missing syscalls

missing syscalls

[Sam Ravnborg]

I have run the "missing syscalls" check for the cross compilers I
have available at the moment.

alpha: 38
arm: 8
mips: 3
sparc: 4
sparc64: 0
x86_64: 2
ia64: 10
i386: 0 (this is the reference anyway)

Result is pasted in below.

	Sam

============== alpha ================
  CALL    /home/sam/kernel/kbuild.git/scripts/checksyscalls.sh
<stdin>:128:2: warning: #warning syscall creat not implemented
<stdin>:148:2: warning: #warning syscall time not implemented
<stdin>:176:2: warning: #warning syscall getpid not implemented
<stdin>:192:2: warning: #warning syscall getuid not implemented
<stdin>:204:2: warning: #warning syscall alarm not implemented
<stdin>:212:2: warning: #warning syscall pause not implemented
<stdin>:216:2: warning: #warning syscall utime not implemented
<stdin>:284:2: warning: #warning syscall getgid not implemented
<stdin>:292:2: warning: #warning syscall geteuid not implemented
<stdin>:296:2: warning: #warning syscall getegid not implemented
<stdin>:352:2: warning: #warning syscall getppid not implemented
<stdin>:688:2: warning: #warning syscall fdatasync not implemented
<stdin>:1220:2: warning: #warning syscall kexec_load not implemented
<stdin>:1260:2: warning: #warning syscall migrate_pages not implemented
<stdin>:1264:2: warning: #warning syscall openat not implemented
<stdin>:1268:2: warning: #warning syscall mkdirat not implemented
<stdin>:1272:2: warning: #warning syscall mknodat not implemented
<stdin>:1276:2: warning: #warning syscall fchownat not implemented
<stdin>:1280:2: warning: #warning syscall futimesat not implemented
<stdin>:1288:2: warning: #warning syscall unlinkat not implemented
<stdin>:1292:2: warning: #warning syscall renameat not implemented
<stdin>:1296:2: warning: #warning syscall linkat not implemented
<stdin>:1300:2: warning: #warning syscall symlinkat not implemented
<stdin>:1304:2: warning: #warning syscall readlinkat not implemented
<stdin>:1308:2: warning: #warning syscall fchmodat not implemented
<stdin>:1312:2: warning: #warning syscall faccessat not implemented
<stdin>:1316:2: warning: #warning syscall pselect6 not implemented
<stdin>:1320:2: warning: #warning syscall ppoll not implemented
<stdin>:1324:2: warning: #warning syscall unshare not implemented
<stdin>:1328:2: warning: #warning syscall set_robust_list not implemented
<stdin>:1332:2: warning: #warning syscall get_robust_list not implemented
<stdin>:1336:2: warning: #warning syscall splice not implemented
<stdin>:1340:2: warning: #warning syscall sync_file_range not implemented
<stdin>:1344:2: warning: #warning syscall tee not implemented
<stdin>:1348:2: warning: #warning syscall vmsplice not implemented
<stdin>:1352:2: warning: #warning syscall move_pages not implemented
<stdin>:1356:2: warning: #warning syscall getcpu not implemented
<stdin>:1360:2: warning: #warning syscall epoll_pwait not implemented

============== arm ================
  CALL    /home/sam/kernel/kbuild.git/scripts/checksyscalls.sh
<stdin>:1092:2: warning: #warning syscall fadvise64 not implemented
<stdin>:1176:2: warning: #warning syscall fadvise64_64 not implemented
<stdin>:1220:2: warning: #warning syscall kexec_load not implemented
<stdin>:1260:2: warning: #warning syscall migrate_pages not implemented
<stdin>:1316:2: warning: #warning syscall pselect6 not implemented
<stdin>:1320:2: warning: #warning syscall ppoll not implemented
<stdin>:1340:2: warning: #warning syscall sync_file_range not implemented
<stdin>:1360:2: warning: #warning syscall epoll_pwait not implemented

============== mips ================
  CALL    /home/sam/kernel/kbuild.git/scripts/checksyscalls.sh
<stdin>:424:2: warning: #warning syscall select not implemented
<stdin>:856:2: warning: #warning syscall vfork not implemented
<stdin>:1176:2: warning: #warning syscall fadvise64_64 not implemented

