Re: [Qemu-devel] [PATCH 10/16] scsi: move request parsing to common code

[Peter Maydell]

On 3 August 2011 09:49, Paolo Bonzini <address@hidden> wrote:
> @@ -157,8 +172,22 @@ SCSIRequest *scsi_req_new(SCSIDevice *d, uint32_t tag, 
> uint32_t lun,
>                           uint8_t *buf, void *hba_private)
>  {
>     SCSIRequest *req;
> -    req = d->info->alloc_req(d, tag, lun, hba_private);
> -    memcpy(req->cmd.buf, buf, 16);
> +    SCSICommand cmd;
> +
> +    if (scsi_req_parse(&cmd, d, buf) != 0) {
> +        trace_scsi_req_parse_bad(d->id, lun, tag, buf[0]);
> +        req = scsi_req_alloc(&reqops_invalid_opcode, d, tag, lun, 
> hba_private);
> +    } else {
> +        trace_scsi_req_parsed(d->id, lun, tag, buf[0],
> +                              cmd.mode, cmd.xfer);
> +        if (req->cmd.lba != -1) {
> +            trace_scsi_req_parsed_lba(d->id, lun, tag, buf[0],
> +                                      cmd.lba);
> +        }
> +        req = d->info->alloc_req(d, tag, lun, hba_private);
> +    }
> +
> +    req->cmd = cmd;
>     return req;
>  }

This patch makes current master fail to compile with optimisation on:

gcc complains:
hw/scsi-bus.c: In function ‘scsi_req_new’:
hw/scsi-bus.c:375: error: ‘req’ may be used uninitialized in this function

because in the 'else' clause we look at req->cmd.lba before we've
called alloc_req().

My guess is that the tracing should just be moved down to after the
allocation?

-- PMM