libata: mask swap internal and hardware tag

hen we're comparing the hardware completion mask passed in from the
driver with the internal tag pending mask, we need to account for the
fact that the internal tag is different from the hardware tag. If not,
then we can end up either prematurely completing the internal tag (since
it's not set in the hw mask), or simply flag an error:

ata2: illegal qc_active transition (100000000->00000001)

If the internal tag is set, then swap that with the hardware tag in this
case before comparing with what the hardware reports.

Fixes: 3996a413557e ("libata: add extra internal command")
Buglink: https://bugzilla.kernel.org/show_bug.cgi?id=201151
Cc: stable@vger.kernel.org
Reported-by: Paul Sbarra <sbarra.paul@gmail.com>
Tested-by: Paul Sbarra <sbarra.paul@gmail.com>
Signed-off-by: Jens Axboe <axboe@kernel.dk>

diff --git a/drivers/ata/libata-core.c b/drivers/ata/libata-core.c
index 172e32840256bef2365a0841950da638900a098e..3893f9bde1e6bc0db2f081943b74233989366540 100644 (file)
--- a/drivers/ata/libata-core.c
+++ b/drivers/ata/libata-core.c
@@ -5359,10 +5359,20 @@ void ata_qc_complete(struct ata_queued_cmd *qc)
  */
 int ata_qc_complete_multiple(struct ata_port *ap, u64 qc_active)
 {
+       u64 done_mask, ap_qc_active = ap->qc_active;
        int nr_done = 0;
-       u64 done_mask;
 
-       done_mask = ap->qc_active ^ qc_active;
+       /*
+        * If the internal tag is set on ap->qc_active, then we care about
+        * bit0 on the passed in qc_active mask. Move that bit up to match
+        * the internal tag.
+        */
+       if (ap_qc_active & (1ULL << ATA_TAG_INTERNAL)) {
+               qc_active |= (qc_active & 0x01) << ATA_TAG_INTERNAL;
+               qc_active ^= qc_active & 0x01;
+       }
+
+       done_mask = ap_qc_active ^ qc_active;
 
        if (unlikely(done_mask & qc_active)) {
                ata_port_err(ap, "illegal qc_active transition (%08llx->%08llx)\n",