[<prev] [next>] [<thread-prev] [day] [month] [year] [list]
Message-ID: <5ac0cc88-fb43-120c-975a-f1ad23ed30fb@suse.com>
Date: Mon, 14 Nov 2022 10:32:48 +0100
From: Jan Beulich <jbeulich@...e.com>
To: Juergen Gross <jgross@...e.com>
Cc: Stefano Stabellini <sstabellini@...nel.org>,
Russell King <linux@...linux.org.uk>,
Boris Ostrovsky <boris.ostrovsky@...cle.com>,
Thomas Gleixner <tglx@...utronix.de>,
Ingo Molnar <mingo@...hat.com>, Borislav Petkov <bp@...en8.de>,
Dave Hansen <dave.hansen@...ux.intel.com>,
"H. Peter Anvin" <hpa@...or.com>,
Oleksandr Tyshchenko <oleksandr_tyshchenko@...m.com>,
xen-devel@...ts.xenproject.org,
linux-arm-kernel@...ts.infradead.org, linux-kernel@...r.kernel.org,
x86@...nel.org
Subject: Re: [PATCH] x86/xen: call xen_set_upcall_vector() only after local
apic setup
On 12.11.2022 14:23, Juergen Gross wrote:
> Calling xen_set_upcall_vector() before the local APIC has been
> initialized results in warnings when setting up the local APIC:
>
> [ 0.037322] APIC: Stale IRR: 00080000, ...
>
> Avoid this problem by moving the call of xen_set_upcall_vector() for
> APs from xen_cpu_up_prepare_hvm() to a new hotplug callback called on
> the AP before interrupts are enabled and after the local APIC has been
> initialized. There is already a Xen specific hotplug state defined at
> a matching position, but right now it is specific to Arm. So rename
> the CPUHP_AP_ARM_XEN_STARTING to CPUHP_AP_XEN_STARTING and use it for
> x86, too.
>
> On the boot processor use the x86_init.irqs.intr_mode_init hook to
> call xen_set_upcall_vector() only after the local APIC initialization.
> As interrupts are on already at the time when the hook is being used,
> modify xen_init_setup_upcall_vector() to disable interrupts. This is
> needed as the xen_percpu_upcall flag is set only after the callback
> has been registered, and the flag is needed for correctly acknowledging
> an interrupt.
>
> Fixes: b1c3497e604d ("x86/xen: Add support for HVMOP_set_evtchn_upcall_vector")
> Reported-by: Jan Beulich <jbeulich@...e.com>
> Signed-off-by: Juergen Gross <jgross@...e.com>
Just to mention it here - as per [1] the issue may instead result from
hypervisor misbehavior, so I guess we want to investigate that route
first.
Jan
[1] https://lists.xen.org/archives/html/xen-devel/2022-11/msg00783.html
Powered by blists - more mailing lists