On Sun, Jan 12, 2025 at 03:19:44PM -0600, Mike Christie wrote:
> On 1/12/25 11:35 AM, michael.chris...@oracle.com wrote:
> > So I think to fix the issue, we would want to:
> >
> > 1. move the
> >
> > memcpy(vs_tpg, vs->vs_tpg, len);
> >
> > to the end of the function after we do the vhost_scsi_flush. This will
> > be more complicated than the current memcpy though. We will want to
> > merge the local vs_tpg and the vs->vs_tpg like:
> >
> > for (i = 0; i < VHOST_SCSI_MAX_TARGET; i++) {
> > if (vs_tpg[i])
> > vs->vs_tpg[i] = vs_tpg[i])
> > }
>
> I think I wrote that in reverse. We would want:
>
> vhost_scsi_flush(vs);
>
> if (vs->vs_tpg) {
> for (i = 0; i < VHOST_SCSI_MAX_TARGET; i++) {
> if (vs->vs_tpg[i])
> vs_tpg[i] = vs->vs_tpg[i])
> }
> }
>
> kfree(vs->vs_tpg);
> vs->vs_tpg = vs_tpg;
>
> or we could just allocate the vs_tpg with the vhost_scsi like:
>
> struct vhost_scsi {
> ....
>
> struct vhost_scsi_tpg *vs_tpg[VHOST_SCSI_MAX_TARGET];
>
> then when we loop in vhost_scsi_set/clear_endpoint set/clear the
> every vs_tpg entry.
Wanna post the patch, Mike?
--
MST
