[PATCH 1/2] drm/amdgpu: Moved fault hash table to amdgpu vm

Zeng, Oak Oak.Zeng at amd.com
Tue Sep 11 14:09:52 UTC 2018


Hi Christian,

I am not sure I understand exactly your thought but see 2 comments inline

Thanks,
Oak

-----Original Message-----
From: Christian König <ckoenig.leichtzumerken at gmail.com> 
Sent: Tuesday, September 11, 2018 3:19 AM
To: Kuehling, Felix <Felix.Kuehling at amd.com>; Koenig, Christian <Christian.Koenig at amd.com>; Zeng, Oak <Oak.Zeng at amd.com>; Oak Zeng <zengshanjun at gmail.com>; amd-gfx at lists.freedesktop.org; Yang, Philip <Philip.Yang at amd.com>
Subject: Re: [PATCH 1/2] drm/amdgpu: Moved fault hash table to amdgpu vm

Hi Felix,

let me try to explain the problem on an example:

1. We have a running job which needs recoverable page faults for accessing the process address space.
2. We run into memory pressure on VRAM and start to evict things.
3. A page tables of the running job is picked up for eviction.
4. We schedule a copy command to move the content of the page table to system memory.
5. The new system memory location of the page table is noted in its BO.
6. We get a fault from the job and swap in the page from the process address space.
7. Now we need to enter the new page address into the page table -> *BAM*

The problem is now that we don't know the address of the page table because the current location was replaced with the future location in step #5.
[Oak]: Step 5, We also mark the corresponding entry of parent page table of the evicted page table to invalid. This invalid bit will cause the utcl2 walker to report a page fault. Depending on implementation, for the evicted page table, its' amdgpu_vm_pt.amdgpu_vm_bo_base will point to a new location in system memory (amdgpu_vm_pt itself is in system memory). In the page fault handler, we allocate a *new* physical page in vram (through ttm), swap in the page content from system memory (using sdma) to this newly allocated page, modify parent page table's corresponding entry to point to this *new allocated* page and set valid bit to 1.

If all the swap-in, swap-out work is done by sdma, we don't need to worry about the case that swap-in happens earlier than swap-out - gpu scheduler will take care the sequence.

If ttm can't turn around for the new page allocation, it can optionally trigger things like OOM. This should be something that need to be considered at ttm physical memory management layer.



We could now argue that we could update the page tables on the fly for the evicted page and never wait for the current job to finish, but this won't work because not all engines can handle that.

I will circumvent this problem for now by blocking the eviction before step #5. 
[Oak] How do you block it? Pin the memory? I agree that swapping out page table should be the last resort during a memory pressure handling. 

The issue with that is that this pipelining is responsible for nearly 20% of the fps in some testcases.

I hope that it won't be that bad when I limit disabling the pipelining to only page tables, but it is certainly possible that we will get pushback on this.

If that happens we probably need to add some flags to limit this workaround even more to only the root PD and all the PDs/PTs which are involved in recoverable faults.

Regards,
Christian.

