[Devel] [PATCH RHEL9 COMMIT] x86/cpu: init_cpu_flags -- use raw spinlock
Konstantin Khorenko
khorenko at virtuozzo.com
Thu Oct 14 19:37:00 MSK 2021
The commit is pushed to "branch-rh9-5.14.vz9.1.x-ovz" and will appear at https://src.openvz.org/scm/ovz/vzkernel.git
after rh9-5.14.0-4.vz9.10.7
------>
commit 1f7b66c334884ee928f2f65f4354b9e375eac089
Author: Cyrill Gorcunov <gorcunov at gmail.com>
Date: Thu Oct 14 19:36:59 2021 +0300
x86/cpu: init_cpu_flags -- use raw spinlock
The @cpu_flags_lock spinlock guards manipulations with
per-cpu @cpu_flags which is used to hide some features
in cpuinfo output inside VE. Still the init_cpu_flags
is called from irq context leading to the following
| [ 13.827635] =============================
| [ 13.827636] [ BUG: Invalid wait context ]
| [ 13.827637] 5.14.0.ovz9.10.1+ #41 Tainted: G C X --------- ---
| [ 13.827638] -----------------------------
| [ 13.827638] systemd/1 is trying to lock:
| [ 13.827639] ffffffffa4c9d258 (cpu_flags_lock){....}-{3:3}, at: init_cpu_flags+0xc8/0x220
| [ 13.827649] other info that might help us debug this:
| [ 13.827651] context-{2:2}
| [ 13.827651] 3 locks held by systemd/1:
| [ 13.827652] #0: ffffffffa56e8c60 (dup_mmap_sem){.+.+}-{0:0}, at: dup_mm+0x83/0x5f0
| [ 13.827660] #1: ffff97574a37d138 (&mm->mmap_lock#2){++++}-{4:4}, at: dup_mm+0x9c/0x5f0
| [ 13.827664] #2: ffff97574489c138 (&mm->mmap_lock/1){+.+.}-{4:4}, at: dup_mm+0xd5/0x5f0
| [ 13.827667] stack backtrace:
| [ 13.827668] CPU: 0 PID: 1 Comm: systemd ve: / Tainted: G C X --------- ---
| [ 13.827670] Hardware name: QEMU Standard PC (Q35 + ICH9, 2009), BIOS 1.14.0-4.fc34 04/01/2014
| [ 13.827672] Call Trace:
| [ 13.827673] <IRQ>
| [ 13.827675] dump_stack_lvl+0x57/0x7d
| [ 13.827686] __lock_acquire.cold+0x28b/0x2cd
| [ 13.827694] lock_acquire+0xca/0x300
| [ 13.827700] ? init_cpu_flags+0xc8/0x220
| [ 13.827703] _raw_spin_lock+0x34/0x80
| [ 13.827708] ? init_cpu_flags+0xc8/0x220
| [ 13.827710] init_cpu_flags+0xc8/0x220
| [ 13.827713] flush_smp_call_function_queue+0x13f/0x1e0
| [ 13.827717] __sysvec_call_function_single+0x43/0x1c0
| [ 13.827722] sysvec_call_function_single+0x9d/0xd0
| [ 13.827724] </IRQ>
| [ 13.827724] asm_sysvec_call_function_single+0x12/0x20
| [ 13.827728] RIP: 0010:lock_release+0x178/0x460
| ...
| [ 13.827741] up_write+0x2f/0x1c0
| [ 13.827743] anon_vma_clone+0x158/0x1f0
| [ 13.827749] anon_vma_fork+0x33/0x180
| [ 13.827751] dup_mm+0x45b/0x5f0
| [ 13.827755] copy_process+0x1e5a/0x2050
| [ 13.827758] kernel_clone+0x9b/0x3f0
| [ 13.827760] ? vfs_statx+0x74/0x130
| [ 13.827766] __do_sys_clone+0x60/0x80
| [ 13.827769] do_syscall_64+0x3b/0x90
| [ 13.827771] entry_SYSCALL_64_after_hwframe+0x44/0xae
The problem is rather coming from rt camp where splinlocks
become sleepable thus can't be used in irq context (and for our kernel
it requires the CONFIG_PROVE_RAW_LOCK_NESTING to be set), thus since
we know that we're operating in irq context lets use raw spinlocks
instead.
https://jira.sw.ru/browse/PSBM-134761
Signed-off-by: Cyrill Gorcunov <gorcunov at gmail.com>
Reviewed-by: Kirill Tkhai <ktkhai at virtuozzo.com>
---
arch/x86/kernel/cpu/proc.c | 10 +++++-----
1 file changed, 5 insertions(+), 5 deletions(-)
diff --git a/arch/x86/kernel/cpu/proc.c b/arch/x86/kernel/cpu/proc.c
index dd9d0811e379..a06281fcbd5c 100644
--- a/arch/x86/kernel/cpu/proc.c
+++ b/arch/x86/kernel/cpu/proc.c
@@ -69,7 +69,7 @@ struct cpu_flags {
};
static DEFINE_PER_CPU(struct cpu_flags, cpu_flags);
-static DEFINE_SPINLOCK(cpu_flags_lock);
+static DEFINE_RAW_SPINLOCK(cpu_flags_lock);
static void init_cpu_flags(void *dummy)
{
@@ -107,9 +107,9 @@ static void init_cpu_flags(void *dummy)
flags.val[10] &= eax;
}
- spin_lock(&cpu_flags_lock);
+ raw_spin_lock(&cpu_flags_lock);
memcpy(&per_cpu(cpu_flags, cpu), &flags, sizeof(flags));
- spin_unlock(&cpu_flags_lock);
+ raw_spin_unlock(&cpu_flags_lock);
}
static int show_cpuinfo(struct seq_file *m, void *v)
@@ -158,9 +158,9 @@ static int show_cpuinfo(struct seq_file *m, void *v)
show_cpuinfo_misc(m, c);
if (!is_super) {
- spin_lock_irq(&cpu_flags_lock);
+ raw_spin_lock_irq(&cpu_flags_lock);
memcpy(&ve_flags, &per_cpu(cpu_flags, cpu), sizeof(ve_flags));
- spin_unlock_irq(&cpu_flags_lock);
+ raw_spin_unlock_irq(&cpu_flags_lock);
}
More information about the Devel
mailing list