MTD API locking and concurrency

MTD API locking and concurrency

[Robert Jarzmik]

Hi,

I came across a concurrent usage of my MTD driver by UBIFS by doing
an IOC_UBI_UPVOL. As my chip doesn't support multiple requests at a time, I need
to protect the access to chip registers.

I wonder if mtd api function can be called in interrupt context. My wish would
be to use mutexes, but I have to be sure there is no interrupt path I have not
seen. The only thing I'm certain is that my MTD drivers doesn't handle
interrupts, so the interrupt won't come from him.

Is there a guarantee that mtd api functions are called with !in_interrupt() ?

Cheers.

-- 
Robert

Re: MTD API locking and concurrency

[Artem Bityutskiy]

[-- Attachment #1: Type: text/plain, Size: 639 bytes --]

On Tue, 2012-03-13 at 06:45 +0100, Robert Jarzmik wrote:
> Hi,
> 
> I came across a concurrent usage of my MTD driver by UBIFS by doing
> an IOC_UBI_UPVOL. As my chip doesn't support multiple requests at a time, I need
> to protect the access to chip registers.
> 
> I wonder if mtd api function can be called in interrupt context.

No, this is not explicitely documented and you are welcome to send a
patch which adds a comment to mtd.h, but all the MTD API functions are
supposed to be called from sleeping context. IOW, the caller should keep
in mind that these functions may sleep.

-- 
Best Regards,
Artem Bityutskiy

[-- Attachment #2: This is a digitally signed message part --]
[-- Type: application/pgp-signature, Size: 836 bytes --]

Re: MTD API locking and concurrency

[Mike Dunn]

On 03/12/2012 10:45 PM, Robert Jarzmik wrote:
> Hi,
> 
> I came across a concurrent usage of my MTD driver by UBIFS by doing
> an IOC_UBI_UPVOL. As my chip doesn't support multiple requests at a time, I need
> to protect the access to chip registers.


FYI, the nand infrastructure code handles this for its devices; see
nand_base.c::nand_get_device()

Mike