Am 10.09.2018 um 21:10 schrieb Felix Kuehling:
> I'm not sure why you need to distinguish current and future state when 
> dealing with page faults. When you get a page fault, you know that the 
> GPU is trying to access memory right now, in the present. So you're 
> always working with the current state. When the CPU page table 
> changes, you get an MMU notifier that allows you to invalidate the 
> corresponding GPU PTEs. If you have a valid GPU PTE, it always 
> represents the current states and is in sync with the CPU page table. 
> If the GPU page table is ever outdated, it should have an invalid entry in that place.
>
> If SDMA is used to update the GPU PTEs, there is a delay. The MMU 
> notifier is synchronous, so it shouldn't be a problem. You just wait 
> for the SDMA job to complete before returning. When updating PTEs with 
> new valid addresses, it's asynchronous. But the GPU will continue 
> retrying on the invalid entry until SDMA finishes. So it's also 
> implicitly synchronized on the GPU side.
>
> Regards,
>    Felix
>
>
> On 2018-09-10 05:42 AM, Christian König wrote:
>> Hi Felix & Oak,
>>
>> over the weekend I had the idea that we could just use the shadow BOs 
>> to have the current state in a page fault. They are GTT BOs and CPU 
>> accessible anyway.
>>
>> Regards,
>> Christian.
>>
>> Am 08.09.2018 um 09:34 schrieb Christian König:
>>> Hi Felix,
>>>
>>>> But why do you want to update page tables when there is no more 
>>>> user mode context that cares about them? Is this just to allow 
>>>> pending work to complete after the user mode context is gone? To prevent hangs?
>>> The problem I'm see is that we still don't have a reliable way of 
>>> killing a command submission while making sure that no more faults 
>>> of it are in the flight.
>>>
>>> E.g. the HWS has a function to kill a running job on the hardware, 
>>> but that only works for compute and there is no way of telling when 
>>> that operation has finished.
>>>
>>> But Oak has already convinced me that we should probably work on 
>>> that instead of trying to keep the VM alive as long as possible.
>>>
>>> Regards,
>>> Christian.
>>>
>>> Am 08.09.2018 um 02:27 schrieb Felix Kuehling:
>>>> Hi Christian,
>>>>
>>>> I'm not sure I get your point here.
>>>>
>>>> As I understand it, the amdgpu_vm structure represents the user 
>>>> mode VM context. It has the pointers to the root page directory and 
>>>> the rest of the page table hierarchy. If there is no amdgpu_vm, 
>>>> there is no user mode context that cares about the page tables.
>>>>
>>>> If we leave the page table pointers in the amdgpu_vm, then handling 
>>>> faults after the VM is destroyed is pointless. We can't find the 
>>>> page tables and we can't update them, so there is nothing we can do 
>>>> in response to VM faults.
>>>>
>>>> So I'm guessing that you want to move the page table pointers into 
>>>> a different structure that exists half-independently of the VM. It 
>>>> would be created when the VM is created (that's where we currently 
>>>> allocate the PASID) but can be destroyed slightly later.
>>>>
>>>> But why do you want to update page tables when there is no more 
>>>> user mode context that cares about them? Is this just to allow 
>>>> pending work to complete after the user mode context is gone? To prevent hangs?
>>>>
>>>> Regards,
>>>>     Felix
>>>>
>>>> On 2018-09-10 07:20 AM, Christian König wrote:
>>>>>> Am I right that the handling of page fault need a valid VM?
>>>>> No, exactly that view is incorrect.
>>>>>
>>>>> The VM is meant to be a memory management structure of page tables 
>>>>> and is completely pointless fault processing because it represent 
>>>>> the future state of the page tables and not the current one.
>>>>>
>>>>> What we need for fault processing is a new structure build around 
>>>>> PASIDs which is feed by the with addresses when page tables are 
>>>>> moved around.
>>>>>
>>>>> Alternatively I hope that we can use the SDMA to walk the page 
>>>>> tables and update the required entries by just using the address.
>>>>>
>>>>> Regards,
>>>>> Christian.
>>>>>
>>>>> Am 07.09.2018 um 16:55 schrieb Zeng, Oak:
>>>>>> Hi Christian,
>>>>>>
>>>>>>
>>>>>> What is the value of delay the destroy of hash to after vm is 
>>>>>> destroyed? Since vm is destroyed, you basically don’t have enough 
>>>>>> information to paging in the correct page to gpuvm. Am I right 
>>>>>> that the handling of page fault need a valid VM? For example, you 
