[PATCH 1/2] mm, Make __get_user_pages return -EHWPOISON for HWPOISON page

[PATCH 1/2] mm, Make __get_user_pages return -EHWPOISON for HWPOISON page

[Huang Ying]

With this patch, the caller of __get_user_pages can distinguish
HWPOISON page from general FAULT page. This is needed by KVM, where
UCR MCE should be relayed to guest for HWPOISON page, while
instruction emulation and MMIO will be tried for general FAULT page.

Signed-off-by: Huang Ying <ying.huang@intel.com>
---
 include/asm-generic/errno.h |    2 ++
 mm/memory.c                 |    5 +++--
 2 files changed, 5 insertions(+), 2 deletions(-)

--- a/include/asm-generic/errno.h
+++ b/include/asm-generic/errno.h
@@ -108,4 +108,6 @@
 
 #define ERFKILL		132	/* Operation not possible due to RF-kill */
 
+#define EHWPOISON	133	/* Memory page has hardware error */
+
 #endif
--- a/mm/memory.c
+++ b/mm/memory.c
@@ -1432,9 +1432,10 @@ int __get_user_pages(struct task_struct
 				if (ret & VM_FAULT_ERROR) {
 					if (ret & VM_FAULT_OOM)
 						return i ? i : -ENOMEM;
-					if (ret &
-					    (VM_FAULT_HWPOISON|VM_FAULT_SIGBUS))
+					if (ret & VM_FAULT_SIGBUS)
 						return i ? i : -EFAULT;
+					if (ret & VM_FAULT_HWPOISON)
+						return i ? i : -EHWPOISON;
 					BUG();
 				}
 				if (ret & VM_FAULT_MAJOR)

Re: [PATCH 1/2] mm, Make __get_user_pages return -EHWPOISON for HWPOISON page

[Andrew Morton]

On Tue, 27 Apr 2010 15:04:18 +0800
Huang Ying <ying.huang@intel.com> wrote:

> With this patch, the caller of __get_user_pages can distinguish
> HWPOISON page from general FAULT page. This is needed by KVM, where
> UCR MCE should be relayed to guest for HWPOISON page, while
> instruction emulation and MMIO will be tried for general FAULT page.
> 
> Signed-off-by: Huang Ying <ying.huang@intel.com>
> ---
>  include/asm-generic/errno.h |    2 ++
>  mm/memory.c                 |    5 +++--
>  2 files changed, 5 insertions(+), 2 deletions(-)
> 
> --- a/include/asm-generic/errno.h
> +++ b/include/asm-generic/errno.h
> @@ -108,4 +108,6 @@
>  
>  #define ERFKILL		132	/* Operation not possible due to RF-kill */
>  
> +#define EHWPOISON	133	/* Memory page has hardware error */
> +
>  #endif
> --- a/mm/memory.c
> +++ b/mm/memory.c
> @@ -1432,9 +1432,10 @@ int __get_user_pages(struct task_struct
>  				if (ret & VM_FAULT_ERROR) {
>  					if (ret & VM_FAULT_OOM)
>  						return i ? i : -ENOMEM;
> -					if (ret &
> -					    (VM_FAULT_HWPOISON|VM_FAULT_SIGBUS))
> +					if (ret & VM_FAULT_SIGBUS)
>  						return i ? i : -EFAULT;
> +					if (ret & VM_FAULT_HWPOISON)
> +						return i ? i : -EHWPOISON;
>  					BUG();
>  				}
>  				if (ret & VM_FAULT_MAJOR)
> 

This will cause various code paths (eg: fs/direct-io.c) to return
-EHWPOISON to userspace, will it not?

