uses of cmpxchg/xadd in spinlock.h and rwsem.h vs. CONFIG_M386

uses of cmpxchg/xadd in spinlock.h and rwsem.h vs. CONFIG_M386

[Jan Beulich]

While looking at a completely different issue I happened to grep for uses of
xadd, and it would appear to me that there got uses added that would make
a M386-configured SMP kernel die on an actual i386.

Oh, for the rwsem case I see - it would use the RWSEM_GENERIC_SPINLOCK case
when selecting M386. But I see nothing similar for the ticket spinlocks - am
I overlooking something?

Jan

Re: uses of cmpxchg/xadd in spinlock.h and rwsem.h vs. CONFIG_M386

[Nick Piggin]

On Tue, May 12, 2009 at 04:19:00PM +0100, Jan Beulich wrote:
> While looking at a completely different issue I happened to grep for uses of
> xadd, and it would appear to me that there got uses added that would make
> a M386-configured SMP kernel die on an actual i386.
> 
> Oh, for the rwsem case I see - it would use the RWSEM_GENERIC_SPINLOCK case
> when selecting M386. But I see nothing similar for the ticket spinlocks - am
> I overlooking something?

Oh, hmm, we do actually attempt to support 386 SMP kernels... I thought
that config wasn't allowed, but looking at the atomic.h code, it has
ifdef fallbacks.

I guess some similar hacks could be added to spinlock.h code.

Re: uses of cmpxchg/xadd in spinlock.h and rwsem.h vs. CONFIG_M386

[Alan Cox]

> Oh, hmm, we do actually attempt to support 386 SMP kernels... I thought
> that config wasn't allowed, but looking at the atomic.h code, it has
> ifdef fallbacks.

We don't support running on SMP 386 hardware but we do in theory support
running uniproc on it.

Re: uses of cmpxchg/xadd in spinlock.h and rwsem.h vs. CONFIG_M386

[Peter Zijlstra]

On Tue, 2009-05-12 at 17:37 +0200, Nick Piggin wrote:
> On Tue, May 12, 2009 at 04:19:00PM +0100, Jan Beulich wrote:
> > While looking at a completely different issue I happened to grep for uses of
> > xadd, and it would appear to me that there got uses added that would make
> > a M386-configured SMP kernel die on an actual i386.
> > 
> > Oh, for the rwsem case I see - it would use the RWSEM_GENERIC_SPINLOCK case
> > when selecting M386. But I see nothing similar for the ticket spinlocks - am
> > I overlooking something?
> 
> Oh, hmm, we do actually attempt to support 386 SMP kernels... I thought
> that config wasn't allowed, but looking at the atomic.h code, it has
> ifdef fallbacks.
> 
> I guess some similar hacks could be added to spinlock.h code.

Last time this came up I thought we decided i386-smp wasn't supported,
and the Kconfig files ought to be adjusted.

Re: uses of cmpxchg/xadd in spinlock.h and rwsem.h vs. CONFIG_M386

[H. Peter Anvin]

Jan Beulich wrote:
> While looking at a completely different issue I happened to grep for uses of
> xadd, and it would appear to me that there got uses added that would make
> a M386-configured SMP kernel die on an actual i386.
> 
> Oh, for the rwsem case I see - it would use the RWSEM_GENERIC_SPINLOCK case
> when selecting M386. But I see nothing similar for the ticket spinlocks - am
> I overlooking something?

It's quite possible we break 386 on a regular basis... I'm not sure how
many people even in the embedded industry run current kernels on
386-compatible hardware.  Most embedded hardware is at least 486.

	-hpa

-- 
H. Peter Anvin, Intel Open Source Technology Center
I work for Intel.  I don't speak on their behalf.

Re: uses of cmpxchg/xadd in spinlock.h and rwsem.h vs. CONFIG_M386

[Nick Piggin]

On Tue, May 12, 2009 at 08:51:12AM -0700, H. Peter Anvin wrote:
> Jan Beulich wrote:
> > While looking at a completely different issue I happened to grep for uses of
> > xadd, and it would appear to me that there got uses added that would make
> > a M386-configured SMP kernel die on an actual i386.
> > 
> > Oh, for the rwsem case I see - it would use the RWSEM_GENERIC_SPINLOCK case
> > when selecting M386. But I see nothing similar for the ticket spinlocks - am
> > I overlooking something?
> 
> It's quite possible we break 386 on a regular basis... I'm not sure how
> many people even in the embedded industry run current kernels on
> 386-compatible hardware.  Most embedded hardware is at least 486.

Is it time to get rid of attempts to support SMP kernels that run on 386, then?
Considering we don't actually support SMP 386, and if there is any 386 systems
then they're going to be running custom kernels anyway.

Re: uses of cmpxchg/xadd in spinlock.h and rwsem.h vs. CONFIG_M386

[H. Peter Anvin]

Nick Piggin wrote:
> 
> Is it time to get rid of attempts to support SMP kernels that run on 386, then?
> Considering we don't actually support SMP 386, and if there is any 386 systems
> then they're going to be running custom kernels anyway.
> 

Makes sense to me.  We have never supported SMP 386 to the best of my
knowledge, and a patch to rip that out would be appreciated.

	-hpa

-- 
H. Peter Anvin, Intel Open Source Technology Center
I work for Intel.  I don't speak on their behalf.

Re: uses of cmpxchg/xadd in spinlock.h and rwsem.h vs. CONFIG_M386

[Jan Beulich]

>>> "H. Peter Anvin" <hpa@zytor.com> 12.05.09 18:04 >>>
>Nick Piggin wrote:
>> 
>> Is it time to get rid of attempts to support SMP kernels that run on 386, then?
>> Considering we don't actually support SMP 386, and if there is any 386 systems
>> then they're going to be running custom kernels anyway.
>> 
>
>Makes sense to me.  We have never supported SMP 386 to the best of my
>knowledge, and a patch to rip that out would be appreciated.

But why shouldn't a distro be allowed to configure a kernel that runs (UP) on
i386 and also MP on modern systems?

Jan

Re: uses of cmpxchg/xadd in spinlock.h and rwsem.h vs. CONFIG_M386

[Andreas Schwab]

"Jan Beulich" <JBeulich@novell.com> writes:

> While looking at a completely different issue I happened to grep for uses of
> xadd, and it would appear to me that there got uses added that would make
> a M386-configured SMP kernel die on an actual i386.
>
> Oh, for the rwsem case I see - it would use the RWSEM_GENERIC_SPINLOCK case
> when selecting M386. But I see nothing similar for the ticket spinlocks - am
> I overlooking something?

This came up before, see
<http://marc.info/?l=linux-kernel&m=123141659309208&w=2>.

Andreas.

-- 
Andreas Schwab, schwab@linux-m68k.org
GPG Key fingerprint = 58CA 54C7 6D53 942B 1756  01D3 44D5 214B 8276 4ED5
"And now for something completely different."