On Fri, Jan 10, 2003 at 03:32:34PM +0100, Maciej W. Rozycki wrote:
> As you might already know there are a few nasty errata in the R4000 and
> the early R4400 that hit 64-bit operation badly. Here is proposed code to
> detect them. If an erratum is found in the processor and no workaround is
> applied to a kernel executable, the kernel refuses to run. In all cases
> the result of the probes is output to the bootstrap log.
> The code has bits that make use of features of non-standard tools
> (binutils and gcc). But it doesn't depend on them -- when built with
> standard tools and run on an affected system, a kernel will simply fail,
> and on good systems it will run normally. Therefore it's safe to apply,
> and if the ultimate implementation in the tools differs, the code may get
> adjusted appropriately later.
> I'd like to apply this code as soon as possible as I consider it a
> prerequisite for integrating 64-bit support for the DECstation (to prevent
> people from running unreliable code), so please tell me if there are any
> doubts about it. Errata descriptions are available at the MIPS site --
> see: 'http://www.mips.com/publications/r400_r5000.html'. Unfortunately,
> despite several attempts to get a permission to duplicate them within
> Linux sources, I failed to get one.
> I'd like to express my gratitude to Karsten and Thiemo for testing the
> code with their hardware. Without their help, I wouldn't be able to
> prepare appropriate tests for errata my hardware doesn't suffer from.
> + __save_and_cli(flags);
> + __restore_flags(flags);
I suggest to replace these with local_irq_save and local_irq_restore.
They're already deprecated for 2.4 and completly gone in 2.5.
Looks ok to me otherwise.