============== sparc ================
  CALL    /home/sam/kernel/kbuild.git/scripts/checksyscalls.sh
<stdin>:752:2: warning: #warning syscall setresuid not implemented
<stdin>:756:2: warning: #warning syscall getresuid not implemented
<stdin>:776:2: warning: #warning syscall setresgid not implemented
<stdin>:780:2: warning: #warning syscall getresgid not implemented

============== sparc64 ================
  CALL    /home/sam/kernel/kbuild.git/scripts/checksyscalls.sh

============== x86_64 ================
  CALL    /home/sam/kernel/kbuild.git/scripts/checksyscalls.sh
<stdin>:1356:2: warning: #warning syscall getcpu not implemented
<stdin>:1360:2: warning: #warning syscall epoll_pwait not implemented

============== ia64 ================
  CALL    /home/sam/kernel/kbuild.git/scripts/checksyscalls.sh
<stdin>:104:2: warning: #warning syscall fork not implemented
<stdin>:148:2: warning: #warning syscall time not implemented
<stdin>:204:2: warning: #warning syscall alarm not implemented
<stdin>:212:2: warning: #warning syscall pause not implemented
<stdin>:216:2: warning: #warning syscall utime not implemented
<stdin>:356:2: warning: #warning syscall getpgrp not implemented
<stdin>:856:2: warning: #warning syscall vfork not implemented
<stdin>:1316:2: warning: #warning syscall pselect6 not implemented
<stdin>:1320:2: warning: #warning syscall ppoll not implemented
<stdin>:1360:2: warning: #warning syscall epoll_pwait not implemented

Re: missing syscalls

[Russell King]

On Sun, Apr 01, 2007 at 06:01:23PM +0200, Sam Ravnborg wrote:
> ============== arm ================
>   CALL    /home/sam/kernel/kbuild.git/scripts/checksyscalls.sh
> <stdin>:1092:2: warning: #warning syscall fadvise64 not implemented

and probably never will be.

> <stdin>:1220:2: warning: #warning syscall kexec_load not implemented

For the ARM kexec-using folk to decide.

> <stdin>:1260:2: warning: #warning syscall migrate_pages not implemented

Does this make sense for ARM?  I've no idea.

> <stdin>:1176:2: warning: #warning syscall fadvise64_64 not implemented
> <stdin>:1340:2: warning: #warning syscall sync_file_range not implemented

These two are replaced by a special ARM version which sane argument
ordering.  My patch for a previous version of the syscall check stuff
shut these warnings up.

> <stdin>:1316:2: warning: #warning syscall pselect6 not implemented
> <stdin>:1320:2: warning: #warning syscall ppoll not implemented
> <stdin>:1360:2: warning: #warning syscall epoll_pwait not implemented

Maybe sometime in the future.

-- 
Russell King
 Linux kernel    2.6 ARM Linux   - http://www.arm.linux.org.uk/
 maintainer of:

Re: missing syscalls

[David Miller]

From: Russell King <rmk@arm.linux.org.uk>
Date: Sun, 1 Apr 2007 20:15:12 +0100

> On Sun, Apr 01, 2007 at 06:01:23PM +0200, Sam Ravnborg wrote:
> > ============== arm ================
> >   CALL    /home/sam/kernel/kbuild.git/scripts/checksyscalls.sh
> > <stdin>:1092:2: warning: #warning syscall fadvise64 not implemented
> 
> and probably never will be.
> 
> > <stdin>:1220:2: warning: #warning syscall kexec_load not implemented
> 
> For the ARM kexec-using folk to decide.
> 
> > <stdin>:1260:2: warning: #warning syscall migrate_pages not implemented
> 
> Does this make sense for ARM?  I've no idea.

You should hook both of these syscalls up even if the config option
that enables them usually is not, or cannot currently be, enabled.

The cond_syscall()'s will make sure they always link properly and
provide a -ENOSYS implementation.

Hooking them up makes it easier to check future missed cases without
us having to add a plethora of ifdefs to the missing syscall checks
for each platform.

