xen/vcpu: ignore VCPU_SSHOTTMR_future
PATCH v2 1/1 xen/vcpu: ignore VCPU_SSHOTTMR_future
The usage of VCPU_SSHOTTMR_future in Linux prior to 4.7 is bogus. When the hypervisor returns -ENOTIME (timeout in the past) Linux keeps retrying to setup the timer with a higher timeout instead of self-injecting a timer interrupt.
On boxes without any hardware assistance for logdirty we have seen HVM Linux guests < 4.7 with 32vCPUs give up trying to setup the timer when logdirty is enabled:
CE: Reprogramming failure. Giving up
CE: xen increased min_delta_ns to 1000000 nsec
CE: Reprogramming failure. Giving up
CE: Reprogramming failure. Giving up
CE: xen increased min_delta_ns to 506250 nsec
CE: xen increased min_delta_ns to 759375 nsec
CE: xen increased min_delta_ns to 1000000 nsec
CE: Reprogramming failure. Giving up
CE: Reprogramming failure. Giving up
CE: Reprogramming failure. Giving up
Freezing user space processes ...
INFO: rcu_sched detected stalls on CPUs/tasks: { 14} (detected by 10, t=60002 jiffies, g=4006, c=4005, q=14130)
Task dump for CPU 14:
swapper/14 R running task 0 0 1 0x00000000
Call Trace:
[<ffffffff90160f5d>] ? rcu_eqs_enter_common.isra.30+0x3d/0xf0
[<ffffffff907b9bde>] ? default_idle+0x1e/0xd0
[<ffffffff90039570>] ? arch_cpu_idle+0x20/0xc0
[<ffffffff9010820a>] ? cpu_startup_entry+0x14a/0x1e0
[<ffffffff9005d3a7>] ? start_secondary+0x1f7/0x270
[<ffffffff900000d5>] ? start_cpu+0x5/0x14
INFO: rcu_sched detected stalls on CPUs/tasks: { 26} (detected by 24, t=60002 jiffies, g=6922, c=6921, q=7013)
Task dump for CPU 26:
swapper/26 R running task 0 0 1 0x00000000
Call Trace:
[<ffffffff90160f5d>] ? rcu_eqs_enter_common.isra.30+0x3d/0xf0
[<ffffffff907b9bde>] ? default_idle+0x1e/0xd0
[<ffffffff90039570>] ? arch_cpu_idle+0x20/0xc0
[<ffffffff9010820a>] ? cpu_startup_entry+0x14a/0x1e0
[<ffffffff9005d3a7>] ? start_secondary+0x1f7/0x270
[<ffffffff900000d5>] ? start_cpu+0x5/0x14
INFO: rcu_sched detected stalls on CPUs/tasks: { 26} (detected by 24, t=60002 jiffies, g=8499, c=8498, q=7664)
Task dump for CPU 26:
swapper/26 R running task 0 0 1 0x00000000
Call Trace:
[<ffffffff90160f5d>] ? rcu_eqs_enter_common.isra.30+0x3d/0xf0
[<ffffffff907b9bde>] ? default_idle+0x1e/0xd0
[<ffffffff90039570>] ? arch_cpu_idle+0x20/0xc0
[<ffffffff9010820a>] ? cpu_startup_entry+0x14a/0x1e0
[<ffffffff9005d3a7>] ? start_secondary+0x1f7/0x270
[<ffffffff900000d5>] ? start_cpu+0x5/0x14Thus leading to CPU stalls and a broken system as a result.
Workaround this bogus usage by ignoring the VCPU_SSHOTTMR_future in the hypervisor. Old Linux versions are the only ones known to have (wrongly) attempted to use the flag, and ignoring it is compatible with the behavior expected by any guests setting that flag.
Note the usage of the flag has been removed from Linux by commit:
c06b6d70feb3 xen/x86: don't lose event interrupts
Which landed in Linux 4.7.
---
Changes since v1:
- Just ignore the flag, as there's no ABI breakage.
