Re: [RFC PATCH glibc 4/8] glibc: Perform rseq(2) registration at C startup and thread creation (v15)

[Mathieu Desnoyers via Libc-alpha <libc-alpha@sourceware.org>]

----- On Mar 19, 2020, at 12:03 PM, Florian Weimer fw@deneb.enyo.de wrote:

> * Mathieu Desnoyers:
> 
>>> Can you use __has_include in <sys/rseq.h>, with a copy of the kernel
>>> definitions if the kernel header is not available?
>>
>> Sure. Should I pull a verbatim copy of uapi linux/rseq.h into glibc ?
>> If so, where should I put it ?
> 
> Probably into <sys/rseq.h>, perhaps with a construct like this
> (untested):
> 
> #ifdef __has_include
> # if __has_include ("linux/rseq.h")
> #   define __GLIBC_HAVE_KERNEL_RSEQ
> # endif
> #else
> # include <linux/version.h>
> # if LINUX_VERSION_CODE >= KERNEL_VERSION (4, 18, 0)
> #   define __GLIBC_HAVE_KERNEL_RSEQ
> # endif
> #endif
> 
> #ifdef __GLIBC_HAVE_KERNEL_RSEQ
> # include <linux/rseq.h>
> #else
> 
> … (fallback goes here)
> #endif

OK will do.

> 
> We have an ongoing debate whether the fallback definition should use
> __u64 or uint64_t.

Then I'll keep including <linux/types.h> in the fallback and use
__u{32,64} for now. If this proves to be an issue we can change it later.
This is the minimal change from the uapi header.

> 
> You also need to add an assert that the compiler supports
> __attribute__ ((aligned)) because ignoring it produces an
> ABI-incompatible header.

Are you aware of some helper macro I should use to do this, or
is it done elsewhere in glibc ?

> The struct rseq/struct rseq_cs definitions
> are broken, they should not try to change the alignment.

AFAIU, this means we should ideally not have used __attribute__((aligned))
in the uapi headers in the first place. Why is it broken ? However, now
that it is in the wild, it's a bit late to change that.

> PS: I have Internet connection trouble.  Nobody should be worried if I
> drop off the net for a while.  I understand this is quite a bad time
> for that. 8-(

Allright, thanks for the heads up! Stay safe!

Thanks,

Mathieu


-- 
Mathieu Desnoyers
EfficiOS Inc.
http://www.efficios.com