Hi,
On Sat, Aug 25, 2018 at 8:45 PM, Terin Stock <terin@xxxxxxxxxxxxxx> wrote:
> Upon upgrading a Raspberry Pi 3B-based project from vanilla 4.14,
> attempts to mount a floppy disk in a generic USB floppy drive would hang
> until the floppy drive was removed from the system.
>
> Tracing shows that during mounting the drive produces a large amount of
> NAKed transactions, but would eventually continue. A previous commit
> added a retry delay on NAKed transactions, using jiffies, that results
> in indefinite NAKs in this scenario.
>
> Modify the wait delay utilize the high resolution timer API to allow for
> more accurately scheduled callbacks.
I think your commit will be more compelling with additional data. As
Allen says it looks like you're not actually changing the delay. You
could include things like:
* On systems with 100 HZ in the ideal case we'd end up delaying for 10
ms - 20 ms when we used jiffies. Now we'll get much closer to 1 ms.
* Timer functions are not very high priority, so if we were running at
a high load then we'd sometimes see much longer delays. (NOTE: if you
say this then please back it up with data--I think I've heard
anecdotally that the normal timer functions can be quite delayed but I
haven't done the research to back it up). Presumably you could use
ktime to measure delays before and after your patch and you could
include this in the commit message.
It would also be good to document what device you were plugging in
that you were having problems with and what system you were running
on. That would help someone else if they ever wanted to modify the
same area of code and re-test. They'd have a better chance of not
re-breaking you.
NOTE: it's possible that the problem here is just that the USB device
you're plugging in is not very forgiving to the kernel taking a long
time to talk to it again after the NAK. Having such a long delay here
isn't common and presumably the device you have just doesn't handle
it. Possibly the device is non-compliant (I'm not enough of an expert
on the spec), but it's still nice to try to support it.
> Fixes: 38d2b5fb75c1 ("usb: dwc2: host: Don't retry NAKed transactions right away")
> Signed-off-by: Terin Stock <terin@xxxxxxxxxxxxxx>
> ---
> drivers/usb/dwc2/hcd.h | 2 +-
> drivers/usb/dwc2/hcd_queue.c | 17 ++++++++++-------
> 2 files changed, 11 insertions(+), 8 deletions(-)
Overall nit: please CC LKML on patches. If nothing else that makes
them easier to find in lore.kernel.org/patchwork
> diff --git a/drivers/usb/dwc2/hcd.h b/drivers/usb/dwc2/hcd.h
> index 5502a501f516..93483dc37801 100644
> --- a/drivers/usb/dwc2/hcd.h
> +++ b/drivers/usb/dwc2/hcd.h
> @@ -366,7 +366,7 @@ struct dwc2_qh {
> u32 desc_list_sz;
> u32 *n_bytes;
> struct timer_list unreserve_timer;
> - struct timer_list wait_timer;
> + struct hrtimer wait_timer;
> struct dwc2_tt *dwc_tt;
> int ttport;
> unsigned tt_buffer_dirty:1;
> diff --git a/drivers/usb/dwc2/hcd_queue.c b/drivers/usb/dwc2/hcd_queue.c
> index 301ced1618f8..2d0cfd7f2cfe 100644
> --- a/drivers/usb/dwc2/hcd_queue.c
> +++ b/drivers/usb/dwc2/hcd_queue.c
> @@ -59,7 +59,7 @@
> #define DWC2_UNRESERVE_DELAY (msecs_to_jiffies(5))
>
> /* If we get a NAK, wait this long before retrying */
> -#define DWC2_RETRY_WAIT_DELAY (msecs_to_jiffies(1))
> +#define DWC2_RETRY_WAIT_DELAY 1*1E6L
>
> /**
> * dwc2_periodic_channel_available() - Checks that a channel is available for a
> @@ -1465,9 +1465,9 @@ static void dwc2_deschedule_periodic(struct dwc2_hsotg *hsotg,
> *
> * @t: Pointer to wait_timer in a qh.
> */
> -static void dwc2_wait_timer_fn(struct timer_list *t)
> +static enum hrtimer_restart dwc2_wait_timer_fn(struct hrtimer *t)
nit: please update function docstring to include a "Return:" clause now.
Other than the above things look pretty good here to me. Thanks for the patch!
-Doug
