RE: Kernel Preemption patch

Eugene,

Sorry, I'm using microsoft outlook and I can't seem to figure out how to
embedded comments in a satisfactory manner, so I have to write my replies at
the top of email threads all the time...

Below is an excerp from your post below. You include it to show the
redefinition of the line "#if !defined(CONFIG_4xx) || defined(CONFIG_440)".

However, I noticed that your 'preempt_enable' at the bottom of the exerp has
been changed from 'preempt_enable' as in the RML patch, to
'preempt_enable_no_reshed'. What's the reason for this inconsistency?

<----------------- cut ------------------->
#if !defined(CONFIG_4xx) || defined(CONFIG_440)
         case 0x900:
#else
         case 0x1000:
#endif
                 ret = timer_interrupt(regs);
                 break;
         default:
                 BUG();
         }

         preempt_enable_no_resched();
         return ret;
}
#endif /* CONFIG_PREEMPT */
<------------------ cut -------------------->

-- Brian

-----Original Message-----
From: Eugene Surovegin [mailto:ebs@ebshome.net]
Sent: Sunday, February 23, 2003 5:54 PM
To: brian.auld@adic.com
Cc: linuxppc-embedded@lists.linuxppc.org
Subject: Re: Kernel Preemption patch

At 02:28 PM 2/23/2003, brian.auld@adic.com wrote:

Hello,

Has anyone successfully applied and run the linux kernel preemption patch

on

a 2.4.20 kernel. I have the following:

- Latest Denx kernel (2_4_devel) running on 440GP ebony
- 'preempt-kernel-rml-2.4.20-1.patch' applied to my kernel source
  (I manually corrected a couple of fails and double checked a few
   'successfully applied with fuzz' instances when applying the patch).
- Kernel compiled with 'Kernel preemption =y' in make oldconfig

When I try to boot this kernel, I get the following error:

<------- snipped
Kernel command line: root=/dev/nfs rw
nfsroot=172.16.77.152:/opt/eldk/ppc_4xx
ip=172.16.77.155:172.16.77.152::255.255.255.0:EBONY-2::off
kernel BUG at irq.c:575!
Oops: Exception in kernel mode, sig: 4
NIP: C0002B04 XER: 00000000 LR: C0002B04 SP: C0125F50 REGS: c0125ea0 TRAP:
0700    Not tainted
MSR: 00001030 EE: 0 PR: 0 FP: 0 ME: 1 IR/DR: 11
TASK = c0124060[0] 'swapper' Last syscall: 0
last math 00000000 last altivec 00000000
PLB0: bear= 0x00008000 acr=   0x9b000000 besr=  0x00000000

GPR00: C0002B04 C0125F50 C0124060 00000019 00001030 00000001 00000020
C0170000
GPR08: 000001D8 00000002 00000000 C0125E70 C012A8A8 71000000 07FE7300
007FFC73
GPR16: 00000000 00000001 007FFC00 FFFFFFFF 00001032 C0125F60 00000000
C00017CC
GPR24: C0002AA0 007FFB90 00000E00 007FFC73 007FFC00 00000000 00000000
007FFB90
Call backtrace:
C0002B04 C00017CC C0133504 C0000224
Kernel panic: Attempted to kill the idle task!
In idle task - not syncing
 <0>Rebooting in 180 seconds..

RML patches doesn't support 2_4_devel tree very well (and especially
PPC440).

If irq.c:575 correspond to preempt_intercept function in your tree, it
should look like:

#ifdef CONFIG_PREEMPT
int
preempt_intercept(struct pt_regs *regs)
{
         int ret;

         preempt_disable();

         switch(regs->trap) {
         case 0x500:
                 ret = do_IRQ(regs);
                 break;
#if !defined(CONFIG_4xx) || defined(CONFIG_440)
         case 0x900:
#else
         case 0x1000:
#endif
                 ret = timer_interrupt(regs);
                 break;
         default:
                 BUG();
         }

         preempt_enable_no_resched();
         return ret;
}
#endif /* CONFIG_PREEMPT */


Please, note additional "defined(CONFIG_440)"

Eugene


** Sent via the linuxppc-embedded mail list. See http://lists.linuxppc.org/

At 07:56 AM 2/24/2003, brian.auld@adic.com wrote:

However, I noticed that your 'preempt_enable' at the bottom of the exerp has
been changed from 'preempt_enable' as in the RML patch, to
'preempt_enable_no_reshed'. What's the reason for this inconsistency?

Well, as I said I don't use RML patch. This code is from MVL.
Frankly, I don't think that this particular change is very important (maybe
it is :).
I didn't notice any difference between these two versions.

It's "if !defined(CONFIG_4xx) || defined(CONFIG_440)" which is more
important :)

Eugene


** Sent via the linuxppc-embedded mail list. See http://lists.linuxppc.org/

At 07:56 AM 2/24/2003, brian.auld@adic.com wrote:

However, I noticed that your 'preempt_enable' at the bottom of the
exerp has
been changed from 'preempt_enable' as in the RML patch, to
'preempt_enable_no_reshed'. What's the reason for this inconsistency?

The preempt_enable_no_reshed avoids a kernel stack overflow seen in
certain cases of heavy preemption, where each nested call to reschedule
on preemption is interrupted and again requests a preemption.  The
proper place to check for another preemption is at the top-level
interrupt/exception handler, so that we iterate and not recurse.

Oops, MontaVista needs to contribute this bug fix back to RML's public
patch, I'll put it on my list.


--
Todd


** Sent via the linuxppc-embedded mail list. See http://lists.linuxppc.org/