Re: [PATCH] selftests: kvm: Fix a compile error in selftests/kvm/rseq_test.c

From: Andrew Jones <andrew.jones@linux.dev>
To: Sean Christopherson <seanjc@google.com>
Cc: Jinrong Liang <ljr.kernel@gmail.com>,
	Paolo Bonzini <pbonzini@redhat.com>,
	Wanpeng Li <wanpengli@tencent.com>,
	Jim Mattson <jmattson@google.com>,
	Vitaly Kuznetsov <vkuznets@redhat.com>,
	Joerg Roedel <joro@8bytes.org>,
	Jinrong Liang <cloudliang@tencent.com>,
	kvm@vger.kernel.org, linux-kernel@vger.kernel.org
Subject: Re: [PATCH] selftests: kvm: Fix a compile error in selftests/kvm/rseq_test.c
Date: Wed, 3 Aug 2022 19:26:27 +0200	[thread overview]
Message-ID: <20220803172627.kccwzda6eshx3vol@kamzik> (raw)
In-Reply-To: <YuqeDetNukKp9lyF@google.com>

On Wed, Aug 03, 2022 at 04:10:53PM +0000, Sean Christopherson wrote:
> On Wed, Aug 03, 2022, Andrew Jones wrote:
> > On Wed, Aug 03, 2022 at 09:58:51PM +0800, Jinrong Liang wrote:
> > > My ldd version is (GNU libc) 2.28, and I get a compilation error in this case.
> > > But I use another ldd (Ubuntu GLIBC 2.31-0ubuntu9.2) 2.31 is compiling fine.
> > > This shows that compilation errors may occur in different GNU libc environments.
> > > Would it be more appropriate to use syscall for better compatibility?
> > 
> > OK, it's a pity, but no big deal to use syscall().
> 
> Ya, https://man7.org/linux/man-pages/man2/gettid.2.html says:
> 
>   The gettid() system call first appeared on Linux in kernel 2.4.11.  Library
>   support was added in glibc 2.30.
> 
> But there are already two other instances of syscall(SYS_gettid) in KVM selftests,
> tools/testing/selftests/kvm/lib/assert.c even adds a _gettid() wrapper.

Ha! And I found four more in selftests...

testing/selftests/powerpc/include/utils.h
testing/selftests/proc/proc.h
testing/selftests/rseq/param_test.c
testing/selftests/sched/cs_prctl_test.c

and even more in tools...

> 
> So rather than having to remember (or discover) to use syscall(SYS_gettid), I wonder
> if it's possible to conditionally define gettid()?  E.g. check for GLIBC version?
> Or do
> 
>   #define gettid() syscall(SYS_gettid)
> 
> so that it's always available and simply overrides the library's gettid() if it's
> provided?

Sounds good to me. Now the question is where to put it? kvm_util.h,
test_util.h, or maybe we should create a new header just for stuff
like this?

It doesn't really "fit" in kvm_util.h, but if we put it there, then we
greatly reduce the chance that we'll have to revisit this issue again.
We could also create a new header just for stuff like this and then
include that from kvm_util.h...

Thanks,
drew