Re: unaligned access using bash on glibc 2.5

On 5/2/07, John David Anglin <[email protected]> wrote:
> This is a bug in __attribute__ ((tls_model ("initial-exec"))).  It
> doesn't express that it uses r19.  As a result, GCC is free to optimize
> the save and restore of r19 across the call to free away.
>
> The same is true of the asm.

You told me that listing r19 as input was problematic. Is this no longer true?

I noticed that the tie_load tgd_load and tld _load patterns in pa.md
are broken and don't specify a dependency on r19 or r27.

Would you be able to cook up a possible fix? :-)

Cheers,
Carlos.