Syscall wrappers breaks things.

Syscall wrappers breaks things.

[David Daney]

In commit ee6a093222549ac0c72cfd296c69fa5e7d6daa34 the syscall wrappers 
infrastructure was introduced.

A typical definition is something like this (from Ralf's patch):

SYSCALL_DEFINE6(32_ipc, u32, call, long, first, long, second, long, third,
	unsigned long, ptr, unsigned long, fifth)
{
.
.
.

This creates a symbol sys_32_ipc.  The old name of the symbol is 
sys32_ipc (note the change in the number of '_').

For the mips64 kernel the syscall table for o32 ABI executable contains:

.
.
.
	PTR	sys32_ipc
.
.
.

This is now linked to the sys32_ipc function in kernel/sys_ni.c which 
kindly returns ENOSYS.

The long and short of the problem is that the is now a kernel wide 
mismatch in function names between the names created by SYSCALL_DEFINE*, 
the syscall tables and kernel/sys_ni.c.

I am not sure what the best solution is, but something should probably 
be done to un-break things.

Thanks,
David Daney

Re: Syscall wrappers breaks things.

[Heiko Carstens]

On Fri, 20 Feb 2009 10:48:58 -0800
David Daney <ddaney@caviumnetworks.com> wrote:

> In commit ee6a093222549ac0c72cfd296c69fa5e7d6daa34 the syscall wrappers 
> infrastructure was introduced.
> 
> A typical definition is something like this (from Ralf's patch):
> 
> SYSCALL_DEFINE6(32_ipc, u32, call, long, first, long, second, long, third,
> 	unsigned long, ptr, unsigned long, fifth)
> {
> .
> 
> This creates a symbol sys_32_ipc.  The old name of the symbol is 
> sys32_ipc (note the change in the number of '_').
> 
> For the mips64 kernel the syscall table for o32 ABI executable contains:
> 	PTR	sys32_ipc
> This is now linked to the sys32_ipc function in kernel/sys_ni.c which 
> kindly returns ENOSYS.

Sure, the SYSCALL_DEFINE macros work only for system calls that follow
the sys_[whatever] naming convention. In this case the original system
call should be renamed. E.g. sys_mips_ipc.

> The long and short of the problem is that the is now a kernel wide 
> mismatch in function names between the names created by SYSCALL_DEFINE*, 
> the syscall tables and kernel/sys_ni.c.

Only if the macros are used in a wrong way.

> I am not sure what the best solution is, but something should probably 
> be done to un-break things.

Fix the patch that converts mips to system call wrappers ;)

Ralf's patch isn't in Linus' repository nor is it in linux-next. Or at
least I couldn't find it. So it looks like you're talking about some
private git tree?