On Mon, Nov 16, 2015 at 05:56:25PM +0100, Krzysztof Opasiak wrote:
>
>
> On 11/13/2015 09:00 AM, Peter Chen wrote:
> >Add queue depth for both iso and bulk transfer, with more queues, we
> >can do performance and stress test using sourcesink, and update g_zero
> >accordingly.
> >
> >Signed-off-by: Peter Chen <peter.chen@xxxxxxxxxxxxx>
> >---
> > drivers/usb/gadget/function/f_sourcesink.c | 144 ++++++++++++++++++++++-------
> > drivers/usb/gadget/function/g_zero.h | 6 ++
> > drivers/usb/gadget/legacy/zero.c | 12 +++
> > 3 files changed, 129 insertions(+), 33 deletions(-)
> >
> >diff --git a/drivers/usb/gadget/function/f_sourcesink.c b/drivers/usb/gadget/function/f_sourcesink.c
> >index d7646d3..4ab603e8 100644
> >--- a/drivers/usb/gadget/function/f_sourcesink.c
> >+++ b/drivers/usb/gadget/function/f_sourcesink.c
> >@@ -34,13 +34,6 @@
> > * plus two that support control-OUT tests. If the optional "autoresume"
> > * mode is enabled, it provides good functional coverage for the "USBCV"
> > * test harness from USB-IF.
> >- *
> >- * Note that because this doesn't queue more than one request at a time,
> >- * some other function must be used to test queueing logic. The network
> >- * link (g_ether) is the best overall option for that, since its TX and RX
> >- * queues are relatively independent, will receive a range of packet sizes,
> >- * and can often be made to run out completely. Those issues are important
> >- * when stress testing peripheral controller drivers.
> > */
> > struct f_sourcesink {
> > struct usb_function function;
> >@@ -57,6 +50,8 @@ struct f_sourcesink {
> > unsigned isoc_mult;
> > unsigned isoc_maxburst;
> > unsigned buflen;
> >+ unsigned bulk_qlen;
> >+ unsigned iso_qlen;
> > };
> >
> > static inline struct f_sourcesink *func_to_ss(struct usb_function *f)
> >@@ -595,31 +590,33 @@ static int source_sink_start_ep(struct f_sourcesink *ss, bool is_in,
> > {
> > struct usb_ep *ep;
> > struct usb_request *req;
> >- int i, size, status;
> >-
> >- for (i = 0; i < 8; i++) {
> >- if (is_iso) {
> >- switch (speed) {
> >- case USB_SPEED_SUPER:
> >- size = ss->isoc_maxpacket *
> >- (ss->isoc_mult + 1) *
> >- (ss->isoc_maxburst + 1);
> >- break;
> >- case USB_SPEED_HIGH:
> >- size = ss->isoc_maxpacket * (ss->isoc_mult + 1);
> >- break;
> >- default:
> >- size = ss->isoc_maxpacket > 1023 ?
> >- 1023 : ss->isoc_maxpacket;
> >- break;
> >- }
> >- ep = is_in ? ss->iso_in_ep : ss->iso_out_ep;
> >- req = ss_alloc_ep_req(ep, size);
> >- } else {
> >- ep = is_in ? ss->in_ep : ss->out_ep;
> >- req = ss_alloc_ep_req(ep, 0);
> >+ int i, size, qlen, status = 0;
> >+
> >+ if (is_iso) {
> >+ switch (speed) {
> >+ case USB_SPEED_SUPER:
> >+ size = ss->isoc_maxpacket *
> >+ (ss->isoc_mult + 1) *
> >+ (ss->isoc_maxburst + 1);
> >+ break;
> >+ case USB_SPEED_HIGH:
> >+ size = ss->isoc_maxpacket * (ss->isoc_mult + 1);
> >+ break;
> >+ default:
> >+ size = ss->isoc_maxpacket > 1023 ?
> >+ 1023 : ss->isoc_maxpacket;
> >+ break;
> > }
> >+ ep = is_in ? ss->iso_in_ep : ss->iso_out_ep;
> >+ qlen = ss->iso_qlen;
> >+ } else {
> >+ ep = is_in ? ss->in_ep : ss->out_ep;
> >+ qlen = ss->bulk_qlen;
> >+ size = 0;
> >+ }
> >
> >+ for (i = 0; i < qlen; i++) {
> >+ req = ss_alloc_ep_req(ep, size);
> > if (!req)
> > return -ENOMEM;
> >
> >@@ -639,9 +636,6 @@ static int source_sink_start_ep(struct f_sourcesink *ss, bool is_in,
> > ep->name, status);
> > free_ep_req(ep, req);
>
> Well... to be honest I would prefer to return here an error instead
> of silently try with another request as we may end up in situation
> that qlen is 8 and we haven't successfully enqueued any request but
> we treat this as a success.
>
Ok, but it seems not related with this patch set, I will change after
this one has queued.
--
Best Regards,
Peter Chen
--
To unsubscribe from this list: send the line "unsubscribe linux-usb" in
the body of a message to majordomo@xxxxxxxxxxxxxxx
More majordomo info at http://vger.kernel.org/majordomo-info.html
