On Mon, 2016-04-04 at 15:44 -0700, Bart Van Assche wrote:
> On 04/04/2016 03:41 PM, Nicholas A. Bellinger wrote:
> > On Mon, 2016-04-04 at 11:40 -0700, Bart Van Assche wrote:
> >> On 04/04/2016 11:28 AM, Nicholas A. Bellinger wrote:
> >>> diff --git a/drivers/infiniband/ulp/srpt/ib_srpt.c b/drivers/infiniband/ulp/srpt/ib_srpt.c
> >>> index 0bd3cb2..c4e0a0a 100644
> >>> --- a/drivers/infiniband/ulp/srpt/ib_srpt.c
> >>> +++ b/drivers/infiniband/ulp/srpt/ib_srpt.c
> >>> @@ -1266,7 +1266,9 @@ static struct srpt_send_ioctx *srpt_get_send_ioctx(struct srpt_rdma_ch *ch)
> >>> {
> >>> struct se_session *se_sess;
> >>> struct srpt_send_ioctx *ioctx;
> >>> - int tag;
> >>> + void *buf;
> >>> + dma_addr_t dma;
> >>> + int tag, index;
> >>>
> >>> BUG_ON(!ch);
> >>> se_sess = ch->sess;
> >>> @@ -1277,12 +1279,19 @@ static struct srpt_send_ioctx *srpt_get_send_ioctx(struct srpt_rdma_ch *ch)
> >>> return NULL;
> >>> }
> >>> ioctx = &((struct srpt_send_ioctx *)se_sess->sess_cmd_map)[tag];
> >>> + buf = ioctx->ioctx.buf;
> >>> + dma = ioctx->ioctx.dma;
> >>> + index = ioctx->ioctx.index;
> >>> +
> >>> memset(ioctx, 0, sizeof(struct srpt_send_ioctx));
> >>> ioctx->ch = ch;
> >>> spin_lock_init(&ioctx->spinlock);
> >>> ioctx->state = SRPT_STATE_NEW;
> >>> init_completion(&ioctx->tx_done);
> >>>
> >>> + ioctx->ioctx.buf = buf;
> >>> + ioctx->ioctx.dma = dma;
> >>> + ioctx->ioctx.index = index;
> >>> ioctx->cmd.map_tag = tag;
> >>>
> >>> return ioctx;
> >>> @@ -1961,6 +1970,24 @@ static void srpt_free_ch(struct kref *kref)
> >>> kfree(ch);
> >>> }
> >>
> >> These assignments should happen once, namely just after I/O context
> >> allocation, instead of performing these assignments during every
> >> srpt_get_send_ioctx() call.
> >
> > No, the entire structure is being cleared each time, just like what
> > every other driver is doing here.
> >
> > Don't try to micro-optimize to avoid doing memset.
>
> But why to keep that memset(ioctx, ...) call? That memset() call is new.
> It was added during the 4.6-rc1 merge window.
>
I like memset over explicit zero assignments after reuse so future
additions to srp_send_ioctx don't need extra code.
Either way, I don't really care about micro-optimizations right now.
Are you going to verify this patch..?
--
To unsubscribe from this list: send the line "unsubscribe linux-rdma" in
the body of a message to majordomo@xxxxxxxxxxxxxxx
More majordomo info at http://vger.kernel.org/majordomo-info.html
