pinctrl: Implicit inclusion fallout fixes
Subsystems (11)
cover pinctrl: Implicit inclusion fallout fixes
From: Thierry Reding <[email protected]>
Hi Linus,
these are two patches to fix build errors I've run into that are fallout from your removal of the pinctrl/devinfo.h include from device.h. I sure can imagine that there are more like this, but these seem to be the common ones, judging by my own build tests and the kernel-build-reports mailing list.
I'm sending these to you directly because I think it makes sense for you to apply these on top of 23c35f48f5fb ("pinctrl: remove include file from <linux/device.h>") rather than have these go in through their respective maintainer trees, so that the build gets fixed as quickly as possible.
Thierry
Thierry Reding (2):
drm/rockchip: lvds: Explicitly include pinctrl headers
mmc: meson-gx-mmc: Explicitly include pinctr/consumer.h
drivers/gpu/drm/rockchip/rockchip_lvds.c | 2 ++ drivers/mmc/host/meson-gx-mmc.c | 1 + 2 files changed, 3 insertions(+)
Signature · 1 line
2.15.1
PATCH 1/2 drm/rockchip: lvds: Explicitly include pinctrl headers
From: Thierry Reding <[email protected]>
The Rockchip LVDS driver fails to build after commit 23c35f48f5fb ("pinctrl: remove include file from <linux/device.h>") because it relies on the pinctrl/consumer.h and pinctrl/devinfo.h being pulled in by the device.h header implicitly.
Include these headers explicitly to avoid the build failure.
--- drivers/gpu/drm/rockchip/rockchip_lvds.c | 2 ++ 1 file changed, 2 insertions(+)
drivers/gpu/drm/rockchip/rockchip_lvds.c +2 -0
@@ -25,6 +25,8 @@#include <linux/clk.h> #include <linux/mfd/syscon.h> #include <linux/of_graph.h> +#include <linux/pinctrl/consumer.h> +#include <linux/pinctrl/devinfo.h> #include <linux/pm_runtime.h> #include <linux/regmap.h> #include <linux/reset.h>
Signature · 1 line
2.15.1
PATCH 2/2 mmc: meson-gx-mmc: Explicitly include pinctr/consumer.h
In reply to Thierry RedingFrom: Thierry Reding <[email protected]>
The Meson GX MMC driver fails to build after commit 23c35f48f5fb ("pinctrl: remove include file from <linux/device.h>") because it relies on the pinctrl/consumer.h being pulled in by the device.h header implicitly.
Include the header explicitly to avoid the build failure.
--- drivers/mmc/host/meson-gx-mmc.c | 1 + 1 file changed, 1 insertion(+)
drivers/mmc/host/meson-gx-mmc.c +1 -0
@@ -37,6 +37,7 @@#include <linux/regulator/consumer.h> #include <linux/interrupt.h> #include <linux/bitfield.h> +#include <linux/pinctrl/consumer.h> #define DRIVER_NAME "meson-gx-mmc"
Signature · 1 line
2.15.1
PATCH 1/1 net: mediatek: Explicitly include pinctrl headers
In reply to Thierry RedingFrom: Thierry Reding <[email protected]>
The Mediatek ethernet driver fails to build after commit 23c35f48f5fb ("pinctrl: remove include file from <linux/device.h>") because it relies on the pinctrl/consumer.h and pinctrl/devinfo.h being pulled in by the device.h header implicitly.
Include these headers explicitly to avoid the build failure.
--- drivers/net/ethernet/mediatek/mtk_eth_soc.c | 2 ++ 1 file changed, 2 insertions(+)
drivers/net/ethernet/mediatek/mtk_eth_soc.c +2 -0
@@ -23,6 +23,8 @@#include <linux/reset.h> #include <linux/tcp.h> #include <linux/interrupt.h> +#include <linux/pinctrl/consumer.h> +#include <linux/pinctrl/devinfo.h> #include "mtk_eth_soc.h"
Signature · 1 line
2.15.1
Re: [PATCH] net: mediatek: Explicitly include pinctrl headers
Quoted 2 lines · Thierry Reding
Include these headers explicitly to avoid the build failure.
I don't think you need to include *both*.
<linux/device.h> used to include just <linux/pinctrl/devinfo.h>.
I'll edit your patches to include just that. <linux/pinctrl/consumer.h> will come in automatically through it.
Linus
Re: [PATCH] net: mediatek: Explicitly include pinctrl headers
Quoted 10 lines · Linus Torvalds
Quoted 2 lines · Thierry Reding
Include these headers explicitly to avoid the build failure.
I don't think you need to include *both*.
<
linux/device.h> used to include just <linux/pinctrl/devinfo.h>.I'll edit your patches to include just that. <
linux/pinctrl/consumer.h> will come in automatically through it.
