[PATCH 4/5] drm/panfrost: Add support for GPU heap allocations
Steven Price
steven.price at arm.com
Fri Jul 19 14:45:21 UTC 2019
On 19/07/2019 15:27, Rob Herring wrote:
> On Thu, Jul 18, 2019 at 9:03 AM Steven Price <steven.price at arm.com> wrote:
>>
>> On 17/07/2019 19:33, Rob Herring wrote:
>>> The midgard/bifrost GPUs need to allocate GPU heap memory which is
>>> allocated on GPU page faults and not pinned in memory. The vendor driver
>>> calls this functionality GROW_ON_GPF.
>>>
>>> This implementation assumes that BOs allocated with the
>>> PANFROST_BO_NOEXEC flag are never mmapped or exported. Both of those may
>>> actually work, but I'm unsure if there's some interaction there. It
>>> would cause the whole object to be pinned in memory which would defeat
>>> the point of this.
>>>
>>> On faults, we map in 2MB at a time in order to utilize huge pages (if
>>> enabled). Currently, once we've mapped pages in, they are only unmapped
>>> if the BO is freed. Once we add shrinker support, we can unmap pages
>>> with the shrinker.
>>>
>>> Cc: Tomeu Vizoso <tomeu.vizoso at collabora.com>
>>> Cc: Boris Brezillon <boris.brezillon at collabora.com>
>>> Cc: Robin Murphy <robin.murphy at arm.com>
>>> Cc: Steven Price <steven.price at arm.com>
>>> Cc: Alyssa Rosenzweig <alyssa at rosenzweig.io>
>>> Signed-off-by: Rob Herring <robh at kernel.org>
>>> ---
>>> drivers/gpu/drm/panfrost/TODO | 2 -
>>> drivers/gpu/drm/panfrost/panfrost_drv.c | 2 +-
>>> drivers/gpu/drm/panfrost/panfrost_gem.c | 14 ++-
>>> drivers/gpu/drm/panfrost/panfrost_gem.h | 8 ++
>>> drivers/gpu/drm/panfrost/panfrost_mmu.c | 114 +++++++++++++++++++++---
>>> include/uapi/drm/panfrost_drm.h | 1 +
>>> 6 files changed, 125 insertions(+), 16 deletions(-)
>>>
>>> diff --git a/drivers/gpu/drm/panfrost/TODO b/drivers/gpu/drm/panfrost/TODO
>>> index c2e44add37d8..64129bf73933 100644
>>> --- a/drivers/gpu/drm/panfrost/TODO
>>> +++ b/drivers/gpu/drm/panfrost/TODO
>>> @@ -14,8 +14,6 @@
>>> The hard part is handling when more address spaces are needed than what
>>> the h/w provides.
>>>
>>> -- Support pinning pages on demand (GPU page faults).
>>> -
>>> - Support userspace controlled GPU virtual addresses. Needed for Vulkan. (Tomeu)
>>>
>>> - Support for madvise and a shrinker.