>>>>>> need the VM to get the corresponding physical address of the faulty page.
>>>>>>
>>>>>>
>>>>>> After vm is destroyed, the retry interrupt can be directly 
>>>>>> discard as you can’t find vm through pasid. You can think this is 
>>>>>> also one kind of prescreen.
>>>>>>
>>>>>>
>>>>>> So I am stand for the point that, there is no need to delay the 
>>>>>> destroy of hash to after vm is destroyed. Prescreening hash can 
>>>>>> be destroyed at the time of vm_fini.
>>>>>>
>>>>>>
>>>>>> Thanks,
>>>>>>
>>>>>> Oak
>>>>>>
>>>>>>
>>>>>> *From:*Christian König <ckoenig.leichtzumerken at gmail.com>
>>>>>> *Sent:* Friday, September 07, 2018 4:39 AM
>>>>>> *To:* Zeng, Oak <Oak.Zeng at amd.com>; Oak Zeng 
>>>>>> <zengshanjun at gmail.com>; amd-gfx at lists.freedesktop.org
>>>>>> *Cc:* Zeng, Oak <Oak.Zeng at amd.com>
>>>>>> *Subject:* Re: [PATCH 1/2] drm/amdgpu: Moved fault hash table to 
>>>>>> amdgpu vm
>>>>>>
>>>>>>
>>>>>> Hi Oak,
>>>>>>
>>>>>> yeah, but this is still a NAK. Making the hash per PASID was not 
>>>>>> a suggestion but rather a must have.
>>>>>>
>>>>>> The VM structures must be destroyed while the processing is still 
>>>>>> ongoing, or otherwise we would not have a clean OOM handling.
>>>>>>
>>>>>> The IDR for PASID lockup can be put into amdgpu_ids.c, you can 
>>>>>> just replace the IDA already used there with an IDR.
>>>>>>
>>>>>> Since the PASID is already freed up delayed we should have the 
>>>>>> grace period for interrupt processing included. If that still 
>>>>>> isn't sufficient we can still add some delayed work for this.
>>>>>>
>>>>>> Regards,
>>>>>> Christian.
>>>>>>
>>>>>> Am 07.09.2018 um 06:16 schrieb ozeng:
>>>>>>
>>>>>>       Hi Christian,
>>>>>>
>>>>>>       In this implementation, fault hash is made per vm, not per 
>>>>>> pasid
>>>>>>       as suggested, based on below considerations:
>>>>>>
>>>>>>         * Delay the destroy of hash introduces more effort like 
>>>>>> how to
>>>>>>           set the proper grace period after which no retry 
>>>>>> interrupt
>>>>>>           will be generated. We want to avoid those complication 
>>>>>> if we
>>>>>>           can.
>>>>>>         * The  problem of the per vm implementation is that it is 
>>>>>> hard
>>>>>>           to delay the destroy of fault hash, because when vm is
>>>>>>           destroyed, prescreen function won't be able to retrieve 
>>>>>> the
>>>>>>           fault hash. But in this case, the prescreen function 
>>>>>> can
>>>>>>           either let the interrupt go through (to gmc) or ignore it.
>>>>>>           From this perspective, it is not very necessary to 
>>>>>> delay the
>>>>>>           destroy of hash table.
>>>>>>         * On the other hand, since ASICs after vega10 have the HW 
>>>>>> CAM
>>>>>>           feature. So the SW IV prescreen is only useful for vega10.
>>>>>>           For all the HW implemented CAM, we can consider the 
>>>>>> delayed
>>>>>>           CAM acknowledgment.
>>>>>>         * Making it per pasid need to introduce another idr to
>>>>>>           correlate pasid and hash table - where to put the idr? 
>>>>>> You
>>>>>>           will have to make it a global variable which is not 
>>>>>> very
>>>>>>           desirable.
>>>>>>
>>>>>>       The main purpose of this patch is to solve the 
>>>>>> inter-process lock
>>>>>>       issue (when hash table is full), while still keep codes simple.
>>>>>>
>>>>>>       Also with this patch, the faults kfifo is not necessary any 
>>>>>> more.
>>>>>>       See patch 2.
>>>>>>
>>>>>>       Regards,
>>>>>>
>>>>>>       Oak
>>>>>>
>>>>>>
>>>>>>       On 09/06/2018 11:28 PM, Oak Zeng wrote:
>>>>>>
>>>>>>           In stead of share one fault hash table per device, make 
>>>>>> it
>>>>>>
>>>>>>           per vm. This can avoid inter-process lock issue when 
>>>>>> fault
>>>>>>
>>>>>>           hash table is full.
>>>>>>
>>>>>>
>>>>>>           Change-Id: I5d1281b7c41eddc8e26113e010516557588d3708
>>>>>>
>>>>>>           Signed-off-by: Oak Zeng <Oak.Zeng at amd.com> 
>>>>>> <mailto:Oak.Zeng at amd.com>
>>>>>>
>>>>>>           Suggested-by: Christian Konig 
>>>>>> <Christian.Koenig at amd.com> <mailto:Christian.Koenig at amd.com>
>>>>>>
>>>>>>           Suggested-by: Felix Kuehling <Felix.Kuehling at amd.com> 
>>>>>> <mailto:Felix.Kuehling at amd.com>
>>>>>>
>>>>>>           ---
>>>>>>
>>>>>>            drivers/gpu/drm/amd/amdgpu/amdgpu_ih.c |  75
>>>>>> ------------------------
>>>>>>
