[PATCH 2/9] drm/amdgpu: Add PASID management
Christian König
deathsimple at vodafone.de
Sat Aug 26 13:27:14 UTC 2017
Am 26.08.2017 um 09:19 schrieb Felix Kuehling:
> Allows assigning a PASID to a VM for identifying VMs involved in page
> faults. The global PASID manager is also exported in the KFD
> interface so that AMDGPU and KFD can share the PASID space.
>
> PASIDs of different sizes can be requested. On APUs, the PASID size
> is deterined by the capabilities of the IOMMU. So KFD must be able
> to allocate PASIDs in a smaller range.
>
> TODO:
> * Actually assign PASIDs to VMs
> * Update the PASID-VMID mapping registers during CS
>
> Change-Id: I88c9357a7c584f10e84b5607ac09eba77c833393
> Signed-off-by: Felix Kuehling <Felix.Kuehling at amd.com>
The patch itself is Reviewed-by: Christian König <christian.koenig at amd.com>.
But I'm a bit confused, doesn't his stuff belong into the IOMMU code?
Regards,
Christian.
> ---
> drivers/gpu/drm/amd/amdgpu/amdgpu_amdkfd_gfx_v7.c | 2 +
> drivers/gpu/drm/amd/amdgpu/amdgpu_amdkfd_gfx_v8.c | 2 +
> drivers/gpu/drm/amd/amdgpu/amdgpu_amdkfd_gfx_v9.c | 2 +
> drivers/gpu/drm/amd/amdgpu/amdgpu_amdkfd_gpuvm.c | 2 +-
> drivers/gpu/drm/amd/amdgpu/amdgpu_kms.c | 2 +-
> drivers/gpu/drm/amd/amdgpu/amdgpu_vm.c | 76 ++++++++++++++++++++++-
> drivers/gpu/drm/amd/amdgpu/amdgpu_vm.h | 14 ++++-
> drivers/gpu/drm/amd/include/kgd_kfd_interface.h | 6 ++
> 8 files changed, 101 insertions(+), 5 deletions(-)
>
> diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_amdkfd_gfx_v7.c b/drivers/gpu/drm/amd/amdgpu/amdgpu_amdkfd_gfx_v7.c
> index 3e28d2b..0807d52 100644
> --- a/drivers/gpu/drm/amd/amdgpu/amdgpu_amdkfd_gfx_v7.c
> +++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_amdkfd_gfx_v7.c
> @@ -188,6 +188,8 @@ static const struct kfd2kgd_calls kfd2kgd = {
> .get_local_mem_info = get_local_mem_info,
> .get_gpu_clock_counter = get_gpu_clock_counter,
> .get_max_engine_clock_in_mhz = get_max_engine_clock_in_mhz,
> + .alloc_pasid = amdgpu_vm_alloc_pasid,
> + .free_pasid = amdgpu_vm_free_pasid,
> .create_process_vm = amdgpu_amdkfd_gpuvm_create_process_vm,
> .destroy_process_vm = amdgpu_amdkfd_gpuvm_destroy_process_vm,
> .get_process_page_dir = amdgpu_amdkfd_gpuvm_get_process_page_dir,
> diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_amdkfd_gfx_v8.c b/drivers/gpu/drm/amd/amdgpu/amdgpu_amdkfd_gfx_v8.c
> index 3b6b4d9..c20c000 100644
> --- a/drivers/gpu/drm/amd/amdgpu/amdgpu_amdkfd_gfx_v8.c
> +++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_amdkfd_gfx_v8.c
> @@ -162,6 +162,8 @@ static const struct kfd2kgd_calls kfd2kgd = {
> .get_local_mem_info = get_local_mem_info,
> .get_gpu_clock_counter = get_gpu_clock_counter,
> .get_max_engine_clock_in_mhz = get_max_engine_clock_in_mhz,
> + .alloc_pasid = amdgpu_vm_alloc_pasid,
> + .free_pasid = amdgpu_vm_free_pasid,
> .create_process_vm = amdgpu_amdkfd_gpuvm_create_process_vm,
> .destroy_process_vm = amdgpu_amdkfd_gpuvm_destroy_process_vm,
