Re: [HACKERS] Erroneous PPC spinlock code

2003-11-09 Thread Bruce Momjian
Marcus Meissner wrote: -- Start of PGP signed section. > On Thu, Nov 06, 2003 at 12:08:56AM +0100, Reinhard Max wrote: > > On Wed, 5 Nov 2003 at 13:28, Tom Lane wrote: > > > > > Peter Eisentraut <[EMAIL PROTECTED]> writes: > > > > > > > The SuSE PPC guru said that the PPC spinlock code we currentl

Re: [HACKERS] Erroneous PPC spinlock code

2003-11-09 Thread Reinhard Max
On Wed, 5 Nov 2003 at 13:28, Tom Lane wrote: > Peter Eisentraut <[EMAIL PROTECTED]> writes: > > > The SuSE PPC guru said that the PPC spinlock code we currently use > > may behave erroneously on multiprocessor systems. > > What's his evidence for that claim? Let's ask himself. > The code we have

Re: [HACKERS] Erroneous PPC spinlock code

2003-11-09 Thread Marcus Meissner
On Thu, Nov 06, 2003 at 12:08:56AM +0100, Reinhard Max wrote: > On Wed, 5 Nov 2003 at 13:28, Tom Lane wrote: > > > Peter Eisentraut <[EMAIL PROTECTED]> writes: > > > > > The SuSE PPC guru said that the PPC spinlock code we currently use > > > may behave erroneously on multiprocessor systems. > > >

Re: [HACKERS] Erroneous PPC spinlock code

2003-11-05 Thread Tom Lane
Peter Eisentraut <[EMAIL PROTECTED]> writes: > The SuSE PPC guru said that the PPC spinlock code we currently use may > behave erroneously on multiprocessor systems. Attached is the proposed > patch, suggested for inclusion in 7.4. Comments? I looked into this some more. The current CVS tip is

Re: [HACKERS] Erroneous PPC spinlock code

2003-11-05 Thread Tom Lane
Peter Eisentraut <[EMAIL PROTECTED]> writes: > The SuSE PPC guru said that the PPC spinlock code we currently use may > behave erroneously on multiprocessor systems. What's his evidence for that claim? The code we have is based directly on the recommendations in the PPC manuals, and has been test

[HACKERS] Erroneous PPC spinlock code

2003-11-05 Thread Peter Eisentraut
The SuSE PPC guru said that the PPC spinlock code we currently use may behave erroneously on multiprocessor systems. Attached is the proposed patch, suggested for inclusion in 7.4. Comments? -- Peter Eisentraut [EMAIL PROTECTED] --- src/include/storage/s_lock.h +++ src/include/storage/s_lock.