Virtualization - Jul 2015 - [PATCH RESEND] virtio: Fix typecast of pointer in vring_init()

On Sun, Jul 05, 2015 at 12:58:53PM +0200, Michael S. Tsirkin
wrote:
> On Thu, Jul 02, 2015 at 09:21:22AM +0200, Thomas Huth wrote:
> > The virtio_ring.h header is used in userspace programs (ie. QEMU),
> > too. Here we can not assume that sizeof(pointer) is the same as
> > sizeof(long), e.g. when compiling for Windows, so the typecast in
> > vring_init() should be done with (uintptr_t) instead of (unsigned
long).
> > 
> > Signed-off-by: Thomas Huth <thuth at redhat.com>
> 
> This seems to break some userspace too:
> 
>   INSTALL usr/include/linux/ (413 files)
>   CHECK   usr/include/linux/ (413 files)
>   HOSTCC  Documentation/accounting/getdelays
>   HOSTCC  Documentation/connector/ucon
>   HOSTCC  Documentation/mic/mpssd/mpssd.o
> In file included from Documentation/mic/mpssd/mpssd.c:36:0:
> ./usr/include/linux/virtio_ring.h: In function ?vring_init?:
> ./usr/include/linux/virtio_ring.h:146:24: error: ?uintptr_t? undeclared
> (first use in this function)
>   vr->used = (void *)(((uintptr_t)&vr->avail->ring[num] +
> sizeof(__virtio16)
>                         ^
> ./usr/include/linux/virtio_ring.h:146:24: note: each undeclared
> identifier is reported only once for each function it appears in
> scripts/Makefile.host:108: recipe for target
> 'Documentation/mic/mpssd/mpssd.o' failed
> make[3]: *** [Documentation/mic/mpssd/mpssd.o] Error 1
> scripts/Makefile.build:403: recipe for target
'Documentation/mic/mpssd'
> failed
> make[2]: *** [Documentation/mic/mpssd] Error 2
> scripts/Makefile.build:403: recipe for target 'Documentation/mic'
failed
> make[1]: *** [Documentation/mic] Error 2
> 
> 
> E.g. fuse.h has this code:
> #ifdef __KERNEL__
> #include <linux/types.h>
> #else
> #include <stdint.h>
> #endif
> 
> Maybe we need something similar.
The following seems to help.  Does it help the windows build?

---
 include/uapi/linux/virtio_ring.h | 3 +++
 1 file changed, 3 insertions(+)

diff --git a/include/uapi/linux/virtio_ring.h b/include/uapi/linux/virtio_ring.h
index 8682551..c072959 100644
--- a/include/uapi/linux/virtio_ring.h
+++ b/include/uapi/linux/virtio_ring.h
@@ -31,6 +31,9 @@
  * SUCH DAMAGE.
  *
  * Copyright Rusty Russell IBM Corporation 2007. */
+#ifndef __KERNEL__
+#include <stdint.h>
+#endif
 #include <linux/types.h>
 #include <linux/virtio_types.h>
 
-- 
MST

On Sun, 5 Jul 2015 14:59:54 +0200
"Michael S. Tsirkin" <mst at redhat.com> wrote:
> On Sun, Jul 05, 2015 at 12:58:53PM +0200, Michael S. Tsirkin wrote:
> > On Thu, Jul 02, 2015 at 09:21:22AM +0200, Thomas Huth wrote:
> > > The virtio_ring.h header is used in userspace programs (ie.
QEMU),
> > > too. Here we can not assume that sizeof(pointer) is the same as
> > > sizeof(long), e.g. when compiling for Windows, so the typecast in
> > > vring_init() should be done with (uintptr_t) instead of (unsigned
long).
> > > 
> > > Signed-off-by: Thomas Huth <thuth at redhat.com>
> > 
> > This seems to break some userspace too:
> > 
> >   INSTALL usr/include/linux/ (413 files)
> >   CHECK   usr/include/linux/ (413 files)
> >   HOSTCC  Documentation/accounting/getdelays
> >   HOSTCC  Documentation/connector/ucon
> >   HOSTCC  Documentation/mic/mpssd/mpssd.o
> > In file included from Documentation/mic/mpssd/mpssd.c:36:0:
> > ./usr/include/linux/virtio_ring.h: In function ?vring_init?:
> > ./usr/include/linux/virtio_ring.h:146:24: error: ?uintptr_t?
undeclared
> > (first use in this function)
> >   vr->used = (void *)(((uintptr_t)&vr->avail->ring[num] +
> > sizeof(__virtio16)
> >                         ^
> > ./usr/include/linux/virtio_ring.h:146:24: note: each undeclared
> > identifier is reported only once for each function it appears in
> > scripts/Makefile.host:108: recipe for target
D'oh, I only built the kernel for powerpc, that's why I did not notice
this breakage in the MIC code. Sorry!
> > E.g. fuse.h has this code:
> > #ifdef __KERNEL__
> > #include <linux/types.h>
> > #else
> > #include <stdint.h>
> > #endif
> > 
> > Maybe we need something similar.
> 
> The following seems to help.  Does it help the windows build?
...
> diff --git a/include/uapi/linux/virtio_ring.h
b/include/uapi/linux/virtio_ring.h
> index 8682551..c072959 100644
> --- a/include/uapi/linux/virtio_ring.h
> +++ b/include/uapi/linux/virtio_ring.h
> @@ -31,6 +31,9 @@
>   * SUCH DAMAGE.
>   *
>   * Copyright Rusty Russell IBM Corporation 2007. */
> +#ifndef __KERNEL__
> +#include <stdint.h>
> +#endif
>  #include <linux/types.h>
>  #include <linux/virtio_types.h>
Since the update-linux-headers.sh script from QEMU replaces the
"#include <linux/types.h>" in virtio_ring.h to include a file
that again
includes stdint.h already, your above patch should not do any
difference for compiling QEMU for Windows, I think. It's really just
about replacing the typecast there.

