platform: allow ATOM PMC code to be optional
Subsystems (4)
cover platform: allow ATOM PMC code to be optional
A few months back I was doing a test build for "defconfig-like" COTS hardware and happened to notice some pmc-atom stuff being built. Fine, I thought - the defconfig isn't minimal - we all know that - what Kconfig do I use to turn that off? Well, imagine my surprise when I found out you couldn't turn it [Atom Power Management Controller] code off!
Normally we all agree to not use "default y" unless unavoidable, but somehow this was added as "def_bool y" which is even worse ; you can see the Kconfig setting but there is *no* way you can turn it off.
After investigating, I found no reason why the atom code couldn't be opt-out with just minor changes that the original addition overlooked.
And so this series addreses that. I tried to be sensitive to not break any existing configs in the process, but the defconfig will now intentionally be different and exclude this atom specific code.
Using a defconfig on today's linux-next, here is the delta summary:
$ ./scripts/bloat-o-meter -c ../pmc-atom-pre/vmlinux ../pmc-atom-post/vmlinux|grep add/remove add/remove: 0/410 grow/shrink: 0/7 up/down: 0/-47659 (-47659) add/remove: 0/105 grow/shrink: 0/1 up/down: 0/-6848 (-6848) add/remove: 0/56 grow/shrink: 0/1 up/down: 0/-10155 (-10155)
$ ./scripts/bloat-o-meter -c ../pmc-atom-pre/vmlinux ../pmc-atom-post/vmlinux|grep Total Total: Before=13626994, After=13579335, chg -0.35% Total: Before=3572137, After=3565289, chg -0.19% Total: Before=1235335, After=1225180, chg -0.82%
It is hard to reclaim anything against the inevitable growth in footprint, so I'd say we should be glad to take whatever we can.
Boot tested defconfig on today's linux-next on older non-atom COTS.
Paul.
---
Paul Gortmaker (4):
platform/x86: pmc_atom: remove unused pmc_atom_write()
ACPI: LPSS: make the Kconfig dependency on PMC_ATOM explicit platform/x86: pmc_atom: dont export pmc_atom_read - no modular users platform/x86: pmc_atom: make the PMC driver actually unselectable
arch/x86/Kconfig | 1 + drivers/platform/x86/Kconfig | 11 ++++++++--- drivers/platform/x86/pmc_atom.c | 13 ------------- include/linux/platform_data/x86/pmc_atom.h | 1 - 4 files changed, 9 insertions(+), 17 deletions(-)
Signature · 1 line
2.17.1
PATCH 1/4 platform/x86: pmc_atom: remove unused pmc_atom_write()
This function isn't used anywhere in the driver or anywhere in tree. So remove it. It can always be re-added if/when a use arises.
--- drivers/platform/x86/pmc_atom.c | 12 ------------ include/linux/platform_data/x86/pmc_atom.h | 1 - 2 files changed, 13 deletions(-)
drivers/platform/x86/pmc_atom.c +0 -12
@@ -223,18 +223,6 @@ int pmc_atom_read(int offset, u32 *value) }EXPORT_SYMBOL_GPL (pmc_atom_read); -int pmc_atom_write (int offset,u32 value) -{ -struct pmc_dev *pmc = &pmc_device; - -if (!pmc->init) -return -ENODEV ; - -pmc_reg_write (pmc, offset, value); -return 0 ; -} -EXPORT_SYMBOL_GPL (pmc_atom_write); -static void pmc_power_off (void ) {u16 pm1_cnt_port;
include/linux/platform_data/x86/pmc_atom.h +0 -1
@@ -144,6 +144,5 @@#define SLEEP_ENABLE 0x2000extern int pmc_atom_read (int offset,u32 *value); -extern int pmc_atom_write (int offset,u32 value);#endif /* PMC_ATOM_H */
Signature · 1 line
2.17.1
Re: [PATCH 1/4] platform/x86: pmc_atom: remove unused pmc_atom_write()
Quoted 2 lines · Paul Gortmaker
This function isn't used anywhere in the driver or anywhere in tree. So remove it. It can always be re-added if/when a use arises.
Can we first move the driver under intel subfolder and then update?
I have no strong opinion, but personally I would keep it (even unexported) just for the sake of consistency. Up to Mark and Hans.
Signature · 2 lines
With Best Regards, Andy Shevchenko
Re: [PATCH 1/4] platform/x86: pmc_atom: remove unused pmc_atom_write()
In reply to Paul GortmakerHi,
Quoted 9 lines · Paul Gortmaker
This function isn't used anywhere in the driver or anywhere in tree. So remove it. It can always be re-added if/when a use arises.
