linux-kernel - Re: [PATCH] KVM: arm/arm64: don't set vtimer->cnt_ctl in kvm_arch_timer

lists.openwall.net		lists / announce owl-users owl-dev john-users john-dev passwdqc-users yescrypt popa3d-users / oss-security kernel-hardening musl sabotage tlsify passwords / crypt-dev xvendor / Bugtraq Full-Disclosure linux-kernel linux-netdev linux-ext4 linux-hardening linux-cve-announce PHC
Open Source and information security mailing list archives

Hash Suite: Windows password security audit tool. GUI, reports in PDF.

[<prev] [next>] [<thread-prev] [thread-next>] [day] [month] [year] [list]

Message-ID: <20171214130954.GV910@cbox>
Date:   Thu, 14 Dec 2017 14:09:54 +0100
From:   Christoffer Dall <christoffer.dall@...aro.org>
To:     Jia He <hejianet@...il.com>
Cc:     Marc Zyngier <marc.zyngier@....com>,
        linux-arm-kernel@...ts.infradead.org, kvmarm@...ts.cs.columbia.edu,
        linux-kernel@...r.kernel.org, Jia He <jia.he@...-semitech.com>
Subject: Re: [PATCH] KVM: arm/arm64: don't set vtimer->cnt_ctl in
 kvm_arch_timer_handler

On Thu, Dec 14, 2017 at 12:57:54PM +0800, Jia He wrote:
Hi Jia,

> 
> I have tried your newer level-mapped-v7 branch, but bug is still there.
> 
> There is no special load in both host and guest. The guest (kernel
> 4.14) is often hanging when booting
> 
> the guest kernel log
> 
> [ OK ] Reached target Remote File Systems.
> Starting File System Check on /dev/mapper/fedora-root...
> [ OK ] Started File System Check on /dev/mapper/fedora-root.
> Mounting /sysroot...
> [ 2.670764] SGI XFS with ACLs, security attributes, no debug enabled
> [ 2.678180] XFS (dm-0): Mounting V5 Filesystem
> [ 2.740364] XFS (dm-0): Ending clean mount
> [ OK ] Mounted /sysroot.
> [ OK ] Reached target Initrd Root File System.
> Starting Reload Configuration from the Real Root...
> [ 61.288215] INFO: rcu_sched detected stalls on CPUs/tasks:
> [ 61.290791] 1-...!: (0 ticks this GP) idle=574/0/0 softirq=5/5 fqs=1
> [ 61.293664] (detected by 0, t=6002 jiffies, g=-263, c=-264, q=39760)
> [ 61.296480] Task dump for CPU 1:
> [ 61.297938] swapper/1 R running task 0 0 1 0x00000020
> [ 61.300643] Call trace:
> [ 61.301260] __switch_to+0x6c/0x78
> [ 61.302095] cpu_number+0x0/0x8
> [ 61.302867] rcu_sched kthread starved for 6000 jiffies!
> g18446744073709551353 c18446744073709551352 f0x0 RCU_GP_WAIT_FQS(3)
> ->state=0x402 ->cpu=1
> [ 61.305941] rcu_sched I 0 8 2 0x00000020
> [ 61.307250] Call trace:
> [ 61.307854] __switch_to+0x6c/0x78
> [ 61.308693] __schedule+0x268/0x8f0
> [ 61.309545] schedule+0x2c/0x88
> [ 61.310325] schedule_timeout+0x84/0x3b8
> [ 61.311278] rcu_gp_kthread+0x4d4/0x7d8
> [ 61.312213] kthread+0x134/0x138
> [ 61.313001] ret_from_fork+0x10/0x1c
> 
> Maybe my previous patch is not perfect enough, thanks for your comments.
> 
> I digged it futher more, do you think below code logic is possibly
> problematic?
> 
> 
> vtimer_save_state           (vtimer->loaded = false, cntv_ctl is 0)
> 
> kvm_arch_timer_handler        (read cntv_ctl and set vtimer->cnt_ctl = 0)
> 
> vtimer_restore_state            (write vtimer->cnt_ctl to cntv_ctl,
> then cntv_ctl will
> 
>                        be 0 forever)
> 
> 
> If above analysis is reasonable

Yes, I think there's something there if the hardware doesn't retire the
signal fast enough...

> how about below patch? already
> tested in my arm64 server.
> 
> diff --git a/virt/kvm/arm/arch_timer.c b/virt/kvm/arm/arch_timer.c
> index f9555b1..ee6dd3f 100644
> --- a/virt/kvm/arm/arch_timer.c
> +++ b/virt/kvm/arm/arch_timer.c
> @@ -99,7 +99,7 @@ static irqreturn_t kvm_arch_timer_handler(int irq,
> void *dev_id)
>         }
>         vtimer = vcpu_vtimer(vcpu);
> 
> -       if (!vtimer->irq.level) {
> +       if (vtimer->loaded && !vtimer->irq.level) {
>                 vtimer->cnt_ctl = read_sysreg_el0(cntv_ctl);
>                 if (kvm_timer_irq_can_fire(vtimer))
>                         kvm_timer_update_irq(vcpu, true, vtimer);
> 

There's nothing really wrong with that patch, I just didn't think it
would be necessary, as we really shouldn't see interrupts if the timer
is not loaded.  Can you confirm that a WARN_ON(!vtimer->loaded) in
kvm_arch_timer_handler() gives you a splat?

Also, could you give the following a try (without your patch):

diff --git a/virt/kvm/arm/arch_timer.c b/virt/kvm/arm/arch_timer.c
index 73d262c4712b..4751255345d1 100644
--- a/virt/kvm/arm/arch_timer.c
+++ b/virt/kvm/arm/arch_timer.c
@@ -367,6 +367,7 @@ static void vtimer_save_state(struct kvm_vcpu *vcpu)
 
 	/* Disable the virtual timer */
 	write_sysreg_el0(0, cntv_ctl);
+	isb();
 
 	vtimer->loaded = false;
 out:

Thanks,
-Christoffer