>>>>>>            drivers/gpu/drm/amd/amdgpu/amdgpu_ih.h |  10 ----
>>>>>>
>>>>>>            drivers/gpu/drm/amd/amdgpu/amdgpu_vm.c | 102
>>>>>> ++++++++++++++++++++++++++++++++-
>>>>>>
>>>>>>            drivers/gpu/drm/amd/amdgpu/amdgpu_vm.h |  12 ++++
>>>>>>
>>>>>>            drivers/gpu/drm/amd/amdgpu/vega10_ih.c |  38 
>>>>>> +++++-------
>>>>>>
>>>>>>            5 files changed, 127 insertions(+), 110 deletions(-)
>>>>>>
>>>>>>
>>>>>>           diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_ih.c
>>>>>> b/drivers/gpu/drm/amd/amdgpu/amdgpu_ih.c
>>>>>>
>>>>>>           index 06373d4..4ed8621 100644
>>>>>>
>>>>>>           --- a/drivers/gpu/drm/amd/amdgpu/amdgpu_ih.c
>>>>>>
>>>>>>           +++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_ih.c
>>>>>>
>>>>>>           @@ -197,78 +197,3 @@ int amdgpu_ih_process(struct 
>>>>>> amdgpu_device *adev)
>>>>>>
>>>>>>              return IRQ_HANDLED;
>>>>>>
>>>>>>            }
>>>>>>
>>>>>>
>>>>>>           -/**
>>>>>>
>>>>>>           - * amdgpu_ih_add_fault - Add a page fault record
>>>>>>
>>>>>>           - *
>>>>>>
>>>>>>           - * @adev: amdgpu device pointer
>>>>>>
>>>>>>           - * @key: 64-bit encoding of PASID and address
>>>>>>
>>>>>>           - *
>>>>>>
>>>>>>           - * This should be called when a retry page fault 
>>>>>> interrupt is
>>>>>>
>>>>>>           - * received. If this is a new page fault, it will be 
>>>>>> added to a hash
>>>>>>
>>>>>>           - * table. The return value indicates whether this is a 
>>>>>> new fault, or
>>>>>>
>>>>>>           - * a fault that was already known and is already being 
>>>>>> handled.
>>>>>>
>>>>>>           - *
>>>>>>
>>>>>>           - * If there are too many pending page faults, this 
>>>>>> will fail. Retry
>>>>>>
>>>>>>           - * interrupts should be ignored in this case until 
>>>>>> there is enough
>>>>>>
>>>>>>           - * free space.
>>>>>>
>>>>>>           - *
>>>>>>
>>>>>>           - * Returns 0 if the fault was added, 1 if the fault 
>>>>>> was already known,
>>>>>>
>>>>>>           - * -ENOSPC if there are too many pending faults.
>>>>>>
>>>>>>           - */
>>>>>>
>>>>>>           -int amdgpu_ih_add_fault(struct amdgpu_device *adev, 
>>>>>> u64
>>>>>> key)
>>>>>>
>>>>>>           -{
>>>>>>
>>>>>>           -  unsigned long flags;
>>>>>>
>>>>>>           -  int r = -ENOSPC;
>>>>>>
>>>>>>           -
>>>>>>
>>>>>>           -  if (WARN_ON_ONCE(!adev->irq.ih.faults))
>>>>>>
>>>>>>           -          /* Should be allocated in <IP>_ih_sw_init on 
>>>>>> GPUs that
>>>>>>
>>>>>>           -           * support retry faults and require retry 
>>>>>> filtering.
>>>>>>
>>>>>>           -           */
>>>>>>
>>>>>>           -          return r;
>>>>>>
>>>>>>           -
>>>>>>
>>>>>>           - spin_lock_irqsave(&adev->irq.ih.faults->lock, flags);
>>>>>>
>>>>>>           -
>>>>>>
>>>>>>           -  /* Only let the hash table fill up to 50% for best 
>>>>>> performance */
>>>>>>
>>>>>>           -  if (adev->irq.ih.faults->count >= (1 <<
>>>>>> (AMDGPU_PAGEFAULT_HASH_BITS-1)))
>>>>>>
>>>>>>           -          goto unlock_out;
>>>>>>
>>>>>>           -
>>>>>>
>>>>>>           -  r = chash_table_copy_in(&adev->irq.ih.faults->hash,
>>>>>> key, NULL);
>>>>>>
>>>>>>           -  if (!r)
>>>>>>
>>>>>>           -          adev->irq.ih.faults->count++;
>>>>>>
>>>>>>           -
>>>>>>
>>>>>>           -  /* chash_table_copy_in should never fail unless 
>>>>>> we're losing count */
>>>>>>
>>>>>>           -  WARN_ON_ONCE(r < 0);
>>>>>>
>>>>>>           -
>>>>>>
>>>>>>           -unlock_out:
>>>>>>
>>>>>>           - spin_unlock_irqrestore(&adev->irq.ih.faults->lock, 
>>>>>> flags);
>>>>>>
>>>>>>           -  return r;
>>>>>>
>>>>>>           -}
>>>>>>
>>>>>>           -
>>>>>>
>>>>>>           -/**
>>>>>>
>>>>>>           - * amdgpu_ih_clear_fault - Remove a page fault record
>>>>>>
>>>>>>           - *
>>>>>>
>>>>>>           - * @adev: amdgpu device pointer
>>>>>>
>>>>>>           - * @key: 64-bit encoding of PASID and address
>>>>>>
>>>>>>           - *
>>>>>>
>>>>>>           - * This should be called when a page fault has been 
>>>>>> handled. Any
>>>>>>
>>>>>>           - * future interrupt with this key will be processed as 
>>>>>> a new
>>>>>>
>>>>>>           - * page fault.
>>>>>>
>>>>>>           - */
>>>>>>
>>>>>>           -void amdgpu_ih_clear_fault(struct amdgpu_device *adev,
>>>>>> u64 key)
>>>>>>
>>>>>>           -{