Cc Andy Shevchenko <[email protected]>Cc Aubrey Li <[email protected]>Cc Hans de Goede <[email protected]>Cc Mark Gross <[email protected]>Signed-off-by Paul Gortmaker <[email protected]>
Thank you for your patch, I've applied this patch to my review-hans branch:
https://git.kernel.org/pub/scm/linux/kernel/git/pdx86/platform-drivers-x86.git/log/?h=review-hans
Note it will show up in my review-hans branch once I've pushed my local branch there, which might take a while.
Once I've run some tests on this branch the patches there will be added to the platform-drivers-x86/for-next branch and eventually will be included in the pdx86 pull-request to Linus for the next merge-window.
Regards,
Hans
Quoted 39 lines
--- drivers/platform/x86/pmc_atom.c | 12 ------------ include/linux/platform_data/x86/pmc_atom.h | 1 - 2 files changed, 13 deletions(-)drivers/platform/x86/pmc_atom.c +0
-12@@ -223,18 +223,6 @@ int pmc_atom_read(int offset, u32 *value) }EXPORT_SYMBOL_GPL (pmc_atom_read); -int pmc_atom_write (int offset,u32 value) -{ -struct pmc_dev *pmc = &pmc_device; - -if (!pmc->init) -return -ENODEV ; - -pmc_reg_write (pmc, offset, value); -return 0 ; -} -EXPORT_SYMBOL_GPL (pmc_atom_write); -static void pmc_power_off (void ) {u16 pm1_cnt_port;include/linux/platform_data/x86/pmc_atom.h +0
-1@@ -144,6 +144,5 @@#define SLEEP_ENABLE 0x2000extern int pmc_atom_read (int offset,u32 *value); -extern int pmc_atom_write (int offset,u32 value);#endif /* PMC_ATOM_H */
PATCH 2/4 ACPI: LPSS: make the Kconfig dependency on PMC_ATOM explicit
In reply to Paul GortmakerThe code in acpi_lpss.c has been unconditionally using pmc_atom_read() for about the past six years. This hasn't been a problem since you currently can't disable PMC_ATOM short of disabling PCI entirely.
But it doesn't need to be that way, and so that we can change Kconfigs in a subsequent commit, we make sure LPSS selects PMC_ATOM in advance, so that existing .config files can live on with "make oldconfig".
In theory, one could make LPSS build w/o PMC_ATOM, similar to what it did six years ago, but I doubt there is any demand for that now.
--- arch/x86/Kconfig | 1 + 1 file changed, 1 insertion(+)
arch/x86/Kconfig +1 -0
@@ -654,6 +654,7 @@ config X86_INTEL_LPSS select COMMON_CLK select PINCTRL select IOSF_MBI + select PMC_ATOM help Select to build support for Intel Low Power Subsystem such as found on Intel Lynxpoint PCH. Selecting this option enables
Signature · 1 line
2.17.1
Re: [PATCH 2/4] ACPI: LPSS: make the Kconfig dependency on PMC_ATOM explicit
Quoted 10 lines · Paul Gortmaker
The code in
acpi_lpss.chas been unconditionally usingpmc_atom_read()for about the past six years. This hasn't been a problem since you currently can't disablePMC_ATOMshort of disabling PCI entirely.But it doesn't need to be that way, and so that we can change Kconfigs in a subsequent commit, we make sure LPSS selects
PMC_ATOMin advance, so that existing .config files can live on with "make oldconfig".In theory, one could make LPSS build w/o
PMC_ATOM, similar to what it did six years ago, but I doubt there is any demand for that now.
I'm wondering if without LPSS we may use those devices to some extend. If it's the case, this patch is half-baked since it's missed the network and audio drivers to also enable it (see PMC clock driver registration and usage).
That said, I'm not sure it's beneficial to spread this selection over the drivers that may be used widely on non-Intel-Atom platforms.
Signature · 2 lines
With Best Regards, Andy Shevchenko
Re: [PATCH 2/4] ACPI: LPSS: make the Kconfig dependency on PMC_ATOM explicit
In reply to Paul GortmakerQuoted 11 lines · Paul Gortmaker
The code in
acpi_lpss.chas been unconditionally usingpmc_atom_read()for about the past six years. This hasn't been a problem since you currently can't disablePMC_ATOMshort of disabling PCI entirely.But it doesn't need to be that way, and so that we can change Kconfigs in a subsequent commit, we make sure LPSS selects
PMC_ATOMin advance, so that existing .config files can live on with "make oldconfig".In theory, one could make LPSS build w/o
PMC_ATOM, similar to what it did six years ago, but I doubt there is any demand for that now.
