[Qemu-devel] [PATCH 02/02] linux-user : add vmsplice system call

[Qemu-devel] [PATCH 02/02] linux-user : add vmsplice system call

[vibi sreenivasan]

linux-user : add vmsplice system call

Signed-off-by: vibi sreenivasan <vibi_sreenivasan@cms.com>

---
 linux-user/syscall.c |   14 ++++++++++++++
 1 files changed, 14 insertions(+), 0 deletions(-)

diff --git a/linux-user/syscall.c b/linux-user/syscall.c
index dc726ad..bb32f84 100644
--- a/linux-user/syscall.c
+++ b/linux-user/syscall.c
@@ -6598,6 +6598,20 @@ abi_long do_syscall(void *cpu_env, int num, abi_long arg1,
         break;
 #endif
 
+#ifdef TARGET_NR_vmsplice
+	case TARGET_NR_vmsplice:
+        {
+            int count = arg3;
+            struct iovec *vec;
+
+            vec = alloca(count * sizeof(struct iovec));
+            if (lock_iovec(VERIFY_READ, vec, arg2, count, 1) < 0)
+                goto efault;
+            ret = get_errno(vmsplice(arg1, vec, count, arg4));
+            unlock_iovec(vec, arg2, count, 0);
+        }
+        break;
+#endif
     default:
     unimplemented:
         gemu_log("qemu: Unsupported syscall: %d\n", num);
-- 
1.6.0

Re: [Qemu-devel] [PATCH 02/02] linux-user : add vmsplice system call

[Martin Mohring]

Hi,

fine that you provide missing syscalls for user mode one after the other.

I have one remark: if the host os kernel is too old to provide the
syscalls, they should be #if'ed out (tee, vmsplice & co).
Or even emulate the syscall - which can be quite hard.

I put together the new user linux patches together with riku, seems the
main qemu staff likes more to make qemu system mode stuff.

Regards, Martin

vibi sreenivasan wrote:
> linux-user : add vmsplice system call
>
> Signed-off-by: vibi sreenivasan <vibi_sreenivasan@cms.com>
>
> ---
>  linux-user/syscall.c |   14 ++++++++++++++
>  1 files changed, 14 insertions(+), 0 deletions(-)
>
> diff --git a/linux-user/syscall.c b/linux-user/syscall.c
> index dc726ad..bb32f84 100644
> --- a/linux-user/syscall.c
> +++ b/linux-user/syscall.c
> @@ -6598,6 +6598,20 @@ abi_long do_syscall(void *cpu_env, int num, abi_long arg1,
>          break;
>  #endif
>  
> +#ifdef TARGET_NR_vmsplice
> +	case TARGET_NR_vmsplice:
> +        {
> +            int count = arg3;
> +            struct iovec *vec;
> +
> +            vec = alloca(count * sizeof(struct iovec));
> +            if (lock_iovec(VERIFY_READ, vec, arg2, count, 1) < 0)
> +                goto efault;
> +            ret = get_errno(vmsplice(arg1, vec, count, arg4));
> +            unlock_iovec(vec, arg2, count, 0);
> +        }
> +        break;
> +#endif
>      default:
>      unimplemented:
>          gemu_log("qemu: Unsupported syscall: %d\n", num);
>   

Re: [Qemu-devel] [PATCH 02/02] linux-user : add vmsplice system call

[vibi sreenivasan]

hi,
	Thanks for your comment.
> fine that you provide missing syscalls for user mode one after the other.
> 
> I have one remark: if the host os kernel is too old to provide the
> syscalls, they should be #if'ed out (tee, vmsplice & co).
if the host kernel is too old , i think it will return -1 & set errno to
ENOSYS.
So isnt that same as putting #ifdef.

> Or even emulate the syscall - which can be quite hard.
Hmm thinking on that. will it be worth?
> 
> I put together the new user linux patches together with riku, seems the
> main qemu staff likes more to make qemu system mode stuff.
> 
Thanks for that.

Thanks & Regards
vibi sreenivasan

> Regards, Martin
> 
> vibi sreenivasan wrote:
> > linux-user : add vmsplice system call
> >
> > Signed-off-by: vibi sreenivasan <vibi_sreenivasan@cms.com>
> >
> > ---
> >  linux-user/syscall.c |   14 ++++++++++++++
> >  1 files changed, 14 insertions(+), 0 deletions(-)
> >
> > diff --git a/linux-user/syscall.c b/linux-user/syscall.c
> > index dc726ad..bb32f84 100644
> > --- a/linux-user/syscall.c
> > +++ b/linux-user/syscall.c
> > @@ -6598,6 +6598,20 @@ abi_long do_syscall(void *cpu_env, int num, abi_long arg1,
> >          break;
> >  #endif
> >  
> > +#ifdef TARGET_NR_vmsplice
> > +	case TARGET_NR_vmsplice:
> > +        {
> > +            int count = arg3;
> > +            struct iovec *vec;
> > +
> > +            vec = alloca(count * sizeof(struct iovec));
> > +            if (lock_iovec(VERIFY_READ, vec, arg2, count, 1) < 0)
> > +                goto efault;
> > +            ret = get_errno(vmsplice(arg1, vec, count, arg4));
> > +            unlock_iovec(vec, arg2, count, 0);
> > +        }
> > +        break;
> > +#endif
> >      default:
> >      unimplemented:
> >          gemu_log("qemu: Unsupported syscall: %d\n", num);
> >   
> 
> 
> 

Re: [Qemu-devel] [PATCH 02/02] linux-user : add vmsplice system call

[Riku Voipio]

On Mon, May 25, 2009 at 11:43:07AM +0530, vibi sreenivasan wrote:
> hi,
> 	Thanks for your comment.
> > fine that you provide missing syscalls for user mode one after the other.
> > 
> > I have one remark: if the host os kernel is too old to provide the
> > syscalls, they should be #if'ed out (tee, vmsplice & co).
> if the host kernel is too old , i think it will return -1 & set errno to
> ENOSYS.
> So isnt that same as putting #ifdef.

Yes, having old kernel is not a problem in this. However, having a
old libc when compiling qemu can be a problem. Think Redhat RHEL 4
or debian etch. vmsplice() is not available, and compiling qemu
would fail.

See how we #ifdef utimensat in linux-user/syscall.c and detect
availability in ./configure

Other than that, your patches looked ok.
 
> > Or even emulate the syscall - which can be quite hard.

> Hmm thinking on that. will it be worth?

This could be in interest for people who want to run qemu linux-user
on non-linux systems (such a *bsd). But that doesn't work yet,
and even when people get it working, new syscalls such as tee/splice
are not the most important ones to provide emulation for. Very
few applications use these syscalls yet.