On Dec 19, 2012 2:13 AM, "Dmitry Antipov" <dmitry.antipov@linaro.org> wrote:
>
> On 12/19/2012 04:55 AM, John Stultz wrote:
>
>> I was playing with your current ashmem unit test and needed the following trivial
>> patch to get it all (seemingly) working against Linus' head kernel w/ ubuntu userland.
>
>
> Ashmem unit test currently targets ashmem kernel driver from:
>
> http://linux-arm.org/git?p=linux-2.6-armdroid.git;a=shortlog;h=refs/heads/linux-3.7-armdroid
>
> which is a bit different from k.org version (we use armdroid version for juice-aosp because
> 64-bit developement goes there). Note that armdroid version has bug: ashmem.c expects that
> _IOC_SIZE (ioctl_number) == sizeof (struct ashmem_pin) for ASHMEM_GET_PIN_STATUS, but ashmem.h
> encodes ASHMEM_GET_PIN_STATUS with _IO(XXX), e.g. with zero _IOC_SIZE.
>
> Since ASHMEM_GET_PIN_STATUS ioctl (a.k.a. ASHMEM_ISPINNED in user-space Android headers)
> uses struct ashmem_pin as 3rd arg, I believe that the correct definition for the user space is:
>
>
> #define ASHMEM_ISPINNED         _IOW(__ASHMEMIOC, 9, struct ashmem_pin)
>
> and for the kernel, respectively:
>
> #define ASHMEM_GET_PIN_STATUS   _IOW(__ASHMEMIOC, 9, struct ashmem_pin)
>
> This is the way I make it work in juice-aosp:
>
> http://git.linaro.org/gitweb?p=people/antipov/linux-aarch64.git;a=commit;h=8bd650de8fbc6afbefd7695f287169a7125701b9
>

So, I'll double check aosp common.bit, but I think the Google kernel and k.org are consistent. Not sure why the arm kernel would be different.

Given the confusion we should check w android devs. I suspect the issue is no one really uses the is pinned/get status interfaces, so this inconsistency has gone unnoticed.

Thanks
-john