linux-kernel - RE: [PATCH v2] gpiolib: Fix GPIO chip IRQ initialization restriction

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]

Date:   Fri, 16 Jun 2023 10:11:43 +0800
From:   Jiawen Wu <jiawenwu@...stnetic.com>
To:     "'Andy Shevchenko'" <andy.shevchenko@...il.com>,
        "'Michael Walle'" <michael@...le.cc>
Cc:     <linus.walleij@...aro.org>, <brgl@...ev.pl>,
        <shreeya.patel@...labora.com>, <linux-gpio@...r.kernel.org>,
        <linux-kernel@...r.kernel.org>
Subject: RE: [PATCH v2] gpiolib: Fix GPIO chip IRQ initialization restriction

On Thursday, June 15, 2023 9:56 PM, Andy Shevchenko wrote:
> On Thu, Jun 15, 2023 at 1:45 PM Michael Walle <michael@...le.cc> wrote:
> > > BTW, I wonder if it has problems when unregistering gpio-regmap.
> > > Call Trace of irq_domain_remove() always exits in my test:
> > > https://lore.kernel.org/all/011c01d98d3d$99e6c6e0$cdb454a0$@trustnetic.com/
> > >
> > > Of course, it could be because there was something wrong with my
> > > test code. But I want to be clear about this.
> >
> > Mh, you've said you don't use the devm_ variant of
> > regmap_add_irq_chip(),
> > correct? Do you call regmap_del_irq_chip() yourself?

Yes, devm_regmap_del_irq_chip() also led to a call trace. I thought it
might be the order of release, so I called it myself without devm_.

> > It seems that gpiolib is already removing the domain itself. Mh.
> > I guess if the the domain is set via gpiochip_irqchip_add_domain()
> > gpiolib must not call irq_domain_remove() because the domain resource
> > is handled externally (i.e. gpiolib doesn't allocate the domain
> > itself) in our case.
> >
> > Nice finding! Looks like it has been broken since the beginning
> > when I've introduced the gpiochip_irqchip_add_domain(). Will you
> > do another fixes patch for that? 

I used to be rough at fixing in my test, I tried to set gc->irq.domain = NULL
after calling irq_domain_remove() in gpiochip_irqchip_remove(). But
there seemed to be some other issue that was causing my device to not
work, so I didn't go further. I wonder what risks such fix introduces.

Sorry I may not be able to do the fix patch for a while. I'm working on
other patches, this test will disrupt my work.

> > I'm not sure where to store
> > that information though. Maybe a new bool "no_domain_free"
> > in struct gpio_irq_chip?
> 
> While reading this I also thought about flag, but please use positive
> notation, e.g. "irq_domain_is_ext".