--- CHANGELOG.md | 2 ++ xen/common/domain.c | 13 ++++++++++--- xen/include/public/vcpu.h | 2 +- 3 files changed, 13 insertions(+), 4 deletions(-)
CHANGELOG.md +2 -0
@@ -9,6 +9,8 @@ The format is based on [Keep a Changelog](https://keepachangelog.com/en/1.0.0/) ### Changed - Repurpose command line gnttab_max_{maptrack_,}frames options so they don't cap toolstack provided values. + - Ignore VCPU_SSHOTTMR_future VCPUOP_set_singleshot_timer flag. The only + known user doesn't use it properly, leading to in-guest breakage. ### Added - On x86, support for features new in Intel Sapphire Rapids CPUs:
xen/common/domain.c +10 -3
@@ -1762,9 +1762,16 @@ long common_vcpu_op(int cmd, struct vcpu *v, XEN_GUEST_HANDLE_PARAM(void) arg)if (copy_from_guest (&set, arg,1 ) )return -EFAULT ; -if ( (set.flags & VCPU_SSHOTTMR_future) && - (set.timeout_abs_ns <NOW ()) ) -return -ETIME ; +if ( set.timeout_abs_ns <NOW () ) + { +/* +* Simplify the logic if the timeout has already expired and just +* inject the event. +*/ +stop_timer (&v->singleshot_timer); +send_timer_event (v); +break ; + }migrate_timer (&v->singleshot_timer,smp_processor_id ());set_timer (&v->singleshot_timer, set.timeout_abs_ns);
xen/include/public/vcpu.h +1 -1
@@ -150,7 +150,7 @@ typedef struct vcpu_set_singleshot_timer vcpu_set_singleshot_timer_t;DEFINE_XEN_GUEST_HANDLE (vcpu_set_singleshot_timer_t);/* Flags to VCPUOP_set_singleshot_timer. */ -/* Require the timeout to be in the future (return -ETIME if it's passed). */ +/* Ignored. */ #define _VCPU_SSHOTTMR_future (0)#define VCPU_SSHOTTMR_future (1U << _VCPU_SSHOTTMR_future)
Signature · 1 line
2.40.0
Hi Roger,
Quoted 65 lines
-----Original Message-----
From: Roger Pau Monne <[email protected]>
Subject: [PATCH v2] xen/vcpu: ignoreVCPU_SSHOTTMR_futureThe usage of
VCPU_SSHOTTMR_futurein Linux prior to 4.7 is bogus. When the hypervisor returns-ENOTIME(timeout in the past) Linux keeps retrying to setup the timer with a higher timeout instead of self-injecting a timer interrupt.On boxes without any hardware assistance for logdirty we have seen HVM Linux guests < 4.7 with 32vCPUs give up trying to setup the timer when logdirty is enabled:
CE: Reprogramming failure. Giving up
CE: xen increasedmin_delta_nsto 1000000 nsec
CE: Reprogramming failure. Giving up
CE: Reprogramming failure. Giving up
CE: xen increasedmin_delta_nsto 506250 nsec
CE: xen increasedmin_delta_nsto 759375 nsec
CE: xen increasedmin_delta_nsto 1000000 nsec
CE: Reprogramming failure. Giving up
CE: Reprogramming failure. Giving up
CE: Reprogramming failure. Giving up
Freezing user space processes ...INFO: rcu_sched detected stalls on CPUs/tasks: { 14} (detected by 10, t=60002 jiffies, g=4006, c=4005, q=14130) Task dump for CPU 14: swapper/14 R running task 0 0 1 0x00000000 Call Trace: [<ffffffff90160f5d>] ? rcu_eqs_enter_common.isra.30+0x3d/0xf0 [<ffffffff907b9bde>] ? default_idle+0x1e/0xd0 [<ffffffff90039570>] ? arch_cpu_idle+0x20/0xc0 [<ffffffff9010820a>] ? cpu_startup_entry+0x14a/0x1e0 [<ffffffff9005d3a7>] ? start_secondary+0x1f7/0x270 [<ffffffff900000d5>] ? start_cpu+0x5/0x14 INFO: rcu_sched detected stalls on CPUs/tasks: { 26} (detected by 24, t=60002 jiffies, g=6922, c=6921, q=7013) Task