On Tue, Jul 30, 2024 at 10:49 AM Jason Wang <jasowang@xxxxxxxxxx> wrote:
>
> On Mon, Jul 29, 2024 at 2:05 PM Barry Song <21cnbao@xxxxxxxxx> wrote:
> >
> > On Mon, Jul 29, 2024 at 3:42 PM Jason Wang <jasowang@xxxxxxxxxx> wrote:
> > >
> > > On Thu, Jul 25, 2024 at 3:00 PM Barry Song <21cnbao@xxxxxxxxx> wrote:
> > > >
> > > > On Thu, Jul 25, 2024 at 6:08 PM Michal Hocko <mhocko@xxxxxxxx> wrote:
> > > > >
> > > > > On Thu 25-07-24 10:50:45, Barry Song wrote:
> > > > > > On Thu, Jul 25, 2024 at 12:27 AM Michal Hocko <mhocko@xxxxxxxx> wrote:
> > > > > > >
> > > > > > > On Wed 24-07-24 20:55:40, Barry Song wrote:
> > > > > [...]
> > > > > > > > diff --git a/drivers/vdpa/vdpa_user/iova_domain.c b/drivers/vdpa/vdpa_user/iova_domain.c
> > > > > > > > index 791d38d6284c..eff700e5f7a2 100644
> > > > > > > > --- a/drivers/vdpa/vdpa_user/iova_domain.c
> > > > > > > > +++ b/drivers/vdpa/vdpa_user/iova_domain.c
> > > > > > > > @@ -287,28 +287,44 @@ void vduse_domain_remove_user_bounce_pages(struct vduse_iova_domain *domain)
> > > > > > > > {
> > > > > > > > struct vduse_bounce_map *map;
> > > > > > > > unsigned long i, count;
> > > > > > > > + struct page **pages = NULL;
> > > > > > > >
> > > > > > > > write_lock(&domain->bounce_lock);
> > > > > > > > if (!domain->user_bounce_pages)
> > > > > > > > goto out;
> > > > > > > > -
> > > > > > > > count = domain->bounce_size >> PAGE_SHIFT;
> > > > > > > > + write_unlock(&domain->bounce_lock);
> > > > > > > > +
> > > > > > > > + pages = kmalloc_array(count, sizeof(*pages), GFP_KERNEL | __GFP_NOFAIL);
> > > > > > > > + for (i = 0; i < count; i++)
> > > > > > > > + pages[i] = alloc_page(GFP_KERNEL | __GFP_NOFAIL);
> > > > > > >
> > > > > > > AFAICS vduse_domain_release calls this function with
> > > > > > > spin_lock(&domain->iotlb_lock) so dropping &domain->bounce_lock is not
> > > > > > > sufficient.
> > > > > >
> > > > > > yes. this is true:
> > > > > >
> > > > > > static int vduse_domain_release(struct inode *inode, struct file *file)
> > > > > > {
> > > > > > struct vduse_iova_domain *domain = file->private_data;
> > > > > >
> > > > > > spin_lock(&domain->iotlb_lock);
> > > > > > vduse_iotlb_del_range(domain, 0, ULLONG_MAX);
> > > > > > vduse_domain_remove_user_bounce_pages(domain);
> > > > > > vduse_domain_free_kernel_bounce_pages(domain);
> > > > > > spin_unlock(&domain->iotlb_lock);
> > > > > > put_iova_domain(&domain->stream_iovad);
> > > > > > put_iova_domain(&domain->consistent_iovad);
> > > > > > vhost_iotlb_free(domain->iotlb);
> > > > > > vfree(domain->bounce_maps);
> > > > > > kfree(domain);
> > > > > >
> > > > > > return 0;
> > > > > > }
> > > > > >
> > > > > > This is quite a pain. I admit I don't have knowledge of this driver, and I don't
> > > > > > think it's safe to release two locks and then reacquire them. The situation is
> > > > > > rather complex. Therefore, I would prefer if the VDPA maintainers could
> > > > > > take the lead in implementing a proper fix.
> > > > >
> > > > > Would it be possible to move all that work to a deferred context?
> > > >
> > > > My understanding is that we need to be aware of both the iotlb_lock and
> > > > bounce_lock to implement the correct changes. As long as we still need
> > > > to acquire these two locks in a deferred context, there doesn't seem to
> > > > be any difference.
> > > >
> > > > I can do the memory pre-allocation before spin_lock(&domain->iotlb_lock),
> > > > but I have no knowledge whether the "count" will change after I make
> > > > the preallocation.