>>>>>>
>>>>>>           -  unsigned long flags;
>>>>>>
>>>>>>           -  int r;
>>>>>>
>>>>>>           -
>>>>>>
>>>>>>           -  if (!adev->irq.ih.faults)
>>>>>>
>>>>>>           -          return;
>>>>>>
>>>>>>           -
>>>>>>
>>>>>>           - spin_lock_irqsave(&adev->irq.ih.faults->lock, flags);
>>>>>>
>>>>>>           -
>>>>>>
>>>>>>           -  r = chash_table_remove(&adev->irq.ih.faults->hash,
>>>>>> key, NULL);
>>>>>>
>>>>>>           -  if (!WARN_ON_ONCE(r < 0)) {
>>>>>>
>>>>>>           -          adev->irq.ih.faults->count--;
>>>>>>
>>>>>>           - WARN_ON_ONCE(adev->irq.ih.faults->count < 0);
>>>>>>
>>>>>>           -  }
>>>>>>
>>>>>>           -
>>>>>>
>>>>>>           - spin_unlock_irqrestore(&adev->irq.ih.faults->lock, 
>>>>>> flags);
>>>>>>
>>>>>>           -}
>>>>>>
>>>>>>           diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_ih.h
>>>>>> b/drivers/gpu/drm/amd/amdgpu/amdgpu_ih.h
>>>>>>
>>>>>>           index a23e1c0..f411ffb 100644
>>>>>>
>>>>>>           --- a/drivers/gpu/drm/amd/amdgpu/amdgpu_ih.h
>>>>>>
>>>>>>           +++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_ih.h
>>>>>>
>>>>>>           @@ -32,13 +32,6 @@ struct amdgpu_device;
>>>>>>
>>>>>>            #define AMDGPU_IH_CLIENTID_LEGACY 0
>>>>>>
>>>>>>            #define AMDGPU_IH_CLIENTID_MAX SOC15_IH_CLIENTID_MAX
>>>>>>
>>>>>>
>>>>>>           -#define AMDGPU_PAGEFAULT_HASH_BITS 8
>>>>>>
>>>>>>           -struct amdgpu_retryfault_hashtable {
>>>>>>
>>>>>>           -  DECLARE_CHASH_TABLE(hash, 
>>>>>> AMDGPU_PAGEFAULT_HASH_BITS, 8, 0);
>>>>>>
>>>>>>           -  spinlock_t     lock;
>>>>>>
>>>>>>           -  int            count;
>>>>>>
>>>>>>           -};
>>>>>>
>>>>>>           -
>>>>>>
>>>>>>            /*
>>>>>>
>>>>>>             * R6xx+ IH ring
>>>>>>
>>>>>>             */
>>>>>>
>>>>>>           @@ -57,7 +50,6 @@ struct amdgpu_ih_ring {
>>>>>>
>>>>>>              bool                   use_doorbell;
>>>>>>
>>>>>>              bool                   use_bus_addr;
>>>>>>
>>>>>>              dma_addr_t             rb_dma_addr; /* only used 
>>>>>> when use_bus_addr = true */
>>>>>>
>>>>>>           -  struct amdgpu_retryfault_hashtable *faults;
>>>>>>
>>>>>>            };
>>>>>>
>>>>>>
>>>>>>            #define AMDGPU_IH_SRC_DATA_MAX_SIZE_DW 4
>>>>>>
>>>>>>           @@ -95,7 +87,5 @@ int amdgpu_ih_ring_init(struct 
>>>>>> amdgpu_device *adev, unsigned ring_size,
>>>>>>
>>>>>>                              bool use_bus_addr);
>>>>>>
>>>>>>            void amdgpu_ih_ring_fini(struct amdgpu_device *adev);
>>>>>>
>>>>>>            int amdgpu_ih_process(struct amdgpu_device *adev);
>>>>>>
>>>>>>           -int amdgpu_ih_add_fault(struct amdgpu_device *adev, 
>>>>>> u64 key);
>>>>>>
>>>>>>           -void amdgpu_ih_clear_fault(struct amdgpu_device *adev,
>>>>>> u64 key);
>>>>>>
>>>>>>
>>>>>>            #endif
>>>>>>
>>>>>>           diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_vm.c
>>>>>> b/drivers/gpu/drm/amd/amdgpu/amdgpu_vm.c
>>>>>>
>>>>>>           index 1d7e3c1..8b220e9 100644
>>>>>>
>>>>>>           --- a/drivers/gpu/drm/amd/amdgpu/amdgpu_vm.c
>>>>>>
>>>>>>           +++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_vm.c
>>>>>>
>>>>>>           @@ -2692,6 +2692,22 @@ void 
>>>>>> amdgpu_vm_adjust_size(struct amdgpu_device *adev, uint32_t 
>>>>>> min_vm_size,
>>>>>>
>>>>>>                       adev->vm_manager.fragment_size);
>>>>>>
>>>>>>            }
>>>>>>
>>>>>>
>>>>>>           +static struct amdgpu_retryfault_hashtable
>>>>>> *init_fault_hash(void)
>>>>>>
>>>>>>           +{
>>>>>>
>>>>>>           +  struct amdgpu_retryfault_hashtable *fault_hash;
>>>>>>
>>>>>>           +
>>>>>>
>>>>>>           +  fault_hash = kmalloc(sizeof(*fault_hash), 
>>>>>> GFP_KERNEL);
>>>>>>
>>>>>>           +  if (!fault_hash)
>>>>>>
>>>>>>           +          return fault_hash;
>>>>>>
>>>>>>           +
>>>>>>
>>>>>>           +  INIT_CHASH_TABLE(fault_hash->hash,
>>>>>>
>>>>>>           +                  AMDGPU_PAGEFAULT_HASH_BITS, 8, 0);
>>>>>>
>>>>>>           +  spin_lock_init(&fault_hash->lock);
>>>>>>
>>>>>>           +  fault_hash->count = 0;
>>>>>>
>>>>>>           +
>>>>>>
>>>>>>           +  return fault_hash;
>>>>>>
>>>>>>           +}
>>>>>>
>>>>>>           +
>>>>>>
>>>>>>            /**
>>>>>>
>>>>>>             * amdgpu_vm_init - initialize a vm instance
>>>>>>
>>>>>>             *
>>>>>>
>>>>>>           @@ -2780,6 +2796,12 @@ int amdgpu_vm_init(struct 
