[uPATCH] 2.4.17-preX compile fix for RedHat gcc 3.1-0.10

[uPATCH] 2.4.17-preX compile fix for RedHat gcc 3.1-0.10

[Wayne Whitney]

Hello,

I recently upgraded to "gcc version 3.1 20011127 (Red Hat Linux Rawhide
3.1-0.10)".  It compiles the recent 2.4.17-preX kernels OK, with one small
hitch:  rpciod_tcp_dispatcher in module net/sunrpc/sunrpc.o is an
unresolved symbol.  The patch below (or an equivalent) is required for
net/sunrpc/sched.c to compile properly with this version of gcc.

The situation is that net/sunrpc/sched.c (the only caller of
rpciod_tcp_dispatcher) includes linux/sunrpc/clnt.h, which includes
linux/sunrpc/xprt.h.  linux/sunrpc/clnt.h defines rpciod_tcp_dispatcher as
extern void, while linux/sunrpc/xprt.h defines it as static inline.  
Apparently this latter definition is the intended one, and while earlier
versions of gcc chose it, the above version of gcc choses the first
definition.

Since I'm no C expert, I don't know what the correct behavior for gcc is.
So maybe this is a bug in the kernel, or maybe it is a bug in the above
version of gcc.

Cheers,
Wayne

--- linux-2.4.17-pre6/include/linux/sunrpc/clnt.h	Mon Dec 10 20:36:26 2001
+++ linux-2.4.17-pre7/include/linux/sunrpc/clnt.h	Mon Dec 10 20:50:39 2001
@@ -136,7 +136,6 @@
 	xprt_set_timeout(&clnt->cl_timeout, retr, incr);
 }
 
-extern void rpciod_tcp_dispatcher(void);
 extern void rpciod_wake_up(void);
 
 /*

2.4.17-pre8 OOPS with RedHat gcc 3.1-0.10

[Wayne Whitney]

On Mon, 10 Dec 2001, Wayne Whitney wrote:

> I recently upgraded to "gcc version 3.1 20011127 (Red Hat Linux
> Rawhide 3.1-0.10)".  It compiles the recent 2.4.17-preX kernels

Well, it compiles them, but I get an oops on booting 2.4.17-pre8, for
example, almost immediately after init is launched.  "gcc version 2.96
20000731 (Red Hat Linux 7.2 2.96-101.9)" works fine.

The output of ksymoops is below, there were two oopses actually.  Since
the oopses occurred just after init started, and I don't have an initrd, I
believe there were no modules loaded.  I captured the oopses via the line
printer console and had to type them back in.  Too bad there is no floppy
disk console.  I did double check my typing, but the standard disclaimers
apply.

Cheers,
Wayne


ksymoops 2.4.3 on i686 2.4.17-pre8.  Options used
     -v /usr/local/src/linux-2.4.17-pre8-gcc-3.1/vmlinux (specified)
     -K (specified)
     -L (specified)
     -O (specified)
     -m /boot/System.map-2.4.17-pre8-gcc-3.1 (specified)

Unable to handle kernel NULL pointer dereference at virtual address 0000002f printing eip:
c0106e56
*pde = 00000000
Oops: 0000
CPU: 0
EIP: 0010:[<c0106e56>]
Using defaults from ksymoops -t elf32-i386 -a i386
EFLAGS: 00010292
eax: cfda5fc4 ebx: cfda4000 ecx: cfc7d2e8 edx: 00000001
esi: 00000080 edi: bffff860 ebp: 00000003 esp: cfda5f1c
ds: 0018 es: 0018 ss: 0018
Process rc.sysinit (pid: 10, stackpage=cfda50000)
Stack: 00000000 c0283100 cfc7d2e8 cfda4000 00000296 cfc7d2e8 c0142e54 00000000
       cfda4000 00000000 00000000 00000000 cfda4000 cff28a40 cff28a40 fffffe00
       cfc7d280 cfda4000 00000080 c0142fb4 cfc7d280 00000014 080708b0 cfc7d2e8
Call Trace: [<c0142e54>] [<c0142fb4>] [<c0139b87>] [<c0107e24>]
Code: 8b 45 2c 83 e0 03 83 f8 03 74 0d 81 c4 94 00 00 00 89 d0 5b

