lkml.org 
[lkml]   [2019]   [Aug]   [8]   [last100]   RSS Feed
Views: [wrap][no wrap]   [headers]  [forward] 
 
Messages in this thread
/
SubjectRe: [PATCH] nouveau/hmm: map pages after migration
From
Date

On 8/8/19 12:07 AM, Christoph Hellwig wrote:
> On Wed, Aug 07, 2019 at 08:02:14AM -0700, Ralph Campbell wrote:
>> When memory is migrated to the GPU it is likely to be accessed by GPU
>> code soon afterwards. Instead of waiting for a GPU fault, map the
>> migrated memory into the GPU page tables with the same access permissions
>> as the source CPU page table entries. This preserves copy on write
>> semantics.
>>
>> Signed-off-by: Ralph Campbell <rcampbell@nvidia.com>
>> Cc: Christoph Hellwig <hch@lst.de>
>> Cc: Jason Gunthorpe <jgg@mellanox.com>
>> Cc: "Jérôme Glisse" <jglisse@redhat.com>
>> Cc: Ben Skeggs <bskeggs@redhat.com>
>> ---
>>
>> This patch is based on top of Christoph Hellwig's 9 patch series
>> https://lore.kernel.org/linux-mm/20190729234611.GC7171@redhat.com/T/#u
>> "turn the hmm migrate_vma upside down" but without patch 9
>> "mm: remove the unused MIGRATE_PFN_WRITE" and adds a use for the flag.
>
> This looks useful. I've already dropped that patch for the pending
> resend.

Thanks.

>
>> static unsigned long nouveau_dmem_migrate_copy_one(struct nouveau_drm *drm,
>> - struct vm_area_struct *vma, unsigned long addr,
>> - unsigned long src, dma_addr_t *dma_addr)
>> + struct vm_area_struct *vma, unsigned long src,
>> + dma_addr_t *dma_addr, u64 *pfn)
>
> I'll pick up the removal of the not needed addr argument for the patch
> introducing nouveau_dmem_migrate_copy_one, thanks,
>
>> static void nouveau_dmem_migrate_chunk(struct migrate_vma *args,
>> - struct nouveau_drm *drm, dma_addr_t *dma_addrs)
>> + struct nouveau_drm *drm, dma_addr_t *dma_addrs, u64 *pfns)
>> {
>> struct nouveau_fence *fence;
>> unsigned long addr = args->start, nr_dma = 0, i;
>>
>> for (i = 0; addr < args->end; i++) {
>> args->dst[i] = nouveau_dmem_migrate_copy_one(drm, args->vma,
>> - addr, args->src[i], &dma_addrs[nr_dma]);
>> + args->src[i], &dma_addrs[nr_dma], &pfns[i]);
>
> Nit: I find the &pfns[i] way to pass the argument a little weird to read.
> Why not "pfns + i"?

OK, will do in v2.
Should I convert to "dma_addrs + nr_dma" too?

>> +u64 *
>> +nouveau_pfns_alloc(unsigned long npages)
>> +{
>> + struct nouveau_pfnmap_args *args;
>> +
>> + args = kzalloc(sizeof(*args) + npages * sizeof(args->p.phys[0]),
>
> Can we use struct_size here?

Yes, good suggestion.

>
>> + int ret;
>> +
>> + if (!svm)
>> + return;
>> +
>> + mutex_lock(&svm->mutex);
>> + svmm = nouveau_find_svmm(svm, mm);
>> + if (!svmm) {
>> + mutex_unlock(&svm->mutex);
>> + return;
>> + }
>> + mutex_unlock(&svm->mutex);
>
> Given that nouveau_find_svmm doesn't take any kind of reference, what
> gurantees svmm doesn't go away after dropping the lock?

I asked Ben and Jerome about this too.
I'm still looking into it.

>
>> @@ -44,5 +49,19 @@ static inline int nouveau_svmm_bind(struct drm_device *device, void *p,
>> {
>> return -ENOSYS;
>> }
>> +
>> +u64 *nouveau_pfns_alloc(unsigned long npages)
>> +{
>> + return NULL;
>> +}
>> +
>> +void nouveau_pfns_free(u64 *pfns)
>> +{
>> +}
>> +
>> +void nouveau_pfns_map(struct nouveau_drm *drm, struct mm_struct *mm,
>> + unsigned long addr, u64 *pfns, unsigned long npages)
>> +{
>> +}
>> #endif /* IS_ENABLED(CONFIG_DRM_NOUVEAU_SVM) */
>
> nouveau_dmem.c and nouveau_svm.c are both built conditional on
> CONFIG_DRM_NOUVEAU_SVM, so there is no need for stubs here.
>

Good point. I'll remove them in v2.

\
 
 \ /
  Last update: 2019-08-08 23:29    [W:0.106 / U:0.380 seconds]
©2003-2020 Jasper Spaans|hosted at Digital Ocean and TransIP|Read the blog|Advertise on this site