Re: Dmesg protocol for C2D E8400&ASUS P5K Premium (stack trace came up twice)

[Jeremy Fitzhardinge <jeremy@goop.org>]

Boris Derzhavets wrote:
>
> [    6.011212] ------------[ cut here ]------------
> [    6.011257] WARNING: at kernel/smp.c:329 
> smp_call_function_many+0x1e9/0x250()
> [    6.011303] Hardware name: P5K Premium
> [    6.011345] Modules linked in: usbhid(+) hid sd_mod crc_t10dif 
> sr_mod cdrom r8169 mii ohci1394 ieee1394 sg sky2 pata_acpi ata_generic 
> ehci_hcd uhci_hcd usbcore dm_mirror dm_region_hash dm_log dm_snapshot 
> dm_mod thermal processor fan fuse
> [    6.012219] Pid: 2303, comm: modprobe Not tainted 2.6.29-rc5-tip #1
> [    6.012263] Call Trace:
> [    6.012309]  [<ffffffff8024c050>] warn_slowpath+0xd0/0x130
> [    6.012356]  [<ffffffff8022d9ae>] ? pvclock_clocksource_read+0x4e/0x90
> [    6.012403]  [<ffffffff8020eeed>] ? xen_force_evtchn_callback+0xd/0x10
> [    6.012450]  [<ffffffff8020f732>] ? check_events+0x12/0x20
> [    6.012495]  [<ffffffff8020f531>] ? xen_clocksource_read+0x21/0x30
> [    6.012542]  [<ffffffff80210617>] ? xen_spin_lock+0xa7/0x110
> [    6.012587]  [<ffffffff8020f71f>] ? xen_restore_fl_direct_end+0x0/0x1
> [    6.012634]  [<ffffffff8052e373>] ? _spin_lock+0x13/0x30
> [    6.012681]  [<ffffffff802a6690>] ? 
> perf_counter_task_sched_in+0x60/0x190
> [    6.012728]  [<ffffffff8022d9ae>] ? pvclock_clocksource_read+0x4e/0x90
> [    6.012776]  [<ffffffff8020bdf9>] ? 
> __raw_callee_save_xen_pmd_val+0x11/0x1e
> [    6.012823]  [<ffffffff802360b0>] ? do_flush_tlb_all+0x0/0x50
> [    6.012869]  [<ffffffff802742a9>] smp_call_function_many+0x1e9/0x250
> [    6.012915]  [<ffffffff802360b0>] ? do_flush_tlb_all+0x0/0x50
> [    6.012961]  [<ffffffff802360b0>] ? do_flush_tlb_all+0x0/0x50
> [    6.013007]  [<ffffffff80274330>] smp_call_function+0x20/0x30
> [    6.013053]  [<ffffffff80251fcf>] on_each_cpu+0x1f/0x50
> [    6.013098]  [<ffffffff80235ea7>] flush_tlb_all+0x17/0x20
>

Could you try this patch:

Subject: [PATCH] x86: implement flush_tlb_all in terms of flush_tlb_others

Modify flush_tlb_others to take a NULL mm, meaning "flush user
and kernel tlbs", and implement flush_tlb_all in terms of that.

The principle motivation for this is to make sure it goes via
paravirt_ops, which allows for more efficient cross-cpu tlb
flushes than a plain IPI.

Signed-off-by: Jeremy Fitzhardinge <jeremy.fitzhardinge@citrix.com>

diff --git a/arch/x86/mm/tlb.c b/arch/x86/mm/tlb.c
index 14c5af4..2d112f9 100644
--- a/arch/x86/mm/tlb.c
+++ b/arch/x86/mm/tlb.c
@@ -148,7 +148,13 @@ void smp_invalidate_interrupt(struct pt_regs *regs)
 		 * BUG();
 		 */
 
-	if (f->flush_mm == percpu_read(cpu_tlbstate.active_mm)) {
+	if (f->flush_mm == NULL) {
+		/* No mm - flush all kernel and user tlbs */
+		__flush_tlb_all();
+		if (percpu_read(cpu_tlbstate.state) == TLBSTATE_LAZY)
+			leave_mm(cpu);
+	} else if (f->flush_mm == percpu_read(cpu_tlbstate.active_mm)) {
+		/* Flushing a specific user mm */
 		if (percpu_read(cpu_tlbstate.state) == TLBSTATE_OK) {
 			if (f->flush_va == TLB_FLUSH_ALL)
 				local_flush_tlb();
@@ -281,16 +287,15 @@ void flush_tlb_page(struct vm_area_struct *vma, unsigned long va)
 	preempt_enable();
 }
 
-static void do_flush_tlb_all(void *info)
+void flush_tlb_all(void)
 {
-	unsigned long cpu = smp_processor_id();
+	preempt_disable();
+
+	flush_tlb_others(cpu_online_mask, NULL, TLB_FLUSH_ALL);
 
 	__flush_tlb_all();
 	if (percpu_read(cpu_tlbstate.state) == TLBSTATE_LAZY)
-		leave_mm(cpu);
-}
+		leave_mm(smp_processor_id());
 
-void flush_tlb_all(void)
-{
-	on_each_cpu(do_flush_tlb_all, NULL, 1);
+	preempt_enable();
 }
diff --git a/arch/x86/xen/mmu.c b/arch/x86/xen/mmu.c
index 7cadab1..dc83b47 100644
--- a/arch/x86/xen/mmu.c
+++ b/arch/x86/xen/mmu.c
@@ -1350,7 +1350,6 @@ static void xen_flush_tlb_others(const struct cpumask *cpus,
 	struct multicall_space mcs;
 
 	BUG_ON(cpumask_empty(cpus));
-	BUG_ON(!mm);
 
 	mcs = xen_mc_entry(sizeof(*args));
 	args = mcs.args;