Re: [PATCH v7 10/25] arm64: compat: Add vDSO

[Vincenzo Frascino]

Hi John,

On 10/07/2019 05:02, John Stultz wrote:
> On Fri, Jun 21, 2019 at 3:18 AM Vincenzo Frascino
> <vincenzo.frascino@xxxxxxx> wrote:
>>
>> Provide the arm64 compat (AArch32) vDSO in kernel/vdso32 in a similar
>> way to what happens in kernel/vdso.
>>
>> The compat vDSO leverages on an adaptation of the arm architecture code
>> with few changes:
>>  - Use of lib/vdso for gettimeofday
>>  - Implementation of syscall based fallback
>>  - Introduction of clock_getres for the compat library
>>  - Implementation of trampolines
>>  - Implementation of elf note
>>
>> To build the compat vDSO a 32 bit compiler is required and needs to be
>> specified via CONFIG_CROSS_COMPILE_COMPAT_VDSO.
>>
> 
> Hey Vincenzo!
>   Congrats on getting this work merged, I know its been a long effort
> over a number of years!
> 
> Though unfortunately, it seems the arm64 vdso code that just landed is
> breaking AOSP for me.
> 
> I see a lot of the following errors:
> 01-01 01:22:14.097   755   755 F libc    : Fatal signal 11 (SIGSEGV),
> code 1 (SEGV_MAPERR), fault addr 0x3cf2c96c in tid 755 (cameraserver),
> pid 755 (cameraserver)
> 01-01 01:22:14.112   759   759 F libc    : Fatal signal 11 (SIGSEGV),
> code 1 (SEGV_MAPERR), fault addr 0x3cf2c96c in tid 759
> (android.hardwar), pid 759 (android.hardwar)
> 01-01 01:22:14.120   756   756 F libc    : Fatal signal 11 (SIGSEGV),
> code 1 (SEGV_MAPERR), fault addr 0x3cf2c96c in tid 756 (drmserver),
> pid 756 (drmserver)
> 
> Which go away if I revert the vdso merge that went in via tip/timers.
> 
> I tried to bisect things down a bit, but as some later fixes are
> required (at one point, date was returning the start epoch and never
> increasing), this hasn't worked too well. But I'm guessing since I
> see: "CROSS_COMPILE_COMPAT not defined or empty, the compat vDSO will
> not be built", and the system is half working, I'm guessing this is an
> issue with just the 32bit code failing.  While I can try to sort out
> the proper CROSS_COMPILE_COMPAT in my build environment, I assume
> userland shouldn't be crashing if that value isn't set.
> 
> Any chance this issue has already been raised?
> 

I do not have Android (bionic/libc) as part of my testing environment hence I
never saw this issue. Thanks for reporting it.

I am investigating the problem and will post a fix as soon as it is ready.

As Will suggested, .config would help the debugging and I would like to ask to
you to test my fix once it is ready. Is that OK for you?

> thanks
> -john
> 

-- 
Regards,
Vincenzo