Re: Oops in 2.4.0-ac5

From: Keith Owens (kaos@ocs.com.au)
Date: Wed Jan 10 2001 - 17:29:07 EST


On Wed, 10 Jan 2001 15:23:14 -0500,
Nathan Walp <faceprint@faceprint.com> wrote:
>Here it is... I opted to cut out the 1200-odd warnings, which from the
>look of them were all because i'm running it under 2.4.0-ac4 (which
>boots fine).

ksymoops defaults to using /proc entries from the current system. This
is fine for most people but you have to override the default when
decoding an oops from another kernel.

  ksymoops -KL -m /boot/System.map-2.4.0-ac5

ignores the running ksyms and lsmod files, removing your warnings.

>>>EIP; c01129ba <setup_apic_nmi_watchdog+a/90> <=====
>Trace; c0100191 <L6+0/2>
>Code; c01129ba <setup_apic_nmi_watchdog+a/90>
>00000000 <_EIP>:
>Code; c01129ba <setup_apic_nmi_watchdog+a/90> <=====
> 0: 0f 30 wrmsr <=====
>Code; c01129bc <setup_apic_nmi_watchdog+c/90>

This is why my original NMI for UP code in kdb uses wrmsr_eio() instead
of wrmsr. wrmsr_eio() catches errors where the APIC does not support
the msr and returns EIO instead of oopsing and taking the kernel with
it. I could never persuade Ingo to use wrmsr_eio() and check the
return code, maybe this will change his mind. Extract from kdb v1.7.

/*
 * {rd,wr}msr_eio by HPA, moved from arch/i386/msr.c to here.
 * Keith Owens.
 */

/* Note: "err" is handled in a funny way below. Otherwise one version
   of gcc or another breaks. */

extern inline int wrmsr_eio(u32 reg, u32 eax, u32 edx)
{
  int err;

  asm volatile(
              "1: wrmsr\n"
              "2:\n"
              ".section .fixup,\"ax\"\n"
              "3: movl %4,%0\n"
              " jmp 2b\n"
              ".previous\n"
              ".section __ex_table,\"a\"\n"
              " .align 4\n"
              " .long 1b,3b\n"
              ".previous"
              : "=&bDS" (err)
              : "a" (eax), "d" (edx), "c" (reg), "i" (-EIO), "0" (0));

  return err;
}

.....

/*
 * Activate or deactivate NMI watchdog via a local APIC.
 */

int setup_apic_nmi_watchdog(int value)
{
       int ret, eax;

       if (!test_bit(X86_FEATURE_APIC, &boot_cpu_data.x86_capability))
               return(-EIO);
       if (nmi_watchdog_source && nmi_watchdog_source != 1)
               return(0); /* Not using local APIC */
       /* Disable performance counters 0, 1 for all NMI changes */
       nmi_watchdog = proc_nmi_watchdog = nmi_watchdog_source = 0;
       if ((ret = wrmsr_eio(EVNTSEL0, 0, 0)) ||
           (ret = wrmsr_eio(EVNTSEL1, 0, 0)) ||
           (ret = wrmsr_eio(PERFCTR0, 0, 0)) ||
           (ret = wrmsr_eio(PERFCTR1, 0, 0)))
               goto exit;
       if (!value)
               return(0);
       /* Must set before activation to catch first NMI */
       nmi_watchdog = proc_nmi_watchdog = nmi_watchdog_source = 1;
       eax = 1 << 20 /* Interrupt on overflow */
           | 1 << 17 /* OS mode */
           | 1 << 16 /* User mode */
           | 0x79; /* Event, cpu clocks not halted */
       if ((ret = wrmsr_eio(EVNTSEL1, eax, 0))
        || (ret = set_nmi_counter_local()))
               goto exit;
       apic_write_around(APIC_LVTPC, APIC_DM_NMI);
       eax = 1 << 22; /* Enable performance counters, only using ctr1 */
       ret = wrmsr_eio(EVNTSEL0, eax, 0);
exit:
       if (ret)
               nmi_watchdog = proc_nmi_watchdog = nmi_watchdog_source = 0;
       else
               printk(KERN_INFO "Activated NMI via local APIC\n");
       return(ret);
}

-
To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
the body of a message to majordomo@vger.kernel.org
Please read the FAQ at http://www.tux.org/lkml/



This archive was generated by hypermail 2b29 : Mon Jan 15 2001 - 21:00:28 EST