Skip to contentSkip to conversation

x86: fix hang when AP bringup is too slow

PATCH Superseded Igor Mammedov 3 participants 11 messages Mar 31, 2014 – Apr 3, 2014 x86

v2 of 7 Superseded Latest is v7 →

lore
Share & tools
lore
#0 lore raw

v2 cover x86: fix hang when AP bringup is too slow

Changes since v1:
* reword comment in cpu_init() as suggested by Prarit
* make master CPU report wakeup error at ERR level instead of DBG level so it would be visible to user.
* while testing found and fixed memory corruption caused by invalid usage of x86_cpu_to_apicid and cpu_present maps on failure path in do_boot_cpu()
--

Hang is observed on virtual machines during CPU hotplug, especially in big guests with many CPUs. (It happens more often if host is over-committed).

Hang happens because master CPU timeouts on waiting till AP boots and 'cancels' CPU online operation assuming AP is not functional but AP may continue run wild later causing various hangs or panics in running kernel that is assuming that AP was offline.

This is an alternative approach, that instead of canceling in-progress AP bringup (https://lkml.org/lkml/2014/3/6/257),
removes timeouts so that AP bringup won't be affected by poor timing and syncs AP with master CPU at early startup making sure that AP won't run wild if master CPU doesn't expect AP to come online.

Below is detailed description of a more often happening hang:
-----
Master CPU may timeout before cpu_callin_mask is set and cancel booting CPU, but being onlined CPU still continues to boot, sets cpu_active_mask (CPU_STARTING notifiers) and spins in check_tsc_sync_target() for master cpu to arrive. Following attempt to online another cpu hangs in stop_machine, initiated from here:
smp_callin ->
smp_store_cpu_info ->

    identify_secondary_cpu ->
      mtrr_ap_init -> set_mtrr_from_inactive_cpu

stop_machine waits on completion of stop_work on all CPUs from cpu_active_mask including a failed CPU that spins in check_tsc_sync_target().

Igor Mammedov (5):
x86: replace timeouts when booting secondary CPU with infinite wait
loop
x86: cleanup not needed cpu_initialized_mask
x86: log error on secondary CPU wakeup failure at ERR level
x86: fix list corruption on CPU hotplug
x86: fix memory corruption in acpi_unmap_lsapic()

 arch/x86/include/asm/cpumask.h |    1 -
 arch/x86/kernel/cpu/common.c   |   28 ++++++++-------
 arch/x86/kernel/smpboot.c      |   73 ++--------------------------------------
 3 files changed, 18 insertions(+), 84 deletions(-)
#1 lore raw

PATCH v2 1/5 x86: replace timeouts when booting secondary CPU with infinite wait loop

Hang is observed on virtual machines during CPU hotplug, especially in big guests with many CPUs. (It reproducible more often if host is over-committed).

It happens because master CPU gives up waiting on secondary CPU and allows it to run wild. As result AP causes locking or crashing system. For example as described here: https://lkml.org/lkml/2014/3/6/257

If master CPU have sent STARTUP IPI successfully, make it wait indefinitely till AP boots. To ensure that AP won't ever run wild, make it wait at early startup till master CPU confirms its intention to wait for it.

---
v2:
- ammend comment in cpu_init()

---
 arch/x86/kernel/cpu/common.c |   17 +++++++++-
 arch/x86/kernel/smpboot.c    |   68 ++----------------------------------------
 2 files changed, 18 insertions(+), 67 deletions(-)
arch/x86/kernel/cpu/common.c +15 -2
index 8e28bf2..248161e 100644
@@ -1235,16 +1235,22 @@ void cpu_init(void)
 	struct task_struct *me;
 	struct tss_struct *t;
 	unsigned long v;
-	int cpu;
+	int cpu = stack_smp_processor_id();
 	int i;
 
 	/*
+	 * wait for ACK from master CPU before continuing
+	 * with AP initialization
+	 */
+	while (!cpumask_test_cpu(cpu, cpu_callout_mask))
+		cpu_relax();
+
+	/*
 	 * Load microcode on this cpu if a valid microcode is available.
 	 * This is early microcode loading procedure.
 	 */
 	load_ucode_ap();
 
-	cpu = stack_smp_processor_id();
 	t = &per_cpu(init_tss, cpu);
 	oist = &per_cpu(orig_ist, cpu);
 
