lkml.org 
[lkml]   [2017]   [Jan]   [12]   [last100]   RSS Feed
Views: [wrap][no wrap]   [headers]  [forward] 
 
Messages in this thread
/
From
SubjectRe: [PATCH] power: supply: axp20x_usb_power: fix 64-bit build warning
Date
On Thursday, January 12, 2017 4:53:55 PM CET Michal Suchánek wrote:
> Hello,
>
> On Thu, 12 Jan 2017 09:32:26 +0100
> Arnd Bergmann <arnd@arndb.de> wrote:
>
> > On Thursday, January 12, 2017 3:58:24 AM CET Sebastian Reichel wrote:
> > > Hi Arnd,
> > >
> > > On Wed, Jan 11, 2017 at 03:51:55PM +0100, Arnd Bergmann wrote:
> > > > Casting a pointer to 'int' is not always valid:
> > > >
> > > > drivers/power/supply/axp20x_usb_power.c: In function
> > > > 'axp20x_usb_power_probe':
> > > > drivers/power/supply/axp20x_usb_power.c:297:21: error: cast from
> > > > pointer to integer of different size [-Werror=pointer-to-int-cast]
> > > >
> > > > This makes the code use uintptr_t explicitly.
> > > >
> > > > Fixes: 0dcc70ca8644 ("power: supply: axp20x_usb_power: use
> > > > of_device_id data field instead of device_is_compatible")
> > > > Signed-off-by: Arnd Bergmann <arnd@arndb.de>
> > >
> > > I queued Michal's patch instead:
> > >
> > > https://git.kernel.org/cgit/linux/kernel/git/sre/linux-power-supply.git/commit/?h=for-next&id=15df6d98ec3b40775918fc6ef73d7f1c2d0cf870
> >
> > Hmm, that doesn't look right: You can't really rely on an 'enum' type
> > to have a specific size, especially not the same size as a pointer,
> > in portable code.
> >
> > IIRC on many architectures it defaults to 'int' rather than 'long',
> > and it might also be 'short' on architectures that default to enums
> > being the smallest integer type that fits.
>
> Technically to fit the pointer into an integer you should be using
> uintptr_t or unsigned long. However, gcc does not issue a warning when
> casting to enum either. The reason might possibly be that when casting
> to an enum you make it clear that you are expecting one of the values
> that are part of the enumeration and the value should fit into the enum
> even if its actual size is char.

Interesting. I've confirmed that here, your patch is fine then!

> Either way handling of casts from the match pointer to integer varies
> between drivers and some that are possibly never built on 64bit even use
> int.
>
> drivers/gpu/drm/bridge/adv7511/adv7511_drv.c:
> adv7511->type = (enum adv7511_type)of_device_get_match_data(dev);
> drivers/i2c/busses/i2c-rcar.c: priv->devtype = (enum
> rcar_i2c_type)of_device_get_match_data(dev);
> drivers/net/ethernet/renesas/ravb_main.c: chip_id = (enum
> ravb_chip_id)of_device_get_match_data(&pdev->dev);
> drivers/pci/host/pci-imx6.c: (enum
> imx6_pcie_variants)of_device_get_match_data(dev);
> drivers/reset/hisilicon/hi6220_reset.c: type = (enum
> hi6220_reset_ctrl_type)of_device_get_match_data(dev);
> drivers/usb/phy/phy-msm-usb.c: pdata->phy_type = (enum
> msm_usb_phy_type)of_device_get_match_data(&pdev->dev);
>
> drivers/firmware/qcom_scm.c: clks = (unsigned
> long)of_device_get_match_data(&pdev->dev);
> drivers/gpu/drm/exynos/exynos5433_drm_decon.c: ctx->out_type =
> (unsigned long)of_device_get_match_data(dev);
> drivers/pinctrl/sunxi/pinctrl-sun5i.c: unsigned long variant =
> (unsigned long)of_device_get_match_data(&pdev->dev);
> drivers/spi/spi-sun6i.c: sspi->fifo_depth = (unsigned
> long)of_device_get_match_data(&pdev->dev);
> drivers/thermal/rcar_thermal.c:#define rcar_of_data(dev)
> ((unsigned long)of_device_get_match_data(dev))
> sound/soc/sh/rcar/core.c: priv->flags = (unsigned
> long)of_device_get_match_data(dev);
>
> drivers/spi/spi-mpc512x-psc.c: mps->type =
> (int)of_device_get_match_data(dev);
> drivers/leds/leds-pm8058.c: led->ledtype =
> (u32)of_device_get_match_data(&pdev->dev);
>
>
> So what is the preferred way to do the cast to be portable across Linux
> architectures?

Traditionally, we use a cast to 'unsigned long' in the kernel, though
I tend to use 'uintptr_t' when I do patches because that makes the
intention clearer. Both of these always work on all architectures that
Linux supports. The two last cases you cite that use 'int' or 'u32'
are obviously nonportable.

Arnd

\
 
 \ /
  Last update: 2017-01-12 17:12    [W:0.228 / U:0.576 seconds]
©2003-2020 Jasper Spaans|hosted at Digital Ocean and TransIP|Read the blog|Advertise on this site