diff options
Diffstat (limited to 'meta-amd-bsp/recipes-kernel/linux/linux-yocto-4.14.71/1294-drm-amdkfd-Add-wavefront-context-save-state-retrieva.patch')
-rw-r--r-- | meta-amd-bsp/recipes-kernel/linux/linux-yocto-4.14.71/1294-drm-amdkfd-Add-wavefront-context-save-state-retrieva.patch | 326 |
1 files changed, 326 insertions, 0 deletions
diff --git a/meta-amd-bsp/recipes-kernel/linux/linux-yocto-4.14.71/1294-drm-amdkfd-Add-wavefront-context-save-state-retrieva.patch b/meta-amd-bsp/recipes-kernel/linux/linux-yocto-4.14.71/1294-drm-amdkfd-Add-wavefront-context-save-state-retrieva.patch new file mode 100644 index 00000000..ca0eb1a4 --- /dev/null +++ b/meta-amd-bsp/recipes-kernel/linux/linux-yocto-4.14.71/1294-drm-amdkfd-Add-wavefront-context-save-state-retrieva.patch @@ -0,0 +1,326 @@ +From 73e2c7822b620b3ba71b33c5fca52fed217640b0 Mon Sep 17 00:00:00 2001 +From: Jay Cornwall <Jay.Cornwall@amd.com> +Date: Tue, 2 May 2017 17:39:37 -0500 +Subject: [PATCH 1294/4131] drm/amdkfd: Add wavefront context save state + retrieval ioctl + +Wavefront context save data is of interest to userspace clients for +debugging static wavefront state. The MQD contains two parameters +required to parse the control stack and the control stack itself +is kept in the MQD from gfx9 onwards. + +Add an ioctl to fetch the context save area and control stack offsets +and to copy the control stack to a userspace address if it is kept in +the MQD. + +Change-Id: I51145096e07e99511a9af67d5aa5cda28f344751 +Signed-off-by: Jay Cornwall <Jay.Cornwall@amd.com> +Signed-off-by: Felix Kuehling <Felix.Kuehling@amd.com> +--- + drivers/gpu/drm/amd/amdkfd/kfd_chardev.c | 23 +++++++++++++- + .../gpu/drm/amd/amdkfd/kfd_device_queue_manager.c | 37 ++++++++++++++++++++++ + .../gpu/drm/amd/amdkfd/kfd_device_queue_manager.h | 8 +++++ + drivers/gpu/drm/amd/amdkfd/kfd_mqd_manager.h | 8 +++++ + drivers/gpu/drm/amd/amdkfd/kfd_mqd_manager_v9.c | 24 ++++++++++++++ + drivers/gpu/drm/amd/amdkfd/kfd_mqd_manager_vi.c | 23 ++++++++++++++ + drivers/gpu/drm/amd/amdkfd/kfd_priv.h | 5 +++ + .../gpu/drm/amd/amdkfd/kfd_process_queue_manager.c | 22 +++++++++++++ + 8 files changed, 149 insertions(+), 1 deletion(-) + +diff --git a/drivers/gpu/drm/amd/amdkfd/kfd_chardev.c b/drivers/gpu/drm/amd/amdkfd/kfd_chardev.c +index 72c3109..fe7de4c 100644 +--- a/drivers/gpu/drm/amd/amdkfd/kfd_chardev.c ++++ b/drivers/gpu/drm/amd/amdkfd/kfd_chardev.c +@@ -2000,6 +2000,24 @@ static int kfd_ioctl_cross_memory_copy(struct file *filep, + } + #endif + ++static int kfd_ioctl_get_queue_wave_state(struct file *filep, ++ struct kfd_process *p, void *data) ++{ ++ struct kfd_ioctl_get_queue_wave_state_args *args = data; ++ int r; ++ ++ down_read(&p->lock); ++ ++ r = pqm_get_wave_state(&p->pqm, args->queue_id, ++ (void __user *)args->ctl_stack_address, ++ &args->ctl_stack_used_size, ++ &args->save_area_used_size); ++ ++ up_read(&p->lock); ++ ++ return r; ++} ++ + #define AMDKFD_IOCTL_DEF(ioctl, _func, _flags) \ + [_IOC_NR(ioctl)] = {.cmd = ioctl, .func = _func, .flags = _flags, \ + .cmd_drv = 0, .name = #ioctl} +@@ -2100,7 +2118,10 @@ static const struct amdkfd_ioctl_desc amdkfd_ioctls[] = { + kfd_ioctl_ipc_export_handle, 0), + + AMDKFD_IOCTL_DEF(AMDKFD_IOC_CROSS_MEMORY_COPY, +- kfd_ioctl_cross_memory_copy, 0) ++ kfd_ioctl_cross_memory_copy, 0), ++ ++ AMDKFD_IOCTL_DEF(AMDKFD_IOC_GET_QUEUE_WAVE_STATE, ++ kfd_ioctl_get_queue_wave_state, 0) + + }; + +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 172eddc..1c1535e 100644 +--- a/drivers/gpu/drm/amd/amdkfd/kfd_device_queue_manager.c ++++ b/drivers/gpu/drm/amd/amdkfd/kfd_device_queue_manager.c +@@ -1441,6 +1441,41 @@ static int process_termination_nocpsch(struct device_queue_manager *dqm, + return retval; + } + ++static int get_wave_state(struct device_queue_manager *dqm, ++ struct queue *q, ++ void __user *ctl_stack, ++ u32 *ctl_stack_used_size, ++ u32 *save_area_used_size) ++{ ++ struct mqd_manager *mqd; ++ int r; ++ ++ mutex_lock(&dqm->lock); ++ ++ if (q->properties.type != KFD_QUEUE_TYPE_COMPUTE || ++ q->properties.is_active || !q->device->cwsr_enabled) { ++ r = -EINVAL; ++ goto dqm_unlock; ++ } ++ ++ mqd = dqm->ops.get_mqd_manager(dqm, KFD_MQD_TYPE_COMPUTE); ++ if (!mqd) { ++ r = -ENOMEM; ++ goto dqm_unlock; ++ } ++ ++ if (!mqd->get_wave_state) { ++ r = -EINVAL; ++ goto dqm_unlock; ++ } ++ ++ r = mqd->get_wave_state(mqd, q->mqd, ctl_stack, ctl_stack_used_size, ++ save_area_used_size); ++ ++dqm_unlock: ++ mutex_unlock(&dqm->lock); ++ return r; ++} + + static int process_termination_cpsch(struct device_queue_manager *dqm, + struct qcm_process_device *qpd) +@@ -1545,6 +1580,7 @@ struct device_queue_manager *device_queue_manager_init(struct kfd_dev *dev) + dqm->ops.set_cache_memory_policy = set_cache_memory_policy; + dqm->ops.set_trap_handler = set_trap_handler; + dqm->ops.process_termination = process_termination_cpsch; ++ dqm->ops.get_wave_state = get_wave_state; + break; + case KFD_SCHED_POLICY_NO_HWS: + /* initialize dqm for no cp scheduling */ +@@ -1561,6 +1597,7 @@ struct device_queue_manager *device_queue_manager_init(struct kfd_dev *dev) + dqm->ops.set_cache_memory_policy = set_cache_memory_policy; + dqm->ops.set_trap_handler = set_trap_handler; + dqm->ops.process_termination = process_termination_nocpsch; ++ dqm->ops.get_wave_state = get_wave_state; + break; + default: + pr_err("Invalid scheduling policy %d\n", dqm->sched_policy); +diff --git a/drivers/gpu/drm/amd/amdkfd/kfd_device_queue_manager.h b/drivers/gpu/drm/amd/amdkfd/kfd_device_queue_manager.h +index 113ac1f..841283a 100644 +--- a/drivers/gpu/drm/amd/amdkfd/kfd_device_queue_manager.h ++++ b/drivers/gpu/drm/amd/amdkfd/kfd_device_queue_manager.h +@@ -85,6 +85,8 @@ struct device_process_node { + * + * @process_termination: Clears all process queues belongs to that device. + * ++ * @get_wave_state: Retrieves context save state and optionally copies the ++ * control