[PATCH] syscall: Take SYSCALL_OFFSET into account

[PATCH] syscall: Take SYSCALL_OFFSET into account

[Markos Chandras]

MIPS and IA64 have their syscall tables starting at non-zero
offsets so take that into account when executing a sycall
otherwise trinity just fails with ENOSYS error messages.

Tested on MIPS 32/LE system.

Signed-off-by: Markos Chandras <markos.chandras@imgtec.com>
---
 syscall.c | 5 +++--
 1 file changed, 3 insertions(+), 2 deletions(-)

diff --git a/syscall.c b/syscall.c
index 968962b..a1ccb6b 100644
--- a/syscall.c
+++ b/syscall.c
@@ -79,6 +79,7 @@ static long syscall32(unsigned int call,
 static unsigned long do_syscall(int childno, int *errno_saved)
 {
 	int nr = shm->syscall[childno].nr;
+	int call = nr + SYSCALL_OFFSET;
 	unsigned long a1, a2, a3, a4, a5, a6;
 	unsigned long ret = 0;
 
@@ -99,9 +100,9 @@ static unsigned long do_syscall(int childno, int *errno_saved)
 	errno = 0;
 
 	if (shm->syscall[childno].do32bit == FALSE)
-		ret = syscall(nr, a1, a2, a3, a4, a5, a6);
+		ret = syscall(call, a1, a2, a3, a4, a5, a6);
 	else
-		ret = syscall32(nr, a1, a2, a3, a4, a5, a6);
+		ret = syscall32(call, a1, a2, a3, a4, a5, a6);
 
 	*errno_saved = errno;
 
-- 
1.9.3

Re: [PATCH] syscall: Take SYSCALL_OFFSET into account

[Dave Jones]

On Wed, May 14, 2014 at 12:35:38PM +0100, Markos Chandras wrote:
 > MIPS and IA64 have their syscall tables starting at non-zero
 > offsets so take that into account when executing a sycall
 > otherwise trinity just fails with ENOSYS error messages.
 > 
 > Tested on MIPS 32/LE system.
 > 
 > Signed-off-by: Markos Chandras <markos.chandras@imgtec.com>
 > ---
 >  syscall.c | 5 +++--
 >  1 file changed, 3 insertions(+), 2 deletions(-)
 > 
 > diff --git a/syscall.c b/syscall.c
 > index 968962b..a1ccb6b 100644
 > --- a/syscall.c
 > +++ b/syscall.c
 > @@ -79,6 +79,7 @@ static long syscall32(unsigned int call,
 >  static unsigned long do_syscall(int childno, int *errno_saved)
 >  {
 >  	int nr = shm->syscall[childno].nr;
 > +	int call = nr + SYSCALL_OFFSET;
 >  	unsigned long a1, a2, a3, a4, a5, a6;
 >  	unsigned long ret = 0;
 >  
 > @@ -99,9 +100,9 @@ static unsigned long do_syscall(int childno, int *errno_saved)
 >  	errno = 0;
 >  
 >  	if (shm->syscall[childno].do32bit == FALSE)
 > -		ret = syscall(nr, a1, a2, a3, a4, a5, a6);
 > +		ret = syscall(call, a1, a2, a3, a4, a5, a6);
 >  	else
 > -		ret = syscall32(nr, a1, a2, a3, a4, a5, a6);
 > +		ret = syscall32(call, a1, a2, a3, a4, a5, a6);
 >  
 >  	*errno_saved = errno;

Oops. This was the intention of the code in mkcall() that calls
do_syscall().

152         /* Some architectures (IA64/MIPS) start their Linux syscalls
153          * At non-zero, and have other ABIs below.
154          */
155         call += SYSCALL_OFFSET;

Looking at that code closer, it seems that the code around line 193
will do the wrong thing on MIPS/IA64 because we've done this addition.

I'm wondering if just removing those lines I just quoted would be
the right thing to do (after applying your patch).

