Re: [PATCH 1/3] picolcd: driver for PicoLCD HID device

[Jiri Kosina]

On Wed, 24 Feb 2010, Rick L. Vinyard, Jr. wrote:

> >> +static int picolcd_raw_event(struct hid_device *hdev,
> >> +               struct hid_report *report, u8 *raw_data, int size)
> >> +{
> >> +       struct picolcd_data *data = hid_get_drvdata(hdev);
> >> +       char hexdata[25];
> >> +       int i;
> >> +
> >> +       if (data == NULL)
> >> +               return 1;
> >> +
> >> +       for (i = 0; i < sizeof(hexdata) / 2; i++)
> >> +               sprintf(hexdata+2*i, "%02hhx", raw_data[i]);
> >> +       if (size >= sizeof(hexdata)/2) {
> >> +               hexdata[sizeof(hexdata)-4] = '.';
> >> +               hexdata[sizeof(hexdata)-3] = '.';
> >> +               hexdata[sizeof(hexdata)-2] = '.';
> >> +               hexdata[sizeof(hexdata)-1] = '\0';
> >> +       } else
> >> +               hexdata[size*2] = '\0';
> >> +
> >> +       switch (report->id) {
> >> +       case REPORT_KEYPAD:
> >> +               if (size == 3 && raw_data[0] == 0x11 &&
> >> data->input_keys) {
> >> +                       return picolcd_raw_keypad(hdev, report,
> >> raw_data+1, size-1);
> >> +               } else {
> >> +                       dbg_hid(PICOLCD_NAME " unsupported key event (%d
> >> bytes): 0x%s\n", size, hexdata);
> >> +                       break;
> >> +               }
> >> +               break;
> >> +       case REPORT_VERSION:
> >> +               if (size == 3)
> >> +                       dev_info(&hdev->dev, "Firmware version is
> >> %hd.%hd\n", raw_data[1], raw_data[2]);
> >> +
> >> +               spin_lock(&data->lock);
> >
> > If I recall correctly raw_event is called in interrupt. 

Yes, that is correct.

> The issue, as I understand it is that non-interrupt code may obtain the 
> lock and then the interrupt code is executed... hence the deadlock and 
> the need to use spin_lock_irqsave() and spin_unlock_irqrestore().

Exactly. All the spinlocks that are aquired in interrupt code-paths must 
be acquired with _irqsave()/_irqrestore() from the non-interrupt code to 
prevent exactly this kind of deadlock.
> 
> The key difference is the replacement of spin_lock() with spin_trylock()
> such that if the non-interrupt code has already obtained the lock, the
> interrupt will not deadlock but instead take the else path and schedule a
> framebuffer update at the next interval.

Why is _irqsave() and/or deferred work not enough? The aproach with 
_trylock() seems to be overly complicated for no good reason (I personally 
become very suspicious every time I see code that is using _trylock()).

[ by the way, Rick, are you planning to resubmit the G13 driver with 
incorporated feedback from the last review round? ]

Thanks,

-- 
Jiri Kosina
SUSE Labs, Novell Inc.
--
To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
the body of a message to majordomo@xxxxxxxxxxxxxxx
More majordomo info at  http://vger.kernel.org/majordomo-info.html
Please read the FAQ at  http://www.tux.org/lkml/