drm/amdkfd: Have kfd driver use same PASID values from graphic driver
Subsystems (3)
PATCH 1/1 drm/amdkfd: Have kfd driver use same PASID values from graphic driver
From: Xiaogang Chen <[email protected]>
Current kfd driver has its own PASID value for a kfd process and uses it to locate vm at interrupt handler or mapping between kfd process and vm. That design is not working when a physical gpu device has multiple spatial partitions, ex: adev in CPX mode. This patch has kfd driver use same pasid values that graphic driver generated which is per vm per pasid.
These pasid values are passed to fw/hardware. We do not need change interrupt handler though more pasid values are used. Also, pasid values at log are replaced by user process pid, or pasid values are not exposed to user, user will see their process pids that have meaning in user space.
--- drivers/gpu/drm/amd/amdgpu/amdgpu_amdkfd.h | 4 +- .../gpu/drm/amd/amdgpu/amdgpu_amdkfd_gpuvm.c | 34 +++++---- drivers/gpu/drm/amd/amdkfd/kfd_chardev.c | 25 +++---- drivers/gpu/drm/amd/amdkfd/kfd_debug.c | 8 +-- .../drm/amd/amdkfd/kfd_device_queue_manager.c | 52 +++++++------- drivers/gpu/drm/amd/amdkfd/kfd_events.c | 17 ++--- .../drm/amd/amdkfd/kfd_packet_manager_v9.c | 4 +- .../drm/amd/amdkfd/kfd_packet_manager_vi.c | 3 +- drivers/gpu/drm/amd/amdkfd/kfd_priv.h | 5 +- drivers/gpu/drm/amd/amdkfd/kfd_process.c | 71 ++++++++----------- .../amd/amdkfd/kfd_process_queue_manager.c | 10 +-- drivers/gpu/drm/amd/amdkfd/kfd_svm.c | 17 +++-- 12 files changed, 127 insertions(+), 123 deletions(-)
drivers/gpu/drm/amd/amdgpu/amdgpu_amdkfd.h +3 -1
@@ -47,6 +47,7 @@ enum TLB_FLUSH_TYPE { };struct amdgpu_device ; +struct kfd_process_device ;struct amdgpu_reset_context ;enum kfd_mem_attachment_type {
@@ -300,7 +301,8 @@ bool amdgpu_amdkfd_compute_active(struct amdgpu_device *adev, uint32_t node_id); ((struct drm_file *)(drm_priv))->driver_priv)->vm)int amdgpu_amdkfd_gpuvm_set_vm_pasid (struct amdgpu_device *adev, -struct amdgpu_vm *avm,u32 pasid); +struct kfd_process_device *pdd, +struct amdgpu_vm *avm); int amdgpu_amdkfd_gpuvm_acquire_process_vm(struct amdgpu_device *adev, struct amdgpu_vm *avm, void **process_info,
drivers/gpu/drm/amd/amdgpu/amdgpu_amdkfd_gpuvm.c +20 -14
@@ -1530,24 +1530,30 @@ static void amdgpu_amdkfd_gpuvm_unpin_bo(struct amdgpu_bo *bo) }int amdgpu_amdkfd_gpuvm_set_vm_pasid (struct amdgpu_device *adev, -struct amdgpu_vm *avm,u32 pasid) +struct kfd_process_device *pdd, +struct amdgpu_vm *avm) { -int ret; - -/* Free the original amdgpu allocated pasid, -* will be replaced with kfd allocated pasid. -*/ -if (avm->pasid) { -amdgpu_pasid_free (avm->pasid); -amdgpu_vm_set_pasid (adev, avm,0 ); +int ret =0 ; +int pasid; + +/* if avm has pasid assigned use it */ +if (avm->pasid) + pdd->pasid = avm->pasid; +else { +/* otherwise generate a new pasid and assign to avm */ + pasid =amdgpu_pasid_alloc (16 ); +if (pasid <0 ) { +dev_warn (adev->dev,"No more PASIDs available!" ); + ret = -ENOMEM ; + }else + ret =xa_err (xa_store_irq (&adev->vm_manager.pasids, pasid, avm, +GFP_KERNEL )); +if (ret >=0 ) + avm->pasid = pasid; } - ret =amdgpu_vm_set_pasid (adev, avm, pasid); -if (ret) -return ret; - -return 0 ; +return ret; } int amdgpu_amdkfd_gpuvm_acquire_process_vm(struct amdgpu_device *adev,
drivers/gpu/drm/amd/amdkfd/kfd_chardev.c +13 -12
@@ -155,8 +155,8 @@ static int kfd_open(struct inode *inode, struct file *filep)/* filep now owns the reference returned by kfd_create_process */ filep->private_data = process; -dev_dbg (kfd_device,"process %d opened, compat mode (32 bit) - %d\n" , - process->pasid, process->is_32bit_user_mode); +dev_dbg (kfd_device,"process pid %d opened kfd node, compat mode (32 bit) - %d\n" , + (int )process->lead_thread->pid, process->is_32bit_user_mode);return 0 ; }
@@ -361,8 +361,8 @@ static int kfd_ioctl_create_queue(struct file *filep, struct kfd_process *p, goto err_acquire_queue_buf; } -pr_debug ("Creating queue for PASID 0x%x on gpu 0x%x\n" , - p->pasid, +pr_debug ("Creating queue for process pid 0x%x on gpu 0x%x\n" , + (int )p->lead_thread->pid, dev->id); err = pqm_create_queue(&p->pqm, dev, &q_properties, &queue_id,
@@ -415,9 +415,9 @@ static int kfd_ioctl_destroy_queue(struct file *filp, struct kfd_process *p,int retval;struct kfd_ioctl_destroy_queue_args *args = data; -pr_debug ("Destroying queue id %d for pasid 0x%x\n" , +pr_debug ("Destroying queue id %d for process pid 0x%x\n" , args->queue_id, - p->pasid); + (int )p->lead_thread->pid);mutex_lock (&p->mutex);
@@ -468,8 +468,8 @@ static int kfd_ioctl_update_queue(struct file *filp, struct kfd_process *p, properties.pm4_target_xcc = (args->queue_percentage >>8 ) &0xFF ; properties.priority = args->queue_priority; -pr_debug ("Updating queue id %d for pasid 0x%x\n" , - args->queue_id, p->pasid); +pr_debug ("Updating queue id %d for process pid 0x%x\n" , + args->queue_id, (int )p->lead_thread->pid);mutex_lock (&p->mutex);
@@ -695,7 +695,7 @@ static int kfd_ioctl_get_process_apertures(struct file *filp,struct kfd_process_device_apertures *pAperture;int i; -dev_dbg (kfd_device,"get apertures for PASID 0x%x" , p->pasid); +dev_dbg (kfd_device,"get apertures for process pid 0x%x" , (int )p->lead_thread->pid); args->num_of_nodes =0 ;
@@ -747,7 +747,8 @@ static int kfd_ioctl_get_process_apertures_new(struct file *filp,int ret;int i; -dev_dbg (kfd_device,"get apertures for PASID 0x%x" , p->pasid); +dev_dbg (kfd_device,"get apertures for process pid 0x%x" , + (int )p->lead_thread->pid);if (args->num_of_nodes ==0 ) {/* Return number of nodes, so that user space can alloacate
@@ -3365,12 +3366,12 @@ static int kfd_mmio_mmap(struct kfd_node *dev, struct kfd_process *process, vma->vm_page_prot =pgprot_noncached (vma->vm_page_prot); -pr_debug ("pasid 0x%x mapping mmio page\n" +pr_debug ("process pid 0x%x mapping mmio page\n" " target user address == 0x%08llX\n" " physical address == 0x%08llX\n" " vm_flags == 0x%04lX\n" " size == 0x%04lX\n" , - process->pasid, (unsigned long long ) vma->vm_start, + (int )process->lead_thread->pid, (unsigned long long ) vma->vm_start, address, vma->vm_flags,PAGE_SIZE ); return io_remap_pfn_range(vma,
drivers/gpu/drm/amd/amdkfd/kfd_debug.c +4 -4
@@ -238,8 +238,8 @@ bool kfd_set_dbg_ev_from_interrupt(struct kfd_node *dev,mutex_unlock (&p->mutex); }else if (trap_mask &KFD_EC_MASK (EC_DEVICE_MEMORY_VIOLATION )) { -kfd_dqm_evict_pasid (dev->dqm, p->pasid); -kfd_signal_vm_fault_event (dev, p->pasid,NULL , +kfd_dqm_evict_pasid (dev->dqm, pasid); +kfd_signal_vm_fault_event (dev, pasid,NULL , exception_data); signaled_to_debugger_or_runtime = true;
@@ -276,8 +276,8 @@ int kfd_dbg_send_exception_to_runtime(struct kfd_process *p, data = (struct kfd_hsa_memory_exception_data *) pdd->vm_fault_exc_data; -kfd_dqm_evict_pasid (pdd->dev->dqm, p->pasid); -kfd_signal_vm_fault_event (pdd->dev, p->pasid,NULL , data); +kfd_dqm_evict_pasid (pdd->dev->dqm, pdd->pasid); +kfd_signal_vm_fault_event (pdd->dev, pdd->pasid,NULL , data); error_reason &= ~KFD_EC_MASK (EC_DEVICE_MEMORY_VIOLATION ); }
drivers/gpu/drm/amd/amdkfd/kfd_device_queue_manager.c +27 -25
133 lines not on this page · raw message
drivers/gpu/drm/amd/amdkfd/kfd_events.c +9 -8
@@ -1128,8 +1128,8 @@ static void lookup_events_by_type_and_signal(struct kfd_process *p,if (type ==KFD_EVENT_TYPE_MEMORY ) {dev_warn (kfd_device, -"Sending SIGSEGV to process %d (pasid 0x%x)" , - p->lead_thread->pid, p->pasid); +"Sending SIGSEGV to process pid %d" , + p->lead_thread->pid);send_sig (SIGSEGV , p->lead_thread,0 ); }
@@ -1137,13 +1137,13 @@ static void lookup_events_by_type_and_signal(struct kfd_process *p,if (send_signal) {if (send_sigterm) {dev_warn (kfd_device, -"Sending SIGTERM to process %d (pasid 0x%x)" , - p->lead_thread->pid, p->pasid); +"Sending SIGTERM to process pid %d" , + p->lead_thread->pid);send_sig (SIGTERM , p->lead_thread,0 ); }else {dev_err (kfd_device, -"Process %d (pasid 0x%x) got unhandled exception" , - p->lead_thread->pid, p->pasid); +"Process pid %d got unhandled exception" , + p->lead_thread->pid); } }
@@ -1253,7 +1253,8 @@ void kfd_signal_reset_event(struct kfd_node *dev) }if (unlikely (!pdd)) { -WARN_ONCE (1 ,"Could not get device data from pasid:0x%x\n" , p->pasid); +WARN_ONCE (1 ,"Could not get device data from process pid:0x%x\n" , + p->lead_thread->pid);continue ; }
@@ -1263,7 +1264,7 @@ void kfd_signal_reset_event(struct kfd_node *dev)if (dev->dqm->detect_hang_count) {struct amdgpu_task_info *ti; - ti =amdgpu_vm_get_task_info_pasid (dev->adev, p->pasid); + ti =amdgpu_vm_get_task_info_pasid (dev->adev, pdd->pasid);if (ti) { dev_err(dev->adev->dev, "Queues reset on process %s tid %d thread %s pid %d\n",
drivers/gpu/drm/amd/amdkfd/kfd_packet_manager_v9.c +2 -2
@@ -47,7 +47,7 @@ static int pm_map_process_v9(struct packet_manager *pm, packet->bitfields2.exec_cleaner_shader =1 ; packet->bitfields2.diq_enable = (qpd->is_debug) ?1 :0 ; packet->bitfields2.process_quantum =10 ; - packet->bitfields2.pasid = qpd->pqm->process->pasid; + packet->bitfields2.pasid = pdd->pasid; packet->bitfields14.gds_size = qpd->gds_size &0x3F ; packet->bitfields14.gds_size_hi = (qpd->gds_size >>6 ) &0xF ; packet->bitfields14.num_gws = (qpd->mapped_gws_queue) ? qpd->num_gws :0 ;
@@ -106,7 +106,7 @@ static int pm_map_process_aldebaran(struct packet_manager *pm, packet->bitfields2.exec_cleaner_shader =1 ; packet->bitfields2.diq_enable = (qpd->is_debug) ?1 :0 ; packet->bitfields2.process_quantum =10 ; - packet->bitfields2.pasid = qpd->pqm->process->pasid; + packet->bitfields2.pasid = pdd->pasid; packet->bitfields14.gds_size = qpd->gds_size &0x3F ; packet->bitfields14.gds_size_hi = (qpd->gds_size >>6 ) &0xF ; packet->bitfields14.num_gws = (qpd->mapped_gws_queue) ? qpd->num_gws :0 ;
drivers/gpu/drm/amd/amdkfd/kfd_packet_manager_vi.c +2 -1
@@ -42,6 +42,7 @@ unsigned int pm_build_pm4_header(unsigned int opcode, size_t packet_size)static int pm_map_process_vi (struct packet_manager *pm,uint32_t *buffer,struct qcm_process_device *qpd) { +struct kfd_process_device *pdd =qpd_to_pdd (qpd);struct pm4_mes_map_process *packet; packet = (struct pm4_mes_map_process *)buffer;
@@ -52,7 +53,7 @@ static int pm_map_process_vi(struct packet_manager *pm, uint32_t *buffer, sizeof(struct pm4_mes_map_process)); packet->bitfields2.diq_enable = (qpd->is_debug) ?1 :0 ; packet->bitfields2.process_quantum =10 ; - packet->bitfields2.pasid = qpd->pqm->process->pasid; + packet->bitfields2.pasid = pdd->pasid; packet->bitfields3.page_table_base = qpd->page_table_base; packet->bitfields10.gds_size = qpd->gds_size; packet->bitfields10.num_gws = qpd->num_gws;
drivers/gpu/drm/amd/amdkfd/kfd_priv.h +2 -3
@@ -849,6 +849,8 @@ struct kfd_process_device {/* Tracks queue reset status */ bool has_reset_queue; + +u32 pasid; };#define qpd_to_pdd (x) container_of(x, struct kfd_process_device, qpd)
@@ -908,8 +910,6 @@ struct kfd_process {/* We want to receive a notification when the mm_struct is destroyed */ struct mmu_notifier mmu_notifier; -u32 pasid; -/* * Array of kfd_process_device pointers, * one for each device the process is using.