>>EIP; c0106e56 <do_signal+16/2d0>   <=====
Trace; c0142e54 <pipe_wait+94/c0>
Trace; c0142fb4 <pipe_read+134/220>
Trace; c0139b86 <sys_read+86/110>
Trace; c0107e24 <io_check_error+34/50>
Code;  c0106e56 <do_signal+16/2d0>
00000000 <_EIP>:
Code;  c0106e56 <do_signal+16/2d0>   <=====
   0:   8b 45 2c                  mov    0x2c(%ebp),%eax   <=====
Code;  c0106e58 <do_signal+18/2d0>
   3:   83 e0 03                  and    $0x3,%eax
Code;  c0106e5c <do_signal+1c/2d0>
   6:   83 f8 03                  cmp    $0x3,%eax
Code;  c0106e5e <do_signal+1e/2d0>
   9:   74 0d                     je     18 <_EIP+0x18> c0106e6e <do_signal+2e/2d0>
Code;  c0106e60 <do_signal+20/2d0>
   b:   81 c4 94 00 00 00         add    $0x94,%esp
Code;  c0106e66 <do_signal+26/2d0>
  11:   89 d0                     mov    %edx,%eax
Code;  c0106e68 <do_signal+28/2d0>
  13:   5b                        pop    %ebx

Unable to handle kernel NULL pointer dereference at virtual address 00000037 printing eip:
c0106e56
*pde = 00000000
Oops: 0000
CPU: 0
EIP: 0010:[<c0106e56>]
EFLAGS: 00010292
eax: c13cdfc4 ebx: c13cc000 ecx: 00000000 edx: 00000001
esi: 00000000 edi: bffff878 ebP; 0000000b esp: c13cdf1c
ds: 0018 es: 0018 ss: 0018
Process init (pid: 1, stackpage=c13cd000)
Stack: fffffdfe c014905c cff03140 00000004 c13cdf8c 000001da 00000001 00000004
       00000060 00002107 00000000 00000000 0000002a 00001180 00000001 00000000
       00000000 00000000 cff03148 00000000 00000000 cff03144 00001000 00000000
Call Trace: [<c014905c>] [<c0107324>]
Code: 8b 45 2c 83 e0 03 83 f8 03 74 0d 81 c4 94 00 00 00 89 d0 5b

>>EIP; c0106e56 <do_signal+16/2d0>   <=====
Trace; c014905c <sys_select+1dc/580>
Trace; c0107324 <signal_return+14/18>
Code;  c0106e56 <do_signal+16/2d0>
00000000 <_EIP>:
Code;  c0106e56 <do_signal+16/2d0>   <=====
   0:   8b 45 2c                  mov    0x2c(%ebp),%eax   <=====
Code;  c0106e58 <do_signal+18/2d0>
   3:   83 e0 03                  and    $0x3,%eax
Code;  c0106e5c <do_signal+1c/2d0>
   6:   83 f8 03                  cmp    $0x3,%eax
Code;  c0106e5e <do_signal+1e/2d0>
   9:   74 0d                     je     18 <_EIP+0x18> c0106e6e <do_signal+2e/2d0>
Code;  c0106e60 <do_signal+20/2d0>
   b:   81 c4 94 00 00 00         add    $0x94,%esp
Code;  c0106e66 <do_signal+26/2d0>
  11:   89 d0                     mov    %edx,%eax
Code;  c0106e68 <do_signal+28/2d0>
  13:   5b                        pop    %ebx

Re: 2.4.17-pre8 OOPS with RedHat gcc 3.1-0.10

[Benjamin LaHaise]

On Tue, Dec 11, 2001 at 11:51:41AM -0800, Wayne Whitney wrote:
> On Mon, 10 Dec 2001, Wayne Whitney wrote:
> 
> > I recently upgraded to "gcc version 3.1 20011127 (Red Hat Linux
> > Rawhide 3.1-0.10)".  It compiles the recent 2.4.17-preX kernels
> 
> Well, it compiles them, but I get an oops on booting 2.4.17-pre8, for
> example, almost immediately after init is launched.  "gcc version 2.96
> 20000731 (Red Hat Linux 7.2 2.96-101.9)" works fine.

This looks like a gcc bug.  Could you disassemble the entirety of do_signal 
and submit that plus the original source as a bug report against gcc in 
Red Hat's bugzilla?  Thanks,

		-ben