> .create_process_gpumem = create_process_gpumem,
> diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_amdkfd_gfx_v9.c b/drivers/gpu/drm/amd/amdgpu/amdgpu_amdkfd_gfx_v9.c
> index 961369d..bb99c64 100644
> --- a/drivers/gpu/drm/amd/amdgpu/amdgpu_amdkfd_gfx_v9.c
> +++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_amdkfd_gfx_v9.c
> @@ -209,6 +209,8 @@ static const struct kfd2kgd_calls kfd2kgd = {
> .get_local_mem_info = get_local_mem_info,
> .get_gpu_clock_counter = get_gpu_clock_counter,
> .get_max_engine_clock_in_mhz = get_max_engine_clock_in_mhz,
> + .alloc_pasid = amdgpu_vm_alloc_pasid,
> + .free_pasid = amdgpu_vm_free_pasid,
> .create_process_vm = amdgpu_amdkfd_gpuvm_create_process_vm,
> .destroy_process_vm = amdgpu_amdkfd_gpuvm_destroy_process_vm,
> .create_process_gpumem = create_process_gpumem,
> diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_amdkfd_gpuvm.c b/drivers/gpu/drm/amd/amdgpu/amdgpu_amdkfd_gpuvm.c
> index 35f7d77..462011c 100644
> --- a/drivers/gpu/drm/amd/amdgpu/amdgpu_amdkfd_gpuvm.c
> +++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_amdkfd_gpuvm.c
> @@ -1397,7 +1397,7 @@ int amdgpu_amdkfd_gpuvm_create_process_vm(struct kgd_dev *kgd, void **vm,
> return -ENOMEM;
>
> /* Initialize the VM context, allocate the page directory and zero it */
> - ret = amdgpu_vm_init(adev, &new_vm->base, AMDGPU_VM_CONTEXT_COMPUTE);
> + ret = amdgpu_vm_init(adev, &new_vm->base, AMDGPU_VM_CONTEXT_COMPUTE, 0);
> if (ret != 0) {
> pr_err("Failed init vm ret %d\n", ret);
> /* Undo everything related to the new VM context */
> diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_kms.c b/drivers/gpu/drm/amd/amdgpu/amdgpu_kms.c
> index e390c01..ba3dc4d 100644
> --- a/drivers/gpu/drm/amd/amdgpu/amdgpu_kms.c
> +++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_kms.c
> @@ -825,7 +825,7 @@ int amdgpu_driver_open_kms(struct drm_device *dev, struct drm_file *file_priv)
> }
>
> r = amdgpu_vm_init(adev, &fpriv->vm,
> - AMDGPU_VM_CONTEXT_GFX);
> + AMDGPU_VM_CONTEXT_GFX, 0);
> if (r) {
> kfree(fpriv);
> goto out_suspend;
> diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_vm.c b/drivers/gpu/drm/amd/amdgpu/amdgpu_vm.c
> index 70d7632..c635699 100644
> --- a/drivers/gpu/drm/amd/amdgpu/amdgpu_vm.c
> +++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_vm.c
> @@ -27,12 +27,59 @@
> */
> #include <linux/dma-fence-array.h>
> #include <linux/interval_tree_generic.h>
> +#include <linux/idr.h>
> #include <drm/drmP.h>
> #include <drm/amdgpu_drm.h>
> #include "amdgpu.h"
> #include "amdgpu_trace.h"
>
> /*
> + * PASID manager
> + *
> + * PASIDs are global address space identifiers that can be shared
> + * between the GPU, an IOMMU and the driver. VMs on different devices
> + * may use the same PASID if they share the same address
> + * space. Therefore PASIDs are allocated using a global IDA. VMs are
> + * looked up from the PASID per amdgpu_device.