@@ -1335,6 +1341,13 @@ void cpu_init(void)
 	struct tss_struct *t = &per_cpu(init_tss, cpu);
 	struct thread_struct *thread = &curr->thread;
 
+	/*
+	 * wait for ACK from master CPU before continuing
+	 * with AP initialization
+	 */
+	while (!cpumask_test_cpu(cpu, cpu_callout_mask))
+		cpu_relax();
+
 	show_ucode_info_early();
 
 	if (cpumask_test_and_set_cpu(cpu, cpu_initialized_mask)) {
arch/x86/kernel/smpboot.c +3 -65

99 lines not on this page · raw message

Signature · 1 line
1.7.1
#2 lore raw

Re: [PATCH v2 1/5] x86: replace timeouts when booting secondary CPU with infinite wait loop

Quoted 11 lines · Igor Mammedov

Hang is observed on virtual machines during CPU hotplug, especially in big guests with many CPUs. (It reproducible more often if host is over-committed).

It happens because master CPU gives up waiting on secondary CPU and allows it to run wild. As result AP causes locking or crashing system. For example as described here: https://lkml.org/lkml/2014/3/6/257

If master CPU have sent STARTUP IPI successfully, make it wait indefinitely till AP boots.

But what happens on a real machine when the other CPU is dead?

I've seen that. Kernel still boots. With your patch it would hang.

I don't think you can do that. It needs to have some timeout. Maybe a longer or configurable one?

-Andi

Signature · 1 line
[email protected] -- Speaking for myself only
#3 lore raw

Re: [PATCH v2 1/5] x86: replace timeouts when booting secondary CPU with infinite wait loop

Quoted 11 lines · On Wed, 02 Apr 2014 10:15:29 -0700 Andi Kleen <[email protected]> wrote:
Quoted 9 lines · Igor Mammedov

Hang is observed on virtual machines during CPU hotplug, especially in big guests with many CPUs. (It reproducible more often if host is over-committed).

It happens because master CPU gives up waiting on secondary CPU and allows it to run wild. As result AP causes locking or crashing system. For example as described here: https://lkml.org/lkml/2014/3/6/257

Quoted 5 lines
Quoted 2 lines

If master CPU have sent STARTUP IPI successfully, make it wait indefinitely till AP boots.

But what happens on a real machine when the other CPU is dead?

One possible way to boot such machine would be to disable dead CPU in kernel parameters.

Quoted 5 lines

I've seen that. Kernel still boots. With your patch it would hang.

I don't think you can do that. It needs to have some timeout. Maybe a longer or configurable one?

there were patch that tried to keep timeouts and 'gracefully' cancel AP boot if master timed out on it.
https://lkml.org/lkml/2014/3/6/257

It's possible to keep timeouts in do_boot_cpu(), is setting trampoline_status sufficient indication that AP is not dead and worth waiting for?

than it could be rewritten like this:

  if (!boot_error) {
      boot_error = 1;
      for (timeout = 0; timeout < 50000; timeout++) {
          /* Wait till AP signals that it's ready to start initialization */
          if (*trampoline_status == 0xA5A5A5A5) {
              boot_error = 0;
              /* allow AP to start initializing. */
              cpumask_set_cpu(cpu, cpu_callout_mask);

              /* wait till AP boots till cpu_callin_mask point */
              while (cpumask_test_cpu(cpu, cpu_callin_mask))
                   schedule();

              break;  /* It has booted */
          }
          udelay(100);
      }

}

it will provide timeout if AP is dead and still keep AP from running wild if master CPU timed out on it.

Quoted 5 lines

-Andi

Signature · 1 line
[email protected] -- Speaking for myself only
Signature · 2 lines
Regards,
  Igor
#4 lore raw

Re: [PATCH v2 1/5] x86: replace timeouts when booting secondary CPU with infinite wait loop

Quoted 14 lines · Igor Mammedov
Quoted 11 lines · On Wed, 02 Apr 2014 10:15:29 -0700 Andi Kleen <[email protected]> wrote:
Quoted 9 lines · Igor Mammedov

Hang is observed on virtual machines during CPU hotplug, especially in big guests with many CPUs. (It reproducible more often if host is over-committed).