But you're right of course for other user space applications which
include this header by other means. So I'm fine if you change my patch
with your hunk above (or add it as an additional patch). Or shall I
rather resend my patch with your above hunk included?

 Thomas

On Mon, Jul 06, 2015 at 11:24:42AM +0200, Thomas Huth
wrote:
> On Sun, 5 Jul 2015 14:59:54 +0200
> "Michael S. Tsirkin" <mst at redhat.com> wrote:
> 
> > On Sun, Jul 05, 2015 at 12:58:53PM +0200, Michael S. Tsirkin wrote:
> > > On Thu, Jul 02, 2015 at 09:21:22AM +0200, Thomas Huth wrote:
> > > > The virtio_ring.h header is used in userspace programs (ie.
QEMU),
> > > > too. Here we can not assume that sizeof(pointer) is the same
as
> > > > sizeof(long), e.g. when compiling for Windows, so the
typecast in
> > > > vring_init() should be done with (uintptr_t) instead of
(unsigned long).
> > > > 
> > > > Signed-off-by: Thomas Huth <thuth at redhat.com>
> > > 
> > > This seems to break some userspace too:
> > > 
> > >   INSTALL usr/include/linux/ (413 files)
> > >   CHECK   usr/include/linux/ (413 files)
> > >   HOSTCC  Documentation/accounting/getdelays
> > >   HOSTCC  Documentation/connector/ucon
> > >   HOSTCC  Documentation/mic/mpssd/mpssd.o
> > > In file included from Documentation/mic/mpssd/mpssd.c:36:0:
> > > ./usr/include/linux/virtio_ring.h: In function ?vring_init?:
> > > ./usr/include/linux/virtio_ring.h:146:24: error: ?uintptr_t?
undeclared
> > > (first use in this function)
> > >   vr->used = (void
*)(((uintptr_t)&vr->avail->ring[num] +
> > > sizeof(__virtio16)
> > >                         ^
> > > ./usr/include/linux/virtio_ring.h:146:24: note: each undeclared
> > > identifier is reported only once for each function it appears in
> > > scripts/Makefile.host:108: recipe for target
> 
> D'oh, I only built the kernel for powerpc, that's why I did not
notice
> this breakage in the MIC code. Sorry!
> 
> > > E.g. fuse.h has this code:
> > > #ifdef __KERNEL__
> > > #include <linux/types.h>
> > > #else
> > > #include <stdint.h>
> > > #endif
> > > 
> > > Maybe we need something similar.
> > 
> > The following seems to help.  Does it help the windows build?
> ...
> > diff --git a/include/uapi/linux/virtio_ring.h
b/include/uapi/linux/virtio_ring.h
> > index 8682551..c072959 100644
> > --- a/include/uapi/linux/virtio_ring.h
> > +++ b/include/uapi/linux/virtio_ring.h
> > @@ -31,6 +31,9 @@
> >   * SUCH DAMAGE.
> >   *
> >   * Copyright Rusty Russell IBM Corporation 2007. */
> > +#ifndef __KERNEL__
> > +#include <stdint.h>
> > +#endif
> >  #include <linux/types.h>
> >  #include <linux/virtio_types.h>
> 
> Since the update-linux-headers.sh script from QEMU replaces the
> "#include <linux/types.h>" in virtio_ring.h to include a
file that again
> includes stdint.h already, your above patch should not do any
> difference for compiling QEMU for Windows, I think. It's really just
> about replacing the typecast there.
> 
> But you're right of course for other user space applications which
> include this header by other means. So I'm fine if you change my patch
> with your hunk above (or add it as an additional patch). Or shall I
> rather resend my patch with your above hunk included?
> 
>  Thomas
You don't have to, but could you confirm that stdint has uintptr_t on
mingw?

-- 
MST