@@ -970,7 +970,6 @@ struct kfd_process {/* Kobj for our procfs */ struct kobject *kobj;struct kobject *kobj_queues; -struct attribute attr_pasid;/* Keep track cwsr init */ bool has_cwsr;
drivers/gpu/drm/amd/amdkfd/kfd_process.c +30 -41
192 lines not on this page · raw message
drivers/gpu/drm/amd/amdkfd/kfd_process_queue_manager.c +5 -5
@@ -69,8 +69,8 @@ static int find_available_queue_slot(struct process_queue_manager *pqm,pr_debug ("The new slot id %lu\n" , found);if (found >=KFD_MAX_NUM_OF_QUEUES_PER_PROCESS ) { -pr_info ("Cannot open more queues for process with pasid 0x%x\n" , - pqm->process->pasid); +pr_info ("Cannot open more queues for process with pid 0x%x\n" , + pqm->process->lead_thread->pid);return -ENOMEM ; }
@@ -425,8 +425,8 @@ int pqm_create_queue(struct process_queue_manager *pqm, }if (retval !=0 ) { -pr_err ("Pasid 0x%x DQM create queue type %d failed. ret %d\n" , - pqm->process->pasid, type, retval); +pr_err ("process pid 0x%x DQM create queue type %d failed. ret %d\n" , + pqm->process->lead_thread->pid, type, retval); goto err_create_queue; }
@@ -520,7 +520,7 @@ int pqm_destroy_queue(struct process_queue_manager *pqm, unsigned int qid) retval = dqm->ops.destroy_queue (dqm, &pdd->qpd, pqn->q);if (retval) {pr_err ("Pasid 0x%x destroy queue %d failed, ret %d\n" , - pqm->process->pasid, + pdd->pasid, pqn->q->properties.queue_id, retval);if (retval != -ETIME ) goto err_destroy_queue;
drivers/gpu/drm/amd/amdkfd/kfd_svm.c +10 -7
@@ -563,7 +563,8 @@ svm_range_vram_node_new(struct kfd_node *node, struct svm_range *prange,int r; p = container_of(prange->svms, struct kfd_process, svms); -pr_debug ("pasid: %x svms 0x%p [0x%lx 0x%lx]\n" , p->pasid, prange->svms, +pr_debug ("process pid: %x svms 0x%p [0x%lx 0x%lx]\n" , + (int )p->lead_thread->pid, prange->svms, prange->start, prange->last);if (svm_range_validate_svm_bo (node, prange))
@@ -3232,7 +3233,8 @@ void svm_range_list_fini(struct kfd_process *p)struct svm_range *prange;struct svm_range *next; -pr_debug ("pasid 0x%x svms 0x%p\n" , p->pasid, &p->svms); +pr_debug ("process pid 0x%x svms 0x%p\n" , (int )p->lead_thread->pid, + &p->svms);cancel_delayed_work_sync (&p->svms.restore_work);
@@ -3255,7 +3257,8 @@ void svm_range_list_fini(struct kfd_process *p)mutex_destroy (&p->svms.lock); -pr_debug ("pasid 0x%x svms 0x%p done\n" , p->pasid, &p->svms); +pr_debug ("process pid 0x%x svms 0x%p done\n" , + (int )p->lead_thread->pid, &p->svms); }int svm_range_list_init (struct kfd_process *p)
@@ -3618,8 +3621,8 @@ svm_range_set_attr(struct kfd_process *p, struct mm_struct *mm,bool flush_tlb;int r, ret =0 ; -pr_debug ("pasid 0x%x svms 0x%p [0x%llx 0x%llx] pages 0x%llx\n" , - p->pasid, &p->svms, start, start + size -1 , size); +pr_debug ("process pid 0x%x svms 0x%p [0x%llx 0x%llx] pages 0x%llx\n" , + (int )p->lead_thread->pid, &p->svms, start, start + size -1 , size); r =svm_range_check_attr (p, nattr, attrs);if (r)
@@ -3727,8 +3730,8 @@ svm_range_set_attr(struct kfd_process *p, struct mm_struct *mm, out:mutex_unlock (&process_info->lock); -pr_debug ("pasid 0x%x svms 0x%p [0x%llx 0x%llx] done, r=%d\n" , p->pasid, - &p->svms, start, start + size -1 , r); +pr_debug ("process pid 0x%x svms 0x%p [0x%llx 0x%llx] done, r=%d\n" , + (int )p->lead_thread->pid, &p->svms, start, start + size -1 , r);return ret ? ret : r; }
Signature · 1 line
2.25.1
Felix Kuehling
maintainerQuoted 56 lines · Xiaogang.Chen
From: Xiaogang Chen <[email protected]>
Current kfd driver has its own PASID value for a kfd process and uses it to locate vm at interrupt handler or mapping between kfd process and vm. That design is not working when a physical gpu device has multiple spatial partitions, ex: adev in CPX mode. This patch has kfd driver use same pasid values that graphic driver generated which is per vm per pasid.
These pasid values are passed to fw/hardware. We do not need change interrupt handler though more pasid values are used. Also, pasid values at log are replaced by user process pid, or pasid values are not exposed to user, user will see their process pids that have meaning in user space.
Signed-off-by Xiaogang Chen <[email protected]>--- drivers/gpu/drm/amd/amdgpu/amdgpu_amdkfd.h | 4 +- .../gpu/drm/amd/amdgpu/amdgpu_amdkfd_gpuvm.c | 34 +++++---- drivers/gpu/drm/amd/amdkfd/kfd_chardev.c | 25 +++---- drivers/gpu/drm/amd/amdkfd/kfd_debug.c | 8 +-- .../drm/amd/amdkfd/kfd_device_queue_manager.c | 52 +++++++------- drivers/gpu/drm/amd/amdkfd/kfd_events.c | 17 ++--- .../drm/amd/amdkfd/kfd_packet_manager_v9.c | 4 +- .../drm/amd/amdkfd/kfd_packet_manager_vi.c | 3 +- drivers/gpu/drm/amd/amdkfd/kfd_priv.h | 5 +- drivers/gpu/drm/amd/amdkfd/kfd_process.c | 71 ++++++++----------- .../amd/amdkfd/kfd_process_queue_manager.c | 10 +-- drivers/gpu/drm/amd/amdkfd/kfd_svm.c | 17 +++-- 12 files changed, 127 insertions(+), 123 deletions(-)drivers/gpu/drm/amd/amdgpu/amdgpu_amdkfd.h +3
-1@@ -47,6 +47,7 @@ enum TLB_FLUSH_TYPE { };struct amdgpu_device ; +struct kfd_process_device ;struct amdgpu_reset_context ;enum kfd_mem_attachment_type {@@ -300,7 +301,8 @@ bool amdgpu_amdkfd_compute_active(struct amdgpu_device *adev, uint32_t node_id); ((struct drm_file *)(drm_priv))->driver_priv)->vm)int amdgpu_amdkfd_gpuvm_set_vm_pasid (struct amdgpu_device *adev, -struct amdgpu_vm *avm,u32 pasid); +struct kfd_process_device *pdd, +struct amdgpu_vm *avm); int amdgpu_amdkfd_gpuvm_acquire_process_vm(struct amdgpu_device *adev, struct amdgpu_vm *avm, void **process_info,drivers/gpu/drm/amd/amdgpu/amdgpu_amdkfd_gpuvm.c +0
-0
drivers/gpu/drm/amd/amdgpu/amdgpu_amdkfd_gpuvm.c static void amdgpu_amdkfd_gpuvm_unpin_bo(struct amdgpu_bo *bo) · line 1534
Quoted 6 lines
drivers/gpu/drm/amd/amdgpu/amdgpu_amdkfd_gpuvm.c +2
-1@@ -1530,24 +1530,30 @@ static void amdgpu_amdkfd_gpuvm_unpin_bo(struct amdgpu_bo *bo) }int amdgpu_amdkfd_gpuvm_set_vm_pasid (struct amdgpu_device *adev, -struct amdgpu_vm *avm,u32 pasid) +struct kfd_process_device *pdd, +struct amdgpu_vm *avm)
I'd prefer if this function got renamed to amdgpu_amdkfd_gpuvm_get_pasid and it returns the pasid.
Quoted 8 lines
drivers/gpu/drm/amd/amdgpu/amdgpu_amdkfd_gpuvm.c +0
-6{ -int ret; - -/* Free the original amdgpu allocated pasid, -* will be replaced with kfd allocated pasid. -*/ -if (avm->pasid) {
drivers/gpu/drm/amd/amdgpu/amdgpu_amdkfd_gpuvm.c static void amdgpu_amdkfd_gpuvm_unpin_bo(struct amdgpu_bo *bo) · line 1544
Quoted 10 lines
drivers/gpu/drm/amd/amdgpu/amdgpu_amdkfd_gpuvm.c +8
-2-amdgpu_pasid_free (avm->pasid); -amdgpu_vm_set_pasid (adev, avm,0 ); +int ret =0 ; +int pasid; + +/* if avm has pasid assigned use it */ +if (avm->pasid) + pdd->pasid = avm->pasid; +else { +/* otherwise generate a new pasid and assign to avm */
Does this ever happen? I think any avm that was acquired properly should already have a PASID. If it doesn't, maybe we can just return an error.
