| Message ID | 20201028145117.1731876-1-geert+renesas@glider.be (mailing list archive) |
|---|---|
| State | New, archived |
| Headers | show |
| Series | pinctrl: Remove hole in pinctrl_gpio_range | expand |
On Wed, Oct 28, 2020 at 3:51 PM Geert Uytterhoeven <geert+renesas@glider.be> wrote: > On 64-bit platforms, pointer size and alignment are 64-bit, hence two > 4-byte holes are present before the pins and gc members of the > pinctrl_gpio_range structure. Get rid of these holes by moving the > pins pointer. > > This reduces kernel size of an arm64 Rockchip kernel by ca. 512 bytes. > > Signed-off-by: Geert Uytterhoeven <geert+renesas@glider.be> > --- > Compile-tested only (arm/multi_v7_defconfig and arm64/defconfig). Patch applied. Do you think it'd be worth it to add a check to checkpatch to suggest to move pointers toward the end of any struct? Yours, Linus Walleij
On 2020-11-05 13:57, Linus Walleij wrote: > On Wed, Oct 28, 2020 at 3:51 PM Geert Uytterhoeven > <geert+renesas@glider.be> wrote: > >> On 64-bit platforms, pointer size and alignment are 64-bit, hence two >> 4-byte holes are present before the pins and gc members of the >> pinctrl_gpio_range structure. Get rid of these holes by moving the >> pins pointer. >> >> This reduces kernel size of an arm64 Rockchip kernel by ca. 512 bytes. >> >> Signed-off-by: Geert Uytterhoeven <geert+renesas@glider.be> >> --- >> Compile-tested only (arm/multi_v7_defconfig and arm64/defconfig). > > Patch applied. > > Do you think it'd be worth it to add a check to checkpatch to suggest > to move pointers toward the end of any struct? For a general rule, I thought that ordering struct members largest-first was the conventional wisdom, since that way no sensible compiler would add padding between any members, only at the end? That said, the trouble with any checkpatch rule is that people will inevitably try to apply it indiscriminately. With structure layout, that could often end up hurting readability and/or performance (via cache effects), while in many cases making no actual difference to the overall size anyway. Robin.
diff --git a/include/linux/pinctrl/pinctrl.h b/include/linux/pinctrl/pinctrl.h index 2aef59df93d70550..70b45d28e7a9293b 100644 --- a/include/linux/pinctrl/pinctrl.h +++ b/include/linux/pinctrl/pinctrl.h @@ -51,8 +51,8 @@ struct pinctrl_pin_desc { * @id: an ID number for the chip in this range * @base: base offset of the GPIO range * @pin_base: base pin number of the GPIO range if pins == NULL - * @pins: enumeration of pins in GPIO range or NULL * @npins: number of pins in the GPIO range, including the base number + * @pins: enumeration of pins in GPIO range or NULL * @gc: an optional pointer to a gpio_chip */ struct pinctrl_gpio_range { @@ -61,8 +61,8 @@ struct pinctrl_gpio_range { unsigned int id; unsigned int base; unsigned int pin_base; - unsigned const *pins; unsigned int npins; + unsigned const *pins; struct gpio_chip *gc; };
On 64-bit platforms, pointer size and alignment are 64-bit, hence two 4-byte holes are present before the pins and gc members of the pinctrl_gpio_range structure. Get rid of these holes by moving the pins pointer. This reduces kernel size of an arm64 Rockchip kernel by ca. 512 bytes. Signed-off-by: Geert Uytterhoeven <geert+renesas@glider.be> --- Compile-tested only (arm/multi_v7_defconfig and arm64/defconfig). --- include/linux/pinctrl/pinctrl.h | 4 ++-- 1 file changed, 2 insertions(+), 2 deletions(-)