>>> diff --git a/drivers/gpu/drm/panfrost/panfrost_drv.c b/drivers/gpu/drm/panfrost/panfrost_drv.c
>>> index b91e991bc6a3..9e87d0060202 100644
>>> --- a/drivers/gpu/drm/panfrost/panfrost_drv.c
>>> +++ b/drivers/gpu/drm/panfrost/panfrost_drv.c
>>> @@ -50,7 +50,7 @@ static int panfrost_ioctl_create_bo(struct drm_device *dev, void *data,
>>> struct drm_panfrost_create_bo *args = data;
>>>
>>> if (!args->size || args->pad ||
>>> - (args->flags & ~PANFROST_BO_NOEXEC))
>>> + (args->flags & ~(PANFROST_BO_NOEXEC | PANFROST_BO_HEAP)))
>>> return -EINVAL;
>>>
>>> bo = panfrost_gem_create_with_handle(file, dev, args->size, args->flags,
>>> diff --git a/drivers/gpu/drm/panfrost/panfrost_gem.c b/drivers/gpu/drm/panfrost/panfrost_gem.c
>>> index 37ffec8391da..528396000038 100644
>>> --- a/drivers/gpu/drm/panfrost/panfrost_gem.c
>>> +++ b/drivers/gpu/drm/panfrost/panfrost_gem.c
>>> @@ -87,7 +87,10 @@ static int panfrost_gem_map(struct panfrost_device *pfdev, struct panfrost_gem_o
>>> if (ret)
>>> return ret;
>>>
>>> - return panfrost_mmu_map(bo);
>>> + if (!bo->is_heap)
>>> + ret = panfrost_mmu_map(bo);
>>> +
>>> + return ret;
>>> }
>>>
>>> struct panfrost_gem_object *
>>> @@ -101,7 +104,11 @@ panfrost_gem_create_with_handle(struct drm_file *file_priv,
>>> struct drm_gem_shmem_object *shmem;
>>> struct panfrost_gem_object *bo;
>>>
>>> - size = roundup(size, PAGE_SIZE);
>>> + /* Round up heap allocations to 2MB to keep fault handling simple */
>>> + if (flags & PANFROST_BO_HEAP)
>>> + size = roundup(size, SZ_2M);
>>> + else
>>> + size = roundup(size, PAGE_SIZE);
>>>
>>> shmem = drm_gem_shmem_create_with_handle(file_priv, dev, size, handle);
>>> if (IS_ERR(shmem))
>>> @@ -109,6 +116,9 @@ panfrost_gem_create_with_handle(struct drm_file *file_priv,
>>>
>>> bo = to_panfrost_bo(&shmem->base);
>>> bo->noexec = !!(flags & PANFROST_BO_NOEXEC);
>>> + bo->is_heap = !!(flags & PANFROST_BO_HEAP);
>>> + if (bo->is_heap)
>>> + bo->noexec = true;
>>
>> While I agree an executable heap is pretty weird, I'd prefer making this
>> explicit - i.e. failing the allocation if the flags don't make sense.
>
> Seems a bit strange too to always have to set NOEXEC when HEAP is set.
> There's not really any reason to reject setting just HEAP.
Personally I prefer an explicit API, so you get exactly what you
requested (no extra properties added on). But I'm not too bothered
because I can't actually see why HEAP would ever be used without NOEXEC.
> [...]
>
>>> +int panfrost_mmu_map_fault_addr(struct panfrost_device *pfdev, int as, u64 addr)
>>> +{
>>> + int ret, i;
>>> + struct drm_mm_node *node;
>>> + struct panfrost_gem_object *bo;
>>> + struct address_space *mapping;
>>> + pgoff_t page_offset;
>>> + struct sg_table sgt = {};
>>> + struct page **pages;
>>> +
>>> + node = addr_to_drm_mm_node(pfdev, as, addr);
>>> + if (!node)
>>> + return -ENOENT;
>>> +
>>> + bo = drm_mm_node_to_panfrost_bo(node);
>>> + if (!bo->is_heap) {
>>> + dev_WARN(pfdev->dev, "matching BO is not heap type (GPU VA = %llx)",
>>> + node->start << PAGE_SHIFT);
>>> + return -EINVAL;
>>> + }
>>> + /* Assume 2MB alignment and size multiple */
>>> + addr &= ~((u64)SZ_2M - 1);
>>> + page_offset = addr >> PAGE_SHIFT;
>>> + page_offset -= node->start;
>>> +
>>> + pages = kvmalloc_array(NUM_FAULT_PAGES, sizeof(struct page *), GFP_KERNEL);
>>> + if (!pages)
>>> + return -ENOMEM;
>>> +
>>> + mapping = bo->base.base.filp->f_mapping;
>>> + mapping_set_unevictable(mapping);
>>> +
>>> + for (i = 0; i < NUM_FAULT_PAGES; i++) {
>>> + pages[i] = shmem_read_mapping_page(mapping, page_offset + i);
>>> + if (IS_ERR(pages[i])) {
>>> + ret = PTR_ERR(pages[i]);
>>> + goto err_pages;
>>> + }
>>> + }
>>> +
>>> + ret = sg_alloc_table_from_pages(&sgt, pages, NUM_FAULT_PAGES, 0,
>>> + SZ_2M, GFP_KERNEL);
>>> + if (ret)
>>> + goto err_pages;
>>> +
>>> + if (dma_map_sg(pfdev->dev, sgt.sgl, sgt.nents, DMA_BIDIRECTIONAL) == 0) {
>>> + ret = -EINVAL;
>>> + goto err_map;
>>> + }
>>> +
>>> + mmu_map_sg(pfdev, addr, IOMMU_WRITE | IOMMU_READ | IOMMU_NOEXEC, &sgt);
>>> +
>>> + mmu_write(pfdev, MMU_INT_CLEAR, BIT(as));
>>> + bo->is_mapped = true;
>>> +
>>> + dev_dbg(pfdev->dev, "mapped page fault @ %llx", addr);
>>> +
>>> + return 0;
>>
>> You still need to free sgt and pages - so this should be:
>>
>> ret = 0;
>>
>> to fall through to the clean up below:
>
> I think I had that then thought I forgot the return...
