[PATCH v05 54/72] include/uapi/xen/privcmd.h: fix compilation in userspace

[PATCH v05 54/72] include/uapi/xen/privcmd.h: fix compilation in userspace

[Mikko Rapeli]

xen/interface/xen.h is not exported from kernel headers so remove the
dependency and provide needed defines for domid_t and xen_pfn_t if they
are not already defined by some other e.g. Xen specific headers.

Suggested by Andrew Cooper <andrew.cooper3@citrix.com> on lkml message
<5569F9C9.8000607@citrix.com>.

The ifdef for ARM is ugly but did not find better solutions for it.

Fixes userspace compilation error:

xen/privcmd.h:38:31: fatal error: xen/interface/xen.h: No such file or directory

Signed-off-by: Mikko Rapeli <mikko.rapeli@iki.fi>
Cc: David Vrabel <david.vrabel@citrix.com>
---
 arch/arm/include/asm/xen/interface.h |  2 +-
 include/uapi/xen/privcmd.h           | 12 +++++++++++-
 2 files changed, 12 insertions(+), 2 deletions(-)

diff --git a/arch/arm/include/asm/xen/interface.h b/arch/arm/include/asm/xen/interface.h
index 75d5968..6898ee1 100644
--- a/arch/arm/include/asm/xen/interface.h
+++ b/arch/arm/include/asm/xen/interface.h
@@ -38,7 +38,7 @@
  * fine since it simply wouldn't be able to create any sure pfns in
  * the first place.
  */
-typedef uint64_t xen_pfn_t;
+typedef __u64 xen_pfn_t;
 #define PRI_xen_pfn "llx"
 typedef uint64_t xen_ulong_t;
 #define PRI_xen_ulong "llx"
diff --git a/include/uapi/xen/privcmd.h b/include/uapi/xen/privcmd.h
index 7ddeeda..16c11f9 100644
--- a/include/uapi/xen/privcmd.h
+++ b/include/uapi/xen/privcmd.h
@@ -35,7 +35,17 @@
 
 #include <linux/types.h>
 #include <linux/compiler.h>
-#include <xen/interface/xen.h>
+
+/* Defined by include/xen/interface/xen.h, but it is not part of Linux uapi */
+#ifndef __XEN_PUBLIC_XEN_H__
+typedef __u16 domid_t;
+
+#if (defined __ARMEL__ || defined __ARMEB__)
+typedef __u64 xen_pfn_t;
+#else
+typedef unsigned long xen_pfn_t;
+#endif /* (defined __ARMEL__ || defined __ARMEB__) */
+#endif /* __XEN_PUBLIC_XEN_H__ */
 
 struct privcmd_hypercall {
 	__u64 op;
-- 
2.8.1

[PATCH v05 60/72] arch/arm/include/uapi/asm/signal.h: use __kernel_size_t instead of size_t

[Mikko Rapeli]

Fixes uapi header compilation error from userspace on ARCH=arm:

asm/signal.h:112:2: error: unknown type name ?size_t?
  size_t ss_size;

Signed-off-by: Mikko Rapeli <mikko.rapeli@iki.fi>
---
 arch/arm/include/uapi/asm/signal.h | 2 +-
 1 file changed, 1 insertion(+), 1 deletion(-)

diff --git a/arch/arm/include/uapi/asm/signal.h b/arch/arm/include/uapi/asm/signal.h
index 33073bd..859f2de 100644
--- a/arch/arm/include/uapi/asm/signal.h
+++ b/arch/arm/include/uapi/asm/signal.h
@@ -113,7 +113,7 @@ struct sigaction {
 typedef struct sigaltstack {
 	void __user *ss_sp;
 	int ss_flags;
-	size_t ss_size;
+	__kernel_size_t ss_size;
 } stack_t;
 
 
-- 
2.8.1

[PATCH v05 54/72] include/uapi/xen/privcmd.h: fix compilation in userspace

[Russell King - ARM Linux]

On Mon, Aug 22, 2016 at 08:33:11PM +0200, Mikko Rapeli wrote:
> xen/interface/xen.h is not exported from kernel headers so remove the
> dependency and provide needed defines for domid_t and xen_pfn_t if they
> are not already defined by some other e.g. Xen specific headers.