You probably are right and it will get some more build test coverage with respect to the other option, so
Quoted 25 lines
Cc Andy Shevchenko <[email protected]>Cc Aubrey Li <[email protected]>Cc "Rafael J. Wysocki" <[email protected]>Cc Len Brown <[email protected]>Signed-off-by Paul Gortmaker <[email protected]>--- arch/x86/Kconfig | 1 + 1 file changed, 1 insertion(+)arch/x86/Kconfig +1
-1@@ -654,6 +654,7 @@ config X86_INTEL_LPSS select COMMON_CLK select PINCTRL select IOSF_MBI + select PMC_ATOM help Select to build support for Intel Low Power Subsystem such as found on Intel Lynxpoint PCH. Selecting this option enables --2.17.1
PATCH 3/4 platform/x86: pmc_atom: dont export pmc_atom_read - no modular users
In reply to Paul GortmakerThere is only one user of pmc_atom_read in tree, and that is in drivers/acpi/acpi_lpss.c -- which can't be anything but built-in.
As such there is no point in adding this function to the global symbol list exported to modules.
Note that there is no <linux/export.h> include removal since the code was getting that header implicitly.
--- drivers/platform/x86/pmc_atom.c | 1 - 1 file changed, 1 deletion(-)
drivers/platform/x86/pmc_atom.c +0 -1
@@ -221,7 +221,6 @@ int pmc_atom_read(int offset, u32 *value) *value =pmc_reg_read (pmc, offset);return 0 ; } -EXPORT_SYMBOL_GPL (pmc_atom_read);static void pmc_power_off (void ) {
Signature · 1 line
2.17.1
Re: [PATCH 3/4] platform/x86: pmc_atom: dont export pmc_atom_read - no modular users
Hi,
Quoted 9 lines · Paul Gortmaker
There is only one user of
pmc_atom_readin tree, and that is indrivers/acpi/acpi_lpss.c-- which can't be anything but built-in.As such there is no point in adding this function to the global symbol list exported to modules.
Note that there is no <
linux/export.h> include removal since the code was getting that header implicitly.
Quoted 6 lines
Cc Andy Shevchenko <[email protected]>Cc Aubrey Li <[email protected]>Cc Hans de Goede <[email protected]>Cc Mark Gross <[email protected]>Signed-off-by Paul Gortmaker <[email protected]>
Thank you for your patch, I've applied this patch to my review-hans branch:
https://git.kernel.org/pub/scm/linux/kernel/git/pdx86/platform-drivers-x86.git/log/?h=review-hans
Note it will show up in my review-hans branch once I've pushed my local branch there, which might take a while.
Once I've run some tests on this branch the patches there will be added to the platform-drivers-x86/for-next branch and eventually will be included in the pdx86 pull-request to Linus for the next merge-window.
Regards,
Hans
Quoted 16 lines
--- drivers/platform/x86/pmc_atom.c | 1 - 1 file changed, 1 deletion(-)drivers/platform/x86/pmc_atom.c +0
-1@@ -221,7 +221,6 @@ int pmc_atom_read(int offset, u32 *value) *value =pmc_reg_read (pmc, offset);return 0 ; } -EXPORT_SYMBOL_GPL (pmc_atom_read);static void pmc_power_off (void ) {
PATCH 4/4 platform/x86: pmc_atom: make the PMC driver actually unselectable
In reply to Paul GortmakerThis caught my eye when I saw it was def_bool and hence largely pointless to have a Kconfig for it at all.
Yet there is no reason why you shouldn't be able to opt out of Atom platform support if you only care about desktop and server class CPUs.
It was introduced as def_bool, but there is no obvious reason as to why it was forcibly built-in for everyone, other than LPSS implicitly relying on it (which is now fixed). So here we fix up the Kconfig and open the door for people to opt out.
Since putting "default y" on anything that isn't absolutely essential is generally frowned upon, I made the default be CONFIG_MATOM. People who use "make oldconfig" or similar won't notice any difference.
The two "unchanged" lines for PCI and COMMON_CLK appear in the diff from fixing a whitespace issue that somehow managed to live on despite being moved between two different Kconfig files since its introduction.
--- drivers/platform/x86/Kconfig | 11 ++++++++--- 1 file changed, 8 insertions(+), 3 deletions(-)
drivers/platform/x86/Kconfig +8 -3
@@ -1163,6 +1163,11 @@ config WINMATE_FM07_KEYS endif # X86_PLATFORM_DEVICES config PMC_ATOM - def_bool y - depends on PCI - select COMMON_CLK + bool "Intel Atom SOC Power Management Controller driver" + default MATOM + depends on PCI + select COMMON_CLK + help + This enables support for the Atom SOC Power Management Controller + driver, and associated platform clocks. If you intend to boot this + kernel on platforms with an intel Atom CPU, say Y here.
Signature · 1 line
2.17.1
Re: [PATCH 4/4] platform/x86: pmc_atom: make the PMC driver actually unselectable
Quoted 20 lines · Paul Gortmaker
This caught my eye when I saw it was
def_booland hence largely pointless to have aKconfigfor it at all.Yet there is no reason why you shouldn't be able to opt out of Atom platform support if you only care about desktop and server class CPUs.