	Dave

Re: [PATCH] syscall: Take SYSCALL_OFFSET into account

[Markos Chandras]

On 05/14/2014 04:22 PM, Dave Jones wrote:
> On Wed, May 14, 2014 at 12:35:38PM +0100, Markos Chandras wrote:
>  > MIPS and IA64 have their syscall tables starting at non-zero
>  > offsets so take that into account when executing a sycall
>  > otherwise trinity just fails with ENOSYS error messages.
>  > 
>  > Tested on MIPS 32/LE system.
>  > 
>  > Signed-off-by: Markos Chandras <markos.chandras@imgtec.com>
>  > ---
>  >  syscall.c | 5 +++--
>  >  1 file changed, 3 insertions(+), 2 deletions(-)
>  > 
>  > diff --git a/syscall.c b/syscall.c
>  > index 968962b..a1ccb6b 100644
>  > --- a/syscall.c
>  > +++ b/syscall.c
>  > @@ -79,6 +79,7 @@ static long syscall32(unsigned int call,
>  >  static unsigned long do_syscall(int childno, int *errno_saved)
>  >  {
>  >  	int nr = shm->syscall[childno].nr;
>  > +	int call = nr + SYSCALL_OFFSET;
>  >  	unsigned long a1, a2, a3, a4, a5, a6;
>  >  	unsigned long ret = 0;
>  >  
>  > @@ -99,9 +100,9 @@ static unsigned long do_syscall(int childno, int *errno_saved)
>  >  	errno = 0;
>  >  
>  >  	if (shm->syscall[childno].do32bit == FALSE)
>  > -		ret = syscall(nr, a1, a2, a3, a4, a5, a6);
>  > +		ret = syscall(call, a1, a2, a3, a4, a5, a6);
>  >  	else
>  > -		ret = syscall32(nr, a1, a2, a3, a4, a5, a6);
>  > +		ret = syscall32(call, a1, a2, a3, a4, a5, a6);
>  >  
>  >  	*errno_saved = errno;
> 
> Oops. This was the intention of the code in mkcall() that calls
> do_syscall().
> 
> 152         /* Some architectures (IA64/MIPS) start their Linux syscalls
> 153          * At non-zero, and have other ABIs below.
> 154          */
> 155         call += SYSCALL_OFFSET;
> 
> Looking at that code closer, it seems that the code around line 193
> will do the wrong thing on MIPS/IA64 because we've done this addition.
> 
> I'm wondering if just removing those lines I just quoted would be
> the right thing to do (after applying your patch).
> 
> 	Dave
> 
Hi Dave,

Removing these four lines may work but it will break the following:

209                 output(1, "%s (%d) returned ENOSYS, marking as
inactive.\n",
210                         entry->name, call);

'call' will have the syscall (shm->syscall[childno].nr) number without
the offset which will simply print the wrong syscall number for mips and
ia64.

I am not that familiar with the code yet to be able to tell whether the
'search_syscall_table' or 'deactivate_syscall' functions need the offset
as well or not.

-- 
markos

Re: [PATCH] syscall: Take SYSCALL_OFFSET into account

[Dave Jones]

On Thu, May 15, 2014 at 09:26:08AM +0100, Markos Chandras wrote:

 > > I'm wondering if just removing those lines I just quoted would be
 > > the right thing to do (after applying your patch).
 > > 
 > 
 > Removing these four lines may work but it will break the following:
 > 
 > 209                 output(1, "%s (%d) returned ENOSYS, marking as
 > inactive.\n",
 > 210                         entry->name, call);
 > 
 > 'call' will have the syscall (shm->syscall[childno].nr) number without
 > the offset which will simply print the wrong syscall number for mips and
 > ia64.

You're right. I just fixed this up in git after removing those other
lines yesterday.

 > I am not that familiar with the code yet to be able to tell whether the
 > 'search_syscall_table' or 'deactivate_syscall' functions need the offset
 > as well or not.

They should be safe, as they use/return the number as index into the tables
we constructed.  But shout if something looks odd, and I'll look into it.

	Dave