dump for CPU 26: swapper/26 R running task 0 0 1 0x00000000 Call Trace: [<ffffffff90160f5d>] ? rcu_eqs_enter_common.isra.30+0x3d/0xf0 [<ffffffff907b9bde>] ? default_idle+0x1e/0xd0 [<ffffffff90039570>] ? arch_cpu_idle+0x20/0xc0 [<ffffffff9010820a>] ? cpu_startup_entry+0x14a/0x1e0 [<ffffffff9005d3a7>] ? start_secondary+0x1f7/0x270 [<ffffffff900000d5>] ? start_cpu+0x5/0x14 INFO: rcu_sched detected stalls on CPUs/tasks: { 26} (detected by 24, t=60002 jiffies, g=8499, c=8498, q=7664) Task dump for CPU 26: swapper/26 R running task 0 0 1 0x00000000 Call Trace: [<ffffffff90160f5d>] ? rcu_eqs_enter_common.isra.30+0x3d/0xf0 [<ffffffff907b9bde>] ? default_idle+0x1e/0xd0 [<ffffffff90039570>] ? arch_cpu_idle+0x20/0xc0 [<ffffffff9010820a>] ? cpu_startup_entry+0x14a/0x1e0 [<ffffffff9005d3a7>] ? start_secondary+0x1f7/0x270 [<ffffffff900000d5>] ? start_cpu+0x5/0x14Thus leading to CPU stalls and a broken system as a result.
Workaround this bogus usage by ignoring the
VCPU_SSHOTTMR_futurein the hypervisor. Old Linux versions are the only ones known to have (wrongly) attempted to use the flag, and ignoring it is compatible with the behavior expected by any guests setting that flag.
Quoted 7 lines
Note the usage of the flag has been removed from Linux by commit:
c06b6d70feb3xen/x86: don't lose event interruptsWhich landed in Linux 4.7.
Signed-off-by Roger Pau Monné <[email protected]>
Kind regards,
Henry
Quoted 2 lines · Roger Pau Monne
The usage of
VCPU_SSHOTTMR_futurein Linux prior to 4.7 is bogus. When the hypervisor returns-ENOTIME(timeout in the past) Linux keeps
Nit: -ETIME
CHANGELOG.md The format is based on [Keep a Changelog](https://keepachangelog.com/en/1.0.0/) · line 13
Quoted 11 lines
retrying to setup the timer with a higher timeout instead of self-injecting a timer interrupt.
[...]CHANGELOG.md +2
-0@@ -9,6 +9,8 @@ The format is based on [Keep a Changelog](https://keepachangelog.com/en/1.0.0/) ### Changed - Repurpose command line gnttab_max_{maptrack_,}frames options so they don't cap toolstack provided values. + - Ignore VCPU_SSHOTTMR_future VCPUOP_set_singleshot_timer flag. The only + known user doesn't use it properly, leading to in-guest breakage.
Might this read a little better as
- Ignore VCPUOP_set_singleshot_timer's VCPU_SSHOTTMR_future flag. The only known user doesn't use it properly, leading to in-guest breakage.
xen/include/public/vcpu.h typedef struct vcpu_set_singleshot_timer vcpu_set_singleshot_timer_t; · line 153
Quoted 8 lines
xen/include/public/vcpu.h +1
-1@@ -150,7 +150,7 @@ typedef struct vcpu_set_singleshot_timer vcpu_set_singleshot_timer_t;DEFINE_XEN_GUEST_HANDLE (vcpu_set_singleshot_timer_t);/* Flags to VCPUOP_set_singleshot_timer. */ -/* Require the timeout to be in the future (return -ETIME if it's passed). */ +/* Ignored. */
I think this could do with something like "as of Xen 4.18", as the public header shouldn't be tied to a specific version (and then perhaps also retaining the original text). Arguably mentioning a specific version may be a little odd in case we'd consider backporting this, but something would imo better be said.
All this said - while I'm likely to ack the final patch, I would still feel a little uneasy doing so.
