On Thu, 19 Jul 2007 14:05:06 +0200
Zoltan Menyhart <[email protected]> wrote:
> KAMEZAWA Hiroyuki wrote:
>
> > Then, what should I do more for fixing this SIGILL problem ?
> >
> > -Kame
>
> I can think of a relatively cheap solution:
>
Maybe I should take performance numbers with the patch.
But is it too costly that flushing icache page only if a page is newly
installed into the system (PG_arch1) && it is mapped as executable ?
I don't want to leak this (stupid) corner case to the file system layer.
Hmm...can't we do clever flushing (like your idea) in VM layer ?
-Kame
> New pages are allocated with the bit PG_arch_1 off, see
> page_cache_read() ... prep_new_page(), i.e. the I-cache is
> not coherent with the D-cache.
>
> page_cache_read() should add a macro, say:
>
> ARCH_PREP_PAGE_BEFORE_READ(page);
>
> before actually calling mapping->a_ops->readpage(file, page).
>
> This macro can be for ia64 something like:
>
> do { if (CPU_has_split_L2_I_cache) set_bit(PG_arch_1, &page->flags); }
>
> and empty for the the architectures non concerned.
>
> The file systems which are identified not to use HW tools to
> avoid split I-cache incoherency, e.g. nfs_readpage(), are required
> to add a macro, say:
>
> ARCH_CHECK_READ_PAGE_COHERENCY(page);
>
> This macro can be for ia64:
>
> do { if (CPU_has_split_L2_I_cache) clear_bit(PG_arch_1, &page->flags); }
>
> and empty for the the architectures non concerned.
>
> Back to do_no_page():
> if the new PTE includes the exec bit and PG_arch_1 is set,
> the page has to be flushed from the I-cache before the PTE is
> made globally visible.
>
> File systems using local block devices with DMA are considered
> to be safe, with the exceptions of the bounce buffers.
>
> When you copy into the destination page, another macro should be
> added, say:
>
> ARCH_CHECK_BOUNCE_READ_COHERENCY(bio_vec);
>
> #define ARCH_CHECK_BOUNCE_READ_COHERENCY(bio_vec) \
> ARCH_CHECK_READ_PAGE_COHERENCY(bio_vec->bv_page)
>
> Remote DMA based network file systems, e.g. Lustre on Quadrics,
> Infiniband are also considered to be safe.
>
> Thanks,
>
> Zoltan
>
-
To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
the body of a message to [email protected]
More majordomo info at http://vger.kernel.org/majordomo-info.html
Please read the FAQ at http://www.tux.org/lkml/
[Index of Archives]
[Kernel Newbies]
[Netfilter]
[Bugtraq]
[Photo]
[Stuff]
[Gimp]
[Yosemite News]
[MIPS Linux]
[ARM Linux]
[Linux Security]
[Linux RAID]
[Video 4 Linux]
[Linux for the blind]
[Linux Resources]