Re: [PATCH 1/2] mm, Make __get_user_pages return -EHWPOISON for HWPOISON page

[Andrew Morton]

On Wed, 28 Apr 2010 09:50:12 +0800 Huang Ying <ying.huang@intel.com> wrote:

> On Wed, 2010-04-28 at 05:47 +0800, Andrew Morton wrote:
> > On Tue, 27 Apr 2010 15:04:18 +0800
> > Huang Ying <ying.huang@intel.com> wrote:
> > 
>
> ...
> 
> > >  				if (ret & VM_FAULT_ERROR) {
> > >  					if (ret & VM_FAULT_OOM)
> > >  						return i ? i : -ENOMEM;
> > > -					if (ret &
> > > -					    (VM_FAULT_HWPOISON|VM_FAULT_SIGBUS))
> > > +					if (ret & VM_FAULT_SIGBUS)
> > >  						return i ? i : -EFAULT;
> > > +					if (ret & VM_FAULT_HWPOISON)
> > > +						return i ? i : -EHWPOISON;
> > >  					BUG();
> > >  				}
> > >  				if (ret & VM_FAULT_MAJOR)
> > > 
> > 
> > This will cause various code paths (eg: fs/direct-io.c) to return
> > -EHWPOISON to userspace, will it not?
> 
> Yes. I think it may be returned to userspace. Do we need to keep the
> userspace errno interface like this?

I don't think we should extend the kernel interface in a non-posix
fashion without good reason and some thought.  manpages and glibc will
need to be updated, etc..

> If it is, we can check whether the virtual page is hwpoisoned via going
> through the page table because the hwpoisoned virtual page has a special
> swap entry in page table.

Spose so.  Or you could perhaps extend __get_user_pages() by adding a
new FOLL_HWPOISON and make it return -EHWPOISON only if FOLL_HWPOISON
was set, and require that the __get_user_pages(FOLL_HWPOISON) caller
rewrite -EHWPOISON to something else.  But I didn't think about that
very hard.

Re: [PATCH 1/2] mm, Make __get_user_pages return -EHWPOISON for HWPOISON page

[Huang Ying]

On Wed, 2010-04-28 at 05:47 +0800, Andrew Morton wrote:
> On Tue, 27 Apr 2010 15:04:18 +0800
> Huang Ying <ying.huang@intel.com> wrote:
> 
> > With this patch, the caller of __get_user_pages can distinguish
> > HWPOISON page from general FAULT page. This is needed by KVM, where
> > UCR MCE should be relayed to guest for HWPOISON page, while
> > instruction emulation and MMIO will be tried for general FAULT page.
> > 
> > Signed-off-by: Huang Ying <ying.huang@intel.com>
> > ---
> >  include/asm-generic/errno.h |    2 ++
> >  mm/memory.c                 |    5 +++--
> >  2 files changed, 5 insertions(+), 2 deletions(-)
> > 
> > --- a/include/asm-generic/errno.h
> > +++ b/include/asm-generic/errno.h
> > @@ -108,4 +108,6 @@
> >  
> >  #define ERFKILL		132	/* Operation not possible due to RF-kill */
> >  
> > +#define EHWPOISON	133	/* Memory page has hardware error */
> > +
> >  #endif
> > --- a/mm/memory.c
> > +++ b/mm/memory.c
> > @@ -1432,9 +1432,10 @@ int __get_user_pages(struct task_struct
> >  				if (ret & VM_FAULT_ERROR) {
> >  					if (ret & VM_FAULT_OOM)
> >  						return i ? i : -ENOMEM;
> > -					if (ret &
> > -					    (VM_FAULT_HWPOISON|VM_FAULT_SIGBUS))
> > +					if (ret & VM_FAULT_SIGBUS)
> >  						return i ? i : -EFAULT;
> > +					if (ret & VM_FAULT_HWPOISON)
> > +						return i ? i : -EHWPOISON;
> >  					BUG();
> >  				}
> >  				if (ret & VM_FAULT_MAJOR)
> > 
> 
> This will cause various code paths (eg: fs/direct-io.c) to return
> -EHWPOISON to userspace, will it not?

Yes. I think it may be returned to userspace. Do we need to keep the
userspace errno interface like this?

If it is, we can check whether the virtual page is hwpoisoned via going
through the page table because the hwpoisoned virtual page has a special
swap entry in page table.

Best Regards,
Huang Ying