On 10/27/20 3:45 PM, Keith Busch wrote: > On Tue, Oct 27, 2020 at 02:55:25PM +0100, Philippe Mathieu-Daudé wrote: >> Instead of displaying warning on stderr, use warn_report() >> which also displays it on the monitor. >> >> Signed-off-by: Philippe Mathieu-Daudé <phi...@redhat.com> >> --- >> block/nvme.c | 4 ++-- >> 1 file changed, 2 insertions(+), 2 deletions(-) >> >> diff --git a/block/nvme.c b/block/nvme.c >> index 739a0a700cb..6f1d7f9b2a1 100644 >> --- a/block/nvme.c >> +++ b/block/nvme.c >> @@ -399,8 +399,8 @@ static bool nvme_process_completion(NVMeQueuePair *q) >> } >> cid = le16_to_cpu(c->cid); > > Not related to your patch, but it stands out as odd that this is treated > as an endian type. The field is just an opaque cookie, so there shouldn't > be a need for byte swapping. It in fact looks like this is broken on a > big-endian host, as the swaping on submission uses a 32-bit value. Won't > that truncate the relavant bits?
You are right, thanks for having a look and catching this bug :) I suppose we never tested on big-endian host yet. > >> if (cid == 0 || cid > NVME_QUEUE_SIZE) { >> - fprintf(stderr, "Unexpected CID in completion queue: %" PRIu32 >> "\n", >> - cid); >> + warn_report("NVMe: Unexpected CID in completion queue: >> %"PRIu32", " >> + "queue size: %u", cid, NVME_QUEUE_SIZE); >> continue; >> } >> trace_nvme_complete_command(s, q->index, cid); >