I'm confused.  How did we end up with a 64-bit PFN number on ARM?  It's
insane - especially as the kernel uses "unsigned long" almost everywhere
for PFNs - we can't have physical addresses more than 44 bits (32 bit
pfn + 4k page size).

> Suggested by Andrew Cooper <andrew.cooper3@citrix.com> on lkml message
> <5569F9C9.8000607@citrix.com>.
> 
> The ifdef for ARM is ugly but did not find better solutions for it.

#ifdef __arm__

maybe?  Even if not, the unsightly parens are not necessary.

-- 
RMK's Patch system: http://www.armlinux.org.uk/developer/patches/
FTTC broadband for 0.8mile line: currently at 9.6Mbps down 400kbps up
according to speedtest.net.

[PATCH v05 54/72] include/uapi/xen/privcmd.h: fix compilation in userspace

[Stefano Stabellini]

On Tue, 23 Aug 2016, Russell King - ARM Linux wrote:
> On Mon, Aug 22, 2016 at 08:33:11PM +0200, Mikko Rapeli wrote:
> > xen/interface/xen.h is not exported from kernel headers so remove the
> > dependency and provide needed defines for domid_t and xen_pfn_t if they
> > are not already defined by some other e.g. Xen specific headers.
> 
> I'm confused.  How did we end up with a 64-bit PFN number on ARM?  It's
> insane - especially as the kernel uses "unsigned long" almost everywhere
> for PFNs - we can't have physical addresses more than 44 bits (32 bit
> pfn + 4k page size).

That's because xen_pfn_t is the type used to store pfns in structures
passed to Xen via hypercalls. The Xen hypercall ABI is shared between
ARM and ARM64. On x86_32 and x86_64 we have different sizes for pfn
types in the hypercall ABI and it caused quite a bit of trouble in the
past as it is possible to run 32bit domains on a 64bit hypervisor.
Having a common type for pfns on ARM and ARM64 helped a lot in that
respect.


> > Suggested by Andrew Cooper <andrew.cooper3@citrix.com> on lkml message
> > <5569F9C9.8000607@citrix.com>.
> > 
> > The ifdef for ARM is ugly but did not find better solutions for it.
> 
> #ifdef __arm__
> 
> maybe?  Even if not, the unsightly parens are not necessary.

Yes, I think it should be:

#if defined(__arm__) || defined(__aarch64__)


> diff --git a/arch/arm/include/asm/xen/interface.h b/arch/arm/include/asm/xen/interface.h
> index 75d5968..6898ee1 100644
> --- a/arch/arm/include/asm/xen/interface.h
> +++ b/arch/arm/include/asm/xen/interface.h
> @@ -38,7 +38,7 @@
>   * fine since it simply wouldn't be able to create any sure pfns in
>   * the first place.
>   */
> -typedef uint64_t xen_pfn_t;
> +typedef __u64 xen_pfn_t;
>  #define PRI_xen_pfn "llx"
>  typedef uint64_t xen_ulong_t;
>  #define PRI_xen_ulong "llx"

Why this change?

[PATCH v05 54/72] include/uapi/xen/privcmd.h: fix compilation in userspace

[Mikko Rapeli]