Re: missing syscalls

[Sam Ravnborg]

On Sun, Apr 01, 2007 at 08:15:12PM +0100, Russell King wrote:
> On Sun, Apr 01, 2007 at 06:01:23PM +0200, Sam Ravnborg wrote:
> > ============== arm ================
> >   CALL    /home/sam/kernel/kbuild.git/scripts/checksyscalls.sh
> > <stdin>:1092:2: warning: #warning syscall fadvise64 not implemented
> 
> and probably never will be.
> 
> > <stdin>:1220:2: warning: #warning syscall kexec_load not implemented
> 
> For the ARM kexec-using folk to decide.
> 
> > <stdin>:1260:2: warning: #warning syscall migrate_pages not implemented
> 
> Does this make sense for ARM?  I've no idea.
> 
> > <stdin>:1176:2: warning: #warning syscall fadvise64_64 not implemented
> > <stdin>:1340:2: warning: #warning syscall sync_file_range not implemented
> 
> These two are replaced by a special ARM version which sane argument
> ordering.  My patch for a previous version of the syscall check stuff
> shut these warnings up.
Should they be ignored only for arm (and arm26?) or in general?

	Sam

Re: missing syscalls

[Matthew Wilcox]

On Sun, Apr 01, 2007 at 12:23:45PM -0700, David Miller wrote:
> You should hook both of these syscalls up even if the config option
> that enables them usually is not, or cannot currently be, enabled.
> 
> The cond_syscall()'s will make sure they always link properly and
> provide a -ENOSYS implementation.
> 
> Hooking them up makes it easier to check future missed cases without
> us having to add a plethora of ifdefs to the missing syscall checks
> for each platform.

I haven't looked at the missing syscall check implementation, but it seems
like it's poorly designed if we have to add ifdefs for each arch.  Why not
allow arches a mechanism to state which syscalls they intentionally
don't implement?

Re: missing syscalls

[David Miller]

From: Matthew Wilcox <matthew@wil.cx>
Date: Sun, 1 Apr 2007 14:09:47 -0600

> I haven't looked at the missing syscall check implementation, but it seems
> like it's poorly designed if we have to add ifdefs for each arch.  Why not
> allow arches a mechanism to state which syscalls they intentionally
> don't implement?

The whole idea is for the arch's to just find out that a new syscall
exists when someone adds a new one and some tries a build on that
platform the next time.  If the arch's still have to do work then the
whole exercise is pointless.

Re: missing syscalls

[Sam Ravnborg]

On Sun, Apr 01, 2007 at 02:09:47PM -0600, Matthew Wilcox wrote:
> On Sun, Apr 01, 2007 at 12:23:45PM -0700, David Miller wrote:
> > You should hook both of these syscalls up even if the config option
> > that enables them usually is not, or cannot currently be, enabled.
> > 
> > The cond_syscall()'s will make sure they always link properly and
> > provide a -ENOSYS implementation.
> > 
> > Hooking them up makes it easier to check future missed cases without
> > us having to add a plethora of ifdefs to the missing syscall checks
> > for each platform.
> 
> I haven't looked at the missing syscall check implementation, but it seems
> like it's poorly designed if we have to add ifdefs for each arch.  Why not
> allow arches a mechanism to state which syscalls they intentionally
> don't implement?

It's a stupid script so it is easy to fool.
The following patch makes x86_64 shut up:

diff --git a/include/asm-x86_64/unistd.h b/include/asm-x86_64/unistd.h
index c5f596e..6f73918 100644
--- a/include/asm-x86_64/unistd.h
+++ b/include/asm-x86_64/unistd.h
@@ -620,6 +620,9 @@ __SYSCALL(__NR_vmsplice, sys_vmsplice)
 #define __NR_move_pages		279
 __SYSCALL(__NR_move_pages, sys_move_pages)
 
+#define __NR_getcpu             /* 318 - not relevant */
+#define __NR_epoll_pwait        /* 319 - not relevant */
+
 #define __NR_syscall_max __NR_move_pages
 
 #ifndef __NO_STUBS

NOTE - I am NOT proposing this patch. It is purely an example how
to in an arch specific fashion to shut up the syscall check.

