Locking in YAFFS/MTD [WAS: Re: [Yaffs] lock held at task exit time]

[Jon Masters]

On 3/14/06, Todd Poynor <tpoynor at mvista.com> wrote:

> Tasks that use yaffs* filesystems may oops when the Realtime Preemption
> patches from Ingo Molnar et al are applied and CONFIG_PREEMPT_RT=y:

> There is a "Semaphore for waiting on erasure" that is initialized to a
> locked state but is not otherwise referenced in the code.  In RT kernels
> this downs an RT mutex.

On a related note, I was recently looking at the general state of
locking in MTD/NAND (mostly on 2.4 kernels though) and discovered a
lack of global locking that will cause problems when using multiple
MTD partitions (multiple filesystems) and where we ultimately
implement properly blocking IO in the MTD layer.

Am I being silly, was this a problem only with older MTDs, or do we
need to look at the whole locking issue a little bit more?

Jon.