On Wed, Jul 31, 2024 at 11:15 AM Barry Song <21cnbao@xxxxxxxxx> wrote:
>
> On Wed, Jul 31, 2024 at 11:10 AM Jason Wang <jasowang@xxxxxxxxxx> wrote:
> >
> > On Wed, Jul 31, 2024 at 8:03 AM Barry Song <21cnbao@xxxxxxxxx> wrote:
> > >
> > > From: Barry Song <v-songbaohua@xxxxxxxx>
> > >
> > > mm doesn't support non-blockable __GFP_NOFAIL allocation. Because
> > > __GFP_NOFAIL without direct reclamation may just result in a busy
> > > loop within non-sleepable contexts.
> > >
> > > static inline struct page *
> > > __alloc_pages_slowpath(gfp_t gfp_mask, unsigned int order,
> > > struct alloc_context *ac)
> > > {
> > > ...
> > > /*
> > > * Make sure that __GFP_NOFAIL request doesn't leak out and make sure
> > > * we always retry
> > > */
> > > if (gfp_mask & __GFP_NOFAIL) {
> > > /*
> > > * All existing users of the __GFP_NOFAIL are blockable, so warn
> > > * of any new users that actually require GFP_NOWAIT
> > > */
> > > if (WARN_ON_ONCE_GFP(!can_direct_reclaim, gfp_mask))
> > > goto fail;
> > > ...
> > > }
> > > ...
> > > fail:
> > > warn_alloc(gfp_mask, ac->nodemask,
> > > "page allocation failure: order:%u", order);
> > > got_pg:
> > > return page;
> > > }
> > >
> > > Let's move the memory allocation out of the atomic context and use
> > > the normal sleepable context to get pages.
> > >
> > > [RFT]: This has only been compile-tested; I'd prefer if the VDPA maintainers
> > > handles it.
> > >
> > > Cc: "Michael S. Tsirkin" <mst@xxxxxxxxxx>
> > > Cc: Jason Wang <jasowang@xxxxxxxxxx>
> > > Cc: Xuan Zhuo <xuanzhuo@xxxxxxxxxxxxxxxxx>
> > > Cc: "Eugenio Pérez" <eperezma@xxxxxxxxxx>
> > > Cc: Maxime Coquelin <maxime.coquelin@xxxxxxxxxx>
> > > Signed-off-by: Barry Song <v-songbaohua@xxxxxxxx>
> > > ---
> > > drivers/vdpa/vdpa_user/iova_domain.c | 31 +++++++++++++++++++++++-----
> > > drivers/vdpa/vdpa_user/iova_domain.h | 5 ++++-
> > > drivers/vdpa/vdpa_user/vduse_dev.c | 4 +++-
> > > 3 files changed, 33 insertions(+), 7 deletions(-)
> > >
> > > diff --git a/drivers/vdpa/vdpa_user/iova_domain.c b/drivers/vdpa/vdpa_user/iova_domain.c
> > > index 791d38d6284c..9318f059a8b5 100644
> > > --- a/drivers/vdpa/vdpa_user/iova_domain.c
> > > +++ b/drivers/vdpa/vdpa_user/iova_domain.c
> > > @@ -283,7 +283,23 @@ int vduse_domain_add_user_bounce_pages(struct vduse_iova_domain *domain,
> > > return ret;
> > > }
> > >
> > > -void vduse_domain_remove_user_bounce_pages(struct vduse_iova_domain *domain)
> > > +struct page **vduse_domain_alloc_pages_to_remove_bounce(struct vduse_iova_domain *domain)
> > > +{
> > > + struct page **pages;
> > > + unsigned long count, i;
> > > +
> > > + if (!domain->user_bounce_pages)
> > > + return NULL;
> > > +
> > > + count = domain->bounce_size >> PAGE_SHIFT;
> > > + pages = kmalloc_array(count, sizeof(*pages), GFP_KERNEL | __GFP_NOFAIL);
> > > + for (i = 0; i < count; i++)
> > > + pages[i] = alloc_page(GFP_KERNEL | __GFP_NOFAIL);
> > > +
> > > + return pages;
> > > +}
> > > +
> > > +void vduse_domain_remove_user_bounce_pages(struct vduse_iova_domain *domain, struct page **pages)
> > > {
> > > struct vduse_bounce_map *map;
> > > unsigned long i, count;
> > > @@ -294,15 +310,16 @@ void vduse_domain_remove_user_bounce_pages(struct vduse_iova_domain *domain)
> > >
> > > count = domain->bounce_size >> PAGE_SHIFT;
> > > for (i = 0; i < count; i++) {
> > > - struct page *page = NULL;
> > > + struct page *page = pages[i];
> > >
> > > map = &domain->bounce_maps[i];
> > > - if (WARN_ON(!map->bounce_page))
> > > + if (WARN_ON(!map->bounce_page)) {
> > > + put_page(page);
> > > continue;
> > > + }
> > >
> > > /* Copy user page to kernel page if it's in use */
> > > if (map->orig_phys != INVALID_PHYS_ADDR) {
> > > - page = alloc_page(GFP_ATOMIC | __GFP_NOFAIL);
> > > memcpy_from_page(page_address(page),
> > > map->bounce_page, 0, PAGE_SIZE);
