Re: 2.6.0-test6: Oops at 'sysexit' on all threads after APM resume.

[Hugh Dickins]

On Wed, 1 Oct 2003, David Woodhouse wrote:
> general protection fault: 0000 [#11]
> CPU:	0
> EIP:	0060:[<c010b3f5>]	Not tainted
> EFLAGS:	00010246
> EIP is at sysenter_past_esp+0x6e/0x71
> eax: 00000001   ebx: 00000000    ecx: bffffd78   edx: ffffe410
> esi: 0804b280   edi: bffffdab    ebp: d6056000   esp: d6057fc4
> 
> 0xc010b3f5 <sysenter_past_esp+110>:     sysexit

Glad to have your company.  Don't know when this started, first saw it
when I started using Dell Latitude C610 with RH9 a couple of weeks ago.
I use the workaround patch below (leaving warning in to remind me it's
not really a fix).  Seems some uses of suspend key go the intended route
(lots of noisy debug messages) e.g. the first after reboot; but some
bypass it completely e.g. the next.  And in that latter case, the vital
MSRs are not reinitialized.  Sorry, no time at present to follow it up.

Hugh

--- 2.6.0-test6/arch/i386/kernel/sysenter.c	2003-08-09 05:44:45.000000000 +0100
+++ linux/arch/i386/kernel/sysenter.c	2003-10-01 23:27:50.576482336 +0100
@@ -34,6 +34,32 @@
 	put_cpu();	
 }
 
+int fixup_sysenter(void)
+{
+#ifdef CONFIG_X86_HIGH_ENTRY
+	struct tss_struct *tss = (struct tss_struct *) __fix_to_virt(FIX_TSS_0);
+#else
+	struct tss_struct *tss = init_tss;
+#endif
+	unsigned int cs, esp, eip, high;
+	int ret = 0;
+
+	if (!boot_cpu_has(X86_FEATURE_SEP))
+		return 0;
+	tss += get_cpu();
+	rdmsr(MSR_IA32_SYSENTER_CS, cs, high);
+	rdmsr(MSR_IA32_SYSENTER_ESP, esp, high);
+	rdmsr(MSR_IA32_SYSENTER_EIP, eip, high);
+	if (cs != __KERNEL_CS || esp != tss->esp1 ||
+	    eip != (unsigned long) sysenter_entry) {
+		enable_sep_cpu(NULL);
+		printk("fixup_sysenter: %x %x %x\n", cs, esp, eip);
+		ret = 1;
+	}
+	put_cpu();
+	return ret;
+}
+
 /*
  * These symbols are defined by vsyscall.o to mark the bounds
  * of the ELF DSO images included therein.
--- 2.6.0-test6/arch/i386/kernel/traps.c	2003-09-28 07:31:58.000000000 +0100
+++ linux/arch/i386/kernel/traps.c	2003-10-01 23:27:50.593479752 +0100
@@ -388,7 +388,7 @@
 	return;
 
 gp_in_kernel:
-	if (!fixup_exception(regs))
+	if (!fixup_sysenter() && !fixup_exception(regs))
 		die("general protection fault", regs, error_code);
 }
 

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