[RESEND PATCH] usb: gadget: at91_udc: move prepare clk into process context

On Fri, Dec 19, 2014 at 03:08:15PM +0100, Nicolas Ferre wrote:

Le 19/12/2014 15:02, Ronald Wahl a ?crit :

quoted

On 19.12.2014 14:51, Luis Henriques wrote:

quoted

Hi Felipe,

On Thu, Nov 20, 2014 at 01:50:49PM -0600, Felipe Balbi wrote:

quoted

On Wed, Nov 19, 2014 at 04:37:27PM +0100, Nicolas Ferre wrote:

quoted

From: Ronald Wahl <redacted>

Commit 7628083227b6bc4a7e33d7c381d7a4e558424b6b (usb: gadget: at91_udc:
prepare clk before calling enable) added clock preparation in interrupt
context. This is not allowed as it might sleep. Also setting the clock
rate is unsafe to call from there for the same reason. Move clock
preparation and setting clock rate into process context (at91udc_probe).

Signed-off-by: Ronald Wahl <redacted>
Acked-by: Alexandre Belloni <redacted>
Acked-by: Boris Brezillon <redacted>
Acked-by: Nicolas Ferre <redacted>
Cc: Felipe Balbi <redacted>
Cc: <redacted> # v3.17+
---
Hi Felipe,

I forgot to answer you on this patch. So I resend it now with the proper
"stable" tag. You can also queue it during this -rc phase if you feel it is
still possible.

I think it's late for v3.18, so it'll go on v3.19 and get backported to
3.17 and 3.18. Sorry :-s

Although this commit (b2ba27a5c56f "usb: gadget: at91_udc: move
prepare clk into process context") is tagged for stable v3.17+, it
seems like it could be applied to earlier kernels.

3.16, 3.13 and 3.12 seem to be affected by the same issue (and they
all include commit 7628083227b6 "usb: gadget: at91_udc: prepare clk
before calling enable").  Is there any reason for not applying it in
these trees?

Not to forget 3.14 (LTS) which was the branch where I primarily found 
the issue...

Yes, of course!  Sorry.

Well it's maybe an issue with the re-naming of the directory to
drivers/usb/gadget/udc/ introduced by patch:
90fccb529d24 (usb: gadget: Gadget directory cleanup - group UDC drivers)

The patch doesn't apply out of the box but it surely can be applied in
those earlier kernels.

Right, the file was renamed but the backport seems to be trivial.
Thanks for confirming.

Cheers,
--
Lu?s

Thanks, bye.
-- 
Nicolas Ferre