[<prev] [next>] [<thread-prev] [day] [month] [year] [list]
Message-ID: <alpine.DEB.2.20.1710170004480.2037@nanos>
Date: Tue, 17 Oct 2017 00:06:56 +0200 (CEST)
From: Thomas Gleixner <tglx@...utronix.de>
To: Dexuan Cui <decui@...rosoft.com>
cc: Bjorn Helgaas <bhelgaas@...gle.com>,
KY Srinivasan <kys@...rosoft.com>,
Stephen Hemminger <sthemmin@...rosoft.com>,
"linux-pci@...r.kernel.org" <linux-pci@...r.kernel.org>,
"devel@...uxdriverproject.org" <devel@...uxdriverproject.org>,
"linux-kernel@...r.kernel.org" <linux-kernel@...r.kernel.org>,
Josh Poulson <jopoulso@...rosoft.com>,
Haiyang Zhang <haiyangz@...rosoft.com>,
Simon Xiao <sixiao@...rosoft.com>,
Saeed Mahameed <saeedm@...lanox.com>,
"Mihai Costache (Cloudbase Solutions SRL)" <v-micos@...rosoft.com>,
Jork Loeser <Jork.Loeser@...rosoft.com>
Subject: Re: [PATCH] PCI: hv: fix "spurious APIC interrupt" due to new vector
reservation mode
On Sat, 14 Oct 2017, Dexuan Cui wrote:
> Last month the vector management code was reworked, and as a result of the
> changes, e.g. commit 22d0b12f3560 ("genirq/irqdomain: Add force
> reactivation flag to irq domains"), commit 4900be83602b ("x86/vector/msi:
> Switch to global reservation mode") etc, now we must add this new flag
> MSI_FLAG_MUST_REACTIVATE when calling pci_msi_create_irq_domain() on x86,
> otherwise Hyper-V PCIe pass-through can't work, and we get:
> "spurious APIC interrupt through vector ff on CPU#0, should never happen",
> this is because no valid MSI/MSI-X vector is allocated.
>
> Signed-off-by: Dexuan Cui <decui@...rosoft.com>
> Cc: Thomas Gleixner <tglx@...utronix.de>
> Cc: Bjorn Helgaas <bhelgaas@...gle.com>
> Cc: K. Y. Srinivasan <kys@...rosoft.com>
> Cc: Haiyang Zhang <haiyangz@...rosoft.com>
> Cc: Stephen Hemminger <sthemmin@...rosoft.com>
> ---
>
> Currently the "reservation mode" patchset has not been in Linus's tree yet, so
> today's mainline hasn't shown the issue, but today's linux-next has shown the
> issue (PCIe pass-through on Hyper-V can't work).
>
> I think any caller of pci_msi_create_irq_domain() that runs on x86 needs to
> be updated, but besides pci-hyper I'm not sure which callers run on x86.
I'm sure I thought about adding something to pci_msi_create_irq_domain() in
the first place, but then forgot about it again. I'll have a look tomorrow
morning.
Thanks,
tglx
Powered by blists - more mailing lists