Quoted 119 lines
drivers/gpu/drm/amd/amdgpu/amdgpu_amdkfd_gpuvm.c +10
-5+ pasid =amdgpu_pasid_alloc (16 ); +if (pasid <0 ) { +dev_warn (adev->dev,"No more PASIDs available!" ); + ret = -ENOMEM ; + }else + ret =xa_err (xa_store_irq (&adev->vm_manager.pasids, pasid, avm, +GFP_KERNEL )); +if (ret >=0 ) + avm->pasid = pasid; } - ret =amdgpu_vm_set_pasid (adev, avm, pasid); -if (ret) -return ret; - -return 0 ; +return ret; } int amdgpu_amdkfd_gpuvm_acquire_process_vm(struct amdgpu_device *adev,drivers/gpu/drm/amd/amdkfd/kfd_chardev.c +13
-12@@ -155,8 +155,8 @@ static int kfd_open(struct inode *inode, struct file *filep)/* filep now owns the reference returned by kfd_create_process */ filep->private_data = process; -dev_dbg (kfd_device,"process %d opened, compat mode (32 bit) - %d\n" , - process->pasid, process->is_32bit_user_mode); +dev_dbg (kfd_device,"process pid %d opened kfd node, compat mode (32 bit) - %d\n" , + (int )process->lead_thread->pid, process->is_32bit_user_mode);return 0 ; }@@ -361,8 +361,8 @@ static int kfd_ioctl_create_queue(struct file *filep, struct kfd_process *p, goto err_acquire_queue_buf; } -pr_debug ("Creating queue for PASID 0x%x on gpu 0x%x\n" , - p->pasid, +pr_debug ("Creating queue for process pid 0x%x on gpu 0x%x\n" , + (int )p->lead_thread->pid, dev->id); err = pqm_create_queue(&p->pqm, dev, &q_properties, &queue_id,@@ -415,9 +415,9 @@ static int kfd_ioctl_destroy_queue(struct file *filp, struct kfd_process *p,int retval;struct kfd_ioctl_destroy_queue_args *args = data; -pr_debug ("Destroying queue id %d for pasid 0x%x\n" , +pr_debug ("Destroying queue id %d for process pid 0x%x\n" , args->queue_id, - p->pasid); + (int )p->lead_thread->pid);mutex_lock (&p->mutex);@@ -468,8 +468,8 @@ static int kfd_ioctl_update_queue(struct file *filp, struct kfd_process *p, properties.pm4_target_xcc = (args->queue_percentage >>8 ) &0xFF ; properties.priority = args->queue_priority; -pr_debug ("Updating queue id %d for pasid 0x%x\n" , - args->queue_id, p->pasid); +pr_debug ("Updating queue id %d for process pid 0x%x\n" , + args->queue_id, (int )p->lead_thread->pid);mutex_lock (&p->mutex);@@ -695,7 +695,7 @@ static int kfd_ioctl_get_process_apertures(struct file *filp,struct kfd_process_device_apertures *pAperture;int i; -dev_dbg (kfd_device,"get apertures for PASID 0x%x" , p->pasid); +dev_dbg (kfd_device,"get apertures for process pid 0x%x" , (int )p->lead_thread->pid); args->num_of_nodes =0 ;@@ -747,7 +747,8 @@ static int kfd_ioctl_get_process_apertures_new(struct file *filp,int ret;int i; -dev_dbg (kfd_device,"get apertures for PASID 0x%x" , p->pasid); +dev_dbg (kfd_device,"get apertures for process pid 0x%x" , + (int )p->lead_thread->pid);if (args->num_of_nodes ==0 ) {/* Return number of nodes, so that user space can alloacate @@ -3365,12 +3366,12 @@ static int kfd_mmio_mmap(struct kfd_node *dev, struct kfd_process *process, vma->vm_page_prot =pgprot_noncached (vma->vm_page_prot); -pr_debug ("pasid 0x%x mapping mmio page\n" +pr_debug ("process pid 0x%x mapping mmio page\n" " target user address == 0x%08llX\n" " physical address == 0x%08llX\n" " vm_flags == 0x%04lX\n" " size == 0x%04lX\n" , - process->pasid, (unsigned long long ) vma->vm_start, + (int )process->lead_thread->pid, (unsigned long long ) vma->vm_start, address, vma->vm_flags,PAGE_SIZE ); return io_remap_pfn_range(vma,drivers/gpu/drm/amd/amdkfd/kfd_debug.c +2
-2@@ -238,8 +238,8 @@ bool kfd_set_dbg_ev_from_interrupt(struct kfd_node *dev,mutex_unlock (&p->mutex); }else if (trap_mask &KFD_EC_MASK (EC_DEVICE_MEMORY_VIOLATION )) { -kfd_dqm_evict_pasid (dev->dqm, p->pasid); -kfd_signal_vm_fault_event (dev, p->pasid,NULL , +kfd_dqm_evict_pasid (dev->dqm, pasid); +kfd_signal_vm_fault_event (dev, pasid,NULL , exception_data); signaled_to_debugger_or_runtime = true;
drivers/gpu/drm/amd/amdkfd/kfd_debug.c int kfd_dbg_send_exception_to_runtime(struct kfd_process *p, · line 279
Quoted 6 lines
drivers/gpu/drm/amd/amdkfd/kfd_debug.c +1
-2@@ -276,8 +276,8 @@ int kfd_dbg_send_exception_to_runtime(struct kfd_process *p, data = (struct kfd_hsa_memory_exception_data *) pdd->vm_fault_exc_data; -kfd_dqm_evict_pasid (pdd->dev->dqm, p->pasid); -kfd_signal_vm_fault_event (pdd->dev, p->pasid,NULL , data); +kfd_dqm_evict_pasid (pdd->dev->dqm, pdd->pasid);
kfd_dqm_evict_pasid just goes and looks up the pdd from the pasid again. Maybe this could be streamlined with a kfd_dqm_evict_pdd that avoids the unnecessary lookup. Could be a follow-up patch.
drivers/gpu/drm/amd/amdkfd/kfd_debug.c int kfd_dbg_send_exception_to_runtime(struct kfd_process *p, · line 280
Quoted 1 line
drivers/gpu/drm/amd/amdkfd/kfd_debug.c +1
-0+kfd_signal_vm_fault_event (pdd->dev, pdd->pasid,NULL , data);
Similarly, this function looks up the process from the PASID. Maybe this could be refactored to avoid the redundant lookup.
Quoted 196 lines
drivers/gpu/drm/amd/amdkfd/kfd_debug.c +0
-0error_reason &= ~KFD_EC_MASK (EC_DEVICE_MEMORY_VIOLATION ); }drivers/gpu/drm/amd/amdkfd/kfd_device_queue_manager.c +27
-25@@ -206,7 +206,7 @@ static int add_queue_mes(struct device_queue_manager *dqm, struct queue *q,return -EIO ;memset (&queue_input,0x0 ,sizeof (struct mes_add_queue_input )); - queue_input.process_id = qpd->pqm->process->pasid; + queue_input.process_id = pdd->pasid; queue_input.page_table_base_addr = qpd->page_table_base; queue_input.process_va_start =0 ; queue_input.process_va_end = adev->vm_manager.max_pfn -1 ;@@ -496,6 +496,7 @@ static int allocate_vmid(struct device_queue_manager *dqm,struct qcm_process_device *qpd, struct queue *q) { + struct kfd_process_device *pdd = qpd_to_pdd(qpd);struct device *dev = dqm->dev->adev->dev;int allocated_vmid =-1 , i;@@ -514,9 +515,9 @@ static int allocate_vmid(struct device_queue_manager *dqm,pr_debug ("vmid allocated: %d\n" , allocated_vmid); - dqm->vmid_pasid[allocated_vmid] = q->process->pasid; + dqm->vmid_pasid[allocated_vmid] = pdd->pasid; -set_pasid_vmid_mapping (dqm, q->process->pasid, allocated_vmid); +set_pasid_vmid_mapping (dqm, pdd->pasid, allocated_vmid); qpd->vmid = allocated_vmid; q->properties.vmid = allocated_vmid;@@ -768,6 +769,11 @@ static int dbgdev_wave_reset_wavefronts(struct kfd_node *dev, struct kfd_processreturn -EOPNOTSUPP ; } +/* taking the VMID for that process on the safe way using PDD */ + pdd =kfd_get_process_device_data (dev, p); +if (!pdd) +return -EFAULT ; +/* Scan all registers in the range ATC_VMID8_PASID_MAPPING .. * ATC_VMID15_PASID_MAPPING * to check which VMID the current process is mapped to. @@ -777,23 +783,19 @@ static int dbgdev_wave_reset_wavefronts(struct kfd_node *dev, struct kfd_process status = dev->kfd2kgd->get_atc_vmid_pasid_mapping_info (dev->adev, vmid, &queried_pasid); -if (status && queried_pasid == p->pasid) { -pr_debug ("Killing wave fronts of vmid %d and pasid 0x%x\n" , - vmid, p->pasid); +if (status && queried_pasid == pdd->pasid) { +pr_debug ("Killing wave fronts of vmid %d and process pid 0x%x\n" , + vmid, p->lead_thread->pid);break ; } }if (vmid > last_vmid_to_scan) { -dev_err (dev->adev->dev,"Didn't find vmid for pasid 0x%x\n" , p->pasid); +dev_err (dev->adev->dev,"Didn't find vmid for process pid 0x%x\n" , + p->lead_thread->pid);return -EFAULT ; } -/* taking the VMID for that process on the safe way using PDD */ - pdd =kfd_get_process_device_data (dev, p); -if (!pdd) -return -EFAULT ; - reg_gfx_index.bits.sh_broadcast_writes =1 ; reg_gfx_index.bits.se_broadcast_writes =1 ; reg_gfx_index.bits.instance_broadcast_writes =1 ;@@ -1029,8 +1031,8 @@ static int suspend_single_queue(struct device_queue_manager *dqm,if (q->properties.is_suspended)return 0 ; -pr_debug ("Suspending PASID %u queue [%i]\n" , - pdd->process->pasid, +pr_debug ("Suspending process pid %u queue [%i]\n" , + pdd->process->lead_thread->pid, q->properties.queue_id); is_new = q->properties.exception_status &KFD_EC_MASK (EC_QUEUE_NEW );@@ -1077,8 +1079,8 @@ static int resume_single_queue(struct device_queue_manager *dqm, pdd =qpd_to_pdd (qpd); -pr_debug ("Restoring from suspend PASID %u queue [%i]\n" , - pdd->process->pasid, +pr_debug ("Restoring from suspend process pid %u queue [%i]\n" , + pdd->process->lead_thread->pid, q->properties.queue_id); q->properties.is_suspended = false;@@ -1111,8 +1113,8 @@ static int evict_process_queues_nocpsch(struct device_queue_manager *dqm, goto out; pdd =qpd_to_pdd (qpd); -pr_debug_ratelimited ("Evicting PASID 0x%x queues\n" , - pdd->process->pasid); +pr_debug_ratelimited ("Evicting process pid 0x%x queues\n" , + pdd->process->lead_thread->pid); pdd->last_evict_timestamp =get_jiffies_64 ();/* Mark all queues as evicted. Deactivate all active queues on @@ -1169,8 +1171,8 @@ static int evict_process_queues_cpsch(struct device_queue_manager *dqm,if (!pdd->drm_priv) goto out; -pr_debug_ratelimited ("Evicting PASID 0x%x queues\n" , - pdd->process->pasid); +pr_debug_ratelimited ("Evicting process pid 0x%x queues\n" , + pdd->process->lead_thread->pid);/* Mark all queues as evicted. Deactivate all active queues on * the qpd. @@ -1228,8 +1230,8 @@ static int restore_process_queues_nocpsch(struct device_queue_manager *dqm, goto out; } -pr_debug_ratelimited ("Restoring PASID 0x%x queues\n" , - pdd->process->pasid); +pr_debug_ratelimited ("Restoring process pid 0x%x queues\n" , + pdd->process->lead_thread->pid);/* Update PD Base in QPD */ qpd->page_table_base = pd_base;@@ -1312,8 +1314,8 @@ static int restore_process_queues_cpsch(struct device_queue_manager *dqm,if (!pdd->drm_priv) goto vm_not_acquired; -pr_debug_ratelimited ("Restoring PASID 0x%x queues\n" , - pdd->process->pasid); +pr_debug_ratelimited ("Restoring process pid 0x%x queues\n" , + pdd->process->lead_thread->pid);/* Update PD Base in QPD */ qpd->page_table_base =amdgpu_amdkfd_gpuvm_get_process_page_dir (pdd->drm_priv);@@ -2102,7 +2104,7 @@ static void set_queue_as_reset(struct device_queue_manager *dqm, struct queue *qstruct kfd_process_device *pdd =qpd_to_pdd (qpd);dev_err (dqm->dev->adev->dev,"queue id 0x%0x at pasid 0x%0x is reset\n" , - q->properties.queue_id, q->process->pasid); + q->properties.queue_id, pdd->process->lead_thread->pid); pdd->has_reset_queue = true;if (q->properties.is_active) {drivers/gpu/drm/amd/amdkfd/kfd_events.c +8
-7@@ -1128,8 +1128,8 @@ static void lookup_events_by_type_and_signal(struct kfd_process *p,if (type ==KFD_EVENT_TYPE_MEMORY ) {dev_warn (kfd_device, -"Sending SIGSEGV to process %d (pasid 0x%x)" , - p->lead_thread->pid, p->pasid); +"Sending SIGSEGV to process pid %d" , + p->lead_thread->pid);send_sig (SIGSEGV , p->lead_thread,0 ); }@@ -1137,13 +1137,13 @@ static void lookup_events_by_type_and_signal(struct kfd_process *p,if (send_signal) {if (send_sigterm) {dev_warn (kfd_device, -"Sending SIGTERM to process %d (pasid 0x%x)" , - p->lead_thread->pid, p->pasid); +"Sending SIGTERM to process pid %d" , + p->lead_thread->pid);send_sig (SIGTERM , p->lead_thread,0 ); }else {dev_err (kfd_device, -"Process %d (pasid 0x%x) got unhandled exception" , - p->lead_thread->pid, p->pasid); +"Process pid %d got unhandled exception" , + p->lead_thread->pid); } }@@ -1253,7 +1253,8 @@ void kfd_signal_reset_event(struct kfd_node *dev) }if (unlikely (!pdd)) { -WARN_ONCE (1 ,"Could not get device data from pasid:0x%x\n" , p->pasid); +WARN_ONCE (1 ,"Could not get device data from process pid:0x%x\n" , + p->lead_thread->pid);continue ; }
drivers/gpu/drm/amd/amdkfd/kfd_events.c void kfd_signal_reset_event(struct kfd_node *dev) · line 1267
Quoted 5 lines
drivers/gpu/drm/amd/amdkfd/kfd_events.c +1
-1@@ -1263,7 +1264,7 @@ void kfd_signal_reset_event(struct kfd_node *dev)if (dev->dqm->detect_hang_count) {struct amdgpu_task_info *ti; - ti =amdgpu_vm_get_task_info_pasid (dev->adev, p->pasid); + ti =amdgpu_vm_get_task_info_pasid (dev->adev, pdd->pasid);
Can we use amdgpu_vm_get_task_info_vm here to avoid a redundant lookup of the VM from the pasid? You'd need a helper to get the vm from pdd->drm_priv (drm_priv_to_vm in amdgpu_amdkfd.h).
