The 'highmem' variable is only set at boot, so we can make it
ro_after_init and prevent it from being corrupted inadvertently, or from
ending up in a contended cacheline.
Signed-off-by: Ard Biesheuvel <[email protected]>
---
mm/memory.c | 2 +-
1 file changed, 1 insertion(+), 1 deletion(-)
diff --git a/mm/memory.c b/mm/memory.c
index c125c4969913..50e82cb94ccc 100644
--- a/mm/memory.c
+++ b/mm/memory.c
@@ -106,7 +106,7 @@ EXPORT_SYMBOL(mem_map);
* highstart_pfn must be the same; there must be no gap between ZONE_NORMAL
* and ZONE_HIGHMEM.
*/
-void *high_memory;
+void *high_memory __ro_after_init;
EXPORT_SYMBOL(high_memory);
/*
--
2.30.2
On Mon, 24 Jan 2022, Ard Biesheuvel wrote:
> diff --git a/mm/memory.c b/mm/memory.c
> index c125c4969913..50e82cb94ccc 100644
> --- a/mm/memory.c
> +++ b/mm/memory.c
> @@ -106,7 +106,7 @@ EXPORT_SYMBOL(mem_map);
> * highstart_pfn must be the same; there must be no gap between ZONE_NORMAL
> * and ZONE_HIGHMEM.
> */
> -void *high_memory;
> +void *high_memory __ro_after_init;
> EXPORT_SYMBOL(high_memory);
>
> /*
Same should be true for the mm/nommu.c variant?
On Mon, Jan 24, 2022 at 06:05:55PM +0100, Ard Biesheuvel wrote:
> The 'highmem' variable is only set at boot, so we can make it
> ro_after_init and prevent it from being corrupted inadvertently, or from
> ending up in a contended cacheline.
I'm not against this patch, but it'd be nice to go further and remove
it entirely for !CONFIG_HIGHMEM builds? Adding something like
static inline bool is_high_pfn(unsigned long pfn)
{
#ifdef CONFIG_HIGHMEM
return pfn <= max_low_pfn;
#else
return false;
#endif
}
static inline bool is_high_phys(phys_addr_t pa)
{
return is_high_pfn(pa / PAGE_SIZE);
}
static inline bool is_high_addr(void *addr)
{
return is_high_phys(virt_to_phys(addr));
}
should be all the primitives you need, from a quick grep.
> Signed-off-by: Ard Biesheuvel <[email protected]>
> ---
> mm/memory.c | 2 +-
> 1 file changed, 1 insertion(+), 1 deletion(-)
>
> diff --git a/mm/memory.c b/mm/memory.c
> index c125c4969913..50e82cb94ccc 100644
> --- a/mm/memory.c
> +++ b/mm/memory.c
> @@ -106,7 +106,7 @@ EXPORT_SYMBOL(mem_map);
> * highstart_pfn must be the same; there must be no gap between ZONE_NORMAL
> * and ZONE_HIGHMEM.
> */
> -void *high_memory;
> +void *high_memory __ro_after_init;
> EXPORT_SYMBOL(high_memory);
>
> /*
> --
> 2.30.2
>
>
On Mon, Jan 24, 2022 at 06:05:55PM +0100, Ard Biesheuvel wrote:
> The 'highmem' variable is only set at boot, so we can make it
> ro_after_init and prevent it from being corrupted inadvertently, or from
> ending up in a contended cacheline.
>
> Signed-off-by: Ard Biesheuvel <[email protected]>
Acked-by: Mike Rapoport <[email protected]>
> ---
> mm/memory.c | 2 +-
> 1 file changed, 1 insertion(+), 1 deletion(-)
>
> diff --git a/mm/memory.c b/mm/memory.c
> index c125c4969913..50e82cb94ccc 100644
> --- a/mm/memory.c
> +++ b/mm/memory.c
> @@ -106,7 +106,7 @@ EXPORT_SYMBOL(mem_map);
> * highstart_pfn must be the same; there must be no gap between ZONE_NORMAL
> * and ZONE_HIGHMEM.
> */
> -void *high_memory;
> +void *high_memory __ro_after_init;
> EXPORT_SYMBOL(high_memory);
>
> /*
> --
> 2.30.2
>
>
--
Sincerely yours,
Mike.
On Wed, 26 Jan 2022 at 22:38, Mike Rapoport <[email protected]> wrote:
>
> On Mon, Jan 24, 2022 at 06:05:55PM +0100, Ard Biesheuvel wrote:
> > The 'highmem' variable is only set at boot, so we can make it
> > ro_after_init and prevent it from being corrupted inadvertently, or from
> > ending up in a contended cacheline.
> >
> > Signed-off-by: Ard Biesheuvel <[email protected]>
>
> Acked-by: Mike Rapoport <[email protected]>
>
> > ---
> > mm/memory.c | 2 +-
> > 1 file changed, 1 insertion(+), 1 deletion(-)
> >
> > diff --git a/mm/memory.c b/mm/memory.c
> > index c125c4969913..50e82cb94ccc 100644
> > --- a/mm/memory.c
> > +++ b/mm/memory.c
> > @@ -106,7 +106,7 @@ EXPORT_SYMBOL(mem_map);
> > * highstart_pfn must be the same; there must be no gap between ZONE_NORMAL
> > * and ZONE_HIGHMEM.
> > */
> > -void *high_memory;
> > +void *high_memory __ro_after_init;
> > EXPORT_SYMBOL(high_memory);
> >
> > /*
> > --
> > 2.30.2
> >
> >
>
This turns out to break CONFIG_MEMORY_HOTPLUG on x86, as it updates
max_pfn, max_low_pfn and high_memory after init.
So probably best to make all of those __read_mostly instead.