virtio: Improve vq->broken access to avoid any compiler optimization
Subsystems (6)
Sasha Levin
maintainerPATCH 1/3 virtio: Improve vq->broken access to avoid any compiler optimization
From: Parav Pandit <[email protected]>
[ Upstream commit 60f0779862e4ab943810187752c462e85f5fa371 ]
Currently vq->broken field is read by virtqueue_is_broken() in busy loop in one context by virtnet_send_command().
vq->broken is set to true in other process context by virtio_break_device(). Reader and writer are accessing it without any synchronization. This may lead to a compiler optimization which may result to optimize reading vq->broken only once.
Hence, force reading vq->broken on each invocation of virtqueue_is_broken() and also force writing it so that such update is visible to the readers.
It is a theoretical fix that isn't yet encountered in the field.
--- drivers/virtio/virtio_ring.c | 6 ++++-- 1 file changed, 4 insertions(+), 2 deletions(-)
drivers/virtio/virtio_ring.c +4 -2
@@ -1150,7 +1150,7 @@ bool virtqueue_is_broken(struct virtqueue *_vq) {struct vring_virtqueue *vq =to_vvq (_vq); -return vq->broken; +return READ_ONCE (vq->broken); }EXPORT_SYMBOL_GPL (virtqueue_is_broken);
@@ -1164,7 +1164,9 @@ void virtio_break_device(struct virtio_device *dev)list_for_each_entry (_vq, &dev->vqs, list) {struct vring_virtqueue *vq =to_vvq (_vq); - vq->broken = true; + +/* Pairs with READ_ONCE() in virtqueue_is_broken(). */ +WRITE_ONCE (vq->broken, true); } }EXPORT_SYMBOL_GPL (virtio_break_device);
Signature · 6 lines
2.30.2 _______________________________________________ Virtualization mailing list [email protected] https://lists.linuxfoundation.org/mailman/listinfo/virtualization
Sasha Levin
maintainerPATCH 2/3 vringh: Use wiov->used to check for read/write desc order
From: Neeraj Upadhyay <[email protected]>
[ Upstream commit e74cfa91f42c50f7f649b0eca46aa049754ccdbd ]
As __vringh_iov() traverses a descriptor chain, it populates each descriptor entry into either read or write vring iov and increments that iov's ->used member. So, as we iterate over a descriptor chain, at any point, (riov/wriov)->used value gives the number of descriptor enteries available, which are to be read or written by the device. As all read iovs must precede the write iovs, wiov->used should be zero when we are traversing a read descriptor. Current code checks for wiov->i, to figure out whether any previous entry in the current descriptor chain was a write descriptor. However, iov->i is only incremented, when these vring iovs are consumed, at a later point, and remain 0 in __vringh_iov(). So, correct the check for read and write descriptor order, to use wiov->used.
--- drivers/vhost/vringh.c | 2 +- 1 file changed, 1 insertion(+), 1 deletion(-)
drivers/vhost/vringh.c +1 -1
@@ -329,7 +329,7 @@ __vringh_iov(struct vringh *vrh, u16 i, iov = wiov; else { iov = riov; -if (unlikely (wiov && wiov->i)) { +if (unlikely (wiov && wiov->used)) {vringh_bad ("Readable desc %p after writable" , &descs[i]); err = -EINVAL ;
Signature · 6 lines
2.30.2 _______________________________________________ Virtualization mailing list [email protected] https://lists.linuxfoundation.org/mailman/listinfo/virtualization
Sasha Levin
maintainerPATCH 3/3 net/rds: dma_map_sg is entitled to merge entries
In reply to Sasha LevinFrom: Gerd Rausch <[email protected]>
[ Upstream commit fb4b1373dcab086d0619c29310f0466a0b2ceb8a ]
Function "dma_map_sg" is entitled to merge adjacent entries and return a value smaller than what was passed as "nents".
Subsequently "ib_map_mr_sg" needs to work with this value ("sg_dma_len") rather than the original "nents" parameter ("sg_len").
This old RDS bug was exposed and reliably causes kernel panics (using RDMA operations "rds-stress -D") on x86_64 starting with:
commit c588072bba6b ("iommu/vt-d: Convert intel iommu driver to the iommu ops")
Simply put: Linux 5.11 and later.
--- net/rds/ib_frmr.c | 4 ++-- 1 file changed, 2 insertions(+), 2 deletions(-)
net/rds/ib_frmr.c +2 -2
@@ -112,9 +112,9 @@ static int rds_ib_post_reg_frmr(struct rds_ib_mr *ibmr)cpu_relax (); } - ret =ib_map_mr_sg_zbva (frmr->mr, ibmr->sg, ibmr->sg_len, + ret =ib_map_mr_sg_zbva (frmr->mr, ibmr->sg, ibmr->sg_dma_len, &off,PAGE_SIZE ); -if (unlikely (ret != ibmr->sg_len)) +if (unlikely (ret != ibmr->sg_dma_len))return ret <0 ? ret : -EINVAL ;/* Perform a WR for the fast_reg_mr. Each individual page
Signature · 1 line
2.30.2