Quoted 182 lines
drivers/gpu/drm/amd/amdkfd/kfd_events.c +0
-0if (ti) { dev_err(dev->adev->dev, "Queues reset on process %s tid %d thread %s pid %d\n",drivers/gpu/drm/amd/amdkfd/kfd_packet_manager_v9.c +2
-2@@ -47,7 +47,7 @@ static int pm_map_process_v9(struct packet_manager *pm, packet->bitfields2.exec_cleaner_shader =1 ; packet->bitfields2.diq_enable = (qpd->is_debug) ?1 :0 ; packet->bitfields2.process_quantum =10 ; - packet->bitfields2.pasid = qpd->pqm->process->pasid; + packet->bitfields2.pasid = pdd->pasid; packet->bitfields14.gds_size = qpd->gds_size &0x3F ; packet->bitfields14.gds_size_hi = (qpd->gds_size >>6 ) &0xF ; packet->bitfields14.num_gws = (qpd->mapped_gws_queue) ? qpd->num_gws :0 ;@@ -106,7 +106,7 @@ static int pm_map_process_aldebaran(struct packet_manager *pm, packet->bitfields2.exec_cleaner_shader =1 ; packet->bitfields2.diq_enable = (qpd->is_debug) ?1 :0 ; packet->bitfields2.process_quantum =10 ; - packet->bitfields2.pasid = qpd->pqm->process->pasid; + packet->bitfields2.pasid = pdd->pasid; packet->bitfields14.gds_size = qpd->gds_size &0x3F ; packet->bitfields14.gds_size_hi = (qpd->gds_size >>6 ) &0xF ; packet->bitfields14.num_gws = (qpd->mapped_gws_queue) ? qpd->num_gws :0 ;drivers/gpu/drm/amd/amdkfd/kfd_packet_manager_vi.c +2
-1@@ -42,6 +42,7 @@ unsigned int pm_build_pm4_header(unsigned int opcode, size_t packet_size)static int pm_map_process_vi (struct packet_manager *pm,uint32_t *buffer,struct qcm_process_device *qpd) { +struct kfd_process_device *pdd =qpd_to_pdd (qpd);struct pm4_mes_map_process *packet; packet = (struct pm4_mes_map_process *)buffer;@@ -52,7 +53,7 @@ static int pm_map_process_vi(struct packet_manager *pm, uint32_t *buffer, sizeof(struct pm4_mes_map_process)); packet->bitfields2.diq_enable = (qpd->is_debug) ?1 :0 ; packet->bitfields2.process_quantum =10 ; - packet->bitfields2.pasid = qpd->pqm->process->pasid; + packet->bitfields2.pasid = pdd->pasid; packet->bitfields3.page_table_base = qpd->page_table_base; packet->bitfields10.gds_size = qpd->gds_size; packet->bitfields10.num_gws = qpd->num_gws;drivers/gpu/drm/amd/amdkfd/kfd_priv.h +2
-3@@ -849,6 +849,8 @@ struct kfd_process_device {/* Tracks queue reset status */ bool has_reset_queue; + +u32 pasid; };#define qpd_to_pdd (x) container_of(x, struct kfd_process_device, qpd)@@ -908,8 +910,6 @@ struct kfd_process {/* We want to receive a notification when the mm_struct is destroyed */ struct mmu_notifier mmu_notifier; -u32 pasid; -/* * Array of kfd_process_device pointers, * one for each device the process is using. @@ -970,7 +970,6 @@ struct kfd_process {/* Kobj for our procfs */ struct kobject *kobj;struct kobject *kobj_queues; -struct attribute attr_pasid;/* Keep track cwsr init */ bool has_cwsr;drivers/gpu/drm/amd/amdkfd/kfd_process.c +7
-24@@ -282,8 +282,8 @@ static int kfd_get_cu_occupancy(struct attribute *attr, char *buffer) cu_cnt =0 ; proc = pdd->process;if (pdd->qpd.queue_count ==0 ) { -pr_debug ("Gpu-Id: %d has no active queues for process %d\n" , - dev->id, proc->pasid); +pr_debug ("Gpu-Id: %d has no active queues for process pid %d\n" , + dev->id, (int )proc->lead_thread->pid);return snprintf (buffer,PAGE_SIZE ,"%d\n" , cu_cnt); }@@ -327,12 +327,7 @@ static int kfd_get_cu_occupancy(struct attribute *attr, char *buffer)static ssize_t kfd_procfs_show (struct kobject *kobj,struct attribute *attr,char *buffer) { -if (strcmp (attr->name,"pasid" ) ==0 ) { - struct kfd_process *p = container_of(attr, struct kfd_process, - attr_pasid); - -return snprintf (buffer,PAGE_SIZE ,"%d\n" , p->pasid); - }else if (strncmp (attr->name,"vram_" ,5 ) ==0 ) { +if (strncmp (attr->name,"vram_" ,5 ) ==0 ) { struct kfd_process_device *pdd = container_of(attr, struct kfd_process_device, attr_vram);return snprintf (buffer,PAGE_SIZE ,"%llu\n" ,atomic64_read (&pdd->vram_usage));@@ -887,9 +882,6 @@ struct kfd_process *kfd_create_process(struct task_struct *thread) goto out; } -kfd_sysfs_create_file (process->kobj, &process->attr_pasid, -"pasid" ); - process->kobj_queues =kobject_create_and_add ("queues" , process->kobj);if (!process->kobj_queues)@@ -1055,8 +1047,8 @@ static void kfd_process_destroy_pdds(struct kfd_process *p)for (i =0 ; i < p->n_pdds; i++) {struct kfd_process_device *pdd = p->pdds[i]; -pr_debug ("Releasing pdd (topology id %d) for process (pasid 0x%x)\n" , - pdd->dev->id, p->pasid); +pr_debug ("Releasing pdd (topology id %d, for pid 0x%0x)\n" , + pdd->dev->id, (int )p->lead_thread->pid);kfd_process_device_destroy_cwsr_dgpu (pdd);kfd_process_device_destroy_ib_mem (pdd);@@ -1102,7 +1094,6 @@ static void kfd_process_remove_sysfs(struct kfd_process *p)if (!p->kobj)return ; -sysfs_remove_file (p->kobj, &p->attr_pasid);kobject_del (p->kobj_queues);kobject_put (p->kobj_queues); p->kobj_queues =NULL ;@@ -1171,7 +1162,6 @@ static void kfd_process_wq_release(struct work_struct *work)kfd_event_free_process (p); -kfd_pasid_free (p->pasid);mutex_destroy (&p->mutex);put_task_struct (p->lead_thread);@@ -1524,12 +1514,6 @@ static struct kfd_process *create_process(const struct task_struct *thread)atomic_set (&process->debugged_process_count,0 );sema_init (&process->runtime_enable_sema,0 ); - process->pasid =kfd_pasid_alloc (); -if (process->pasid ==0 ) { - err = -ENOSPC ; - goto err_alloc_pasid; - } - err =pqm_init (&process->pqm, process);if (err !=0 ) goto err_process_pqm_init;@@ -1583,8 +1567,6 @@ static struct kfd_process *create_process(const struct task_struct *thread) err_init_apertures:pqm_uninit (&process->pqm); err_process_pqm_init: -kfd_pasid_free (process->pasid); -err_alloc_pasid:kfd_event_free_process (process); err_event_init:mutex_destroy (&process->mutex);@@ -1723,7 +1705,7 @@ int kfd_process_device_init_vm(struct kfd_process_device *pdd,if (ret) goto err_init_cwsr; - ret =amdgpu_amdkfd_gpuvm_set_vm_pasid (dev->adev, avm, p->pasid); + ret =amdgpu_amdkfd_gpuvm_set_vm_pasid (dev->adev, pdd, avm);if (ret) goto err_set_pasid;@@ -1822,15 +1804,20 @@ struct kfd_process *kfd_lookup_process_by_pasid(u32 pasid) {struct kfd_process *p, *ret_p =NULL ;unsigned int temp; +int i;int idx =srcu_read_lock (&kfd_processes_srcu);hash_for_each_rcu (kfd_processes_table, temp, p, kfd_processes) {
drivers/gpu/drm/amd/amdkfd/kfd_process.c struct kfd_process *kfd_lookup_process_by_pasid(u32 pasid) · line 1817
Quoted 10 lines
drivers/gpu/drm/amd/amdkfd/kfd_process.c +6
-4-if (p->pasid == pasid) { -kref_get (&p->ref); - ret_p = p; -break ; +for (i =0 ; i < p->n_pdds; i++) { +if (p->pdds[i]->pasid == pasid) { +kref_get (&p->ref); + ret_p = p; +break ; + }
I think this won't work correctly. The same PASID can be used for different processes on different GPUs because each adev manages its own PASID->amdgpu_vm lookup table. So kfd_lookup_process_by_pasid needs a new parameter that identifies the GPU adev, and you should only compare pasids, if the adev matches.
