Re: [uclibc-ng-devel] arc patch from glibc repo

[Waldemar Brodkorb]

Hi Vineet,
Vineet Gupta wrote,

> On 11/05/2017 02:03 AM, Waldemar Brodkorb wrote:
> >Hi Alexey,
> >
> >the attached patch fixes at least 3 test suite errors for me.
> >tst-cancel20/21/4.
> >Tested with latest 2017.09 ARC binutils/gcc.
> >
> >Okay to apply?
> >
> >best regards
> >  Waldemar
> 
> Are you sure it is the list above.
> 
> I presume, 
> https://tests.embedded-test.org/uClibc-ng/1.0.26/REPORT.arcv2.libc.uClibc-ng-1.0.26
> didn't have this fix and yet the tests mentioned above are PASS there.
> 
> The reason I ask is because in my experiments here with ARC GNU 2017.09 +
> uClibc with and w/o this patch (actually not exactly same, but similar -
> attached), I don't see any consistent PASS(es)
> There are some changes to say tst-kill6, tst-oncex3 etc but if you run them
> 5 times they seem to PASS/FAIL randomly - in either setups
> 
> So this patch is not fixing anything IMO. We need to really fix the inherent
> race in these tests.
> 
> BTW I did post a test case fix for tst-cancel2 and that is now consistently 
> PASS.

Indeed it seems really to depend on which system I have run the
tests with and without the patch.

On my IBM X200 notebook the tests passed after adding the patch.
But on the gcc farm server the tests were still failing.

So it is like you said, the patch does not fix the issue.

best regards
 Waldemar
___
devel mailing list
devel@uclibc-ng.org
https://mailman.uclibc-ng.org/cgi-bin/mailman/listinfo/devel

Re: [uclibc-ng-devel] arc patch from glibc repo

[Vineet Gupta]

On 11/05/2017 02:03 AM, Waldemar Brodkorb wrote:

Hi Alexey,

the attached patch fixes at least 3 test suite errors for me.
tst-cancel20/21/4.
Tested with latest 2017.09 ARC binutils/gcc.

Okay to apply?

best regards
  Waldemar


Are you sure it is the list above.

I presume, 
https://tests.embedded-test.org/uClibc-ng/1.0.26/REPORT.arcv2.libc.uClibc-ng-1.0.26

didn't have this fix and yet the tests mentioned above are PASS there.

The reason I ask is because in my experiments here with ARC GNU 2017.09 + uClibc 
with and w/o this patch (actually not exactly same, but similar - attached), I 
don't see any consistent PASS(es)
There are some changes to say tst-kill6, tst-oncex3 etc but if you run them 5 
times they seem to PASS/FAIL randomly - in either setups


So this patch is not fixing anything IMO. We need to really fix the inherent race 
in these tests.


BTW I did post a test case fix for tst-cancel2 and that is now consistently 
PASS.

-Vineet

>From 055c35bed580c8946feaba4ba8a1f61b550f48f6 Mon Sep 17 00:00:00 2001
From: Cupertino Miranda 
Date: Tue, 21 Feb 2017 17:05:22 +0100
Subject: [PATCH] Patch to clobber blink that works.

---
 libc/sysdeps/linux/arc/bits/syscalls.h | 2 ++
 1 file changed, 2 insertions(+)

diff --git a/libc/sysdeps/linux/arc/bits/syscalls.h b/libc/sysdeps/linux/arc/bits/syscalls.h
index 248ef78..92c1035 100644
--- a/libc/sysdeps/linux/arc/bits/syscalls.h
+++ b/libc/sysdeps/linux/arc/bits/syscalls.h
@@ -68,6 +68,7 @@ extern int __syscall_error (int);
 #define INLINE_SYSCALL(name, nr_args, args...)\
 ({	\
 	register int __res __asm__("r0");\
+	__asm__ volatile ("" : : : "blink");  \
 	__res = INTERNAL_SYSCALL_NCS(__NR_##name, , nr_args, args);	\
 	if (__builtin_expect (INTERNAL_SYSCALL_ERROR_P ((__res), ), 0))	\
 	{\
@@ -81,6 +82,7 @@ extern int __syscall_error (int);
 #define INLINE_SYSCALL_NCS(num, nr_args, args...)			\
 ({	\
 	register int __res __asm__("r0");\
+	__asm__ volatile ("" : : : "blink");  \
 	__res = INTERNAL_SYSCALL_NCS(num, , nr_args, args);		\
 	if (__builtin_expect (INTERNAL_SYSCALL_ERROR_P ((__res), ), 0))	\
 	{\
-- 
2.9.0

___
devel mailing list
devel@uclibc-ng.org
https://mailman.uclibc-ng.org/cgi-bin/mailman/listinfo/devel