It happens because master CPU gives up waiting on secondary CPU and allows it to run wild. As result AP causes locking or crashing system. For example as described here: https://lkml.org/lkml/2014/3/6/257

Quoted 7 lines
Quoted 5 lines
Quoted 2 lines

If master CPU have sent STARTUP IPI successfully, make it wait indefinitely till AP boots.

But what happens on a real machine when the other CPU is dead?

One possible way to boot such machine would be to disable dead CPU in kernel parameters.

That would need explicit user action. It's much better to recover automatically, even if somewhat crippled.

-Andi

#5 lore raw

Re: [PATCH v2 1/5] x86: replace timeouts when booting secondary CPU with infinite wait loop

In reply to Igor Mammedov
Quoted 1 line · Igor Mammedov
Quoted 1 line

I've seen that. Kernel still boots. With your patch it would hang.

Nonsense, not booting is OK when critical hardware is genuinely bad - this isn't a disk drive or networking where bad IO 'happens sometimes' and failure is something we have to engineer for - this is the CPU!

If a critical piece of hardware like the CPU or RAM is non-functional then it should be excluded by the user explicitly, not worked around after some ugly, non-deterministic and fragile timeout.

The timeout in the SMP bringup code was really an ancient property, introduced back more than a decade ago when hardware makers were ignorant of Linux we were ignorant of how to properly interface with SMP hardware.

Today a 'timeout' means one of 3 things:

- bad, fragile hardware - this we don't want to hide, unless explicitly told so by the user. I've seen such symptoms related to overclocking for example - so not booting is perfectly justified, it can prevent reporting a bogus kernel crash down the line.

- buggy SMP bringup. That is a bug that needs to be fixed, not worked around.

- timeout fragility in virtualized environments

I'm not aware of any genuine case where timing out is the correct thing to do.

So the patches look fine to me as-is, I planned on looking at them more closely after the merge window.

Thanks,

Ingo

#6 lore raw

Re: [PATCH v2 1/5] x86: replace timeouts when booting secondary CPU with infinite wait loop

Quoted 5 lines · Ingo Molnar
Quoted 1 line · Igor Mammedov
Quoted 1 line

I've seen that. Kernel still boots. With your patch it would hang.

Quoted 7 lines

Nonsense, not booting is OK when critical hardware is genuinely bad - this isn't a disk drive or networking where bad IO 'happens sometimes' and failure is something we have to engineer for - this is the CPU!

If a critical piece of hardware like the CPU or RAM is non-functional then it should be excluded by the user explicitly, not worked around after some ugly, non-deterministic and fragile timeout.

That's generally not true. We try to recover as best as we can and continue.

That's true for RCU stalls, and RAM errors (hwpoison) and other error conditions. It's true for kernel problems (we try to oops and continue, not to panic etc.)

Hanging forever is not recovering, it's just poor and broken error handling and generally not acceptable these days.

-Andi

#7 lore raw

PATCH v2 2/5 x86: cleanup not needed cpu_initialized_mask

In reply to Igor Mammedov

cpu_initialized_mask is used in cpu_init() for detecting if AP has been already started. But now test could never return true since master CPU doesn't give up waiting on AP startup and native_cpu_up() checks for:

physid_isset(apicid, phys_cpu_present_map) && cpu_callin_mask

so that it would not attempt to call do_boot_cpu() on already started AP.
And since cpu_initialized_mask isn't used for anything else just remove it altogether.

---
 arch/x86/include/asm/cpumask.h |    1 -
 arch/x86/kernel/cpu/common.c   |   11 -----------
 arch/x86/kernel/smpboot.c      |    2 --
 3 files changed, 0 insertions(+), 14 deletions(-)
arch/x86/include/asm/cpumask.h +0 -1
index 61c852f..c64c5f5 100644
@@ -5,7 +5,6 @@
 
 extern cpumask_var_t cpu_callin_mask;
 extern cpumask_var_t cpu_callout_mask;
-extern cpumask_var_t cpu_initialized_mask;
 extern cpumask_var_t cpu_sibling_setup_mask;
 
 extern void setup_cpu_local_masks(void);
arch/x86/kernel/cpu/common.c +0 -11
index 248161e..7dfac93 100644
@@ -47,7 +47,6 @@
 #include "cpu.h"
 