Regards,
Felix
Quoted 190 lines
drivers/gpu/drm/amd/amdkfd/kfd_process.c +17
-13} +if (ret_p) +break ; }srcu_read_unlock (&kfd_processes_srcu, idx);@@ -1991,7 +1978,7 @@ static void evict_process_worker(struct work_struct *work)*/ p = container_of(dwork, struct kfd_process, eviction_work); -pr_debug ("Started evicting pasid 0x%x\n" , p->pasid); +pr_debug ("Started evicting process pid 0x%x\n" , (int )p->lead_thread->pid); ret =kfd_process_evict_queues (p,KFD_QUEUE_EVICTION_TRIGGER_TTM );if (!ret) {/* If another thread already signaled the eviction fence, @@ -2003,9 +1990,9 @@ static void evict_process_worker(struct work_struct *work) msecs_to_jiffies(PROCESS_RESTORE_TIME_MS)))kfd_process_restore_queues (p); -pr_debug ("Finished evicting pasid 0x%x\n" , p->pasid); +pr_debug ("Finished evicting process pid 0x%x\n" , (int )p->lead_thread->pid); }else -pr_err ("Failed to evict queues of pasid 0x%x\n" , p->pasid); +pr_err ("Failed to evict queues of process pid 0x%x\n" , (int )p->lead_thread->pid); }static int restore_process_helper (struct kfd_process *p)@@ -2022,9 +2009,11 @@ static int restore_process_helper(struct kfd_process *p) ret =kfd_process_restore_queues (p);if (!ret) -pr_debug ("Finished restoring pasid 0x%x\n" , p->pasid); +pr_debug ("Finished restoring process pid 0x%x\n" , + (int )p->lead_thread->pid);else -pr_err ("Failed to restore queues of pasid 0x%x\n" , p->pasid); +pr_err ("Failed to restore queues of process pid 0x%x\n" , + (int )p->lead_thread->pid);return ret; }@@ -2041,7 +2030,7 @@ static void restore_process_worker(struct work_struct *work)* lifetime of this thread, kfd_process p will be valid */ p = container_of(dwork, struct kfd_process, restore_work); -pr_debug ("Started restoring pasid 0x%x\n" , p->pasid); +pr_debug ("Started restoring process pasid 0x%x\n" , (int )p->lead_thread->pid);/* Setting last_restore_timestamp before successful restoration. * Otherwise this would have to be set by KGD (restore_process_bos) @@ -2057,8 +2046,8 @@ static void restore_process_worker(struct work_struct *work) ret =restore_process_helper (p);if (ret) { -pr_debug ("Failed to restore BOs of pasid 0x%x, retry after %d ms\n" , - p->pasid,PROCESS_BACK_OFF_TIME_MS ); +pr_debug ("Failed to restore BOs of process pid 0x%x, retry after %d ms\n" , + (int )p->lead_thread->pid,PROCESS_BACK_OFF_TIME_MS );if (mod_delayed_work (kfd_restore_wq, &p->restore_work,msecs_to_jiffies (PROCESS_RESTORE_TIME_MS )))kfd_process_restore_queues (p);@@ -2074,7 +2063,7 @@ void kfd_suspend_all_processes(void)WARN (debug_evictions,"Evicting all processes" );hash_for_each_rcu (kfd_processes_table, temp, p, kfd_processes) {if (kfd_process_evict_queues (p,KFD_QUEUE_EVICTION_TRIGGER_SUSPEND )) -pr_err ("Failed to suspend process 0x%x\n" , p->pasid); +pr_err ("Failed to suspend process pid 0x%x\n" , (int )p->lead_thread->pid);signal_eviction_fence (p); }srcu_read_unlock (&kfd_processes_srcu, idx);@@ -2088,8 +2077,8 @@ int kfd_resume_all_processes(void)hash_for_each_rcu (kfd_processes_table, temp, p, kfd_processes) {if (restore_process_helper (p)) { -pr_err ("Restore process %d failed during resume\n" , - p->pasid); +pr_err ("Restore process pid %d failed during resume\n" , + (int )p->lead_thread->pid); ret = -EFAULT ; } }@@ -2144,7 +2133,7 @@ int kfd_process_drain_interrupts(struct kfd_process_device *pdd)memset (irq_drain_fence,0 ,sizeof (irq_drain_fence)); irq_drain_fence[0 ] = (KFD_IRQ_FENCE_SOURCEID <<8 ) |KFD_IRQ_FENCE_CLIENTID ; - irq_drain_fence[3 ] = pdd->process->pasid; + irq_drain_fence[3 ] = pdd->pasid;/* * For GFX 9.4.3, send the NodeId also in IH cookie DW[3] @@ -2296,7 +2285,7 @@ int kfd_debugfs_mqds_by_process(struct seq_file *m, void *data)hash_for_each_rcu (kfd_processes_table, temp, p, kfd_processes) {seq_printf (m,"Process %d PASID 0x%x:\n" , - p->lead_thread->tgid, p->pasid); + p->lead_thread->tgid, (int )p->lead_thread->pid);mutex_lock (&p->mutex); r =pqm_debugfs_mqds (m, &p->pqm);drivers/gpu/drm/amd/amdkfd/kfd_process_queue_manager.c +5
-5@@ -69,8 +69,8 @@ static int find_available_queue_slot(struct process_queue_manager *pqm,pr_debug ("The new slot id %lu\n" , found);if (found >=KFD_MAX_NUM_OF_QUEUES_PER_PROCESS ) { -pr_info ("Cannot open more queues for process with pasid 0x%x\n" , - pqm->process->pasid); +pr_info ("Cannot open more queues for process with pid 0x%x\n" , + pqm->process->lead_thread->pid);return -ENOMEM ; }@@ -425,8 +425,8 @@ int pqm_create_queue(struct process_queue_manager *pqm, }if (retval !=0 ) { -pr_err ("Pasid 0x%x DQM create queue type %d failed. ret %d\n" , - pqm->process->pasid, type, retval); +pr_err ("process pid 0x%x DQM create queue type %d failed. ret %d\n" , + pqm->process->lead_thread->pid, type, retval); goto err_create_queue; }@@ -520,7 +520,7 @@ int pqm_destroy_queue(struct process_queue_manager *pqm, unsigned int qid) retval = dqm->ops.destroy_queue (dqm, &pdd->qpd, pqn->q);if (retval) {pr_err ("Pasid 0x%x destroy queue %d failed, ret %d\n" , - pqm->process->pasid, + pdd->pasid, pqn->q->properties.queue_id, retval);if (retval != -ETIME ) goto err_destroy_queue;drivers/gpu/drm/amd/amdkfd/kfd_svm.c +10
-7@@ -563,7 +563,8 @@ svm_range_vram_node_new(struct kfd_node *node, struct svm_range *prange,int r; p = container_of(prange->svms, struct kfd_process, svms); -pr_debug ("pasid: %x svms 0x%p [0x%lx 0x%lx]\n" , p->pasid, prange->svms, +pr_debug ("process pid: %x svms 0x%p [0x%lx 0x%lx]\n" , + (int )p->lead_thread->pid, prange->svms, prange->start, prange->last);if (svm_range_validate_svm_bo (node, prange))@@ -3232,7 +3233,8 @@ void svm_range_list_fini(struct kfd_process *p)struct svm_range *prange;struct svm_range *next; -pr_debug ("pasid 0x%x svms 0x%p\n" , p->pasid, &p->svms); +pr_debug ("process pid 0x%x svms 0x%p\n" , (int )p->lead_thread->pid, + &p->svms);cancel_delayed_work_sync (&p->svms.restore_work);@@ -3255,7 +3257,8 @@ void svm_range_list_fini(struct kfd_process *p)mutex_destroy (&p->svms.lock); -pr_debug ("pasid 0x%x svms 0x%p done\n" , p->pasid, &p->svms); +pr_debug ("process pid 0x%x svms 0x%p done\n" , + (int )p->lead_thread->pid, &p->svms); }int svm_range_list_init (struct kfd_process *p)@@ -3618,8 +3621,8 @@ svm_range_set_attr(struct kfd_process *p, struct mm_struct *mm,bool flush_tlb;int r, ret =0 ; -pr_debug ("pasid 0x%x svms 0x%p [0x%llx 0x%llx] pages 0x%llx\n" , - p->pasid, &p->svms, start, start + size -1 , size); +pr_debug ("process pid 0x%x svms 0x%p [0x%llx 0x%llx] pages 0x%llx\n" , + (int )p->lead_thread->pid, &p->svms, start, start + size -1 , size); r =svm_range_check_attr (p, nattr, attrs);if (r)@@ -3727,8 +3730,8 @@ svm_range_set_attr(struct kfd_process *p, struct mm_struct *mm, out:mutex_unlock (&process_info->lock); -pr_debug ("pasid 0x%x svms 0x%p [0x%llx 0x%llx] done, r=%d\n" , p->pasid, - &p->svms, start, start + size -1 , r); +pr_debug ("process pid 0x%x svms 0x%p [0x%llx 0x%llx] done, r=%d\n" , + (int )p->lead_thread->pid, &p->svms, start, start + size -1 , r);return ret ? ret : r; }
Quoted 60 lines · Felix Kuehling
Quoted 58 lines · Xiaogang.Chen
From: Xiaogang Chen <[email protected]>
Current kfd driver has its own PASID value for a kfd process and uses it to locate vm at interrupt handler or mapping between kfd process and vm. That design is not working when a physical gpu device has multiple spatial partitions, ex: adev in CPX mode. This patch has kfd driver use same pasid values that graphic driver generated which is per vm per pasid.
These pasid values are passed to fw/hardware. We do not need change interrupt handler though more pasid values are used. Also, pasid values at log are replaced by user process pid, or pasid values are not exposed to user, user will see their process pids that have meaning in user space.
Signed-off-by Xiaogang Chen <[email protected]>--- drivers/gpu/drm/amd/amdgpu/amdgpu_amdkfd.h | 4 +- .../gpu/drm/amd/amdgpu/amdgpu_amdkfd_gpuvm.c | 34 +++++---- drivers/gpu/drm/amd/amdkfd/kfd_chardev.c | 25 +++---- drivers/gpu/drm/amd/amdkfd/kfd_debug.c | 8 +-- .../drm/amd/amdkfd/kfd_device_queue_manager.c | 52 +++++++------- drivers/gpu/drm/amd/amdkfd/kfd_events.c | 17 ++--- .../drm/amd/amdkfd/kfd_packet_manager_v9.c | 4 +- .../drm/amd/amdkfd/kfd_packet_manager_vi.c | 3 +- drivers/gpu/drm/amd/amdkfd/kfd_priv.h | 5 +- drivers/gpu/drm/amd/amdkfd/kfd_process.c | 71 ++++++++----------- .../amd/amdkfd/kfd_process_queue_manager.c | 10 +-- drivers/gpu/drm/amd/amdkfd/kfd_svm.c | 17 +++-- 12 files changed, 127 insertions(+), 123 deletions(-)drivers/gpu/drm/amd/amdgpu/amdgpu_amdkfd.h +3
-1@@ -47,6 +47,7 @@ enum TLB_FLUSH_TYPE { };struct amdgpu_device ; +struct kfd_process_device ;struct amdgpu_reset_context ;enum kfd_mem_attachment_type {@@ -300,7 +301,8 @@ bool amdgpu_amdkfd_compute_active(struct amdgpu_device *adev, uint32_t node_id); ((struct drm_file *)(drm_priv))->driver_priv)->vm)int amdgpu_amdkfd_gpuvm_set_vm_pasid (struct amdgpu_device *adev, -struct amdgpu_vm *avm,u32 pasid); +struct kfd_process_device *pdd, +struct amdgpu_vm *avm); int amdgpu_amdkfd_gpuvm_acquire_process_vm(struct amdgpu_device *adev, struct amdgpu_vm *avm, void **process_info,drivers/gpu/drm/amd/amdgpu/amdgpu_amdkfd_gpuvm.c +0
-0@@ -1530,24 +1530,30 @@ static void amdgpu_amdkfd_gpuvm_unpin_bo(struct amdgpu_bo *bo) }
Quoted 5 lines
drivers/gpu/drm/amd/amdgpu/amdgpu_amdkfd_gpuvm.c static void amdgpu_amdkfd_gpuvm_unpin_bo(struct amdgpu_bo *bo) · line 1534
Quoted 4 lines
drivers/gpu/drm/amd/amdgpu/amdgpu_amdkfd_gpuvm.c +2
-1int amdgpu_amdkfd_gpuvm_set_vm_pasid (struct amdgpu_device *adev, -struct amdgpu_vm *avm,u32 pasid) +struct kfd_process_device *pdd, +struct amdgpu_vm *avm)I'd prefer if this function got renamed to
amdgpu_amdkfd_gpuvm_get_pasidand it returns the pasid.
ok.
Quoted 11 lines
Quoted 11 lines
drivers/gpu/drm/amd/amdgpu/amdgpu_amdkfd_gpuvm.c +1
-8{ -int ret; - -/* Free the original amdgpu allocated pasid, -* will be replaced with kfd allocated pasid. -*/ -if (avm->pasid) { -amdgpu_pasid_free (avm->pasid); -amdgpu_vm_set_pasid (adev, avm,0 ); +int ret =0 ;
Quoted 8 lines
drivers/gpu/drm/amd/amdgpu/amdgpu_amdkfd_gpuvm.c static void amdgpu_amdkfd_gpuvm_unpin_bo(struct amdgpu_bo *bo) · line 1544
Quoted 7 lines
drivers/gpu/drm/amd/amdgpu/amdgpu_amdkfd_gpuvm.c +7
-0+int pasid; + +/* if avm has pasid assigned use it */ +if (avm->pasid) + pdd->pasid = avm->pasid; +else { +/* otherwise generate a new pasid and assign to avm */ Does this ever happen? I think any avm that was acquired properly should already have a PASID. If it doesn't, maybe we can just return an error.
Right, this case should not happen as app should open render node first, then acquire vm through kfd node. I handled the case here just in case, but we can return an error here.
