On Sun, 2010-05-30 at 12:02 +0300, Boaz Harrosh wrote:
> On 05/18/2010 08:13 AM, Nicholas A. Bellinger wrote:
> > From: Nicholas Bellinger <nab@xxxxxxxxxxxxxxx>
> >
> > This patch changes usr/scsi.c:scsi_cmd_perform() to first check to see if
> > the struct backingstore_template->bs_passthrough=1 for SG_IO, then
> > check for a valid struct device_type_template->cmd_passthrough()
> > function pointer to determine if the device supports SG_IO.
> >
> > This patch then updates the device_type in usr/sb.c to enable the functionality
> > for TYPE_DISK.
> >
> > Signed-off-by: Nicholas A. Bellinger <nab@xxxxxxxxxxxxxxx>
> > ---
> > usr/sbc.c | 1 +
> > usr/scsi.c | 11 +++++++++++
> > usr/tgtd.h | 1 +
> > 3 files changed, 13 insertions(+), 0 deletions(-)
> >
> > diff --git a/usr/sbc.c b/usr/sbc.c
> > index a048d53..d53f7f9 100644
> > --- a/usr/sbc.c
> > +++ b/usr/sbc.c
> > @@ -269,6 +269,7 @@ static struct device_type_template sbc_template = {
> > .lu_online = spc_lu_online,
> > .lu_offline = spc_lu_offline,
> > .lu_exit = spc_lu_exit,
> > + .cmd_passthrough = sbc_rw,
> > .ops = {
> > {spc_test_unit,},
> > {spc_illegal_op,},
> > diff --git a/usr/scsi.c b/usr/scsi.c
> > index cef231b..1db786e 100644
> > --- a/usr/scsi.c
> > +++ b/usr/scsi.c
> > @@ -242,6 +242,17 @@ int scsi_cmd_perform(int host_no, struct scsi_cmd *cmd)
> >
> > if (spc_access_check(cmd))
> > return SAM_STAT_RESERVATION_CONFLICT;
> > + /*
> > + * Use dev_type_template->cmd_passthrough() for usr/bs_sg.c devices
> > + */
> > + if (cmd->dev->bst->bs_passthrough) {
> > + if (!(cmd->dev->dev_type_template.cmd_passthrough)) {
> > + eprintf("cmd->dev->dev_type_template.cmd_passthrough()"
> > + " is NULL!\n");
> > + return SAM_STAT_CHECK_CONDITION;
> > + }
> > + return cmd->dev->dev_type_template.cmd_passthrough(host_no, cmd);
> > + }
> >
>
> Can't you just perform this check once somewhere and set cmd_perform to the
> new cmd_passthrough? why check for every command?
>
> This is not accepted it is the hot path.
Good point..
>
> > return cmd->dev->dev_type_template.ops[op].cmd_perform(host_no, cmd);
> > }
> > diff --git a/usr/tgtd.h b/usr/tgtd.h
> > index 4f26a29..f8ee47f 100644
> > --- a/usr/tgtd.h
> > +++ b/usr/tgtd.h
> > @@ -108,6 +108,7 @@ struct device_type_template {
> > int (*lu_config)(struct scsi_lu *lu, char *args);
> > int (*lu_online)(struct scsi_lu *lu);
> > int (*lu_offline)(struct scsi_lu *lu);
> > + int (*cmd_passthrough)(int, struct scsi_cmd *);
> >
>
> Actually you can do more:
> If you leave the checking and decisions to the target driver
> then you don't need to touch this file. Just let the device_type
> decide which vector to set once it knows what target is used.
>
> (Like start with a vector that check and sets the vector to
> the passthrough or the regular path, on first IO)
>
Ok, this makes alot more sense for a STGT backstore passthrough. I will
respin the patch to let device_type decide to setup which flavour of
->cmd_perform() it would like to use and drop the scsi_cmd_perform()
passthrough check, et al. for the fast path.
Thanks for your comments Boaz!
--nab
--
To unsubscribe from this list: send the line "unsubscribe stgt" in
the body of a message to majordomo@xxxxxxxxxxxxxxx
More majordomo info at http://vger.kernel.org/majordomo-info.html
