RE: [PATCH v3 6/6] PCI: hv: fix 2 hang issues in hv_compose_msi_msg()

[Dexuan Cui]

> From: Dexuan Cui
> Sent: Wednesday, March 7, 2018 13:40
> To: Lorenzo Pieralisi <lorenzo.pieralisi@xxxxxxx>
> Cc: bhelgaas@xxxxxxxxxx; linux-pci@xxxxxxxxxxxxxxx; KY Srinivasan
> <kys@xxxxxxxxxxxxx>; Stephen Hemminger <sthemmin@xxxxxxxxxxxxx>;
> olaf@xxxxxxxxx; apw@xxxxxxxxxxxxx; jasowang@xxxxxxxxxx; linux-
> kernel@xxxxxxxxxxxxxxx; driverdev-devel@xxxxxxxxxxxxxxxxxxxxxx; Haiyang
> Zhang <haiyangz@xxxxxxxxxxxxx>; vkuznets@xxxxxxxxxx;
> marcelo.cerri@xxxxxxxxxxxxx; Michael Kelley (EOSG)
> <Michael.H.Kelley@xxxxxxxxxxxxx>; stable@xxxxxxxxxxxxxxx; Jack
> Morgenstein <jackm@xxxxxxxxxxxx>
> Subject: RE: [PATCH v3 6/6] PCI: hv: fix 2 hang issues in hv_compose_msi_msg()
> 
> > From: Lorenzo Pieralisi <lorenzo.pieralisi@xxxxxxx>
> > Sent: Wednesday, March 7, 2018 04:35
> > On Tue, Mar 06, 2018 at 06:21:56PM +0000, Dexuan Cui wrote:
> > > 1. With the patch "x86/vector/msi: Switch to global reservation mode"
> > > (4900be8360), the recent v4.15 and newer kernels always hang for 1-vCPU
> > > Hyper-V VM with SR-IOV. This is because when we reach
> > hv_compose_msi_msg()
> > > by request_irq()  -> request_threaded_irq() -> __setup_irq()->irq_startup()
> > >  -> __irq_startup() -> irq_domain_activate_irq() -> ... ->
> > > msi_domain_activate() -> ... -> hv_compose_msi_msg(), local irq is
> > > disabled in __setup_irq().
> > >
> > > Fix this by polling the channel.
> > >
> > > 2. If the host is ejecting the VF device before we reach
> > > hv_compose_msi_msg(), in a UP VM, we can hang in
> hv_compose_msi_msg()
> > > forever, because at this time the host doesn't respond to the
> > > CREATE_INTERRUPT request. This issue also happens to old kernels like
> > > v4.14, v4.13, etc.
> >
> > If you are fixing a problem you should report what commit you are fixing
> > with a Fixes: tag and add a CC: stable@xxxxxxxxxxxxxxx to the commit log
> > to send it to stable kernels to which it should be applied; mentioning
> > kernel versions in the commit log is useless and should be omitted.
> 
> Hi Lorenzo,
> Thanks for your comments!
> This patch does have a "Cc: stable@xxxxxxxxxxxxxxx" in the sign-off area. :-)
> 
> Here the patch is made to resolve 2 issues:
> #1 is triggered by the x86 global reservation mode (4900be8360) patch.
> 4900be8360 in itself is good. It's just that drivers/pci/host/pci-hyperv.c
> should be fixed.
> 
> #2 is a longstanding issue since the first day the pci-hyperv driver was
> accepted into the kernel.
> 
> So IMO actually we don't really need to add a Fixes: tag, which is usually
> used to specify a specific commit that introduces a bug that is being fixed.
> 
> > Side note: you should not have stable@xxxxxxxxxxxxxxx in the email
> > addresses CC list you are sending the patches to (you mark patches for
> > stable by adding an appropriate CC tag in the commit log).
> 
> Sorry, I didn't know this, but actually I didn't add stable@xxxxxxxxxxxxxxx
> manually. Instead I used "git send-email" to send this patchset, and it told
> me "The Cc list above has been expanded by additional addresses found
> in the patch commit message."
> 
> I didn't find a way to disable this behavior of "git send-email" by checking
> its manual and googling it. This is strange.
> 
> > Here:
> >
> > git.kernel.org/.../Documentation/process/stable-kernel-rules.rst
> >
> > Last but not least, most of the patches in this series do not justify
> > sending them to stable kernels at all so you should remove the
> > corresponding tag from the patches.
> 
> I hope at least these 2 patches can go into the stable kernels:
> [PATCH v3 3/6] PCI: hv: serialize the present/eject work items
> [PATCH v3 6/6] PCI: hv: fix 2 hang issues in hv_compose_msi_msg()
> Especially the second one, which fixes a real hang issue for UP virtual
> machines running v4.15 and newer.
> And,  IMO the patches are small enough (<100 lines) , but definitely
> the maintainers make the final call.
> 
> >
> > Thanks,
> > Lorenzo
> >
> > > Fix this by polling the channel for the PCI_EJECT message and
> > > hpdev->state, and by checking the PCI vendor ID.
> > >
> > > Note: actually the above issues also happen to a SMP VM, if
> > > "hbus->hdev->channel->target_cpu == smp_processor_id()" is true.
> > >
> > > Signed-off-by: Dexuan Cui <decui@xxxxxxxxxxxxx>
> > > Tested-by: Adrian Suhov <v-adsuho@xxxxxxxxxxxxx>
> > > Tested-by: Chris Valean <v-chvale@xxxxxxxxxxxxx>
> > > Cc: stable@xxxxxxxxxxxxxxx
> > > Cc: Stephen Hemminger <sthemmin@xxxxxxxxxxxxx>
> > > Cc: K. Y. Srinivasan <kys@xxxxxxxxxxxxx>
> > > Cc: Vitaly Kuznetsov <vkuznets@xxxxxxxxxx>
> > > Cc: Jack Morgenstein <jackm@xxxxxxxxxxxx>
> > > ---
> > >  drivers/pci/host/pci-hyperv.c | 58
> 
> 
> Thanks,
> -- Dexuan

Hi Lorenzo, Bjorn, and all,
Do you need more ACKs? Currently Michael and Haiyang reviewed and ack'd 
the patchset.

Should I send a v4 that just removes the "CC: stable@xxxxxxxxxxxxxxx" tag
for patches 1, 2, 4 and 5? I tend to avoid a v4 as I supppose it would be 
easier if you just remove the tags if you belive it's necessary (IMHO all the
6 paches are not big and it would be great if we can have all of them in 
the old stable kernels, but I respect your decision).

Please let me know if I missed something when addressing the comments,
 and if I should send a v4.

Thanks!
-- Dexuan