 /* all of these masks are initialized in setup_cpu_local_masks() */
-cpumask_var_t cpu_initialized_mask;
 cpumask_var_t cpu_callout_mask;
 cpumask_var_t cpu_callin_mask;
 
@@ -57,7 +56,6 @@ cpumask_var_t cpu_sibling_setup_mask;
 /* correctly size the local cpu masks */
 void __init setup_cpu_local_masks(void)
 {
-	alloc_bootmem_cpumask_var(&cpu_initialized_mask);
 	alloc_bootmem_cpumask_var(&cpu_callin_mask);
 	alloc_bootmem_cpumask_var(&cpu_callout_mask);
 	alloc_bootmem_cpumask_var(&cpu_sibling_setup_mask);
@@ -1262,9 +1260,6 @@ void cpu_init(void)
 
 	me = current;
 
-	if (cpumask_test_and_set_cpu(cpu, cpu_initialized_mask))
-		panic("CPU#%d already initialized!\n", cpu);
-
 	pr_debug("Initializing CPU#%d\n", cpu);
 
 	clear_in_cr4(X86_CR4_VME|X86_CR4_PVI|X86_CR4_TSD|X86_CR4_DE);
@@ -1350,12 +1345,6 @@ void cpu_init(void)
 
 	show_ucode_info_early();
 
-	if (cpumask_test_and_set_cpu(cpu, cpu_initialized_mask)) {
-		printk(KERN_WARNING "CPU#%d already initialized!\n", cpu);
-		for (;;)
-			local_irq_enable();
-	}
-
 	printk(KERN_INFO "Initializing CPU#%d\n", cpu);
 
 	if (cpu_has_vme || cpu_has_tsc || cpu_has_de)
arch/x86/kernel/smpboot.c +0 -2
index 8bf67bd..803f658 100644
@@ -1237,8 +1237,6 @@ static void __ref remove_cpu_from_maps(int cpu)
 	set_cpu_online(cpu, false);
 	cpumask_clear_cpu(cpu, cpu_callout_mask);
 	cpumask_clear_cpu(cpu, cpu_callin_mask);
-	/* was set by cpu_init() */
-	cpumask_clear_cpu(cpu, cpu_initialized_mask);
 	numa_remove_cpu(cpu);
 }
 
Signature · 1 line
1.7.1
#8 lore raw

PATCH v2 3/5 x86: log error on secondary CPU wakeup failure at ERR level

In reply to Igor Mammedov

If system is running without debug level logging, it will not log error if do_boot_cpu() failed to wakeup AP. It may lead to silent AP bringup failures at boot time.
Change message level to KERN_ERR to make error visible to user as it's done on other architectures.

---
 arch/x86/kernel/smpboot.c |    2 +-
 1 files changed, 1 insertions(+), 1 deletions(-)
arch/x86/kernel/smpboot.c +1 -1
index 803f658..e540b44 100644
@@ -858,7 +858,7 @@ int native_cpu_up(unsigned int cpu, struct task_struct *tidle)
 
 	err = do_boot_cpu(apicid, cpu, tidle);
 	if (err) {
-		pr_debug("do_boot_cpu failed %d\n", err);
+		pr_err("do_boot_cpu failed(%d) to wakeup CPU#%u\n", err, cpu);
 		return -EIO;
 	}
 
Signature · 1 line
1.7.1
#9 lore raw

PATCH v2 4/5 x86: fix list corruption on CPU hotplug

In reply to Igor Mammedov

currently if AP wake up is failed, master CPU marks AP as not present in do_boot_cpu() by calling set_cpu_present(cpu, false). That leads to following list corruption on the next physical CPU hotplug:

