfd: Replace cpu_relax() with schedule() in DQM In order not to occupy the current core and thus prevent the core from servicing IOMMU PPR requests, this patch replaces the call in DQM to cpu_relax() with a call to schedule(). Signed-off-by: Oded Gabbay Acked-by: Alex Deucher --- drivers/gpu/drm/amd/amdkfd/kfd_device_queue_manager.c | 3 ++- 1 file changed, 2 insertions(+), 1 deletion(-) (limited to 'drivers/gpu/drm/amd/amdkfd/kfd_device_queue_manager.c') 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 fad6877d33ad..e804e871ff82 100644 --- a/drivers/gpu/drm/amd/amdkfd/kfd_device_queue_manager.c +++ b/drivers/gpu/drm/amd/amdkfd/kfd_device_queue_manager.c @@ -26,6 +26,7 @@ #include #include #include +#include #include "kfd_priv.h" #include "kfd_device_queue_manager.h" #include "kfd_mqd_manager.h" @@ -829,7 +830,7 @@ 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; -- cgit v1.2.3 From 010b82e754e20b47bfaf6202c5b696f4b2849713 Mon Sep 17 00:00:00 2001 From: Oded Gabbay Date: Thu, 22 Jan 2015 11:09:27 +0200 Subject: drm/amdkfd: Remove negative check of uint variable Signed-off-by: Oded Gabbay Reviewed-by: Jammy Zhou --- drivers/gpu/drm/amd/amdkfd/kfd_device_queue_manager.c | 2 +- 1 file changed, 1 insertion(+), 1 deletion(-) (limited to 'drivers/gpu/drm/amd/amdkfd/kfd_device_queue_manager.c') 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 a5c69e96ba6f..23a1e95f32c3 100644 --- a/drivers/gpu/drm/amd/amdkfd/kfd_device_queue_manager.c +++ b/drivers/gpu/drm/amd/amdkfd/kfd_device_queue_manager.c @@ -587,7 +587,7 @@ static int allocate_sdma_queue(struct device_queue_manager *dqm, static void deallocate_sdma_queue(struct device_queue_manager *dqm, unsigned int sdma_queue_id) { - if (sdma_queue_id < 0 || sdma_queue_id >= CIK_SDMA_QUEUES) + if (sdma_queue_id >= CIK_SDMA_QUEUES) return; set_bit(sdma_queue_id, (unsigned long *)&dqm->sdma_bitmap); } -- cgit v1.2.3 From 300dec9578f77111d5fe0c236681b9e25a00cee7 Mon Sep 17 00:00:00 2001 From: Oded Gabbay Date: Thu, 22 Jan 2015 11:15:51 +0200 Subject: drm/amdkfd: Add break at the end of case Signed-off-by: Oded Gabbay Reviewed-by: Jammy Zhou --- drivers/gpu/drm/amd/amdkfd/kfd_device_queue_manager.c | 3 +++ 1 file changed, 3 insertions(+) (limited to 'drivers/gpu/drm/amd/amdkfd/kfd_device_queue_manager.c') 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 23a1e95f32c3..85387c8b296c 100644 --- a/drivers/gpu/drm/amd/amdkfd/kfd_device_queue_manager.c +++ b/drivers/gpu/drm/amd/amdkfd/kfd_device_queue_manager.c @@ -1114,8 +1114,11 @@ struct device_queue_manager *device_queue_manager_init(struct kfd_dev *dev) 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) { -- cgit v1.2.3 From 7113cd652969fe1a59a9560e700000633df6a3e1 Mon Sep 17 00:00:00 2001 From: Oded Gabbay Date: Thu, 22 Jan 2015 11:40:06 +0200 Subject: drm/amdkfd: Handle case of invalid queue type This patch handles a case where amdkfd tries to destroy a queue but the queue type is invalid. This case occurs in non-HWS path. Signed-off-by: Oded Gabbay Reviewed-by: Jammy Zhou --- drivers/gpu/drm/amd/amdkfd/kfd_device_queue_manager.c | 5 +++++ 1 file changed, 5 insertions(+) (limited to 'drivers/gpu/drm/amd/amdkfd/kfd_device_queue_manager.c') 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 