> Date: Wed, 8 Dec 2010 06:07:30 +0000
> From: Jacob Meuser <jake...@sdf.lonestar.org>
>
> I recently got a hp officejet 4500. it's a 3-in-1 printer/scanner/fax.
> printing works great with the hplip packages. scanning doesn't work at
> all.
>
> I tracked the problem to read() failing in usb_bulk_read() in libusb.
> errno was EINTR. so I made this function just continue when read() got
> EINTR. I could then scan, but not very reliably.
>
> so I went into the kernel, and tracked it to usbdi_util.c:
>
> @ usbd_bulk_transfer(usbd_xfer_handle xfer, usbd_pipe_ha
>
> error = tsleep((caddr_t)xfer, PZERO | PCATCH, lbl, 0);
> splx(s);
> if (error) {
> DPRINTF(("usbd_bulk_transfer: tsleep=%d\n", error));
> usbd_abort_pipe(pipe);
> return (USBD_INTERRUPTED);
> }
>
> on a hunch, I changed 'PZERO | PCATCH' to 'PWAIT'. then scanning became
> 100% reliable. even full-page 1200dpi (~525MB of data) worked
> perfectly.
>
> great. but, what if we need to interrupt the transfer. we don't want
> to hang here.
>
> well, this function takes a timeout. so, it's possible to make it
> return, even if the transfer stalls. but is this used? I looked
> at the ports that use libusb. setting 0/infinite timeout for bulk
> transfers is extremely rare, only set in code marked experimental
> or problematic. in the kernel, there are however some callers that
> use USBD_NO_TIMEOUT.
>
> so, I offer the following. if there's no timeout, behave like we do
> now and catch signals, and if there's a timeout, ignore signals.
>
> I did the same for interrupt transfers because the situation seems
> to be the same.
>
> thoughts?
Sorry, but this very much feels like you're hacking around a bug in
the application you're using. If it installs signal handler without
specifying the SA_RESTART flag, it has to deal with read(2) failing
with EINTR.
> Index: usbdi_util.c
> ===================================================================
> RCS file: /cvs/src/sys/dev/usb/usbdi_util.c,v
> retrieving revision 1.25
> diff -u -p usbdi_util.c
> --- usbdi_util.c 26 Jun 2008 05:42:19 -0000 1.25
> +++ usbdi_util.c 8 Dec 2010 04:58:49 -0000
> @@ -426,7 +426,7 @@ usbd_bulk_transfer(usbd_xfer_handle xfer, usbd_pipe_ha
> u_int16_t flags, u_int32_t timeout, void *buf, u_int32_t *size, char
> *lbl)
> {
> usbd_status err;
> - int s, error;
> + int s, error, pri;
>
> usbd_setup_xfer(xfer, pipe, 0, buf, *size, flags, timeout,
> usbd_bulk_transfer_cb);
> @@ -437,7 +437,8 @@ usbd_bulk_transfer(usbd_xfer_handle xfer, usbd_pipe_ha
> splx(s);
> return (err);
> }
> - error = tsleep((caddr_t)xfer, PZERO | PCATCH, lbl, 0);
> + pri = timeout == USBD_NO_TIMEOUT ? (PZERO | PCATCH) : PWAIT;
> + error = tsleep((caddr_t)xfer, pri, lbl, 0);
> splx(s);
> if (error) {
> DPRINTF(("usbd_bulk_transfer: tsleep=%d\n", error));
> @@ -467,7 +468,7 @@ usbd_intr_transfer(usbd_xfer_handle xfer, usbd_pipe_ha
> u_int16_t flags, u_int32_t timeout, void *buf, u_int32_t *size, char
> *lbl)
> {
> usbd_status err;
> - int s, error;
> + int s, error, pri;
>
> usbd_setup_xfer(xfer, pipe, 0, buf, *size, flags, timeout,
> usbd_intr_transfer_cb);
> @@ -478,7 +479,8 @@ usbd_intr_transfer(usbd_xfer_handle xfer, usbd_pipe_ha
> splx(s);
> return (err);
> }
> - error = tsleep(xfer, PZERO | PCATCH, lbl, 0);
> + pri = timeout == USBD_NO_TIMEOUT ? (PZERO | PCATCH) : PWAIT;
> + error = tsleep(xfer, pri, lbl, 0);
> splx(s);
> if (error) {
> DPRINTF(("usbd_intr_transfer: tsleep=%d\n", error));