I was trying to avoid any implicit inclusion, but looking at pinctrl/devinfo.h it has a comment right above the pinctrl/consumer.h include that makes it clear that pinctrl/devinfo.h is the consumer of pinctrl for the core, so I guess the implicit include is fine here.
I do question, though, if drivers have any business including this pinctrl/devinfo.h in the first place. For the Mediatek ethernet it seems like selecting the default state is redundant (the core should already have taken care of that, and the driver never selects a different state anywhere).
The same is true of drm/rockchip, which also only seems to select a state which the pinctrl core should've selected by default already. See arch/arm/boot/dts/rk3288.dtsi which sets up the "lcdc" state as the only state for the LVDS output.
Anyway, I think going with the pinctrl/devinfo.h include only is fine for now. If it turns out that the Mediatek ethernet and Rockchip LVDS drivers can just omit the bits fiddling with struct dev_pin_info, we can swap out the pinctrl/devinfo.h include for pinctrl/consumer.h at that time.
LinusW: what are your thoughts on the struct dev_pin_info usage by these drivers? Does their code seem redundant to you, too?
Thierry
Re: [PATCH] net: mediatek: Explicitly include pinctrl headers
Quoted 8 lines · Thierry Reding
Anyway, I think going with the
pinctrl/devinfo.hinclude only is fine for now. If it turns out that the Mediatek ethernet and Rockchip LVDS drivers can just omit the bits fiddling with structdev_pin_info, we can swap out thepinctrl/devinfo.hinclude forpinctrl/consumer.hat that time.LinusW: what are your thoughts on the struct
dev_pin_infousage by these drivers? Does their code seem redundant to you, too?
I don't think they should use struct dev_pin_info at all, that thing is for the device core only.
I like to think that <linux/pinctrl/consumer.h> is for drivers that explicitly grab and control pin control states so this driver should only include <linux/pinctrl/consumer.h>.
Torvalds: can you do it like that instead? Either way will make the compile work again, we can also tidy it up later. (i.e. I will grep for includes of pinctrl/dev_info.h and replace it with consumer.h) at some point.
It wasn't pretty that these drivers were relying on implicit #includes in the first place so either is anyway prettier than what it used to look like.
In a way I kind of like this sideffect even if it broke some compiles.
Yours,
Linus Walleij
Re: [PATCH] net: mediatek: Explicitly include pinctrl headers
Quoted 14 lines · Linus Walleij
Quoted 8 lines · Thierry Reding
Anyway, I think going with the
pinctrl/devinfo.hinclude only is fine for now. If it turns out that the Mediatek ethernet and Rockchip LVDS drivers can just omit the bits fiddling with structdev_pin_info, we can swap out thepinctrl/devinfo.hinclude forpinctrl/consumer.hat that time.LinusW: what are your thoughts on the struct
dev_pin_infousage by these drivers? Does their code seem redundant to you, too?I don't think they should use struct
dev_pin_infoat all, that thing is for the device core only.
Quoted 8 lines
I like to think that <
linux/pinctrl/consumer.h> is for drivers that explicitly grab and control pin control states so this driver should only include <linux/pinctrl/consumer.h>.Torvalds: can you do it like that instead? Either way will make the compile work again, we can also tidy it up later. (i.e. I will grep for includes of
pinctrl/dev_info.hand replace it withconsumer.h) at some point.
That won't work, unfortunately, because these drivers actually try to dereference pointers to struct dev_pin_info and hence need that include for the structure's definition. Those are the only two drivers I can see that access this structure directly (other than the device core).
Thierry
Re: [PATCH] net: mediatek: Explicitly include pinctrl headers
Quoted 24 lines · Thierry Reding
Quoted 14 lines · Linus Walleij
Quoted 8 lines · Thierry Reding
Anyway, I think going with the
pinctrl/devinfo.hinclude only is fine for now. If it turns out that the Mediatek ethernet and Rockchip LVDS drivers can just omit the bits fiddling with structdev_pin_info, we can swap out thepinctrl/devinfo.hinclude forpinctrl/consumer.hat that time.LinusW: what are your thoughts on the struct
dev_pin_infousage by these drivers? Does their code seem redundant to you, too?I don't think they should use struct
dev_pin_infoat all, that thing is for the device core only.Quoted 8 lines
I like to think that <
linux/pinctrl/consumer.h> is for drivers that explicitly grab and control pin control states so this driver should only include <linux/pinctrl/consumer.h>.Torvalds: can you do it like that instead? Either way will make the compile work again, we can also tidy it up later. (i.e. I will grep for includes of
pinctrl/dev_info.hand replace it withconsumer.h) at some point.
Quoted 4 lines
That won't work, unfortunately, because these drivers actually try to dereference pointers to struct
dev_pin_infoand hence need that include for the structure's definition. Those are the only two drivers I can see that access this structure directly (other than the device core).
Grrr OK I thought I was aiming for encapsulation here but then what can I do, sigh.
Yours,
Linus Walleij