Quoted 123 lines
Quoted 123 lines
drivers/gpu/drm/amd/amdgpu/amdgpu_amdkfd_gpuvm.c +10
-5+ pasid =amdgpu_pasid_alloc (16 ); +if (pasid <0 ) { +dev_warn (adev->dev,"No more PASIDs available!" ); + ret = -ENOMEM ; + }else + ret =xa_err (xa_store_irq (&adev->vm_manager.pasids, pasid, avm, +GFP_KERNEL )); +if (ret >=0 ) + avm->pasid = pasid; } - ret =amdgpu_vm_set_pasid (adev, avm, pasid); -if (ret) -return ret; - -return 0 ; +return ret; } int amdgpu_amdkfd_gpuvm_acquire_process_vm(struct amdgpu_device *adev,drivers/gpu/drm/amd/amdkfd/kfd_chardev.c +13
-12@@ -155,8 +155,8 @@ static int kfd_open(struct inode *inode, struct file *filep)/* filep now owns the reference returned by kfd_create_process */ filep->private_data = process; -dev_dbg (kfd_device,"process %d opened, compat mode (32 bit) - %d\n" , - process->pasid, process->is_32bit_user_mode); +dev_dbg (kfd_device,"process pid %d opened kfd node, compat mode (32 bit) - %d\n" , + (int )process->lead_thread->pid, process->is_32bit_user_mode);return 0 ; }@@ -361,8 +361,8 @@ static int kfd_ioctl_create_queue(struct file *filep, struct kfd_process *p, goto err_acquire_queue_buf; } -pr_debug ("Creating queue for PASID 0x%x on gpu 0x%x\n" , - p->pasid, +pr_debug ("Creating queue for process pid 0x%x on gpu 0x%x\n" , + (int )p->lead_thread->pid, dev->id); err = pqm_create_queue(&p->pqm, dev, &q_properties, &queue_id,@@ -415,9 +415,9 @@ static int kfd_ioctl_destroy_queue(struct file *filp, struct kfd_process *p,int retval;struct kfd_ioctl_destroy_queue_args *args = data; -pr_debug ("Destroying queue id %d for pasid 0x%x\n" , +pr_debug ("Destroying queue id %d for process pid 0x%x\n" , args->queue_id, - p->pasid); + (int )p->lead_thread->pid);mutex_lock (&p->mutex);@@ -468,8 +468,8 @@ static int kfd_ioctl_update_queue(struct file *filp, struct kfd_process *p, properties.pm4_target_xcc = (args->queue_percentage >>8 ) &0xFF ; properties.priority = args->queue_priority; -pr_debug ("Updating queue id %d for pasid 0x%x\n" , - args->queue_id, p->pasid); +pr_debug ("Updating queue id %d for process pid 0x%x\n" , + args->queue_id, (int )p->lead_thread->pid);mutex_lock (&p->mutex);@@ -695,7 +695,7 @@ static int kfd_ioctl_get_process_apertures(struct file *filp,struct kfd_process_device_apertures *pAperture;int i; -dev_dbg (kfd_device,"get apertures for PASID 0x%x" , p->pasid); +dev_dbg (kfd_device,"get apertures for process pid 0x%x" , (int )p->lead_thread->pid); args->num_of_nodes =0 ;@@ -747,7 +747,8 @@ static int kfd_ioctl_get_process_apertures_new(struct file *filp,int ret;int i; -dev_dbg (kfd_device,"get apertures for PASID 0x%x" , p->pasid); +dev_dbg (kfd_device,"get apertures for process pid 0x%x" , + (int )p->lead_thread->pid);if (args->num_of_nodes ==0 ) {/* Return number of nodes, so that user space can alloacate @@ -3365,12 +3366,12 @@ static int kfd_mmio_mmap(struct kfd_node *dev, struct kfd_process *process, vma->vm_page_prot =pgprot_noncached (vma->vm_page_prot); -pr_debug ("pasid 0x%x mapping mmio page\n" +pr_debug ("process pid 0x%x mapping mmio page\n" " target user address == 0x%08llX\n" " physical address == 0x%08llX\n" " vm_flags == 0x%04lX\n" " size == 0x%04lX\n" , - process->pasid, (unsigned long long ) vma->vm_start, + (int )process->lead_thread->pid, (unsigned long long ) vma->vm_start, address, vma->vm_flags,PAGE_SIZE ); return io_remap_pfn_range(vma,drivers/gpu/drm/amd/amdkfd/kfd_debug.c +2
-2@@ -238,8 +238,8 @@ bool kfd_set_dbg_ev_from_interrupt(struct kfd_node *dev,mutex_unlock (&p->mutex); }else if (trap_mask &KFD_EC_MASK (EC_DEVICE_MEMORY_VIOLATION )) { -kfd_dqm_evict_pasid (dev->dqm, p->pasid); -kfd_signal_vm_fault_event (dev, p->pasid,NULL , +kfd_dqm_evict_pasid (dev->dqm, pasid); +kfd_signal_vm_fault_event (dev, pasid,NULL , exception_data); signaled_to_debugger_or_runtime = true;@@ -276,8 +276,8 @@ int kfd_dbg_send_exception_to_runtime(struct kfd_process *p, data = (struct kfd_hsa_memory_exception_data *) pdd->vm_fault_exc_data;
Quoted 4 lines
drivers/gpu/drm/amd/amdkfd/kfd_debug.c int kfd_dbg_send_exception_to_runtime(struct kfd_process *p, · line 279
Quoted 3 lines
drivers/gpu/drm/amd/amdkfd/kfd_debug.c +1
-2-kfd_dqm_evict_pasid (pdd->dev->dqm, p->pasid); -kfd_signal_vm_fault_event (pdd->dev, p->pasid,NULL , data); +kfd_dqm_evict_pasid (pdd->dev->dqm, pdd->pasid);
kfd_dqm_evict_pasidjust goes and looks up the pdd from the pasid again. Maybe this could be streamlined with akfd_dqm_evict_pddthat avoids the unnecessary lookup. Could be a follow-up patch.
This is a different thing that we loop up kfd process or pdd more than one time(redundant). This issue happened at several places. I will remove some redundant loop up in next version.
Quoted 3 lines
drivers/gpu/drm/amd/amdkfd/kfd_debug.c int kfd_dbg_send_exception_to_runtime(struct kfd_process *p, · line 280
Quoted 2 lines
drivers/gpu/drm/amd/amdkfd/kfd_debug.c +1
-0+kfd_signal_vm_fault_event (pdd->dev, pdd->pasid,NULL , data);Similarly, this function looks up the process from the PASID. Maybe this could be refactored to avoid the redundant lookup.
Same as above.
Quoted 196 lines
Quoted 196 lines
drivers/gpu/drm/amd/amdkfd/kfd_debug.c +0
-0error_reason &= ~KFD_EC_MASK (EC_DEVICE_MEMORY_VIOLATION ); }drivers/gpu/drm/amd/amdkfd/kfd_device_queue_manager.c +27
-25@@ -206,7 +206,7 @@ static int add_queue_mes(struct device_queue_manager *dqm, struct queue *q,return -EIO ;memset (&queue_input,0x0 ,sizeof (struct mes_add_queue_input )); - queue_input.process_id = qpd->pqm->process->pasid; + queue_input.process_id = pdd->pasid; queue_input.page_table_base_addr = qpd->page_table_base; queue_input.process_va_start =0 ; queue_input.process_va_end = adev->vm_manager.max_pfn -1 ;@@ -496,6 +496,7 @@ static int allocate_vmid(struct device_queue_manager *dqm,struct qcm_process_device *qpd, struct queue *q) { + struct kfd_process_device *pdd = qpd_to_pdd(qpd);struct device *dev = dqm->dev->adev->dev;int allocated_vmid =-1 , i;@@ -514,9 +515,9 @@ static int allocate_vmid(struct device_queue_manager *dqm,pr_debug ("vmid allocated: %d\n" , allocated_vmid); - dqm->vmid_pasid[allocated_vmid] = q->process->pasid; + dqm->vmid_pasid[allocated_vmid] = pdd->pasid; -set_pasid_vmid_mapping (dqm, q->process->pasid, allocated_vmid); +set_pasid_vmid_mapping (dqm, pdd->pasid, allocated_vmid); qpd->vmid = allocated_vmid; q->properties.vmid = allocated_vmid;@@ -768,6 +769,11 @@ static int dbgdev_wave_reset_wavefronts(struct kfd_node *dev, struct kfd_processreturn -EOPNOTSUPP ; } +/* taking the VMID for that process on the safe way using PDD */ + pdd =kfd_get_process_device_data (dev, p); +if (!pdd) +return -EFAULT ; +/* Scan all registers in the range ATC_VMID8_PASID_MAPPING .. * ATC_VMID15_PASID_MAPPING * to check which VMID the current process is mapped to. @@ -777,23 +783,19 @@ static int dbgdev_wave_reset_wavefronts(struct kfd_node *dev, struct kfd_process status = dev->kfd2kgd->get_atc_vmid_pasid_mapping_info (dev->adev, vmid, &queried_pasid); -if (status && queried_pasid == p->pasid) { -pr_debug ("Killing wave fronts of vmid %d and pasid 0x%x\n" , - vmid, p->pasid); +if (status && queried_pasid == pdd->pasid) { +pr_debug ("Killing wave fronts of vmid %d and process pid 0x%x\n" , + vmid, p->lead_thread->pid);break ; } }if (vmid > last_vmid_to_scan) { -dev_err (dev->adev->dev,"Didn't find vmid for pasid 0x%x\n" , p->pasid); +dev_err (dev->adev->dev,"Didn't find vmid for process pid 0x%x\n" , + p->lead_thread->pid);return -EFAULT ; } -/* taking the VMID for that process on the safe way using PDD */ - pdd =kfd_get_process_device_data (dev, p); -if (!pdd) -return -EFAULT ; - reg_gfx_index.bits.sh_broadcast_writes =1 ; reg_gfx_index.bits.se_broadcast_writes =1 ; reg_gfx_index.bits.instance_broadcast_writes =1 ;@@ -1029,8 +1031,8 @@ static int suspend_single_queue(struct device_queue_manager *dqm,if (q->properties.is_suspended)return 0 ; -pr_debug ("Suspending PASID %u queue [%i]\n" , - pdd->process->pasid, +pr_debug ("Suspending process pid %u queue [%i]\n" , + pdd->process->lead_thread->pid, q->properties.queue_id); is_new = q->properties.exception_status &KFD_EC_MASK (EC_QUEUE_NEW );@@ -1077,8 +1079,8 @@ static int resume_single_queue(struct device_queue_manager *dqm, pdd =qpd_to_pdd (qpd); -pr_debug ("Restoring from suspend PASID %u queue [%i]\n" , - pdd->process->pasid, +pr_debug ("Restoring from suspend process pid %u queue [%i]\n" , + pdd->process->lead_thread->pid, q->properties.queue_id); q->properties.is_suspended = false;@@ -1111,8 +1113,8 @@ static int evict_process_queues_nocpsch(struct device_queue_manager *dqm, goto out; pdd =qpd_to_pdd (qpd); -pr_debug_ratelimited ("Evicting PASID 0x%x queues\n" , - pdd->process->pasid); +pr_debug_ratelimited ("Evicting process pid 0x%x queues\n" , + pdd->process->lead_thread->pid); pdd->last_evict_timestamp =get_jiffies_64 ();/* Mark all queues as evicted. Deactivate all active queues on @@ -1169,8 +1171,8 @@ static int evict_process_queues_cpsch(struct device_queue_manager *dqm,if (!pdd->drm_priv) goto out; -pr_debug_ratelimited ("Evicting PASID 0x%x queues\n" , - pdd->process->pasid); +pr_debug_ratelimited ("Evicting process pid 0x%x queues\n" , + pdd->process->lead_thread->pid);/* Mark all queues as evicted. Deactivate all active queues on * the qpd. @@ -1228,8 +1230,8 @@ static int restore_process_queues_nocpsch(struct device_queue_manager *dqm, goto out; } -pr_debug_ratelimited ("Restoring PASID 0x%x queues\n" , - pdd->process->pasid); +pr_debug_ratelimited ("Restoring process pid 0x%x queues\n" , + pdd->process->lead_thread->pid);/* Update PD Base in QPD */ qpd->page_table_base = pd_base;@@ -1312,8 +1314,8 @@ static int restore_process_queues_cpsch(struct device_queue_manager *dqm,if (!pdd->drm_priv) goto vm_not_acquired; -pr_debug_ratelimited ("Restoring PASID 0x%x queues\n" , - pdd->process->pasid); +pr_debug_ratelimited ("Restoring process pid 0x%x queues\n" , + pdd->process->lead_thread->pid);/* Update PD Base in QPD */ qpd->page_table_base =amdgpu_amdkfd_gpuvm_get_process_page_dir (pdd->drm_priv);@@ -2102,7 +2104,7 @@ static void set_queue_as_reset(struct device_queue_manager *dqm, struct queue *qstruct kfd_process_device *pdd =qpd_to_pdd (qpd);dev_err (dqm->dev->adev->dev,"queue id 0x%0x at pasid 0x%0x is reset\n" , - q->properties.queue_id, q->process->pasid); + q->properties.queue_id, pdd->process->lead_thread->pid); pdd->has_reset_queue = true;if (q->properties.is_active) {drivers/gpu/drm/amd/amdkfd/kfd_events.c +8
-7@@ -1128,8 +1128,8 @@ static void lookup_events_by_type_and_signal(struct kfd_process *p,if (type ==KFD_EVENT_TYPE_MEMORY ) {dev_warn (kfd_device, -"Sending SIGSEGV to process %d (pasid 0x%x)" , - p->lead_thread->pid, p->pasid); +"Sending SIGSEGV to process pid %d" , + p->lead_thread->pid);send_sig (SIGSEGV , p->lead_thread,0 ); }@@ -1137,13 +1137,13 @@ static void lookup_events_by_type_and_signal(struct kfd_process *p,if (send_signal) {if (send_sigterm) {dev_warn (kfd_device, -"Sending SIGTERM to process %d (pasid 0x%x)" , - p->lead_thread->pid, p->pasid); +"Sending SIGTERM to process pid %d" , + p->lead_thread->pid);send_sig (SIGTERM , p->lead_thread,0 ); }else {dev_err (kfd_device, -"Process %d (pasid 0x%x) got unhandled exception" , - p->lead_thread->pid, p->pasid); +"Process pid %d got unhandled exception" , + p->lead_thread->pid); } }@@ -1253,7 +1253,8 @@ void kfd_signal_reset_event(struct kfd_node *dev) }if (unlikely (!pdd)) { -WARN_ONCE (1 ,"Could not get device data from pasid:0x%x\n" , p->pasid); +WARN_ONCE (1 ,"Could not get device data from process pid:0x%x\n" , + p->lead_thread->pid);continue ; }
Quoted 7 lines
drivers/gpu/drm/amd/amdkfd/kfd_events.c void kfd_signal_reset_event(struct kfd_node *dev) · line 1267
Quoted 6 lines
drivers/gpu/drm/amd/amdkfd/kfd_events.c +1
-1@@ -1263,7 +1264,7 @@ void kfd_signal_reset_event(struct kfd_node *dev)if (dev->dqm->detect_hang_count) {struct amdgpu_task_info *ti; - ti =amdgpu_vm_get_task_info_pasid (dev->adev, p->pasid); + ti =amdgpu_vm_get_task_info_pasid (dev->adev, pdd->pasid);Can we use
amdgpu_vm_get_task_info_vmhere to avoid a redundant lookup of the VM from the pasid? You'd need a helper to get the vm frompdd->drm_priv(drm_priv_to_vminamdgpu_amdkfd.h).