>>>>>> amdgpu_device *adev, struct amdgpu_vm *vm,
>>>>>>
>>>>>>                      vm->pasid = pasid;
>>>>>>
>>>>>>              }
>>>>>>
>>>>>>
>>>>>>           +  vm->fault_hash = init_fault_hash();
>>>>>>
>>>>>>           +  if (!vm->fault_hash) {
>>>>>>
>>>>>>           +          r = -ENOMEM;
>>>>>>
>>>>>>           +          goto error_free_root;
>>>>>>
>>>>>>           +  }
>>>>>>
>>>>>>           +
>>>>>>
>>>>>>              INIT_KFIFO(vm->faults);
>>>>>>
>>>>>>              vm->fault_credit = 16;
>>>>>>
>>>>>>
>>>>>>           @@ -2973,7 +2995,7 @@ void amdgpu_vm_fini(struct 
>>>>>> amdgpu_device *adev, struct amdgpu_vm *vm)
>>>>>>
>>>>>>
>>>>>>              /* Clear pending page faults from IH when the VM is 
>>>>>> destroyed */
>>>>>>
>>>>>>              while (kfifo_get(&vm->faults, &fault))
>>>>>>
>>>>>>           -          amdgpu_ih_clear_fault(adev, fault);
>>>>>>
>>>>>>           +          amdgpu_vm_clear_fault(vm->fault_hash, 
>>>>>> fault);
>>>>>>
>>>>>>
>>>>>>              if (vm->pasid) {
>>>>>>
>>>>>>                      unsigned long flags;
>>>>>>
>>>>>>           @@ -2983,6 +3005,9 @@ void amdgpu_vm_fini(struct 
>>>>>> amdgpu_device *adev, struct amdgpu_vm *vm)
>>>>>>
>>>>>> spin_unlock_irqrestore(&adev->vm_manager.pasid_lock, flags);
>>>>>>
>>>>>>              }
>>>>>>
>>>>>>
>>>>>>           +  kfree(vm->fault_hash);
>>>>>>
>>>>>>           +  vm->fault_hash = NULL;
>>>>>>
>>>>>>           +
>>>>>>
>>>>>>              drm_sched_entity_destroy(&vm->entity);
>>>>>>
>>>>>>
>>>>>>              if (!RB_EMPTY_ROOT(&vm->va.rb_root)) {
>>>>>>
>>>>>>           @@ -3183,3 +3208,78 @@ void 
>>>>>> amdgpu_vm_set_task_info(struct amdgpu_vm *vm)
>>>>>>
>>>>>>                      }
>>>>>>
>>>>>>              }
>>>>>>
>>>>>>            }
>>>>>>
>>>>>>           +
>>>>>>
>>>>>>           +/**
>>>>>>
>>>>>>           + * amdgpu_vm_add_fault - Add a page fault record to 
>>>>>> fault hash table
>>>>>>
>>>>>>           + *
>>>>>>
>>>>>>           + * @fault_hash: fault hash table
>>>>>>
>>>>>>           + * @key: 64-bit encoding of PASID and address
>>>>>>
>>>>>>           + *
>>>>>>
>>>>>>           + * This should be called when a retry page fault 
>>>>>> interrupt is
>>>>>>
>>>>>>           + * received. If this is a new page fault, it will be 
>>>>>> added to a hash
>>>>>>
>>>>>>           + * table. The return value indicates whether this is a 
>>>>>> new fault, or
>>>>>>
>>>>>>           + * a fault that was already known and is already being 
>>>>>> handled.
>>>>>>
>>>>>>           + *
>>>>>>
>>>>>>           + * If there are too many pending page faults, this 
>>>>>> will fail. Retry
>>>>>>
>>>>>>           + * interrupts should be ignored in this case until 
>>>>>> there is enough
>>>>>>
>>>>>>           + * free space.
>>>>>>
>>>>>>           + *
>>>>>>
>>>>>>           + * Returns 0 if the fault was added, 1 if the fault 
>>>>>> was already known,
>>>>>>
>>>>>>           + * -ENOSPC if there are too many pending faults.
>>>>>>
>>>>>>           + */
>>>>>>
>>>>>>           +int amdgpu_vm_add_fault(struct 
>>>>>> amdgpu_retryfault_hashtable *fault_hash, u64 key)
>>>>>>
>>>>>>           +{
>>>>>>
>>>>>>           +  unsigned long flags;
>>>>>>
>>>>>>           +  int r = -ENOSPC;
>>>>>>
>>>>>>           +
>>>>>>
>>>>>>           +  if (WARN_ON_ONCE(!fault_hash))
>>>>>>
>>>>>>           +          /* Should be allocated in amdgpu_vm_init
>>>>>>
>>>>>>           +           */
>>>>>>
>>>>>>           +          return r;
>>>>>>
>>>>>>           +
>>>>>>
>>>>>>           +  spin_lock_irqsave(&fault_hash->lock, flags);
>>>>>>
>>>>>>           +
>>>>>>
>>>>>>           +  /* Only let the hash table fill up to 50% for best 
>>>>>> performance */
>>>>>>
>>>>>>           +  if (fault_hash->count >= (1 <<
>>>>>> (AMDGPU_PAGEFAULT_HASH_BITS-1)))
>>>>>>
>>>>>>           +          goto unlock_out;
>>>>>>
>>>>>>           +
>>>>>>
>>>>>>           +  r = chash_table_copy_in(&fault_hash->hash, key, 
>>>>>> NULL);
>>>>>>
>>>>>>           +  if (!r)
>>>>>>
>>>>>>           +          fault_hash->count++;
>>>>>>
>>>>>>           +
>>>>>>
>>>>>>           +  /* chash_table_copy_in should never fail unless 
>>>>>> we're losing count */
>>>>>>
>>>>>>           +  WARN_ON_ONCE(r < 0);
>>>>>>
>>>>>>           +
>>>>>>
>>>>>>           +unlock_out:
>>>>>>
>>>>>>           +  spin_unlock_irqrestore(&fault_hash->lock, flags);
>>>>>>
>>>>>>           +  return r;
