On Wed, 4 Mar 2009 01:01:04 +0900
Akinobu Mita <akinobu.mita@xxxxxxxxx> wrote:
> +static void unpoison_page(struct page *page)
> +{
> + unsigned char *mem;
> + int i;
> +
> + if (!page->poison)
> + return;
> +
> + mem = kmap_atomic(page, KM_USER0);
> + for (i = 0; i < PAGE_SIZE; i++) {
> + if (mem[i] != PAGE_POISON) {
> + dump_broken_mem(mem);
> + break;
> + }
> + }
> + kunmap_atomic(mem, KM_USER0);
> + page->poison = false;
> +}
> +
> +static void unpoison_pages(struct page *page, int n)
> +{
> + int i;
> +
> + for (i = 0; i < n; i++)
> + unpoison_page(page + i);
> +}
> +
> +void kernel_map_pages(struct page *page, int numpages, int enable)
> +{
> + if (!debug_pagealloc_enabled)
> + return;
> +
> + if (enable)
> + unpoison_pages(page, numpages);
> + else
> + poison_pages(page, numpages);
> +}
kernel_map_pages() is called from the memory-allocation and
memory-freeing paths. Hence it can be called from interrupt contexts.
KM_USER0 must not be used from interrupt context - it will corrupt the
non-interrupt context's pte, causing unpleasing very hard to track down
memory corruption. Often memory which is getting written to the user's
disk. This makes users unhappy.
We could use KM_IRQ0 here. The code should disable local interrupts
when holding a KM_IRQ0 kmap.
If this code were to switch to using KM_IRQ0 then we still have a
problem - if any other place in the kernel does a memory allcoation or
free while holding a KM_IRQ0 kmap, then this new code will corrupt the
caller's pte.
So I guess we'll need to create a new kmap_atomic slot for this
application. It will need interrupt protection - the page allocator can
be called from interrupt context while it is already running in
non-interrupt context.
Alternatively, we could just not do the kmap_atomic() at all. i386
won't be using this code and IIRC the only other highmem architecture
is powerpc32, and ppc32 appears to also have its own DEBUG_PAGEALLOC
implementation. So you could remove the kmap_atomic() stuff and put
#ifdef CONFIG_HIGHMEM
#error i goofed
#endif
in there.
--
To unsubscribe from this list: send the line "unsubscribe linux-arch" in
the body of a message to majordomo@xxxxxxxxxxxxxxx
More majordomo info at http://vger.kernel.org/majordomo-info.html