stack, if kept in the MQD, to the given userspace address. + */ + + struct device_queue_manager_ops { +@@ -136,6 +138,12 @@ struct device_queue_manager_ops { + + int (*process_termination)(struct device_queue_manager *dqm, + struct qcm_process_device *qpd); ++ ++ int (*get_wave_state)(struct device_queue_manager *dqm, ++ struct queue *q, ++ void __user *ctl_stack, ++ u32 *ctl_stack_used_size, ++ u32 *save_area_used_size); + }; + + struct device_queue_manager_asic_ops { +diff --git a/drivers/gpu/drm/amd/amdkfd/kfd_mqd_manager.h b/drivers/gpu/drm/amd/amdkfd/kfd_mqd_manager.h +index 6f33b79..dcaeda8 100644 +--- a/drivers/gpu/drm/amd/amdkfd/kfd_mqd_manager.h ++++ b/drivers/gpu/drm/amd/amdkfd/kfd_mqd_manager.h +@@ -43,6 +43,9 @@ + * + * @is_occupied: Checks if the relevant HQD slot is occupied. + * ++ * @get_wave_state: Retrieves context save state and optionally copies the ++ * control stack, if kept in the MQD, to the given userspace address. ++ * + * @mqd_mutex: Mqd manager mutex. + * + * @dev: The kfd device structure coupled with this module. +@@ -86,6 +89,11 @@ struct mqd_manager { + uint64_t queue_address, uint32_t pipe_id, + uint32_t queue_id); + ++ int (*get_wave_state)(struct mqd_manager *mm, void *mqd, ++ void __user *ctl_stack, ++ u32 *ctl_stack_used_size, ++ u32 *save_area_used_size); ++ + #if defined(CONFIG_DEBUG_FS) + int (*debugfs_show_mqd)(struct seq_file *m, void *data); + #endif +diff --git a/drivers/gpu/drm/amd/amdkfd/kfd_mqd_manager_v9.c b/drivers/gpu/drm/amd/amdkfd/kfd_mqd_manager_v9.c +index 7327cda..1e956ff 100644 +--- a/drivers/gpu/drm/amd/amdkfd/kfd_mqd_manager_v9.c ++++ b/drivers/gpu/drm/amd/amdkfd/kfd_mqd_manager_v9.c +@@ -23,6 +23,7 @@ + + #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" +@@ -292,6 +293,28 @@ static bool is_occupied(struct mqd_manager *mm, void *mqd, + pipe_id, queue_id); + } + ++static int get_wave_state(struct mqd_manager *mm, void *mqd, ++ void __user *ctl_stack, ++ u32 *ctl_stack_used_size, ++ u32 *save_area_used_size) ++{ ++ struct v9_mqd *m; ++ ++ /* Control stack is located one page after MQD. */ ++ void *mqd_ctl_stack = (void *)((uintptr_t)mqd + PAGE_SIZE); ++ ++ m = get_mqd(mqd); ++ ++ *ctl_stack_used_size = m->cp_hqd_cntl_stack_size - ++ m->cp_hqd_cntl_stack_offset; ++ *save_area_used_size = m->cp_hqd_wg_state_offset; ++ ++ if (copy_to_user(ctl_stack, mqd_ctl_stack, m->cp_hqd_cntl_stack_size)) ++ return -EFAULT; ++ ++ return 0; ++} ++ + static int init_mqd_hiq(struct mqd_manager *mm, void **mqd, + struct kfd_mem_obj **mqd_mem_obj, uint64_t *gart_addr, + struct queue_properties *q) +@@ -468,6 +491,7 @@ struct mqd_manager *mqd_manager_init_v9(enum KFD_MQD_TYPE type, + mqd->update_mqd = update_mqd; + mqd->destroy_mqd = destroy_mqd; + mqd->is_occupied = is_occupied; ++ mqd->get_wave_state = get_wave_state; + #if defined(CONFIG_DEBUG_FS) + mqd->debugfs_show_mqd = debugfs_show_mqd; + #endif +diff --git a/drivers/gpu/drm/amd/amdkfd/kfd_mqd_manager_vi.c