> + */
> +static DEFINE_IDA(amdgpu_vm_pasid_ida);
> +
> +/**
> + * amdgpu_vm_alloc_pasid - Allocate a PASID
> + * @bits: Maximum width of the PASID in bits, must be at least 1
> + *
> + * Allocates a PASID of the given width while keeping smaller PASIDs
> + * available if possible.
> + *
> + * Returns a positive integer on success. Returns %-EINVAL if bits==0.
> + * Returns %-ENOSPC if no PASID was avaliable. Returns %-ENOMEM on
> + * memory allocation failure.
> + */
> +int amdgpu_vm_alloc_pasid(unsigned int bits)
> +{
> + int pasid = -EINVAL;
> +
> + for (bits = min(bits, 31U); bits > 0; bits--) {
> + pasid = ida_simple_get(&amdgpu_vm_pasid_ida,
> + 1U << (bits - 1), 1U << bits,
> + GFP_KERNEL);
> + if (pasid != -ENOSPC)
> + break;
> + }
> +
> + return pasid;
> +}
> +
> +/**
> + * amdgpu_vm_free_pasid - Free a PASID
> + * @pasid: PASID to free
> + */
> +void amdgpu_vm_free_pasid(unsigned int pasid)
> +{
> + ida_simple_remove(&amdgpu_vm_pasid_ida, pasid);
> +}
> +
> +/*
> * GPUVM
> * GPUVM is similar to the legacy gart on older asics, however
> * rather than there being a single global gart table
> @@ -2482,7 +2529,7 @@ void amdgpu_vm_adjust_size(struct amdgpu_device *adev, uint64_t vm_size, uint32_
> * Init @vm fields.
> */
> int amdgpu_vm_init(struct amdgpu_device *adev, struct amdgpu_vm *vm,
> - int vm_context)
> + int vm_context, unsigned int pasid)
> {
> const unsigned align = min(AMDGPU_VM_PTB_ALIGN_SIZE,
> AMDGPU_VM_PTE_COUNT(adev) * 8);
> @@ -2562,6 +2609,19 @@ int amdgpu_vm_init(struct amdgpu_device *adev, struct amdgpu_vm *vm,
> if (r)
> goto error_free_root;
>
> + if (pasid) {
> + unsigned long flags;
> +
> + spin_lock_irqsave(&adev->vm_manager.pasid_lock, flags);
> + r = idr_alloc(&adev->vm_manager.pasid_idr, vm, pasid, pasid + 1,
> + GFP_ATOMIC);
> + spin_unlock_irqrestore(&adev->vm_manager.pasid_lock, flags);
> + if (r < 0)
> + goto error_free_root;
> +
> + vm->pasid = pasid;
> + }
> +
> vm->vm_context = vm_context;
> if (vm_context == AMDGPU_VM_CONTEXT_COMPUTE) {
> mutex_lock(&id_mgr->lock);
> @@ -2650,6 +2710,14 @@ void amdgpu_vm_fini(struct amdgpu_device *adev, struct amdgpu_vm *vm)
> mutex_unlock(&id_mgr->lock);
> }
>
> + if (vm->pasid) {
> + unsigned long flags;
> +
> + spin_lock_irqsave(&adev->vm_manager.pasid_lock, flags);
> + idr_remove(&adev->vm_manager.pasid_idr, vm->pasid);
> + spin_unlock_irqrestore(&adev->vm_manager.pasid_lock, flags);
> + }
> +
> amd_sched_entity_fini(vm->entity.sched, &vm->entity);
>
> if (!RB_EMPTY_ROOT(&vm->va)) {
> @@ -2729,6 +2797,9 @@ void amdgpu_vm_manager_init(struct amdgpu_device *adev)
> adev->vm_manager.vm_update_mode = 0;
> #endif
>
> + idr_init(&adev->vm_manager.pasid_idr);
> + spin_lock_init(&adev->vm_manager.pasid_lock);
> +
> adev->vm_manager.n_compute_vms = 0;
> }
>
> @@ -2743,6 +2814,9 @@ void amdgpu_vm_manager_fini(struct amdgpu_device *adev)
> {
> unsigned i, j;