Re: missing syscalls

[Matthew Wilcox]

On Sun, Apr 01, 2007 at 01:16:19PM -0700, David Miller wrote:
> From: Matthew Wilcox <matthew@wil.cx>
> Date: Sun, 1 Apr 2007 14:09:47 -0600
> 
> > I haven't looked at the missing syscall check implementation, but it seems
> > like it's poorly designed if we have to add ifdefs for each arch.  Why not
> > allow arches a mechanism to state which syscalls they intentionally
> > don't implement?
> 
> The whole idea is for the arch's to just find out that a new syscall
> exists when someone adds a new one and some tries a build on that
> platform the next time.  If the arch's still have to do work then the
> whole exercise is pointless.

By and large, the missing syscalls are older ones, eg socketcall or
utime, and they tend to be missing for good reason.  I think it's
entirely appropriate that the arch has to do work when a new syscall is
added -- either implement it, or state that it's not supposed to be
implemented.

Re: missing syscalls

[Andi Kleen]

> ============== x86_64 ================
>   CALL    /home/sam/kernel/kbuild.git/scripts/checksyscalls.sh
> <stdin>:1356:2: warning: #warning syscall getcpu not implemented

It's implemented as a vsyscall.

> <stdin>:1360:2: warning: #warning syscall epoll_pwait not implemented

That one is missing.

-Andi

Re: missing syscalls

[Sam Ravnborg]

On Sun, Apr 01, 2007 at 10:54:53PM +0200, Andi Kleen wrote:
> 
> > ============== x86_64 ================
> >   CALL    /home/sam/kernel/kbuild.git/scripts/checksyscalls.sh
> > <stdin>:1356:2: warning: #warning syscall getcpu not implemented
> 
> It's implemented as a vsyscall.

Can I assume you take care of silence the warning?
A dummy entry in unistd will do the trick.

	Sam

Re: missing syscalls

[Russell King]

On Sun, Apr 01, 2007 at 10:19:29PM +0200, Sam Ravnborg wrote:
> On Sun, Apr 01, 2007 at 02:09:47PM -0600, Matthew Wilcox wrote:
> > On Sun, Apr 01, 2007 at 12:23:45PM -0700, David Miller wrote:
> > > You should hook both of these syscalls up even if the config option
> > > that enables them usually is not, or cannot currently be, enabled.
> > > 
> > > The cond_syscall()'s will make sure they always link properly and
> > > provide a -ENOSYS implementation.
> > > 
> > > Hooking them up makes it easier to check future missed cases without
> > > us having to add a plethora of ifdefs to the missing syscall checks
> > > for each platform.
> > 
> > I haven't looked at the missing syscall check implementation, but it seems
> > like it's poorly designed if we have to add ifdefs for each arch.  Why not
> > allow arches a mechanism to state which syscalls they intentionally
> > don't implement?
> 
> It's a stupid script so it is easy to fool.
> The following patch makes x86_64 shut up:
> 
> diff --git a/include/asm-x86_64/unistd.h b/include/asm-x86_64/unistd.h
> index c5f596e..6f73918 100644
> --- a/include/asm-x86_64/unistd.h
> +++ b/include/asm-x86_64/unistd.h
> @@ -620,6 +620,9 @@ __SYSCALL(__NR_vmsplice, sys_vmsplice)
>  #define __NR_move_pages		279
>  __SYSCALL(__NR_move_pages, sys_move_pages)
>  
> +#define __NR_getcpu             /* 318 - not relevant */
> +#define __NR_epoll_pwait        /* 319 - not relevant */
> +

If a syscall is not relevant, define __IGNORE_getcpu etc instead.

-- 
Russell King
 Linux kernel    2.6 ARM Linux   - http://www.arm.linux.org.uk/
 maintainer of:

Re: missing syscalls

[Russell King]

On Sun, Apr 01, 2007 at 09:23:55PM +0200, Sam Ravnborg wrote:
> On Sun, Apr 01, 2007 at 08:15:12PM +0100, Russell King wrote:
> > On Sun, Apr 01, 2007 at 06:01:23PM +0200, Sam Ravnborg wrote:
> > > ============== arm ================
> > >   CALL    /home/sam/kernel/kbuild.git/scripts/checksyscalls.sh
> > > <stdin>:1092:2: warning: #warning syscall fadvise64 not implemented
> > 
> > and probably never will be.
> > 
> > > <stdin>:1220:2: warning: #warning syscall kexec_load not implemented
> > 
> > For the ARM kexec-using folk to decide.

