On Mon, Jan 07, 2019 at 10:22:07AM +0800, Hongbo Yao wrote:
> There is an out of bounds array access in nvme_cqe_peding().
>
> When enable irq_thread for nvme interrupt, there is racing between the
> nvmeq->cq_head updating and reading.
Just curious: why did you enable this option? Do you have a workload
where it matters?
> diff --git a/drivers/nvme/host/pci.c b/drivers/nvme/host/pci.c
> index d668682..68375d4 100644
> --- a/drivers/nvme/host/pci.c
> +++ b/drivers/nvme/host/pci.c
> @@ -908,9 +908,11 @@ static void nvme_complete_cqes(struct nvme_queue *nvmeq,
> u16 start, u16 end)
>
> static inline void nvme_update_cq_head(struct nvme_queue *nvmeq)
> {
> - if (++nvmeq->cq_head == nvmeq->q_depth) {
> + if (nvmeq->cq_head == (nvmeq->q_depth - 1)) {
> nvmeq->cq_head = 0;
> nvmeq->cq_phase = !nvmeq->cq_phase;
> + } else {
> + ++nvmeq->cq_head;
No need for the braces above, but otherwise this looks fine. I'll apply
it to nvme-4.21.