> > > >
> > > > diff --git a/drivers/vdpa/vdpa_user/iova_domain.c
> > > > b/drivers/vdpa/vdpa_user/iova_domain.c
> > > > index 791d38d6284c..7ec87ef33d42 100644
> > > > --- a/drivers/vdpa/vdpa_user/iova_domain.c
> > > > +++ b/drivers/vdpa/vdpa_user/iova_domain.c
> > > > @@ -544,9 +544,12 @@ static int vduse_domain_release(struct inode
> > > > *inode, struct file *file)
> > > > {
> > > > struct vduse_iova_domain *domain = file->private_data;
> > > >
> > > > + struct page **pages;
> > > > + spin_lock(&domain->iotlb_lock); maybe also + bounce_lock?
> > > > + count = domain->bounce_size >> PAGE_SHIFT;
> > > > + spin_unlock(&domain->iotlb_lock);
> > >
> > > We probably don't need any lock here as bounce_size won't be changed .
> > >
> > > > +
> > > > + preallocate_count_pages(pages, count);
> > > > +
> > > > ....
> > > > 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);
> > >
> > > This seems to work.
> >
> > Thanks, Jason. I personally have no knowledge of vDPA. Could you please help
> > review and test the patch below?
> >
> > From 1f3cae091159bfcaffdb4a999a4a8e37db2eacf1 Mon Sep 17 00:00:00 2001
> > From: Barry Song <v-songbaohua@xxxxxxxx>
> > Date: Wed, 24 Jul 2024 20:55:40 +1200
> > Subject: [PATCH RFC v2] vpda: try to fix the potential crash due to misusing
> > __GFP_NOFAIL
> > MIME-Version: 1.0
> > Content-Type: text/plain; charset=UTF-8
> > Content-Transfer-Encoding: 8bit
> >
> > 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.
> >
> > [RFC]: 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 | 21 ++++++++++++++++-----
> > drivers/vdpa/vdpa_user/iova_domain.h | 3 ++-
> > drivers/vdpa/vdpa_user/vduse_dev.c | 13 ++++++++++++-
> > 3 files changed, 30 insertions(+), 7 deletions(-)
> >
> > diff --git a/drivers/vdpa/vdpa_user/iova_domain.c b/drivers/vdpa/vdpa_user/iova_domain.c
> > index 791d38d6284c..014809ac2b7c 100644
> > --- a/drivers/vdpa/vdpa_user/iova_domain.c
> > +++ b/drivers/vdpa/vdpa_user/iova_domain.c
> > @@ -283,7 +283,7 @@ 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)
> > +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 +294,16 @@ 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);
> > }
> > @@ -543,10 +544,19 @@ 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 = NULL;
> > + unsigned long count, i;
> > +
> > + if (domain->user_bounce_pages) {
> > + 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);
> > + }
> >
> > 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);
> > @@ -554,6 +564,7 @@ static int vduse_domain_release(struct inode *inode, struct file *file)
> > vhost_iotlb_free(domain->iotlb);
> > vfree(domain->bounce_maps);
> > kfree(domain);
> > + kfree(pages);
> >
> > return 0;
> > }
> > diff --git a/drivers/vdpa/vdpa_user/iova_domain.h b/drivers/vdpa/vdpa_user/iova_domain.h
> > index f92f22a7267d..db0b793d86db 100644
> > --- a/drivers/vdpa/vdpa_user/iova_domain.h
> > +++ b/drivers/vdpa/vdpa_user/iova_domain.h
> > @@ -74,7 +74,8 @@ 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);
> >
> > 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..df7c1b6f1350 100644
> > --- a/drivers/vdpa/vdpa_user/vduse_dev.c
> > +++ b/drivers/vdpa/vdpa_user/vduse_dev.c
> > @@ -1030,6 +1030,8 @@ 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 = NULL;
> > + unsigned long count, i;
> > int ret;
> >
> > mutex_lock(&dev->mem_lock);
> > @@ -1044,13 +1046,22 @@ 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);
> > + if (dev->domain->user_bounce_pages) {
> > + count = dev->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);
> > + }
>
> Nit: there's some code duplication with vduse_domain_release().
>
> Others look good to me.
>
> Would you like to post a formal patch?
Jason, thanks!
I haven't tested this patch and I don't have the setup to test. I
wonder if you can
post a tested-by tag before I send a formal patch.
BTW, if we want to have a common function to remove the duplicated code,
what is the name you suggest to have for this function?
>
> Thanks
>
> > +
> > + 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);
> > mmdrop(dev->umem->mm);
> > vfree(dev->umem->pages);
> > kfree(dev->umem);
> > + kfree(pages);
> > dev->umem = NULL;
> > ret = 0;
> > unlock:
> > --
> > 2.34.1
> >
> > >
> > > Thanks
> > >
> > > >
> > > >
> > > > > --
> > > > > Michal Hocko
> > > > > SUSE Labs
> > > >
> >
> > Thanks
> > Barry
> >
>
Thanks
Barry