Same as above, redundant loop up at original code. Will change that in next version.
Quoted 182 lines
Quoted 182 lines
drivers/gpu/drm/amd/amdkfd/kfd_events.c +0
-0if (ti) { dev_err(dev->adev->dev, "Queues reset on process %s tid %d thread %s pid %d\n",drivers/gpu/drm/amd/amdkfd/kfd_packet_manager_v9.c +2
-2@@ -47,7 +47,7 @@ static int pm_map_process_v9(struct packet_manager *pm, packet->bitfields2.exec_cleaner_shader =1 ; packet->bitfields2.diq_enable = (qpd->is_debug) ?1 :0 ; packet->bitfields2.process_quantum =10 ; - packet->bitfields2.pasid = qpd->pqm->process->pasid; + packet->bitfields2.pasid = pdd->pasid; packet->bitfields14.gds_size = qpd->gds_size &0x3F ; packet->bitfields14.gds_size_hi = (qpd->gds_size >>6 ) &0xF ; packet->bitfields14.num_gws = (qpd->mapped_gws_queue) ? qpd->num_gws :0 ;@@ -106,7 +106,7 @@ static int pm_map_process_aldebaran(struct packet_manager *pm, packet->bitfields2.exec_cleaner_shader =1 ; packet->bitfields2.diq_enable = (qpd->is_debug) ?1 :0 ; packet->bitfields2.process_quantum =10 ; - packet->bitfields2.pasid = qpd->pqm->process->pasid; + packet->bitfields2.pasid = pdd->pasid; packet->bitfields14.gds_size = qpd->gds_size &0x3F ; packet->bitfields14.gds_size_hi = (qpd->gds_size >>6 ) &0xF ; packet->bitfields14.num_gws = (qpd->mapped_gws_queue) ? qpd->num_gws :0 ;drivers/gpu/drm/amd/amdkfd/kfd_packet_manager_vi.c +2
-1@@ -42,6 +42,7 @@ unsigned int pm_build_pm4_header(unsigned int opcode, size_t packet_size)static int pm_map_process_vi (struct packet_manager *pm,uint32_t *buffer,struct qcm_process_device *qpd) { +struct kfd_process_device *pdd =qpd_to_pdd (qpd);struct pm4_mes_map_process *packet; packet = (struct pm4_mes_map_process *)buffer;@@ -52,7 +53,7 @@ static int pm_map_process_vi(struct packet_manager *pm, uint32_t *buffer, sizeof(struct pm4_mes_map_process)); packet->bitfields2.diq_enable = (qpd->is_debug) ?1 :0 ; packet->bitfields2.process_quantum =10 ; - packet->bitfields2.pasid = qpd->pqm->process->pasid; + packet->bitfields2.pasid = pdd->pasid; packet->bitfields3.page_table_base = qpd->page_table_base; packet->bitfields10.gds_size = qpd->gds_size; packet->bitfields10.num_gws = qpd->num_gws;drivers/gpu/drm/amd/amdkfd/kfd_priv.h +2
-3@@ -849,6 +849,8 @@ struct kfd_process_device {/* Tracks queue reset status */ bool has_reset_queue; + +u32 pasid; };#define qpd_to_pdd (x) container_of(x, struct kfd_process_device, qpd)@@ -908,8 +910,6 @@ struct kfd_process {/* We want to receive a notification when the mm_struct is destroyed */ struct mmu_notifier mmu_notifier; -u32 pasid; -/* * Array of kfd_process_device pointers, * one for each device the process is using. @@ -970,7 +970,6 @@ struct kfd_process {/* Kobj for our procfs */ struct kobject *kobj;struct kobject *kobj_queues; -struct attribute attr_pasid;/* Keep track cwsr init */ bool has_cwsr;drivers/gpu/drm/amd/amdkfd/kfd_process.c +7
-24@@ -282,8 +282,8 @@ static int kfd_get_cu_occupancy(struct attribute *attr, char *buffer) cu_cnt =0 ; proc = pdd->process;if (pdd->qpd.queue_count ==0 ) { -pr_debug ("Gpu-Id: %d has no active queues for process %d\n" , - dev->id, proc->pasid); +pr_debug ("Gpu-Id: %d has no active queues for process pid %d\n" , + dev->id, (int )proc->lead_thread->pid);return snprintf (buffer,PAGE_SIZE ,"%d\n" , cu_cnt); }@@ -327,12 +327,7 @@ static int kfd_get_cu_occupancy(struct attribute *attr, char *buffer)static ssize_t kfd_procfs_show (struct kobject *kobj,struct attribute *attr,char *buffer) { -if (strcmp (attr->name,"pasid" ) ==0 ) { - struct kfd_process *p = container_of(attr, struct kfd_process, - attr_pasid); - -return snprintf (buffer,PAGE_SIZE ,"%d\n" , p->pasid); - }else if (strncmp (attr->name,"vram_" ,5 ) ==0 ) { +if (strncmp (attr->name,"vram_" ,5 ) ==0 ) { struct kfd_process_device *pdd = container_of(attr, struct kfd_process_device, attr_vram);return snprintf (buffer,PAGE_SIZE ,"%llu\n" ,atomic64_read (&pdd->vram_usage));@@ -887,9 +882,6 @@ struct kfd_process *kfd_create_process(struct task_struct *thread) goto out; } -kfd_sysfs_create_file (process->kobj, &process->attr_pasid, -"pasid" ); - process->kobj_queues =kobject_create_and_add ("queues" , process->kobj);if (!process->kobj_queues)@@ -1055,8 +1047,8 @@ static void kfd_process_destroy_pdds(struct kfd_process *p)for (i =0 ; i < p->n_pdds; i++) {struct kfd_process_device *pdd = p->pdds[i]; -pr_debug ("Releasing pdd (topology id %d) for process (pasid 0x%x)\n" , - pdd->dev->id, p->pasid); +pr_debug ("Releasing pdd (topology id %d, for pid 0x%0x)\n" , + pdd->dev->id, (int )p->lead_thread->pid);kfd_process_device_destroy_cwsr_dgpu (pdd);kfd_process_device_destroy_ib_mem (pdd);@@ -1102,7 +1094,6 @@ static void kfd_process_remove_sysfs(struct kfd_process *p)if (!p->kobj)return ; -sysfs_remove_file (p->kobj, &p->attr_pasid);kobject_del (p->kobj_queues);kobject_put (p->kobj_queues); p->kobj_queues =NULL ;@@ -1171,7 +1162,6 @@ static void kfd_process_wq_release(struct work_struct *work)kfd_event_free_process (p); -kfd_pasid_free (p->pasid);mutex_destroy (&p->mutex);put_task_struct (p->lead_thread);@@ -1524,12 +1514,6 @@ static struct kfd_process *create_process(const struct task_struct *thread)atomic_set (&process->debugged_process_count,0 );sema_init (&process->runtime_enable_sema,0 ); - process->pasid =kfd_pasid_alloc (); -if (process->pasid ==0 ) { - err = -ENOSPC ; - goto err_alloc_pasid; - } - err =pqm_init (&process->pqm, process);if (err !=0 ) goto err_process_pqm_init;@@ -1583,8 +1567,6 @@ static struct kfd_process *create_process(const struct task_struct *thread) err_init_apertures:pqm_uninit (&process->pqm); err_process_pqm_init: -kfd_pasid_free (process->pasid); -err_alloc_pasid:kfd_event_free_process (process); err_event_init:mutex_destroy (&process->mutex);@@ -1723,7 +1705,7 @@ int kfd_process_device_init_vm(struct kfd_process_device *pdd,if (ret) goto err_init_cwsr; - ret =amdgpu_amdkfd_gpuvm_set_vm_pasid (dev->adev, avm, p->pasid); + ret =amdgpu_amdkfd_gpuvm_set_vm_pasid (dev->adev, pdd, avm);if (ret) goto err_set_pasid;@@ -1822,15 +1804,20 @@ struct kfd_process *kfd_lookup_process_by_pasid(u32 pasid) {struct kfd_process *p, *ret_p =NULL ;unsigned int temp; +int i;int idx =srcu_read_lock (&kfd_processes_srcu);
Quoted 12 lines
drivers/gpu/drm/amd/amdkfd/kfd_process.c struct kfd_process *kfd_lookup_process_by_pasid(u32 pasid) · line 1817
Quoted 11 lines
drivers/gpu/drm/amd/amdkfd/kfd_process.c +6
-4hash_for_each_rcu (kfd_processes_table, temp, p, kfd_processes) { -if (p->pasid == pasid) { -kref_get (&p->ref); - ret_p = p; -break ; +for (i =0 ; i < p->n_pdds; i++) { +if (p->pdds[i]->pasid == pasid) { +kref_get (&p->ref); + ret_p = p; +break ; + }I think this won't work correctly. The same PASID can be used for different processes on different GPUs because each adev manages its own
PASID->amdgpu_vmlookup table. Sokfd_lookup_process_by_pasidneeds a new parameter that identifies the GPU adev, and you should only compare pasids, if the adev matches.
I think it is the main concern here: the pasid used here is global in driver by amdgpu_pasid_alloc(16) at amdgpu_driver_open_kms. Each time a render node(partition) got opened, a new pasid value is generated. Its lifetime is until render node got closed. A pdd just uses this pasid. And each adev has its own xarray which saves pasids for this adev.