>>>>>>
>>>>>>           +}
>>>>>>
>>>>>>           +
>>>>>>
>>>>>>           +/**
>>>>>>
>>>>>>           + * amdgpu_vm_clear_fault - Remove a page fault record
>>>>>>
>>>>>>           + *
>>>>>>
>>>>>>           + * @fault_hash: fault hash table
>>>>>>
>>>>>>           + * @key: 64-bit encoding of PASID and address
>>>>>>
>>>>>>           + *
>>>>>>
>>>>>>           + * This should be called when a page fault has been 
>>>>>> handled. Any
>>>>>>
>>>>>>           + * future interrupt with this key will be processed as 
>>>>>> a new
>>>>>>
>>>>>>           + * page fault.
>>>>>>
>>>>>>           + */
>>>>>>
>>>>>>           +void amdgpu_vm_clear_fault(struct 
>>>>>> amdgpu_retryfault_hashtable *fault_hash, u64 key)
>>>>>>
>>>>>>           +{
>>>>>>
>>>>>>           +  unsigned long flags;
>>>>>>
>>>>>>           +  int r;
>>>>>>
>>>>>>           +
>>>>>>
>>>>>>           +  if (!fault_hash)
>>>>>>
>>>>>>           +          return;
>>>>>>
>>>>>>           +
>>>>>>
>>>>>>           +  spin_lock_irqsave(&fault_hash->lock, flags);
>>>>>>
>>>>>>           +
>>>>>>
>>>>>>           +  r = chash_table_remove(&fault_hash->hash, key, 
>>>>>> NULL);
>>>>>>
>>>>>>           +  if (!WARN_ON_ONCE(r < 0)) {
>>>>>>
>>>>>>           +          fault_hash->count--;
>>>>>>
>>>>>>           +          WARN_ON_ONCE(fault_hash->count < 0);
>>>>>>
>>>>>>           +  }
>>>>>>
>>>>>>           +
>>>>>>
>>>>>>           +  spin_unlock_irqrestore(&fault_hash->lock, flags);
>>>>>>
>>>>>>           +}
>>>>>>
>>>>>>           diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_vm.h
>>>>>> b/drivers/gpu/drm/amd/amdgpu/amdgpu_vm.h
>>>>>>
>>>>>>           index e275ee7..6eb1da1 100644
>>>>>>
>>>>>>           --- a/drivers/gpu/drm/amd/amdgpu/amdgpu_vm.h
>>>>>>
>>>>>>           +++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_vm.h
>>>>>>
>>>>>>           @@ -178,6 +178,13 @@ struct amdgpu_task_info {
>>>>>>
>>>>>>              pid_t   tgid;
>>>>>>
>>>>>>            };
>>>>>>
>>>>>>
>>>>>>           +#define AMDGPU_PAGEFAULT_HASH_BITS 8
>>>>>>
>>>>>>           +struct amdgpu_retryfault_hashtable {
>>>>>>
>>>>>>           +  DECLARE_CHASH_TABLE(hash, 
>>>>>> AMDGPU_PAGEFAULT_HASH_BITS, 8, 0);
>>>>>>
>>>>>>           +  spinlock_t     lock;
>>>>>>
>>>>>>           +  int            count;
>>>>>>
>>>>>>           +};
>>>>>>
>>>>>>           +
>>>>>>
>>>>>>            struct amdgpu_vm {
>>>>>>
>>>>>>              /* tree of virtual addresses mapped */
>>>>>>
>>>>>>              struct rb_root_cached  va;
>>>>>>
>>>>>>           @@ -240,6 +247,7 @@ struct amdgpu_vm {
>>>>>>
>>>>>>              struct ttm_lru_bulk_move lru_bulk_move;
>>>>>>
>>>>>>              /* mark whether can do the bulk move */
>>>>>>
>>>>>>              bool                   bulk_moveable;
>>>>>>
>>>>>>           +  struct amdgpu_retryfault_hashtable *fault_hash;
>>>>>>
>>>>>>            };
>>>>>>
>>>>>>
>>>>>>            struct amdgpu_vm_manager {
>>>>>>
>>>>>>           @@ -355,4 +363,8 @@ void amdgpu_vm_set_task_info(struct 
>>>>>> amdgpu_vm *vm);
>>>>>>
>>>>>>            void amdgpu_vm_move_to_lru_tail(struct amdgpu_device 
>>>>>> *adev,
>>>>>>
>>>>>>                                     struct amdgpu_vm *vm);
>>>>>>
>>>>>>
>>>>>>           +int amdgpu_vm_add_fault(struct 
>>>>>> amdgpu_retryfault_hashtable *fault_hash, u64 key);
>>>>>>
>>>>>>           +
>>>>>>
>>>>>>           +void amdgpu_vm_clear_fault(struct 
>>>>>> amdgpu_retryfault_hashtable *fault_hash, u64 key);
>>>>>>
>>>>>>           +
>>>>>>
>>>>>>            #endif
>>>>>>
>>>>>>           diff --git a/drivers/gpu/drm/amd/amdgpu/vega10_ih.c
>>>>>> b/drivers/gpu/drm/amd/amdgpu/vega10_ih.c
>>>>>>
>>>>>>           index 5ae5ed2..acbe5a7 100644
>>>>>>
>>>>>>           --- a/drivers/gpu/drm/amd/amdgpu/vega10_ih.c
>>>>>>
>>>>>>           +++ b/drivers/gpu/drm/amd/amdgpu/vega10_ih.c
>>>>>>
>>>>>>           @@ -265,35 +265,36 @@ static bool 
>>>>>> vega10_ih_prescreen_iv(struct amdgpu_device *adev)
>>>>>>
>>>>>>                      return true;
>>>>>>
>>>>>>              }
>>>>>>
>>>>>>
>>>>>>           -  addr = ((u64)(dw5 & 0xf) << 44) | ((u64)dw4 << 12);
>>>>>>
>>>>>>           -  key = AMDGPU_VM_FAULT(pasid, addr);
>>>>>>
>>>>>>           -  r = amdgpu_ih_add_fault(adev, key);
>>>>>>
>>>>>>           -
>>>>>>
>>>>>>           -  /* Hash table is full or the fault is already being 
>>>>>> processed,
>>>>>>
>>>>>>           -   * ignore further page faults