On Tue, Aug 23, 2016 at 11:13:52AM -0700, Stefano Stabellini wrote:
> On Tue, 23 Aug 2016, Russell King - ARM Linux wrote:
> > On Mon, Aug 22, 2016 at 08:33:11PM +0200, Mikko Rapeli wrote:
> > > xen/interface/xen.h is not exported from kernel headers so remove the
> > > dependency and provide needed defines for domid_t and xen_pfn_t if they
> > > are not already defined by some other e.g. Xen specific headers.
> > 
> > I'm confused.  How did we end up with a 64-bit PFN number on ARM?  It's
> > insane - especially as the kernel uses "unsigned long" almost everywhere
> > for PFNs - we can't have physical addresses more than 44 bits (32 bit
> > pfn + 4k page size).
> 
> That's because xen_pfn_t is the type used to store pfns in structures
> passed to Xen via hypercalls. The Xen hypercall ABI is shared between
> ARM and ARM64. On x86_32 and x86_64 we have different sizes for pfn
> types in the hypercall ABI and it caused quite a bit of trouble in the
> past as it is possible to run 32bit domains on a 64bit hypervisor.
> Having a common type for pfns on ARM and ARM64 helped a lot in that
> respect.
> 
> 
> > > Suggested by Andrew Cooper <andrew.cooper3@citrix.com> on lkml message
> > > <5569F9C9.8000607@citrix.com>.
> > > 
> > > The ifdef for ARM is ugly but did not find better solutions for it.
> > 
> > #ifdef __arm__
> > 
> > maybe?  Even if not, the unsightly parens are not necessary.
> 
> Yes, I think it should be:
> 
> #if defined(__arm__) || defined(__aarch64__)

Thanks, I will send a new version with this change.
 
> > diff --git a/arch/arm/include/asm/xen/interface.h b/arch/arm/include/asm/xen/interface.h
> > index 75d5968..6898ee1 100644
> > --- a/arch/arm/include/asm/xen/interface.h
> > +++ b/arch/arm/include/asm/xen/interface.h
> > @@ -38,7 +38,7 @@
> >   * fine since it simply wouldn't be able to create any sure pfns in
> >   * the first place.
> >   */
> > -typedef uint64_t xen_pfn_t;
> > +typedef __u64 xen_pfn_t;
> >  #define PRI_xen_pfn "llx"
> >  typedef uint64_t xen_ulong_t;
> >  #define PRI_xen_ulong "llx"
> 
> Why this change?

I will double check but I think the other fix exposed then the common
"<stdint.h> definitions not available in userspace <linux/types.h>" and this
was needed to please the compiler. If so, I'll add this to commit message.

If you would prefer to include libc's <stdint.h> in userspace, well, join the
club. drm and fuse maintainers want the same but my patches with those fixes got
rejected in the past. Example: https://lkml.org/lkml/2015/6/1/160

-Mikko

[PATCH v05 54/72] include/uapi/xen/privcmd.h: fix compilation in userspace

[Stefano Stabellini]

On Tue, 23 Aug 2016, Mikko Rapeli wrote:
> On Tue, Aug 23, 2016 at 11:13:52AM -0700, Stefano Stabellini wrote:
> > On Tue, 23 Aug 2016, Russell King - ARM Linux wrote:
> > > On Mon, Aug 22, 2016 at 08:33:11PM +0200, Mikko Rapeli wrote:
> > > > xen/interface/xen.h is not exported from kernel headers so remove the
> > > > dependency and provide needed defines for domid_t and xen_pfn_t if they
> > > > are not already defined by some other e.g. Xen specific headers.
> > > 
> > > I'm confused.  How did we end up with a 64-bit PFN number on ARM?  It's
> > > insane - especially as the kernel uses "unsigned long" almost everywhere
> > > for PFNs - we can't have physical addresses more than 44 bits (32 bit
> > > pfn + 4k page size).
> > 
> > That's because xen_pfn_t is the type used to store pfns in structures
> > passed to Xen via hypercalls. The Xen hypercall ABI is shared between
> > ARM and ARM64. On x86_32 and x86_64 we have different sizes for pfn
> > types in the hypercall ABI and it caused quite a bit of trouble in the
> > past as it is possible to run 32bit domains on a 64bit hypervisor.
> > Having a common type for pfns on ARM and ARM64 helped a lot in that
> > respect.
> > 
> > 
> > > > Suggested by Andrew Cooper <andrew.cooper3@citrix.com> on lkml message
> > > > <5569F9C9.8000607@citrix.com>.
> > > > 
> > > > The ifdef for ARM is ugly but did not find better solutions for it.
> > > 
> > > #ifdef __arm__
> > > 
> > > maybe?  Even if not, the unsightly parens are not necessary.
> > 
> > Yes, I think it should be:
> > 
> > #if defined(__arm__) || defined(__aarch64__)
> 
> Thanks, I will send a new version with this change.
>  
> > > diff --git a/arch/arm/include/asm/xen/interface.h b/arch/arm/include/asm/xen/interface.h
> > > index 75d5968..6898ee1 100644
> > > --- a/arch/arm/include/asm/xen/interface.h
> > > +++ b/arch/arm/include/asm/xen/interface.h
> > > @@ -38,7 +38,7 @@
> > >   * fine since it simply wouldn't be able to create any sure pfns in
> > >   * the first place.
> > >   */
> > > -typedef uint64_t xen_pfn_t;
> > > +typedef __u64 xen_pfn_t;
> > >  #define PRI_xen_pfn "llx"
> > >  typedef uint64_t xen_ulong_t;
> > >  #define PRI_xen_ulong "llx"
> > 
> > Why this change?
> 
> I will double check but I think the other fix exposed then the common
> "<stdint.h> definitions not available in userspace <linux/types.h>" and this
> was needed to please the compiler. If so, I'll add this to commit message.