> > > }
> > > @@ -310,6 +327,7 @@ void vduse_domain_remove_user_bounce_pages(struct vduse_iova_domain *domain)
> > > map->bounce_page = page;
> > > }
> > > domain->user_bounce_pages = false;
> > > + kfree(pages);
> > > out:
> > > write_unlock(&domain->bounce_lock);
> > > }
> > > @@ -543,10 +561,13 @@ static int vduse_domain_mmap(struct file *file, struct vm_area_struct *vma)
> > > static int vduse_domain_release(struct inode *inode, struct file *file)
> > > {
> > > struct vduse_iova_domain *domain = file->private_data;
> > > + struct page **pages;
> > > +
> > > + pages = vduse_domain_alloc_pages_to_remove_bounce(domain);
> > >
> > > spin_lock(&domain->iotlb_lock);
> > > vduse_iotlb_del_range(domain, 0, ULLONG_MAX);
> > > - vduse_domain_remove_user_bounce_pages(domain);
> > > + vduse_domain_remove_user_bounce_pages(domain, pages);
> > > vduse_domain_free_kernel_bounce_pages(domain);
> > > spin_unlock(&domain->iotlb_lock);
> > > put_iova_domain(&domain->stream_iovad);
> > > diff --git a/drivers/vdpa/vdpa_user/iova_domain.h b/drivers/vdpa/vdpa_user/iova_domain.h
> > > index f92f22a7267d..17efa5555b3f 100644
> > > --- a/drivers/vdpa/vdpa_user/iova_domain.h
> > > +++ b/drivers/vdpa/vdpa_user/iova_domain.h
> > > @@ -74,7 +74,10 @@ void vduse_domain_reset_bounce_map(struct vduse_iova_domain *domain);
> > > int vduse_domain_add_user_bounce_pages(struct vduse_iova_domain *domain,
> > > struct page **pages, int count);
> > >
> > > -void vduse_domain_remove_user_bounce_pages(struct vduse_iova_domain *domain);
> > > +void vduse_domain_remove_user_bounce_pages(struct vduse_iova_domain *domain,
> > > + struct page **pages);
> > > +
> > > +struct page **vduse_domain_alloc_pages_to_remove_bounce(struct vduse_iova_domain *domain);
> > >
> > > void vduse_domain_destroy(struct vduse_iova_domain *domain);
> > >
> > > diff --git a/drivers/vdpa/vdpa_user/vduse_dev.c b/drivers/vdpa/vdpa_user/vduse_dev.c
> > > index 7ae99691efdf..5d8d5810df57 100644
> > > --- a/drivers/vdpa/vdpa_user/vduse_dev.c
> > > +++ b/drivers/vdpa/vdpa_user/vduse_dev.c
> > > @@ -1030,6 +1030,7 @@ static int vduse_dev_queue_irq_work(struct vduse_dev *dev,
> > > static int vduse_dev_dereg_umem(struct vduse_dev *dev,
> > > u64 iova, u64 size)
> > > {
> > > + struct page **pages;
> > > int ret;
> > >
> > > mutex_lock(&dev->mem_lock);
> > > @@ -1044,7 +1045,8 @@ static int vduse_dev_dereg_umem(struct vduse_dev *dev,
> > > if (dev->umem->iova != iova || size != dev->domain->bounce_size)
> > > goto unlock;
> > >
> > > - vduse_domain_remove_user_bounce_pages(dev->domain);
> > > + pages = vduse_domain_alloc_pages_to_remove_bounce(dev->domain);
> > > + vduse_domain_remove_user_bounce_pages(dev->domain, pages);
> > > unpin_user_pages_dirty_lock(dev->umem->pages,
> > > dev->umem->npages, true);
> > > atomic64_sub(dev->umem->npages, &dev->umem->mm->pinned_vm);
> >
> > We miss a kfree(pages); here?
> no.
> i've moved it into vduse_domain_remove_user_bounce_pages.
Ok, but it seems tricky e.g allocated by the caller but freed in
callee. And I think I missed some important issues in the previous
review: The check of user_bounce_pages must be done under the
bounce_lock, otherwise it might race with umem_reg.
So in the case of release(), we know the device is gone, so there's no
need to allocate pages that will be released soon. So we can pass NULL
as a hint and just assign bounce_page to NULL in
vduse_domain_remove_user_bounce_pages().
And in the case of vduse_dev_dereg_umem(), we need to allocate the
pages without checking user_bounce_pages. So in
vduse_domain_remove_user_bounce_pages() if we can free the allocated
pages as well as the pages in the following check
if (!domain->user_bounce_pages)
goto out;
What do you think?
Thanks
>
> >
> > Thanks
> >
> > > --
> > > 2.34.1
> > >
> >
>