Might be an idea to rename the labels to "out_xxx" to make it look less
like the return is missing? :)
>>
>>> +
>>> +err_map:
>>> + sg_free_table(&sgt);
>>> +err_pages:
>>> + kvfree(pages);
>>> + return ret;
>>> +}
>>> +
>>
>> But actually, you need to store the pages allocated in the buffer object
>> so that they can be freed later. At the moment you have a big memory leak.
>
> Ah yes, now I see the memory ends up in 'Unevictable' bucket. I'd been
> looking at the shmem counts and it seemed fine. I have it working now,
> but still need to figure out how to do a dma_unmap.
I was purely looking at the memory usage in 'top'. I was hoping it would
be lower using HEAP, but instead I saw the leak. I'm afraid I gave up
trying to figure out the best way of implementing the dma_unmap - so
good luck! :)
>>> static const char *access_type_name(struct panfrost_device *pfdev,
>>> u32 fault_status)
>>> {
>>> @@ -323,13 +405,11 @@ static irqreturn_t panfrost_mmu_irq_handler(int irq, void *data)
>>> {
>>> struct panfrost_device *pfdev = data;
>>> u32 status = mmu_read(pfdev, MMU_INT_STAT);
>>> - int i;
>>> + int i, ret;
>>>
>>> if (!status)
>>> return IRQ_NONE;
>>>
>>> - dev_err(pfdev->dev, "mmu irq status=%x\n", status);
>>> -
>>> for (i = 0; status; i++) {
>>> u32 mask = BIT(i) | BIT(i + 16);
>>> u64 addr;
>>> @@ -350,6 +430,17 @@ static irqreturn_t panfrost_mmu_irq_handler(int irq, void *data)
>>> access_type = (fault_status >> 8) & 0x3;
>>> source_id = (fault_status >> 16);
>>>
>>> + /* Page fault only */
>>> + if ((status & mask) == BIT(i)) {
>>> + WARN_ON(exception_type < 0xC1 || exception_type > 0xC4);
>>> +
>>> + ret = panfrost_mmu_map_fault_addr(pfdev, i, addr);
>>> + if (!ret) {
>>> + status &= ~mask;
>>> + continue;
>>
>> In this case the IRQ isn't handled and will remain asserted, which
>> probably isn't going to end particularly well.
>
> This is the success condition. We've already cleared the interrupt in
> panfrost_mmu_map_fault_addr. On failure, we fall thru printing out the
> next message and clear the interrupt. Maybe it would be better to
> clear the interrupt in 1 place...
You're absolutely right - sorry for the noise. I think I must have
misread and assumed this was the failure condition. Like you say though
it might be cleaner to only have the one place that the interrupt is
cleared.
>>
>> Ideally you would switch the address space to UNMAPPED to kill off the
>> job, but at the very least we should acknowledge the interrupt and let
>> the GPU timeout reset the GPU to recover (which is equivalent while we
>> still only use the one AS on the GPU).
>
> I'll hopefully remember this detail when doing multiple AS.
A fair bit of the complexity of kbase comes from trying to avoid the
possibility of one process DoSing another by submitting malicious jobs.
The MMU in particular will basically just pause until it is kicked when
a page fault happens, so UNMAPPED is needed to trigger a failure back
into the execution units to allow the threads to progress and report
failure - which is important if there's another job competing for the
same execution units.
Steve
More information about the dri-devel
mailing list