That would be strange, because I don't think
arch/arm/include/asm/xen/interface.h is exposed to userspace. If it
was, we would need to replace the other definitions there too.

[PATCH v05 60/72] arch/arm/include/uapi/asm/signal.h: use __kernel_size_t instead of size_t

[Arnd Bergmann]

On Monday, August 22, 2016 8:33:17 PM CEST Mikko Rapeli wrote:
> diff --git a/arch/arm/include/uapi/asm/signal.h b/arch/arm/include/uapi/asm/signal.h
> index 33073bd..859f2de 100644
> --- a/arch/arm/include/uapi/asm/signal.h
> +++ b/arch/arm/include/uapi/asm/signal.h
> @@ -113,7 +113,7 @@ struct sigaction {
>  typedef struct sigaltstack {
>         void __user *ss_sp;
>         int ss_flags;
> -       size_t ss_size;
> +       __kernel_size_t ss_size;
>  } stack_t;

I was going to reply with an Ack, but on further consideration,
I'm not sure if we can't do this in general: size_t may be either
'unsigned int' or 'unsigned long' (depending on the architecture
and toolchain), and if kernel and glibc disagree on this, we
have a problem with any user space code that expects sigaltstack->ss_size
to be the same type as size_t (as mandated by the man page).

I wonder if there is another way to address this.

	Arnd

[PATCH v05 60/72] arch/arm/include/uapi/asm/signal.h: use __kernel_size_t instead of size_t

[Mikko Rapeli]

On Wed, Aug 24, 2016 at 05:02:56PM +0200, Arnd Bergmann wrote:
> On Monday, August 22, 2016 8:33:17 PM CEST Mikko Rapeli wrote:
> > diff --git a/arch/arm/include/uapi/asm/signal.h b/arch/arm/include/uapi/asm/signal.h
> > index 33073bd..859f2de 100644
> > --- a/arch/arm/include/uapi/asm/signal.h
> > +++ b/arch/arm/include/uapi/asm/signal.h
> > @@ -113,7 +113,7 @@ struct sigaction {
> >  typedef struct sigaltstack {
> >         void __user *ss_sp;
> >         int ss_flags;
> > -       size_t ss_size;
> > +       __kernel_size_t ss_size;
> >  } stack_t;
> 
> I was going to reply with an Ack, but on further consideration,
> I'm not sure if we can't do this in general: size_t may be either
> 'unsigned int' or 'unsigned long' (depending on the architecture
> and toolchain), and if kernel and glibc disagree on this, we
> have a problem with any user space code that expects sigaltstack->ss_size
> to be the same type as size_t (as mandated by the man page).
> 
> I wonder if there is another way to address this.

I presume that kernel headers need to follow libc in this case and include
<stddef.h>?

-Mikko