linux-kernel - Re: [RFC] gpio: about the need to manage irq mapping dynamically.

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] [day] [month] [year] [list]

Date:   Sun, 2 Jul 2017 17:01:11 +0200
From:   Linus Walleij <linus.walleij@...aro.org>
To:     Grygorii Strashko <grygorii.strashko@...com>
Cc:     Jerome Brunet <jbrunet@...libre.com>,
        "linux-gpio@...r.kernel.org" <linux-gpio@...r.kernel.org>,
        Marc Zyngier <marc.zyngier@....com>,
        "open list:ARM/Amlogic Meson..." <linux-amlogic@...ts.infradead.org>,
        Kevin Hilman <khilman@...libre.com>,
        Linux Kernel <linux-kernel@...r.kernel.org>
Subject: Re: [RFC] gpio: about the need to manage irq mapping dynamically.

On Fri, Jun 30, 2017 at 9:54 PM, Grygorii Strashko
<grygorii.strashko@...com> wrote:
> On 06/29/2017 09:16 AM, Linus Walleij wrote:
>> On Tue, Jun 27, 2017 at 10:43 PM, Grygorii Strashko
>> <grygorii.strashko@...com> wrote:
>>
>>> And my opinion is still the same here - It should be perfectly valid to create
>>> mappings from gpio_to_irq() to handle properly orthogonality of gpiochip and
>>> gpio-irqchip functionality and satisfy SPARSE_IRQ goal (allocate Linux virq and
>>> irq descriptors on demand).
>>
>> You are right.
>>
>> I would rather say: GPIO drivers that have a 1-to-1 mapping between GPIO
>> lines and IRQs should not do it, they should map up them all at probe().
>>
>
> Sry, can't completely agree here :( There could be 300 (or even thousands)
> of gpios and only dozen of them will be used as GPIO IRQ, so statical mapping will
> just waste system resources. So, better not define such kind of restrictions -
> it seems platform/system specific.

If there could, yeah.

But the majority of the worlds systems use a hardcoded value of 512 GPIOs.

See include/asm-generic/gpio.h:

#ifndef ARCH_NR_GPIOS
#if defined(CONFIG_ARCH_NR_GPIO) && CONFIG_ARCH_NR_GPIO > 0
#define ARCH_NR_GPIOS CONFIG_ARCH_NR_GPIO
#else
#define ARCH_NR_GPIOS           512
#endif
#endif

The only arch that overrides this is ARM, which has CONFIG_ARCH_NR_GPIO:

# The GPIO number here must be sorted by descending number. In case of
# a multiplatform kernel, we just want the highest value required by the
# selected platforms.
config ARCH_NR_GPIO
        int
        default 1024 if ARCH_BRCMSTB || ARCH_SHMOBILE || ARCH_TEGRA || \
                ARCH_ZYNQ
        default 512 if ARCH_EXYNOS || ARCH_KEYSTONE || SOC_OMAP5 || \
                SOC_DRA7XX || ARCH_S3C24XX || ARCH_S3C64XX || ARCH_S5PV210
        default 416 if ARCH_SUNXI
        default 392 if ARCH_U8500
        default 352 if ARCH_VT8500
        default 288 if ARCH_ROCKCHIP
        default 264 if MACH_H4700
        default 0
        help
          Maximum number of GPIOs in the system.

          If unsure, leave the default value.

So actually, noone use more than 1024 GPIOs.

For each GPIO a descriptor of 16 bytes is allocated. So in worst case 16KiB.

Also this is an upper cap: it just means the nax we have on any platform is 1024
statically allocated GPIO descriptors, not that they all get mapped to IRQs.
That happens on a per-gpiochip basis.

But every irq descriptor is pretty big (somewhere around 64+ bytes), so you have
a point.

If every GPIO allocates an IRQ descriptor, it may add up to as much as 64KiB
on a machine with 1024 GPIOs.

Yours,
Linus Walleij