[<prev] [next>] [<thread-prev] [day] [month] [year] [list]
Message-ID: <c4023681-3899-ad3b-82d8-53cc42312970@linux.intel.com>
Date: Fri, 14 Mar 2025 13:18:07 +0200 (EET)
From: Ilpo Järvinen <ilpo.jarvinen@...ux.intel.com>
To: Lukas Wunner <lukas@...ner.de>
cc: Bjorn Helgaas <bhelgaas@...gle.com>, linux-pci@...r.kernel.org,
Guenter Roeck <groeck@...iper.net>,
Mika Westerberg <mika.westerberg@...ux.intel.com>,
"Rafael J. Wysocki" <rafael.j.wysocki@...el.com>,
Rajat Jain <rajatxjain@...il.com>,
Joel Mathew Thomas <proxy0@...amail.com>,
LKML <linux-kernel@...r.kernel.org>,
Jonathan Cameron <Jonathan.Cameron@...wei.com>
Subject: Re: [PATCH 3/4] PCI/hotplug: reset_lock is not required synchronizing
with irq thread
On Fri, 14 Mar 2025, Lukas Wunner wrote:
> On Thu, Mar 13, 2025 at 04:23:32PM +0200, Ilpo Järvinen wrote:
> > Disabling HPIE (Hot-Plug Interrupt Enable) and synchronizing with irq
> > handling in pciehp_reset_slot() is enough to ensure no pending events
> > are processed during the slot reset. Thus, there is no need to take
> > reset_lock in the IRQ thread.
> [...]
> > --- a/drivers/pci/hotplug/pciehp_hpc.c
> > +++ b/drivers/pci/hotplug/pciehp_hpc.c
> > @@ -748,12 +748,10 @@ static irqreturn_t pciehp_ist(int irq, void *dev_id)
> > * Disable requests have higher priority than Presence Detect Changed
> > * or Data Link Layer State Changed events.
> > */
> > - down_read_nested(&ctrl->reset_lock, ctrl->depth);
> > if (events & DISABLE_SLOT)
> > pciehp_handle_disable_request(ctrl);
> > else if (events & (PCI_EXP_SLTSTA_PDC | PCI_EXP_SLTSTA_DLLSC))
> > pciehp_handle_presence_or_link_change(ctrl, events);
> > - up_read(&ctrl->reset_lock);
> >
> > ret = IRQ_HANDLED;
> > out:
>
> The release and re-acquisition of reset_lock in
> pciehp_configure_device() and pciehp_unconfigure_device()
> needs to be removed as well if the above hunk is applied.
Ah, right. I also now removed reset_lock from
pciehp_ignore_dpc_link_change() that is directly called from pciehp_ist().
This leaves reset_lock only into pciehp_free_irq(), pciehp_reset_slot(),
and pciehp_check_presence(). It seems to me pciehp_check_presence()
requires keeping reset_lock as is. The former two could have synchronized
with a mutex and shorter critical sections in pciehp_reset_slot() but it
doesn't look worth the effort (IMO) to divide reset_lock into two to
realize that.
> But please wait a little while before respinning so that I can
> think through the whole series.
Sure, take your time. :-)
--
i.
Powered by blists - more mailing lists