On Tue, Mar 01, 2016 at 02:37:57AM +0000, Paul Burton wrote:
> When flush_dcache_page is called on an executable page, that page is
> about to be provided to userland & we can presume that the icache
> contains no valid entries for its address range. However if the icache
> does not fill from the dcache then we cannot presume that the pages
> content has been written back as far as the memories that the dcache
> will fill from (ie. L2 or further out).
>
> This was being done for lowmem pages, but not for highmem which can lead
> to icache corruption. Fix this by mapping highmem pages & flushing their
> content from the dcache in __flush_dcache_page before providing the page
> to userland, just as is done for lowmem pages.
>
> Signed-off-by: Paul Burton <[email protected]>
> Cc: Lars Persson <[email protected]>
> ---
>
> arch/mips/mm/cache.c | 12 +++++++++---
> 1 file changed, 9 insertions(+), 3 deletions(-)
>
> diff --git a/arch/mips/mm/cache.c b/arch/mips/mm/cache.c
> index 3f159ca..5a67d8c 100644
> --- a/arch/mips/mm/cache.c
> +++ b/arch/mips/mm/cache.c
> @@ -16,6 +16,7 @@
> #include <linux/mm.h>
>
> #include <asm/cacheflush.h>
> +#include <asm/highmem.h>
> #include <asm/processor.h>
> #include <asm/cpu.h>
> #include <asm/cpu-features.h>
> @@ -83,8 +84,6 @@ void __flush_dcache_page(struct page *page)
> struct address_space *mapping = page_mapping(page);
> unsigned long addr;
>
> - if (PageHighMem(page))
> - return;
> if (mapping && !mapping_mapped(mapping)) {
> SetPageDcacheDirty(page);
> return;
> @@ -95,8 +94,15 @@ void __flush_dcache_page(struct page *page)
> * case is for exec env/arg pages and those are %99 certainly going to
> * get faulted into the tlb (and thus flushed) anyways.
> */
> - addr = (unsigned long) page_address(page);
> + if (PageHighMem(page))
> + addr = (unsigned long)kmap_atomic(page);
> + else
> + addr = (unsigned long)page_address(page);
> +
> flush_data_cache_page(addr);
> +
> + if (PageHighMem(page))
> + __kunmap_atomic((void *)addr);
> }
>
> EXPORT_SYMBOL(__flush_dcache_page);
I don't see how this should work with cache aliases. If the page is unmapped
kmap_atomic will pick a deterministic address only under some circumstances,
kmap won't. As the result the wrong cache way will be flushed out, I think.
Ralf
On Tue, Mar 29, 2016 at 10:35:43AM +0200, Ralf Baechle wrote:
> On Tue, Mar 01, 2016 at 02:37:57AM +0000, Paul Burton wrote:
>
> > When flush_dcache_page is called on an executable page, that page is
> > about to be provided to userland & we can presume that the icache
> > contains no valid entries for its address range. However if the icache
> > does not fill from the dcache then we cannot presume that the pages
> > content has been written back as far as the memories that the dcache
> > will fill from (ie. L2 or further out).
> >
> > This was being done for lowmem pages, but not for highmem which can lead
> > to icache corruption. Fix this by mapping highmem pages & flushing their
> > content from the dcache in __flush_dcache_page before providing the page
> > to userland, just as is done for lowmem pages.
> >
> > Signed-off-by: Paul Burton <[email protected]>
> > Cc: Lars Persson <[email protected]>
> > ---
> >
> > arch/mips/mm/cache.c | 12 +++++++++---
> > 1 file changed, 9 insertions(+), 3 deletions(-)
> >
> > diff --git a/arch/mips/mm/cache.c b/arch/mips/mm/cache.c
> > index 3f159ca..5a67d8c 100644
> > --- a/arch/mips/mm/cache.c
> > +++ b/arch/mips/mm/cache.c
> > @@ -16,6 +16,7 @@
> > #include <linux/mm.h>
> >
> > #include <asm/cacheflush.h>
> > +#include <asm/highmem.h>
> > #include <asm/processor.h>
> > #include <asm/cpu.h>
> > #include <asm/cpu-features.h>
> > @@ -83,8 +84,6 @@ void __flush_dcache_page(struct page *page)
> > struct address_space *mapping = page_mapping(page);
> > unsigned long addr;
> >
> > - if (PageHighMem(page))
> > - return;
> > if (mapping && !mapping_mapped(mapping)) {
> > SetPageDcacheDirty(page);
> > return;
> > @@ -95,8 +94,15 @@ void __flush_dcache_page(struct page *page)
> > * case is for exec env/arg pages and those are %99 certainly going to
> > * get faulted into the tlb (and thus flushed) anyways.
> > */
> > - addr = (unsigned long) page_address(page);
> > + if (PageHighMem(page))
> > + addr = (unsigned long)kmap_atomic(page);
> > + else
> > + addr = (unsigned long)page_address(page);
> > +
> > flush_data_cache_page(addr);
> > +
> > + if (PageHighMem(page))
> > + __kunmap_atomic((void *)addr);
> > }
> >
> > EXPORT_SYMBOL(__flush_dcache_page);
>
> I don't see how this should work with cache aliases. If the page is unmapped
> kmap_atomic will pick a deterministic address only under some circumstances,
> kmap won't. As the result the wrong cache way will be flushed out, I think.
>
> Ralf
Hi Ralf,
None of the systems I tested this on have cache aliases, and highmem on
systems with cache aliases is currently unsupported to the extent that
we BUG_ON such cases in flush_kernel_dcache_page.
So hopefully you won't require that this code making highmem without
aliases also fix the currently-unsupported highmem with aliases case?
Thanks,
Paul