On Wed, Feb 19, 2020 at 08:34:38PM -0800, Bart Van Assche wrote:
> Instead of changing endianness in-place, write the data in CPU endian format
> in another buffer and copy that buffer back. This patch does not change any
> functionality but silences some sparse endianness warnings.
>
> Reviewed-by: Daniel Wagner <dwagner@xxxxxxx>
> Cc: Quinn Tran <qutran@xxxxxxxxxxx>
> Cc: Martin Wilck <mwilck@xxxxxxxx>
> Cc: Roman Bolshakov <r.bolshakov@xxxxxxxxx>
> Signed-off-by: Bart Van Assche <bvanassche@xxxxxxx>
> ---
> drivers/scsi/qla2xxx/qla_def.h | 2 +-
> drivers/scsi/qla2xxx/qla_init.c | 11 +++++------
> 2 files changed, 6 insertions(+), 7 deletions(-)
>
> diff --git a/drivers/scsi/qla2xxx/qla_init.c b/drivers/scsi/qla2xxx/qla_init.c
> index 9e6b56527b25..880f2be062a9 100644
> --- a/drivers/scsi/qla2xxx/qla_init.c
> +++ b/drivers/scsi/qla2xxx/qla_init.c
> @@ -5068,13 +5068,12 @@ qla2x00_configure_local_loop(scsi_qla_host_t *vha)
> rval = qla24xx_get_port_login_templ(vha,
> ha->init_cb_dma, (void *)ha->init_cb, sz);
> if (rval == QLA_SUCCESS) {
> + __be32 *q = &ha->plogi_els_payld.data[0];
> +
> bp = (uint32_t *)ha->init_cb;
> - for (i = 0; i < sz/4 ; i++, bp++)
> - *bp = cpu_to_be32(*bp);
> + cpu_to_be32_array(q, bp, sz / 4);
>
> - memcpy(&ha->plogi_els_payld.data,
> - (void *)ha->init_cb,
> - sizeof(ha->plogi_els_payld.data));
> + memcpy(bp, q, sizeof(ha->plogi_els_payld.data));
Hi Bart,
Honestly, I'd prefer to drop the memcpy as it has no purpose, but since
you replicated a side-effect of the previous code and fixed sparse
warnings,
Reviewed-by: Roman Bolshakov <r.bolshakov@xxxxxxxxx>
Thank you,
Roman
P.S.
During the review, I looked again where init_cb is used. It seems the
reuse and rewriting of init_cb contents by qla24xx_get_port_login_templ
may corrupt init_cb values expected by qla2x00_fdmi_rpa and
qla2x00_fdmiv2_rpa (they retrieve max frame size from Initialization
Control Block), and qla2x00_async_event (recovers WWPN from init_cb in
case of Fabric-assigned WWPN).
That _rpa functions are invoked if a port is in fabric topology and FDMI
registration is enabled, so a change of topology from P2P to Switched
Fabric may result in FDMI registration with wrong max frame size. I
don't know how FC switches use the value though.
If the fabric assigns WWPNs (FA-WWPN) and a disconnect happens (Loop
Down Event), the driver will try to recover WWPN from init_cb that is
corrupted by qla24xx_get_port_login_templ.
Probably, the wrong state may be recovered by reclaiming the underlying
PCI device. init_cb is set to correct values in probe() handler.
