x86-32: improve atomic64_t functions (v2)
in [Kernel]
|
Prev: Security: add static to security_ops and default_security_ops variable
Next: qla2xxx UNDERRUNS and failed multipath paths
From: H. Peter Anvin on 1 Mar 2010 02:40 On 02/26/2010 03:23 AM, Luca Barbieri wrote: > Sent patches, both to conditionally perform the test and implement the > functions for x86 and x86-64. Yes, and with the test turned on, the kernel crashes immediately on boot on x86-64. Some minor investigation reveals the following: lib/atomic64.c has the wrong return value for atomic64_add_unless(). With "wrong" I mean it is the opposite sense compared to atomic_add_unless(), not just on x86 but on all architectures. Accordingly, I have to conclude that lib/atomic64.c is buggy, and that since your test matches that bug, I will have to conclude that your x86-32 implementation is also buggy. Thus, please send patches to fix your test and your 32-bit implementations (and preferrably lib/atomic64.c too, but I can do that just fine.) Cc: Paul Mackerras who did the generic atomic64_t implementation for verification that this is indeed a bug. -hpa -- H. Peter Anvin, Intel Open Source Technology Center I work for Intel. I don't speak on their behalf. -- To unsubscribe from this list: send the line "unsubscribe linux-kernel" in the body of a message to majordomo(a)vger.kernel.org More majordomo info at http://vger.kernel.org/majordomo-info.html Please read the FAQ at http://www.tux.org/lkml/
From: Paul Mackerras on 1 Mar 2010 03:50 On Sun, Feb 28, 2010 at 11:35:31PM -0800, H. Peter Anvin wrote: > On 02/26/2010 03:23 AM, Luca Barbieri wrote: > > Sent patches, both to conditionally perform the test and implement the > > functions for x86 and x86-64. > > Yes, and with the test turned on, the kernel crashes immediately on boot > on x86-64. > > Some minor investigation reveals the following: > > lib/atomic64.c has the wrong return value for atomic64_add_unless(). > With "wrong" I mean it is the opposite sense compared to > atomic_add_unless(), not just on x86 but on all architectures. > > Accordingly, I have to conclude that lib/atomic64.c is buggy, and that > since your test matches that bug, I will have to conclude that your > x86-32 implementation is also buggy. Thus, please send patches to fix > your test and your 32-bit implementations (and preferrably > lib/atomic64.c too, but I can do that just fine.) > > Cc: Paul Mackerras who did the generic atomic64_t implementation for > verification that this is indeed a bug. Yes, it sure looks like it. *blush* Paul. -- To unsubscribe from this list: send the line "unsubscribe linux-kernel" in the body of a message to majordomo(a)vger.kernel.org More majordomo info at http://vger.kernel.org/majordomo-info.html Please read the FAQ at http://www.tux.org/lkml/
From: Luca Barbieri on 1 Mar 2010 12:20 > Yes, and with the test turned on, the kernel crashes immediately on boot > on x86-64. > > Some minor investigation reveals the following: > > lib/atomic64.c has the wrong return value for atomic64_add_unless(). > With "wrong" I mean it is the opposite sense compared to > atomic_add_unless(), not just on x86 but on all architectures. > > Accordingly, I have to conclude that lib/atomic64.c is buggy, and that > since your test matches that bug, I will have to conclude that your > x86-32 implementation is also buggy. �Thus, please send patches to fix > your test and your 32-bit implementations (and preferrably > lib/atomic64.c too, but I can do that just fine.) You are right: sent a patchset to fix it. -- To unsubscribe from this list: send the line "unsubscribe linux-kernel" in the body of a message to majordomo(a)vger.kernel.org More majordomo info at http://vger.kernel.org/majordomo-info.html Please read the FAQ at http://www.tux.org/lkml/
From: Luca Barbieri on 1 Mar 2010 12:40
Upon further inspection, atomic64_inc_not_zero was broken too. The generic implementation implements it in terms of atomic64_add_unless and thus does not need a specific fix for it. Sent another patchset to fix that. -- To unsubscribe from this list: send the line "unsubscribe linux-kernel" in the body of a message to majordomo(a)vger.kernel.org More majordomo info at http://vger.kernel.org/majordomo-info.html Please read the FAQ at http://www.tux.org/lkml/ |