>>>>>>
>>>>>>           -   */
>>>>>>
>>>>>>           -  if (r != 0)
>>>>>>
>>>>>>           -          goto ignore_iv;
>>>>>>
>>>>>>           -
>>>>>>
>>>>>>              /* Track retry faults in per-VM fault FIFO. */
>>>>>>
>>>>>>              spin_lock(&adev->vm_manager.pasid_lock);
>>>>>>
>>>>>>              vm = idr_find(&adev->vm_manager.pasid_idr, pasid);
>>>>>>
>>>>>>           +  addr = ((u64)(dw5 & 0xf) << 44) | ((u64)dw4 << 12);
>>>>>>
>>>>>>           +  key = AMDGPU_VM_FAULT(pasid, addr);
>>>>>>
>>>>>>              if (!vm) {
>>>>>>
>>>>>>                      /* VM not found, process it normally */
>>>>>>
>>>>>> spin_unlock(&adev->vm_manager.pasid_lock);
>>>>>>
>>>>>>           -          amdgpu_ih_clear_fault(adev, key);
>>>>>>
>>>>>>                      return true;
>>>>>>
>>>>>>           +  } else {
>>>>>>
>>>>>>           +          r = amdgpu_vm_add_fault(vm->fault_hash, 
>>>>>> key);
>>>>>>
>>>>>>           +
>>>>>>
>>>>>>           +          /* Hash table is full or the fault is 
>>>>>> already being processed,
>>>>>>
>>>>>>           +           * ignore further page faults
>>>>>>
>>>>>>           +           */
>>>>>>
>>>>>>           +          if (r != 0) {
>>>>>>
>>>>>>           + spin_unlock(&adev->vm_manager.pasid_lock);
>>>>>>
>>>>>>           +                  goto ignore_iv;
>>>>>>
>>>>>>           +          }
>>>>>>
>>>>>>              }
>>>>>>
>>>>>>              /* No locking required with single writer and single 
>>>>>> reader */
>>>>>>
>>>>>>              r = kfifo_put(&vm->faults, key);
>>>>>>
>>>>>>              if (!r) {
>>>>>>
>>>>>>                      /* FIFO is full. Ignore it until there is 
>>>>>> space */
>>>>>>
>>>>>>           +          amdgpu_vm_clear_fault(vm->fault_hash, key);
>>>>>>
>>>>>> spin_unlock(&adev->vm_manager.pasid_lock);
>>>>>>
>>>>>>           -          amdgpu_ih_clear_fault(adev, key);
>>>>>>
>>>>>>                      goto ignore_iv;
>>>>>>
>>>>>>              }
>>>>>>
>>>>>>           - spin_unlock(&adev->vm_manager.pasid_lock);
>>>>>>
>>>>>>
>>>>>>           + spin_unlock(&adev->vm_manager.pasid_lock);
>>>>>>
>>>>>>              /* It's the first fault for this address, process it 
>>>>>> normally */
>>>>>>
>>>>>>              return true;
>>>>>>
>>>>>>
>>>>>>           @@ -386,14 +387,6 @@ static int vega10_ih_sw_init(void
>>>>>> *handle)
>>>>>>
>>>>>>              adev->irq.ih.use_doorbell = true;
>>>>>>
>>>>>>              adev->irq.ih.doorbell_index = AMDGPU_DOORBELL64_IH 
>>>>>> << 1;
>>>>>>
>>>>>>
>>>>>>           -  adev->irq.ih.faults = 
>>>>>> kmalloc(sizeof(*adev->irq.ih.faults), GFP_KERNEL);
>>>>>>
>>>>>>           -  if (!adev->irq.ih.faults)
>>>>>>
>>>>>>           -          return -ENOMEM;
>>>>>>
>>>>>>           - INIT_CHASH_TABLE(adev->irq.ih.faults->hash,
>>>>>>
>>>>>>           -                   AMDGPU_PAGEFAULT_HASH_BITS, 8, 0);
>>>>>>
>>>>>>           - spin_lock_init(&adev->irq.ih.faults->lock);
>>>>>>
>>>>>>           -  adev->irq.ih.faults->count = 0;
>>>>>>
>>>>>>           -
>>>>>>
>>>>>>              r = amdgpu_irq_init(adev);
>>>>>>
>>>>>>
>>>>>>              return r;
>>>>>>
>>>>>>           @@ -406,9 +399,6 @@ static int vega10_ih_sw_fini(void
>>>>>> *handle)
>>>>>>
>>>>>>              amdgpu_irq_fini(adev);
>>>>>>
>>>>>>              amdgpu_ih_ring_fini(adev);
>>>>>>
>>>>>>
>>>>>>           -  kfree(adev->irq.ih.faults);
>>>>>>
>>>>>>           -  adev->irq.ih.faults = NULL;
>>>>>>
>>>>>>           -
>>>>>>
>>>>>>              return 0;
>>>>>>
>>>>>>            }
>>>>>>
>>>>>>
>>>>>>
>>>>>>
>>>>>>
>>>>>>
>>>>>>       _______________________________________________
>>>>>>
>>>>>>       amd-gfx mailing list
>>>>>>
>>>>>>       amd-gfx at lists.freedesktop.org 
>>>>>> <mailto:amd-gfx at lists.freedesktop.org>
>>>>>>
>>>>>>       https://lists.freedesktop.org/mailman/listinfo/amd-gfx
>>>>>>
>>>>>>
>>>>>
>>>>> _______________________________________________
>>>>> amd-gfx mailing list
>>>>> amd-gfx at lists.freedesktop.org
>>>>> https://lists.freedesktop.org/mailman/listinfo/amd-gfx
>>> _______________________________________________
>>> amd-gfx mailing list
>>> amd-gfx at lists.freedesktop.org
>>> https://lists.freedesktop.org/mailman/listinfo/amd-gfx
> _______________________________________________
> amd-gfx mailing list
> amd-gfx at lists.freedesktop.org
> https://lists.freedesktop.org/mailman/listinfo/amd-gfx



More information about the amd-gfx mailing list