b/drivers/gpu/drm/amd/amdkfd/kfd_mqd_manager_vi.c +index c1d35f4..670bbba9 100644 +--- a/drivers/gpu/drm/amd/amdkfd/kfd_mqd_manager_vi.c ++++ b/drivers/gpu/drm/amd/amdkfd/kfd_mqd_manager_vi.c +@@ -309,6 +309,28 @@ static bool is_occupied(struct mqd_manager *mm, void *mqd, + pipe_id, queue_id); + } + ++static int get_wave_state(struct mqd_manager *mm, void *mqd, ++ void __user *ctl_stack, ++ u32 *ctl_stack_used_size, ++ u32 *save_area_used_size) ++{ ++ struct vi_mqd *m; ++ ++ m = get_mqd(mqd); ++ ++ *ctl_stack_used_size = m->cp_hqd_cntl_stack_size - ++ m->cp_hqd_cntl_stack_offset; ++ *save_area_used_size = m->cp_hqd_wg_state_offset - ++ m->cp_hqd_cntl_stack_size; ++ ++ /* Control stack is not copied to user mode for GFXv8 because ++ * it's part of the context save area that is already ++ * accessible to user mode ++ */ ++ ++ return 0; ++} ++ + 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) +@@ -480,6 +502,7 @@ struct mqd_manager *mqd_manager_init_vi(enum KFD_MQD_TYPE type, + mqd->update_mqd = update_mqd; + mqd->destroy_mqd = destroy_mqd; + mqd->is_occupied = is_occupied; ++ mqd->get_wave_state = get_wave_state; + #if defined(CONFIG_DEBUG_FS) + mqd->debugfs_show_mqd = debugfs_show_mqd; + #endif +diff --git a/drivers/gpu/drm/amd/amdkfd/kfd_priv.h b/drivers/gpu/drm/amd/amdkfd/kfd_priv.h +index 9777543..ad84996 100644 +--- a/drivers/gpu/drm/amd/amdkfd/kfd_priv.h ++++ b/drivers/gpu/drm/amd/amdkfd/kfd_priv.h +@@ -935,6 +935,11 @@ int pqm_set_cu_mask(struct process_queue_manager *pqm, unsigned int qid, + struct queue_properties *p); + struct kernel_queue *pqm_get_kernel_queue(struct process_queue_manager *pqm, + unsigned int qid); ++int pqm_get_wave_state(struct process_queue_manager *pqm, ++ unsigned int qid, ++ void __user *ctl_stack, ++ u32 *ctl_stack_used_size, ++ u32 *save_area_used_size); + int kgd2kfd_quiesce_mm(struct kfd_dev *kfd, struct mm_struct *mm); + int kgd2kfd_resume_mm(struct kfd_dev *kfd, struct mm_struct *mm); + +diff --git a/drivers/gpu/drm/amd/amdkfd/kfd_process_queue_manager.c b/drivers/gpu/drm/amd/amdkfd/kfd_process_queue_manager.c +index f249f96..0f82684 100644 +--- a/drivers/gpu/drm/amd/amdkfd/kfd_process_queue_manager.c ++++ b/drivers/gpu/drm/amd/amdkfd/kfd_process_queue_manager.c +@@ -421,6 +421,28 @@ struct kernel_queue *pqm_get_kernel_queue( + return NULL; + } + ++int pqm_get_wave_state(struct process_queue_manager *pqm, ++ unsigned int qid, ++ void __user *ctl_stack, ++ u32 *ctl_stack_used_size, ++ u32 *save_area_used_size) ++{ ++ struct process_queue_node *pqn; ++ ++ pqn = get_queue_by_qid(pqm, qid); ++ if (!pqn) { ++ pr_debug("amdkfd: No queue %d exists for operation\n", ++ qid); ++ return -EFAULT; ++ } ++ ++ return pqn->q->device->dqm->ops.get_wave_state(pqn->q->device->dqm, ++ pqn->q, ++ ctl_stack, ++ ctl_stack_used_size, ++ save_area_used_size); ++} ++ + #if defined(CONFIG_DEBUG_FS) + + int pqm_debugfs_mqds(struct seq_file *m, void *data) +-- +2.7.4 + |