[U-Boot] usb_stor_BBB_transport 5 ms delay - performance
Marek Vasut
marex at denx.de
Fri Jul 27 14:59:29 CEST 2012
Dear Benoît Thébaudeau,
> Hi Jim,
>
> On Thu, Jul 26, 2012 at 10:20:48 PM, Jim Shimer wrote:
> > I'm seeing a 5ms delay in usb_stor_BBB_transport, which occurs every
> > 10K of
> > data for fatload usb or 500ms of delay per 1MB of image size. This
> > adds up
> > to quite a bit of delay if you're loading a large ramdisk.
> >
> > Does anyone know what the reason for the 5ms delay really is? I'm
> > assuming
> > that this delay is to debounce the 5V/100ma USB power up. I made
> > some
> > modification, where the delay is skipped if the device has already
> > been
> > queried as ready. This has save me 500ms/M on fatload times (eg,
> > 140M=70seconds). Is there anything wrong with this tweak?
> >
> > Here's a diff of what I've done to get the performance I need:
> >
> > --- usb_storage.c.orig 2012-07-26 16:06:40.775251000 -0400
> > +++ usb_storage.c 2012-07-26 13:49:36.000000000 -0400
> > @@ -132,6 +132,7 @@ static block_dev_desc_t usb_dev_desc[USB
> >
> > struct us_data;
> > typedef int (*trans_cmnd)(ccb *cb, struct us_data *data);
> > typedef int (*trans_reset)(struct us_data *data);
> >
> > +typedef enum us_status { USB_NOT_READY, USB_READY} us_status;
> >
> > struct us_data {
> >
> > struct usb_device *pusb_dev; /* this usb_device */
> >
> > @@ -154,6 +155,7 @@ struct us_data {
> >
> > ccb *srb; /* current srb */
> > trans_reset transport_reset; /* reset routine */
> > trans_cmnd transport; /* transport routine
> > */
> >
> > + us_status status;
> >
> > };
> >
> > static struct us_data usb_stor[USB_MAX_STOR_DEV];
> >
> > @@ -691,7 +693,10 @@ int usb_stor_BBB_transport(ccb *srb, str
> >
> > usb_stor_BBB_reset(us);
> > return USB_STOR_TRANSPORT_FAILED;
> >
> > }
> >
> > - wait_ms(5);
> > + if(us->status != USB_READY)
> > + {
> > + wait_ms(5);
> > + }
> >
> > pipein = usb_rcvbulkpipe(us->pusb_dev, us->ep_in);
> > pipeout = usb_sndbulkpipe(us->pusb_dev, us->ep_out);
> > /* DATA phase + error handling */
> >
> > @@ -957,7 +962,10 @@ static int usb_test_unit_ready(ccb *srb,
> >
> > srb->datalen = 0;
> > srb->cmdlen = 12;
> > if (ss->transport(srb, ss) ==
> > USB_STOR_TRANSPORT_GOOD)
> >
> > + {
> > + ss->status = USB_READY;
> >
> > return 0;
> >
> > + }
> >
> > usb_request_sense(srb, ss);
> > wait_ms(100);
> >
> > } while (retries--);
> >
> > @@ -965,6 +973,11 @@ static int usb_test_unit_ready(ccb *srb,
> >
> > return -1;
> >
> > }
> >
> > +static void usb_set_unit_not_ready(struct us_data *ss)
> > +{
> > + ss->status = USB_NOT_READY;
> > +}
> > +
> >
> > static int usb_read_capacity(ccb *srb, struct us_data *ss)
> > {
> >
> > int retry;
> >
> > @@ -1108,6 +1121,7 @@ retry_it:
> > blks -= smallblks;
> > buf_addr += srb->datalen;
> >
> > } while (blks != 0);
> >
> > + usb_set_unit_not_ready((struct us_data *)dev->privptr);
> >
> > USB_STOR_PRINTF("usb_read: end startblk %lx, blccnt %x buffer
> >
> > %lx\n",
> >
> > start, smallblks, buf_addr);
> >
> > @@ -1188,6 +1202,7 @@ retry_it:
> > blks -= smallblks;
> > buf_addr += srb->datalen;
> >
> > } while (blks != 0);
> >
> > + usb_set_unit_not_ready((struct us_data *)dev->privptr);
> >
> > USB_STOR_PRINTF("usb_write: end startblk %lx, blccnt %x
> > buffer
> >
> > %lx\n",
> >
> > start, smallblks, buf_addr);
> >
> > @@ -1398,6 +1413,7 @@ int usb_stor_get_info(struct usb_device
> >
> > cap[0] = 2880;
> > cap[1] = 0x200;
> >
> > }
> >
> > + usb_set_unit_not_ready((struct us_data *)dev->privptr);
> >
> > USB_STOR_PRINTF("Read Capacity returns: 0x%lx, 0x%lx\n",
> > cap[0],
> >
> > cap[1]);
> >
> > #if 0
> >
> > I'd appreciate any feedback.
> > Regards
>
> I have not looked into this delay issue, but I had similar performance
> issues that I fixed with the following series:
> http://patchwork.ozlabs.org/patch/172052/
> http://patchwork.ozlabs.org/patch/172204/
> http://patchwork.ozlabs.org/patch/172054/
> http://patchwork.ozlabs.org/patch/172055/
> http://patchwork.ozlabs.org/patch/172056/
>
> Your suggestion is interesting and might be a complement to my series. I
> don't have time to check its correctness right now, but I'll try soon.
Will you two have time to work these into V2 of your series somehow please?
> Best regards,
> Benoît
Best regards,
Marek Vasut
More information about the U-Boot
mailing list