>
> + WARN_ON(!idr_is_empty(&adev->vm_manager.pasid_idr));
> + idr_destroy(&adev->vm_manager.pasid_idr);
> +
> for (i = 0; i < AMDGPU_MAX_VMHUBS; ++i) {
> struct amdgpu_vm_id_manager *id_mgr =
> &adev->vm_manager.id_mgr[i];
> diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_vm.h b/drivers/gpu/drm/amd/amdgpu/amdgpu_vm.h
> index 49e15d7..692b05c 100644
> --- a/drivers/gpu/drm/amd/amdgpu/amdgpu_vm.h
> +++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_vm.h
> @@ -25,6 +25,7 @@
> #define __AMDGPU_VM_H__
>
> #include <linux/rbtree.h>
> +#include <linux/idr.h>
>
> #include "gpu_scheduler.h"
> #include "amdgpu_sync.h"
> @@ -143,8 +144,9 @@ struct amdgpu_vm {
> /* Scheduler entity for page table updates */
> struct amd_sched_entity entity;
>
> - /* client id */
> + /* client id and PASID (TODO: replace client_id with PASID) */
> u64 client_id;
> + unsigned int pasid;
> /* dedicated to vm */
> struct amdgpu_vm_id *reserved_vmid[AMDGPU_MAX_VMHUBS];
>
> @@ -219,14 +221,22 @@ struct amdgpu_vm_manager {
> */
> int vm_update_mode;
>
> + /* PASID to VM mapping, will be used in interrupt context to
> + * look up VM of a page fault
> + */
> + struct idr pasid_idr;
> + spinlock_t pasid_lock;
> +
> /* Number of Compute VMs, used for detecting Compute activity */
> unsigned n_compute_vms;
> };
>
> +int amdgpu_vm_alloc_pasid(unsigned int bits);
> +void amdgpu_vm_free_pasid(unsigned int pasid);
> void amdgpu_vm_manager_init(struct amdgpu_device *adev);
> void amdgpu_vm_manager_fini(struct amdgpu_device *adev);
> int amdgpu_vm_init(struct amdgpu_device *adev, struct amdgpu_vm *vm,
> - int vm_context);
> + int vm_context, unsigned int pasid);
> void amdgpu_vm_fini(struct amdgpu_device *adev, struct amdgpu_vm *vm);
> void amdgpu_vm_get_pd_bo(struct amdgpu_vm *vm,
> struct list_head *validated,
> diff --git a/drivers/gpu/drm/amd/include/kgd_kfd_interface.h b/drivers/gpu/drm/amd/include/kgd_kfd_interface.h
> index e8027b3..5833ef7 100644
> --- a/drivers/gpu/drm/amd/include/kgd_kfd_interface.h
> +++ b/drivers/gpu/drm/amd/include/kgd_kfd_interface.h
> @@ -188,6 +188,9 @@ struct tile_config {
> *
> * @get_max_engine_clock_in_mhz: Retrieves maximum GPU clock in MHz
> *
> + * @alloc_pasid: Allocate a PASID
> + * @free_pasid: Free a PASID
> + *
> * @program_sh_mem_settings: A function that should initiate the memory
> * properties such as main aperture memory type (cache / non cached) and
> * secondary aperture base address, size and memory type.
> @@ -264,6 +267,9 @@ struct kfd2kgd_calls {
>
> uint32_t (*get_max_engine_clock_in_mhz)(struct kgd_dev *kgd);
>
> + int (*alloc_pasid)(unsigned int bits);
> + void (*free_pasid)(unsigned int pasid);
> +
> int (*create_process_vm)(struct kgd_dev *kgd, void **vm,
> void **process_info);
> void (*destroy_process_vm)(struct kgd_dev *kgd, void *vm);
More information about the amd-gfx
mailing list