Actually, this is the result of a typo, and should be fixed.

> > 
> > > <stdin>:1260:2: warning: #warning syscall migrate_pages not implemented
> > 
> > Does this make sense for ARM?  I've no idea.
> > 
> > > <stdin>:1176:2: warning: #warning syscall fadvise64_64 not implemented
> > > <stdin>:1340:2: warning: #warning syscall sync_file_range not implemented
> > 
> > These two are replaced by a special ARM version which sane argument
> > ordering.  My patch for a previous version of the syscall check stuff
> > shut these warnings up.
> Should they be ignored only for arm (and arm26?) or in general?

for ARM only, as per my original patch.

+#define __IGNORE_fadvise64_64          /* we provide our own - see above */
+#define __IGNORE_sync_file_range       /* we provide our own - see above */


-- 
Russell King
 Linux kernel    2.6 ARM Linux   - http://www.arm.linux.org.uk/
 maintainer of:

Re: missing syscalls

[Arnd Bergmann]

On Sunday 01 April 2007 22:19:29 Sam Ravnborg wrote:
> +#define __NR_getcpu             /* 318 - not relevant */
> +#define __NR_epoll_pwait        /* 319 - not relevant */

This looks like it is potentially harmful, since it breaks
user space code that does things like

int my_getcpy(void)
{
#ifdef __NR_getcpu
	return syscall(__NR_getcpu);
#endif
	return -1;
}

	Arnd <><

Re: missing syscalls

[Andi Kleen]

On Sunday 01 April 2007 23:10, Sam Ravnborg wrote:
> On Sun, Apr 01, 2007 at 10:54:53PM +0200, Andi Kleen wrote:
> > 
> > > ============== x86_64 ================
> > >   CALL    /home/sam/kernel/kbuild.git/scripts/checksyscalls.sh
> > > <stdin>:1356:2: warning: #warning syscall getcpu not implemented
> > 
> > It's implemented as a vsyscall.
> 
> Can I assume you take care of silence the warning?
> A dummy entry in unistd will do the trick.

As Arnd points out that would break reasonable user space checks.

-Andi

Re: missing syscalls

[Matthew Wilcox]

On Mon, Apr 02, 2007 at 07:58:51AM +0200, Andi Kleen wrote:
> On Sunday 01 April 2007 23:10, Sam Ravnborg wrote:
> > On Sun, Apr 01, 2007 at 10:54:53PM +0200, Andi Kleen wrote:
> > > 
> > > > ============== x86_64 ================
> > > >   CALL    /home/sam/kernel/kbuild.git/scripts/checksyscalls.sh
> > > > <stdin>:1356:2: warning: #warning syscall getcpu not implemented
> > > 
> > > It's implemented as a vsyscall.
> > 
> > Can I assume you take care of silence the warning?
> > A dummy entry in unistd will do the trick.
> 
> As Arnd points out that would break reasonable user space checks.

Actually, this specific case would also be wrong -- you wouldn't want
programs to return -1, you'd want them to call the vsyscall.

Re: missing syscalls

[David Miller]

From: Sam Ravnborg <sam@ravnborg.org>
Date: Sun, 1 Apr 2007 18:01:23 +0200

> ============== sparc ================
>   CALL    /home/sam/kernel/kbuild.git/scripts/checksyscalls.sh
> <stdin>:752:2: warning: #warning syscall setresuid not implemented
> <stdin>:756:2: warning: #warning syscall getresuid not implemented
> <stdin>:776:2: warning: #warning syscall setresgid not implemented
> <stdin>:780:2: warning: #warning syscall getresgid not implemented

I figured out what was going on here.

Sparc 32-bit only has the "setresuid32", "getresuid32" variants,
it never had the plain ones and there is no value to adding those
old versions into the syscall table.