Regards
Xiaogang
Quoted 194 lines
Regards,
FelixQuoted 190 lines
drivers/gpu/drm/amd/amdkfd/kfd_process.c +17
-13} +if (ret_p) +break ; }srcu_read_unlock (&kfd_processes_srcu, idx);@@ -1991,7 +1978,7 @@ static void evict_process_worker(struct work_struct *work)*/ p = container_of(dwork, struct kfd_process, eviction_work); -pr_debug ("Started evicting pasid 0x%x\n" , p->pasid); +pr_debug ("Started evicting process pid 0x%x\n" , (int )p->lead_thread->pid); ret =kfd_process_evict_queues (p,KFD_QUEUE_EVICTION_TRIGGER_TTM );if (!ret) {/* If another thread already signaled the eviction fence, @@ -2003,9 +1990,9 @@ static void evict_process_worker(struct work_struct *work) msecs_to_jiffies(PROCESS_RESTORE_TIME_MS)))kfd_process_restore_queues (p); -pr_debug ("Finished evicting pasid 0x%x\n" , p->pasid); +pr_debug ("Finished evicting process pid 0x%x\n" , (int )p->lead_thread->pid); }else -pr_err ("Failed to evict queues of pasid 0x%x\n" , p->pasid); +pr_err ("Failed to evict queues of process pid 0x%x\n" , (int )p->lead_thread->pid); }static int restore_process_helper (struct kfd_process *p)@@ -2022,9 +2009,11 @@ static int restore_process_helper(struct kfd_process *p) ret =kfd_process_restore_queues (p);if (!ret) -pr_debug ("Finished restoring pasid 0x%x\n" , p->pasid); +pr_debug ("Finished restoring process pid 0x%x\n" , + (int )p->lead_thread->pid);else -pr_err ("Failed to restore queues of pasid 0x%x\n" , p->pasid); +pr_err ("Failed to restore queues of process pid 0x%x\n" , + (int )p->lead_thread->pid);return ret; }@@ -2041,7 +2030,7 @@ static void restore_process_worker(struct work_struct *work)* lifetime of this thread, kfd_process p will be valid */ p = container_of(dwork, struct kfd_process, restore_work); -pr_debug ("Started restoring pasid 0x%x\n" , p->pasid); +pr_debug ("Started restoring process pasid 0x%x\n" , (int )p->lead_thread->pid);/* Setting last_restore_timestamp before successful restoration. * Otherwise this would have to be set by KGD (restore_process_bos) @@ -2057,8 +2046,8 @@ static void restore_process_worker(struct work_struct *work) ret =restore_process_helper (p);if (ret) { -pr_debug ("Failed to restore BOs of pasid 0x%x, retry after %d ms\n" , - p->pasid,PROCESS_BACK_OFF_TIME_MS ); +pr_debug ("Failed to restore BOs of process pid 0x%x, retry after %d ms\n" , + (int )p->lead_thread->pid,PROCESS_BACK_OFF_TIME_MS );if (mod_delayed_work (kfd_restore_wq, &p->restore_work,msecs_to_jiffies (PROCESS_RESTORE_TIME_MS )))kfd_process_restore_queues (p);@@ -2074,7 +2063,7 @@ void kfd_suspend_all_processes(void)WARN (debug_evictions,"Evicting all processes" );hash_for_each_rcu (kfd_processes_table, temp, p, kfd_processes) {if (kfd_process_evict_queues (p,KFD_QUEUE_EVICTION_TRIGGER_SUSPEND )) -pr_err ("Failed to suspend process 0x%x\n" , p->pasid); +pr_err ("Failed to suspend process pid 0x%x\n" , (int )p->lead_thread->pid);signal_eviction_fence (p); }srcu_read_unlock (&kfd_processes_srcu, idx);@@ -2088,8 +2077,8 @@ int kfd_resume_all_processes(void)hash_for_each_rcu (kfd_processes_table, temp, p, kfd_processes) {if (restore_process_helper (p)) { -pr_err ("Restore process %d failed during resume\n" , - p->pasid); +pr_err ("Restore process pid %d failed during resume\n" , + (int )p->lead_thread->pid); ret = -EFAULT ; } }@@ -2144,7 +2133,7 @@ int kfd_process_drain_interrupts(struct kfd_process_device *pdd)memset (irq_drain_fence,0 ,sizeof (irq_drain_fence)); irq_drain_fence[0 ] = (KFD_IRQ_FENCE_SOURCEID <<8 ) |KFD_IRQ_FENCE_CLIENTID ; - irq_drain_fence[3 ] = pdd->process->pasid; + irq_drain_fence[3 ] = pdd->pasid;/* * For GFX 9.4.3, send the NodeId also in IH cookie DW[3] @@ -2296,7 +2285,7 @@ int kfd_debugfs_mqds_by_process(struct seq_file *m, void *data)hash_for_each_rcu (kfd_processes_table, temp, p, kfd_processes) {seq_printf (m,"Process %d PASID 0x%x:\n" , - p->lead_thread->tgid, p->pasid); + p->lead_thread->tgid, (int )p->lead_thread->pid);mutex_lock (&p->mutex); r =pqm_debugfs_mqds (m, &p->pqm);drivers/gpu/drm/amd/amdkfd/kfd_process_queue_manager.c +5
-5@@ -69,8 +69,8 @@ static int find_available_queue_slot(struct process_queue_manager *pqm,pr_debug ("The new slot id %lu\n" , found);if (found >=KFD_MAX_NUM_OF_QUEUES_PER_PROCESS ) { -pr_info ("Cannot open more queues for process with pasid 0x%x\n" , - pqm->process->pasid); +pr_info ("Cannot open more queues for process with pid 0x%x\n" , + pqm->process->lead_thread->pid);return -ENOMEM ; }@@ -425,8 +425,8 @@ int pqm_create_queue(struct process_queue_manager *pqm, }if (retval !=0 ) { -pr_err ("Pasid 0x%x DQM create queue type %d failed. ret %d\n" , - pqm->process->pasid, type, retval); +pr_err ("process pid 0x%x DQM create queue type %d failed. ret %d\n" , + pqm->process->lead_thread->pid, type, retval); goto err_create_queue; }@@ -520,7 +520,7 @@ int pqm_destroy_queue(struct process_queue_manager *pqm, unsigned int qid) retval = dqm->ops.destroy_queue (dqm, &pdd->qpd, pqn->q);if (retval) {pr_err ("Pasid 0x%x destroy queue %d failed, ret %d\n" , - pqm->process->pasid, + pdd->pasid, pqn->q->properties.queue_id, retval);if (retval != -ETIME ) goto err_destroy_queue;drivers/gpu/drm/amd/amdkfd/kfd_svm.c +10
-7@@ -563,7 +563,8 @@ svm_range_vram_node_new(struct kfd_node *node, struct svm_range *prange,int r; p = container_of(prange->svms, struct kfd_process, svms); -pr_debug ("pasid: %x svms 0x%p [0x%lx 0x%lx]\n" , p->pasid, prange->svms, +pr_debug ("process pid: %x svms 0x%p [0x%lx 0x%lx]\n" , + (int )p->lead_thread->pid, prange->svms, prange->start, prange->last);if (svm_range_validate_svm_bo (node, prange))@@ -3232,7 +3233,8 @@ void svm_range_list_fini(struct kfd_process *p)struct svm_range *prange;struct svm_range *next; -pr_debug ("pasid 0x%x svms 0x%p\n" , p->pasid, &p->svms); +pr_debug ("process pid 0x%x svms 0x%p\n" , (int )p->lead_thread->pid, + &p->svms);cancel_delayed_work_sync (&p->svms.restore_work);@@ -3255,7 +3257,8 @@ void svm_range_list_fini(struct kfd_process *p)mutex_destroy (&p->svms.lock); -pr_debug ("pasid 0x%x svms 0x%p done\n" , p->pasid, &p->svms); +pr_debug ("process pid 0x%x svms 0x%p done\n" , + (int )p->lead_thread->pid, &p->svms); }int svm_range_list_init (struct kfd_process *p)@@ -3618,8 +3621,8 @@ svm_range_set_attr(struct kfd_process *p, struct mm_struct *mm,bool flush_tlb;int r, ret =0 ; -pr_debug ("pasid 0x%x svms 0x%p [0x%llx 0x%llx] pages 0x%llx\n" , - p->pasid, &p->svms, start, start + size -1 , size); +pr_debug ("process pid 0x%x svms 0x%p [0x%llx 0x%llx] pages 0x%llx\n" , + (int )p->lead_thread->pid, &p->svms, start, start + size -1 , size); r =svm_range_check_attr (p, nattr, attrs);if (r)@@ -3727,8 +3730,8 @@ svm_range_set_attr(struct kfd_process *p, struct mm_struct *mm, out:mutex_unlock (&process_info->lock); -pr_debug ("pasid 0x%x svms 0x%p [0x%llx 0x%llx] done, r=%d\n" , p->pasid, - &p->svms, start, start + size -1 , r); +pr_debug ("process pid 0x%x svms 0x%p [0x%llx 0x%llx] done, r=%d\n" , + (int )p->lead_thread->pid, &p->svms, start, start + size -1 , r);return ret ? ret : r; }
Felix Kuehling
maintainerQuoted 17 lines · Xiaogang
Quoted 17 lines
Quoted 17 lines
@@ -1822,15 +1804,20 @@ struct kfd_process *kfd_lookup_process_by_pasid(u32 pasid) {
struct kfd_process *p, *ret_p = NULL;
unsigned int temp;
+ int i;
int idx = srcu_read_lock(&kfd_processes_srcu); hash_for_each_rcu(kfd_processes_table, temp, p, kfd_processes) { - if (p->pasid == pasid) {
- kref_get(&p->ref);
- ret_p = p;
- break;
+ for (i = 0; i < p->n_pdds; i++) {
+ if (p->pdds[i]->pasid == pasid) {
+ kref_get(&p->ref);
+ ret_p = p;
+ break;
+ }
Quoted 3 lines
Quoted 1 line
I think this won't work correctly. The same PASID can be used for different processes on different GPUs because each adev manages its own PASID->amdgpu_vm lookup table. So kfd_lookup_process_by_pasid needs a new parameter that identifies the GPU adev, and you should only compare pasids, if the adev matches.
I think it is the main concern here: the pasid used here is global in driver by
amdgpu_pasid_alloc(16)at amdgpu_driver_open_kms. Each time a render node(partition) got opened, a new pasid value is generated. Its lifetime is until render node got closed. A pdd just uses this pasid. And each adev has its own xarray which saves pasids for this adev.
I think I had a misunderstanding here. I saw the xarray in adev and assumed that each adev allocated PASIDs independently. But there is also a global amdgpu_pasid_ida that I missed. If the PASID allocation is global in the amdgpu_pasid_ida, then each PASID uniquely identifies a VM your code should be fine.
@Christian, we discussed the number of PASIDs consumed on systems with many GPUs and partitions. If the PASIDs are managed globally, then running out of PASIDs is a concern. Do you think we should change this?
Regards,
Felix
Quoted 4 lines
Regards
Xiaogang
Christian König
maintainerAm 21.11.24 um 16:22 schrieb Felix Kuehling:
Quoted 18 lines · Xiaogang
Quoted 17 lines
Quoted 17 lines
Quoted 17 lines
@@ -1822,15 +1804,20 @@ struct kfd_process *kfd_lookup_process_by_pasid(u32 pasid) {
struct kfd_process *p, *ret_p = NULL;
unsigned int temp;
+ int i;
int idx = srcu_read_lock(&kfd_processes_srcu); hash_for_each_rcu(kfd_processes_table, temp, p, kfd_processes) {
- if (p->pasid == pasid) {
- kref_get(&p->ref);
- ret_p = p;
- break;
+ for (i = 0; i < p->n_pdds; i++) {
+ if (p->pdds[i]->pasid == pasid) {
+ kref_get(&p->ref);
+ ret_p = p;
+ break;
+ }
Quoted 5 lines
Quoted 2 lines
Quoted 1 line
I think this won't work correctly. The same PASID can be used for different processes on different GPUs because each adev manages its own PASID->amdgpu_vm lookup table. So kfd_lookup_process_by_pasid needs a new parameter that identifies the GPU adev, and you should only compare pasids, if the adev matches.
I think it is the main concern here: the pasid used here is global in driver by
amdgpu_pasid_alloc(16)at amdgpu_driver_open_kms. Each time a render node(partition) got opened, a new pasid value is generated. Its lifetime is until render node got closed. A pdd just uses this pasid. And each adev has its own xarray which saves pasids for this adev.I think I had a misunderstanding here. I saw the xarray in adev and assumed that each adev allocated PASIDs independently. But there is also a global
amdgpu_pasid_idathat I missed. If the PASID allocation is global in theamdgpu_pasid_ida, then each PASID uniquely identifies a VM your code should be fine.@Christian, we discussed the number of PASIDs consumed on systems with many GPUs and partitions. If the PASIDs are managed globally, then running out of PASIDs is a concern. Do you think we should change this?
I think we could change it, but I'm not sure if we should.
IIRC there used to be some KFD requirement that PASIDs were global unique, but could be that this was dropped together with ATC support.
Regards,
Christian.
Quoted 7 lines
Regards,
FelixQuoted 3 lines
Regards
Xiaogang
Quoted 19 lines · Felix Kuehling
Quoted 17 lines · Xiaogang
Quoted 17 lines
Quoted 17 lines
@@ -1822,15 +1804,20 @@ struct kfd_process *kfd_lookup_process_by_pasid(u32 pasid) {
struct kfd_process *p, *ret_p = NULL;
unsigned int temp;
+ int i;
int idx = srcu_read_lock(&kfd_processes_srcu); hash_for_each_rcu(kfd_processes_table, temp, p, kfd_processes) {
- if (p->pasid == pasid) {
- kref_get(&p->ref);
- ret_p = p;
- break;
+ for (i = 0; i < p->n_pdds; i++) {
+ if (p->pdds[i]->pasid == pasid) {
+ kref_get(&p->ref);
+ ret_p = p;
+ break;
+ }
Quoted 3 lines
Quoted 2 lines
Quoted 1 line
I think this won't work correctly. The same PASID can be used for different processes on different GPUs because each adev manages its own PASID->amdgpu_vm lookup table. So kfd_lookup_process_by_pasid needs a new parameter that identifies the GPU adev, and you should only compare pasids, if the adev matches.
I think it is the main concern here: the pasid used here is global in driver by
amdgpu_pasid_alloc(16)at amdgpu_driver_open_kms. Each time a render node(partition) got opened, a new pasid value is generated. Its lifetime is until render node got closed. A pdd just uses this pasid. And each adev has its own xarray which saves pasids for this adev.I think I had a misunderstanding here. I saw the xarray in adev and assumed that each adev allocated PASIDs independently. But there is also a global
amdgpu_pasid_idathat I missed. If the PASID allocation is global in theamdgpu_pasid_ida, then each PASID uniquely identifies a VM your code should be fine.
We use 16 bits to generate pasid, so could have 64k pasid values. Suppose system has 64 total partitions and each process use all of them, in theory we could have 1k processes.
Regards
Xiaogang
Quoted 9 lines
@Christian, we discussed the number of PASIDs consumed on systems with many GPUs and partitions. If the PASIDs are managed globally, then running out of PASIDs is a concern. Do you think we should change this?
Regards,
FelixQuoted 3 lines
Regards
Xiaogang