| Message ID | 1568030756-17428-1-git-send-email-loberman@redhat.com (mailing list archive) |
|---|---|
| State | Superseded |
| Headers | show |
| Series | bnx2fc: Handle scope bits when array returns BUSY or TASK_SET_FULL | expand |
On 9/9/19 5:05 AM, Laurence Oberman wrote: > The qla2xxx driver had this issue as well when the newer array > firmware returned the retry_delay_timer in the fcp_rsp. > The bnx2fc is not handling the masking of the scope bits either > so the retry_delay_timestamp value lands up being a large value > added to the timer timestamp delaying I/O for up to 27 Minutes. > This patch adds similar code to handle this to the > bnx2fc driver to avoid the huge delay. > > V2. Indent comments as suggested > > Signed-off-by: Laurence Oberman <loberman@redhat.com> > --- > drivers/scsi/bnx2fc/bnx2fc_io.c | 23 ++++++++++++++++++++--- > 1 file changed, 20 insertions(+), 3 deletions(-) > > diff --git a/drivers/scsi/bnx2fc/bnx2fc_io.c b/drivers/scsi/bnx2fc/bnx2fc_io.c > index 9e50e5b..39f4aeb 100644 > --- a/drivers/scsi/bnx2fc/bnx2fc_io.c > +++ b/drivers/scsi/bnx2fc/bnx2fc_io.c > @@ -1928,6 +1928,7 @@ void bnx2fc_process_scsi_cmd_compl(struct bnx2fc_cmd *io_req, > struct bnx2fc_rport *tgt = io_req->tgt; > struct scsi_cmnd *sc_cmd; > struct Scsi_Host *host; > + u16 scope, qualifier = 0; > > > /* scsi_cmd_cmpl is called with tgt lock held */ > @@ -1997,12 +1998,28 @@ void bnx2fc_process_scsi_cmd_compl(struct bnx2fc_cmd *io_req, > > if (io_req->cdb_status == SAM_STAT_TASK_SET_FULL || > io_req->cdb_status == SAM_STAT_BUSY) { > + /* Newer array firmware with BUSY or > + * TASK_SET_FULL may return a status that needs > + * the scope bits masked. > + * Or a huge delay timestamp up to 27 minutes > + * can result. > + */ > + if (fcp_rsp->retry_delay_timer) { > + /* Upper 2 bits */ > + scope = fcp_rsp->retry_delay_timer > + & 0xC000; > + /* Lower 14 bits */ > + qualifier = fcp_rsp->retry_delay_timer > + & 0x3FFF; > + } > + if (scope > 0 && qualifier > 0 && > + qualifier <= 0x3FEF) { > /* Set the jiffies + > retry_delay_timer * 100ms > for the rport/tgt > */ > - tgt->retry_delay_timestamp = jiffies + > - fcp_rsp->retry_delay_timer * HZ / 10; > + tgt->retry_delay_timestamp = jiffies + > + (qualifier * HZ / 10); > + } > } > - > } > if (io_req->fcp_resid) > scsi_set_resid(sc_cmd, io_req->fcp_resid); > Reviewed-by: Lee Duncan <lduncan@suse.com>
On Mon, 2019-09-09 at 08:05 -0400, Laurence Oberman wrote: > The qla2xxx driver had this issue as well when the newer array > firmware returned the retry_delay_timer in the fcp_rsp. > The bnx2fc is not handling the masking of the scope bits either > so the retry_delay_timestamp value lands up being a large value > added to the timer timestamp delaying I/O for up to 27 Minutes. > This patch adds similar code to handle this to the > bnx2fc driver to avoid the huge delay. > > V2. Indent comments as suggested > > Signed-off-by: Laurence Oberman <loberman@redhat.com> > --- > drivers/scsi/bnx2fc/bnx2fc_io.c | 23 ++++++++++++++++++++--- > 1 file changed, 20 insertions(+), 3 deletions(-) > > diff --git a/drivers/scsi/bnx2fc/bnx2fc_io.c > b/drivers/scsi/bnx2fc/bnx2fc_io.c > index 9e50e5b..39f4aeb 100644 > --- a/drivers/scsi/bnx2fc/bnx2fc_io.c > +++ b/drivers/scsi/bnx2fc/bnx2fc_io.c > @@ -1928,6 +1928,7 @@ void bnx2fc_process_scsi_cmd_compl(struct > bnx2fc_cmd *io_req, > struct bnx2fc_rport *tgt = io_req->tgt; > struct scsi_cmnd *sc_cmd; > struct Scsi_Host *host; > + u16 scope, qualifier = 0; > > > /* scsi_cmd_cmpl is called with tgt lock held */ > @@ -1997,12 +1998,28 @@ void bnx2fc_process_scsi_cmd_compl(struct > bnx2fc_cmd *io_req, > > if (io_req->cdb_status == > SAM_STAT_TASK_SET_FULL || > io_req->cdb_status == SAM_STAT_BUSY) { > + /* Newer array firmware with BUSY or > + * TASK_SET_FULL may return a status > that needs > + * the scope bits masked. > + * Or a huge delay timestamp up to 27 > minutes > + * can result. > + */ > + if (fcp_rsp->retry_delay_timer) { > + /* Upper 2 bits */ > + scope = fcp_rsp- > >retry_delay_timer > + & 0xC000; > + /* Lower 14 bits */ > + qualifier = fcp_rsp- > >retry_delay_timer > + & 0x3FFF; > + } > + if (scope > 0 && qualifier > 0 && > + qualifier <= 0x3FEF) { > /* Set the jiffies + > retry_delay_timer * 100ms > for the rport/tgt > */ > - tgt->retry_delay_timestamp = jiffies + > - fcp_rsp->retry_delay_timer * HZ > / 10; > + tgt->retry_delay_timestamp = > jiffies + > + (qualifier * HZ / 10); > + } > } > - > } > if (io_req->fcp_resid) > scsi_set_resid(sc_cmd, io_req->fcp_resid); V2 looks good. Reviewed-by: Chad Dupuis <cdupuis1@gmail.com>
Laurence, > The qla2xxx driver had this issue as well when the newer array > firmware returned the retry_delay_timer in the fcp_rsp. The bnx2fc is > not handling the masking of the scope bits either so the > retry_delay_timestamp value lands up being a large value added to the > timer timestamp delaying I/O for up to 27 Minutes. This patch adds > similar code to handle this to the bnx2fc driver to avoid the huge > delay. Does not apply to 5.4/scsi-queue. When you resubmit, please use git send-email -v3 [...] and put the changelog commentary after a "---" separator. > V2. Indent comments as suggested > > Signed-off-by: Laurence Oberman <loberman@redhat.com>
diff --git a/drivers/scsi/bnx2fc/bnx2fc_io.c b/drivers/scsi/bnx2fc/bnx2fc_io.c index 9e50e5b..39f4aeb 100644 --- a/drivers/scsi/bnx2fc/bnx2fc_io.c +++ b/drivers/scsi/bnx2fc/bnx2fc_io.c @@ -1928,6 +1928,7 @@ void bnx2fc_process_scsi_cmd_compl(struct bnx2fc_cmd *io_req, struct bnx2fc_rport *tgt = io_req->tgt; struct scsi_cmnd *sc_cmd; struct Scsi_Host *host; + u16 scope, qualifier = 0; /* scsi_cmd_cmpl is called with tgt lock held */ @@ -1997,12 +1998,28 @@ void bnx2fc_process_scsi_cmd_compl(struct bnx2fc_cmd *io_req, if (io_req->cdb_status == SAM_STAT_TASK_SET_FULL || io_req->cdb_status == SAM_STAT_BUSY) { + /* Newer array firmware with BUSY or + * TASK_SET_FULL may return a status that needs + * the scope bits masked. + * Or a huge delay timestamp up to 27 minutes + * can result. + */ + if (fcp_rsp->retry_delay_timer) { + /* Upper 2 bits */ + scope = fcp_rsp->retry_delay_timer + & 0xC000; + /* Lower 14 bits */ + qualifier = fcp_rsp->retry_delay_timer + & 0x3FFF; + } + if (scope > 0 && qualifier > 0 && + qualifier <= 0x3FEF) { /* Set the jiffies + retry_delay_timer * 100ms for the rport/tgt */ - tgt->retry_delay_timestamp = jiffies + - fcp_rsp->retry_delay_timer * HZ / 10; + tgt->retry_delay_timestamp = jiffies + + (qualifier * HZ / 10); + } } - }
The qla2xxx driver had this issue as well when the newer array firmware returned the retry_delay_timer in the fcp_rsp. The bnx2fc is not handling the masking of the scope bits either so the retry_delay_timestamp value lands up being a large value added to the timer timestamp delaying I/O for up to 27 Minutes. This patch adds similar code to handle this to the bnx2fc driver to avoid the huge delay. V2. Indent comments as suggested Signed-off-by: Laurence Oberman <loberman@redhat.com> --- drivers/scsi/bnx2fc/bnx2fc_io.c | 23 ++++++++++++++++++++--- 1 file changed, 20 insertions(+), 3 deletions(-) if (io_req->fcp_resid) scsi_set_resid(sc_cmd, io_req->fcp_resid);