VDSO

VDSO

[Alex Smith]

Hi,

I'm intending to start work on a proper VDSO implementation for MIPS that can provide user implementations of gettimeofday() and such, as well as take over from the current signal return trampoline code.

Just wondering if there's anyone else who's doing any work towards this so we can avoid duplication of effort if so?

Thanks,
Alex

VDSO

[Alex Smith]

Hi,

I'm intending to start work on a proper VDSO implementation for MIPS that can provide user implementations of gettimeofday() and such, as well as take over from the current signal return trampoline code.

Just wondering if there's anyone else who's doing any work towards this so we can avoid duplication of effort if so?

Thanks,
Alex

Re: VDSO

[David Daney]

On 08/12/2015 03:21 AM, Alex Smith wrote:
> Hi,
>
> I'm intending to start work on a proper VDSO implementation for MIPS that can provide user implementations of gettimeofday() and such, as well as take over from the current signal return trampoline code.
>
> Just wondering if there's anyone else who's doing any work towards this so we can avoid duplication of effort if so?
>

I am not working on it, but have a small laundry list of what could be done:

1) Perhaps the VA of the vdso needs to be randomized.

2) Provide proper .eh_frame data for the signal return trampolines.

3) gettimeofday (as you suggest).

4) Provide optimized versions of memcpy, etc.

David Daney

Re: VDSO

[Alex Smith]

On 12/08/2015 17:14, David Daney wrote:
> On 08/12/2015 03:21 AM, Alex Smith wrote:
>> Hi,
>>
>> I'm intending to start work on a proper VDSO implementation for MIPS that can provide user implementations of gettimeofday() and such, as well as take over from the current signal return trampoline code.
>>
>> Just wondering if there's anyone else who's doing any work towards this so we can avoid duplication of effort if so?
>>
> 
> I am not working on it, but have a small laundry list of what could be done:
> 
> 1) Perhaps the VA of the vdso needs to be randomized.
> 
> 2) Provide proper .eh_frame data for the signal return trampolines.
> 
> 3) gettimeofday (as you suggest).

Yes, I intend to do all of these.

> 4) Provide optimized versions of memcpy, etc.

Is this something that belongs in a VDSO? memcpy and friends aren't syscalls, this belongs in libc IMO. ELF HWCAP flags provided by the kernel could be used by libc to decide which CPU features it can make use of to implement them.

Thanks,
Alex

> 
> David Daney
> 
> 

Re: VDSO

[Alex Smith]

On 12/08/2015 17:14, David Daney wrote:
> On 08/12/2015 03:21 AM, Alex Smith wrote:
>> Hi,
>>
>> I'm intending to start work on a proper VDSO implementation for MIPS that can provide user implementations of gettimeofday() and such, as well as take over from the current signal return trampoline code.
>>
>> Just wondering if there's anyone else who's doing any work towards this so we can avoid duplication of effort if so?
>>
> 
> I am not working on it, but have a small laundry list of what could be done:
> 
> 1) Perhaps the VA of the vdso needs to be randomized.
> 
> 2) Provide proper .eh_frame data for the signal return trampolines.
> 
> 3) gettimeofday (as you suggest).

Yes, I intend to do all of these.

> 4) Provide optimized versions of memcpy, etc.

Is this something that belongs in a VDSO? memcpy and friends aren't syscalls, this belongs in libc IMO. ELF HWCAP flags provided by the kernel could be used by libc to decide which CPU features it can make use of to implement them.

Thanks,
Alex

> 
> David Daney
> 
>