Jan
Quoted 5 lines · Jan Beulich
Quoted 2 lines · Roger Pau Monne
The usage of
VCPU_SSHOTTMR_futurein Linux prior to 4.7 is bogus. When the hypervisor returns-ENOTIME(timeout in the past) Linux keepsNit:
-ETIME
Oh, thanks.
Quoted 15 lines
CHANGELOG.md The format is based on [Keep a Changelog](https://keepachangelog.com/en/1.0.0/) · line 13
Quoted 12 lines
retrying to setup the timer with a higher timeout instead of self-injecting a timer interrupt.
[...]CHANGELOG.md +2
-0@@ -9,6 +9,8 @@ The format is based on [Keep a Changelog](https://keepachangelog.com/en/1.0.0/) ### Changed - Repurpose command line gnttab_max_{maptrack_,}frames options so they don't cap toolstack provided values. + - Ignore VCPU_SSHOTTMR_future VCPUOP_set_singleshot_timer flag. The only + known user doesn't use it properly, leading to in-guest breakage.Might this read a little better as
Quoted 2 lines
- Ignore VCPUOP_set_singleshot_timer's
VCPU_SSHOTTMR_futureflag. The only known user doesn't use it properly, leading to in-guest breakage.
Sure.
Quoted 5 lines
Quoted 5 lines
xen/include/public/vcpu.h +0
-0@@ -150,7 +150,7 @@ typedef struct vcpu_set_singleshot_timer vcpu_set_singleshot_timer_t;DEFINE_XEN_GUEST_HANDLE (vcpu_set_singleshot_timer_t);
Quoted 9 lines
xen/include/public/vcpu.h typedef struct vcpu_set_singleshot_timer vcpu_set_singleshot_timer_t; · line 153
Quoted 3 lines
xen/include/public/vcpu.h +1
-1/* Flags to VCPUOP_set_singleshot_timer. */ -/* Require the timeout to be in the future (return -ETIME if it's passed). */ +/* Ignored. */ I think this could do with something like "as of Xen 4.18", as the public header shouldn't be tied to a specific version (and then perhaps also retaining the original text). Arguably mentioning a specific version may be a little odd in case we'd consider backporting this, but something would imo better be said.
Hm, at least for XenServer we will backport this to 4.13, other vendors might backport to different versions, at which point I'm not sure the comment is very helpful. It can be misleading because it might seem to infer that checking the Xen version will tell you whether the flag is ignored or not.
What about:
/*
* Request the timeout to be in the future (return -ETIME if it's passed)
* but can be ignored by the hypervisor.
*/
Quoted 2 lines
All this said - while I'm likely to ack the final patch, I would still feel a little uneasy doing so.
Thanks, Roger.
Quoted 23 lines · Roger Pau Monné
Quoted 6 lines · Jan Beulich
Quoted 5 lines · Roger Pau Monne
xen/include/public/vcpu.h +0
-0@@ -150,7 +150,7 @@ typedef struct vcpu_set_singleshot_timer vcpu_set_singleshot_timer_t;DEFINE_XEN_GUEST_HANDLE (vcpu_set_singleshot_timer_t);Quoted 9 lines
xen/include/public/vcpu.h typedef struct vcpu_set_singleshot_timer vcpu_set_singleshot_timer_t; · line 153
Quoted 3 lines
xen/include/public/vcpu.h +1
-1/* Flags to VCPUOP_set_singleshot_timer. */ -/* Require the timeout to be in the future (return -ETIME if it's passed). */ +/* Ignored. */ I think this could do with something like "as of Xen 4.18", as the public header shouldn't be tied to a specific version (and then perhaps also retaining the original text). Arguably mentioning a specific version may be a little odd in case we'd consider backporting this, but something would imo better be said.
Hm, at least for XenServer we will backport this to 4.13, other vendors might backport to different versions, at which point I'm not sure the comment is very helpful. It can be misleading because it might seem to infer that checking the Xen version will tell you whether the flag is ignored or not.
Quoted 6 lines
What about:
/*
* Request the timeout to be in the future (return-ETIMEif it's passed)
* but can be ignored by the hypervisor.
*/
Would be fine with me.
Jan