85387c8b296c..99e2dbbb4862 100644 --- a/drivers/gpu/drm/amd/amdkfd/kfd_device_queue_manager.c +++ b/drivers/gpu/drm/amd/amdkfd/kfd_device_queue_manager.c @@ -301,6 +301,11 @@ static int destroy_queue_nocpsch(struct device_queue_manager *dqm, } 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, -- cgit v1.2.3 From 0b3674ae1c8b9539dde694a70391e974aedde8c2 Mon Sep 17 00:00:00 2001 From: Oded Gabbay Date: Thu, 22 Jan 2015 13:42:28 +0200 Subject: drm/amdkfd: Fix sparse errors Signed-off-by: Oded Gabbay Reviewed-by: Alex Deucher --- .../gpu/drm/amd/amdkfd/kfd_device_queue_manager.c | 28 ++-------------------- 1 file changed, 2 insertions(+), 26 deletions(-) (limited to 'drivers/gpu/drm/amd/amdkfd/kfd_device_queue_manager.c') 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 99e2dbbb4862..b189f9791c90 100644 --- a/drivers/gpu/drm/amd/amdkfd/kfd_device_queue_manager.c +++ b/drivers/gpu/drm/amd/amdkfd/kfd_device_queue_manager.c @@ -62,12 +62,6 @@ enum KFD_MQD_TYPE get_mqd_type_from_queue_type(enum kfd_queue_type type) return KFD_MQD_TYPE_CP; } -inline unsigned int get_pipes_num(struct device_queue_manager *dqm) -{ - BUG_ON(!dqm || !dqm->dev); - return dqm->dev->shared_resources.compute_pipe_count; -} - static inline unsigned int get_first_pipe(struct device_queue_manager *dqm) { BUG_ON(!dqm); @@ -79,25 +73,6 @@ static inline unsigned int get_pipes_num_cpsch(void) return PIPE_PER_ME_CP_SCHEDULING; } -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; -} - -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; -} - void program_sh_mem_settings(struct device_queue_manager *dqm, struct qcm_process_device *qpd) { @@ -336,7 +311,8 @@ static int update_queue(struct device_queue_manager *dqm, struct queue *q) BUG_ON(!dqm || !q || !q->mqd); mutex_lock(&dqm->lock); - mqd = dqm->ops.get_mqd_manager(dqm, q->properties.type); + 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; -- cgit v1.2.3 From b6819cec29cc0088e7398db9ed8377ab8328e012 Mon Sep 17 00:00:00 2001 From: Jay Cornwall Date: Mon, 19 Jan 2015 16:08:14 -0600 Subject: drm/amdkfd: Fix dqm->queue_count tracking dqm->queue_count tracks queues in the active state only. In a few places this count is modified unconditionally, leading to an incorrect value when the UPDATE_QUEUE ioctl is used to make a queue inactive. Signed-off-by: Jay Cornwall Reviewed-by: Ben Goz Signed-off-by: Oded Gabbay --- drivers/gpu/drm/amd/amdkfd/kfd_device_queue_manager.c | 9 ++++++--- 1 file changed, 6 insertions(+), 3 deletions(-) (limited to 'drivers/gpu/drm/amd/amdkfd/kfd_device_queue_manager.c') 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 ecc78ece634c..36b95e16eab1 100644 --- a/drivers/gpu/drm/amd/amdkfd/kfd_device_queue_manager.c +++ b/drivers/gpu/drm/amd/amdkfd/kfd_device_queue_manager.c @@ -167,7 +167,8 @@ 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++; @@ -313,7 +314,8 @@ static int destroy_queue_nocpsch(struct device_queue_manager *dqm, 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 @@ -1018,7 +1020,8 @@ static int destroy_queue_cpsch(struct device_queue_manager *dqm, dqm->sdma_queue_count--; list_del(&q->list); - dqm->queue_count--; + if (q->properties.is_active) + dqm->queue_count--; execute_queues_cpsch(dqm, false); -- cgit v1.2.3