[  418.107336] WARNING: CPU: 1 PID: 45 at lib/list_debug.c:33 __list_add+0xbe/0xd0()
[  418.115268] list_add corruption. prev->next should be next (ffff88003dc57600), but was ffff88003e20c3a0. (prev=ffff88003e20c3a0).
[  418.123693] Modules linked in: nf_conntrack_netbios_ns nf_conntrack_broadcast ipt_MASQUERADE ip6t_REJECT ipt_REJECT cfg80211 xt_conntrack rfkill ee
[  418.138979] CPU: 1 PID: 45 Comm: kworker/u10:1 Not tainted 3.14.0-rc6+ #387
[  418.149989] Hardware name: Red Hat KVM, BIOS 0.5.1 01/01/2007
[  418.165750] Workqueue: kacpi_hotplug acpi_hotplug_work_fn
[  418.166433]  0000000000000021 ffff880038ca7988 ffffffff8159b22d 0000000000000021
[  418.176460]  ffff880038ca79d8 ffff880038ca79c8 ffffffff8106942c ffff880038ca79e8
[  418.177453]  ffff88003e20c3a0 ffff88003dc57600 ffff88003e20c3a0 00000000ffffffea
[  418.178445] Call Trace:
[  418.185811]  [<ffffffff8159b22d>] dump_stack+0x49/0x5c
[  418.186440]  [<ffffffff8106942c>] warn_slowpath_common+0x8c/0xc0
[  418.187192]  [<ffffffff81069516>] warn_slowpath_fmt+0x46/0x50
[  418.191231]  [<ffffffff8136ef51>] ? acpi_ns_get_node+0xb7/0xc7
[  418.193889]  [<ffffffff812f796e>] __list_add+0xbe/0xd0
[  418.196649]  [<ffffffff812e2aa9>] kobject_add_internal+0x79/0x200
[  418.208610]  [<ffffffff812e2e18>] kobject_add_varg+0x38/0x60
[  418.213831]  [<ffffffff812e2ef4>] kobject_add+0x44/0x70
[  418.229961]  [<ffffffff813e2c60>] device_add+0xd0/0x550
[  418.234991]  [<ffffffff813f0e95>] ? pm_runtime_init+0xe5/0xf0
[  418.250226]  [<ffffffff813e32be>] device_register+0x1e/0x30
[  418.255296]  [<ffffffff813e82a3>] register_cpu+0xe3/0x130
[  418.266539]  [<ffffffff81592be5>] arch_register_cpu+0x65/0x150
[  418.285845]  [<ffffffff81355c0d>] acpi_processor_hotadd_init+0x5a/0x9b

...
Which is caused by the fact that generic_processor_info() allocates logical CPU id by calling:

cpu = cpumask_next_zero(-1, cpu_present_mask);

which returns id of previously failed to wake up CPU, since its bit is cleared by do_boot_cpu() and as result register_cpu() tries to register another CPU with the same id as already present but failed to be onlined CPU.

Taking in account that AP will not do anything if master CPU failed to wake it up, there is no reason to mark that AP as not present and break next cpu hotplug attempts. As a side effect of not marking AP as not present, user would be allowed to online it again later.

---
 arch/x86/kernel/smpboot.c |    1 -
 1 files changed, 0 insertions(+), 1 deletions(-)
arch/x86/kernel/smpboot.c +0 -1
index e540b44..0400418 100644
@@ -797,7 +797,6 @@ static int do_boot_cpu(int apicid, int cpu, struct task_struct *idle)
 			schedule();
 		}
 	} else {
-		set_cpu_present(cpu, false);
 		per_cpu(x86_cpu_to_apicid, cpu) = BAD_APICID;
 	}
 
Signature · 1 line
1.7.1
#10 lore raw

PATCH v2 5/5 x86: fix memory corruption in acpi_unmap_lsapic()

In reply to Igor Mammedov

if during CPU hotplug master CPU failed to wake up AP it set percpu x86_cpu_to_apicid to BAD_APICID=0xFFFF for AP.

However following attempt to unplug that CPU will lead to out of bound write access to __apicid_to_node[] which is 32768 items long on x86_64 kernel.

So drop setting x86_cpu_to_apicid to BAD_APICID in do_boot_cpu() and allow acpi_processor_remove()->acpi_unmap_lsapic() cleanly remove CPU.

---
 arch/x86/kernel/smpboot.c |    2 --
 1 files changed, 0 insertions(+), 2 deletions(-)
arch/x86/kernel/smpboot.c +0 -2
index 0400418..2b0b8ee 100644
@@ -796,8 +796,6 @@ static int do_boot_cpu(int apicid, int cpu, struct task_struct *idle)
 			 */
 			schedule();
 		}
-	} else {
-		per_cpu(x86_cpu_to_apicid, cpu) = BAD_APICID;
 	}
 
 	/* mark "stuck" area as not stuck */
Signature · 1 line
1.7.1