It was introduced as
def_bool, but there is no obvious reason as to why it was forcibly built-in for everyone, other than LPSS implicitly relying on it (which is now fixed). So here we fix up theKconfigand open the door for people to opt out.Since putting "default y" on anything that isn't absolutely essential is generally frowned upon, I made the default be
CONFIG_MATOM. People who use "make oldconfig" or similar won't notice any difference.The two "unchanged" lines for PCI and
COMMON_CLKappear in the diff from fixing a whitespace issue that somehow managed to live on despite being moved between two differentKconfigfiles since its introduction.config
PMC_ATOM
drivers/platform/x86/Kconfig config WINMATE_FM07_KEYS · line 1166
Quoted 4 lines
drivers/platform/x86/Kconfig +1
-3- def_bool y - depends on PCI - select COMMON_CLK + bool "Intel Atom SOC Power Management Controller driver"
s/SOC/SoC/ here and everywhere else in this patch.
drivers/platform/x86/Kconfig config WINMATE_FM07_KEYS · line 1172
Quoted 6 lines
drivers/platform/x86/Kconfig +6
-0+ default MATOM + depends on PCI + select COMMON_CLK + help + This enables support for the Atom SOC Power Management Controller + driver, and associated platform clocks. If you intend to boot this
One space is enough.
drivers/platform/x86/Kconfig config WINMATE_FM07_KEYS · line 1173
Quoted 1 line
drivers/platform/x86/Kconfig +1
-0+ kernel on platforms with an intel Atom CPU, say Y here.
Please, list the Atoms that really need this, for example, Broxton doesn't.
I believe the list should include (but not limited to):
Intel Bay Trail
Intel Braswell
Intel Cherry Trail
Signature · 2 lines
With Best Regards, Andy Shevchenko
Re: [PATCH 4/4] platform/x86: pmc_atom: make the PMC driver actually unselectable
In reply to Paul GortmakerHi,
Quoted 34 lines · Paul Gortmaker
This caught my eye when I saw it was
def_booland hence largely pointless to have aKconfigfor it at all.Yet there is no reason why you shouldn't be able to opt out of Atom platform support if you only care about desktop and server class CPUs.
It was introduced as
def_bool, but there is no obvious reason as to why it was forcibly built-in for everyone, other than LPSS implicitly relying on it (which is now fixed). So here we fix up theKconfigand open the door for people to opt out.Since putting "default y" on anything that isn't absolutely essential is generally frowned upon, I made the default be
CONFIG_MATOM. People who use "make oldconfig" or similar won't notice any difference.The two "unchanged" lines for PCI and
COMMON_CLKappear in the diff from fixing a whitespace issue that somehow managed to live on despite being moved between two differentKconfigfiles since its introduction.Cc Aubrey Li <[email protected]>Cc Andy Shevchenko <[email protected]>Cc Hans de Goede <[email protected]>Cc Mark Gross <[email protected]>Signed-off-by Paul Gortmaker <[email protected]>--- drivers/platform/x86/Kconfig | 11 ++++++++--- 1 file changed, 8 insertions(+), 3 deletions(-)drivers/platform/x86/Kconfig +0
-0
drivers/platform/x86/Kconfig config WINMATE_FM07_KEYS · line 1167
Quoted 8 lines
drivers/platform/x86/Kconfig +2
-3@@ -1163,6 +1163,11 @@ config WINMATE_FM07_KEYS endif # X86_PLATFORM_DEVICES config PMC_ATOM - def_bool y - depends on PCI - select COMMON_CLK + bool "Intel Atom SOC Power Management Controller driver" + default MATOM
Besides the remarks from Andy, this does seem like a weird default, MATOM means that gcc is passed -march=atom nothing more and nothing less. For a distro kernel which may e.g. set CONFIG_GENERIC_CPU we still want this enabled. But as said it is brought in by CONFIG_X86_INTEL_LPSS when that is set.
Thinking more about this I think it might be best to just do this instead, replacing patch 2 + 4 of this set:
drivers/clk/x86/Makefile +1 -3
@@ -1,6 +1,4 @@ # SPDX-License-Identifier: GPL-2.0-only -obj-$(CONFIG_PMC_ATOM) += clk-pmc-atom.o obj-$(CONFIG_X86_AMD_PLATFORM_DEVICE) += clk-fch.o -clk-x86-lpss-y := clk-lpss-atom.o -obj-$(CONFIG_X86_INTEL_LPSS) += clk-x86-lpss.o +obj-$(CONFIG_X86_INTEL_LPSS) += clk-lpss-atom.o clk-pmc-atom.o obj-$(CONFIG_CLK_LGM_CGU) += clk-cgu.o clk-cgu-pll.o clk-lgm.o
drivers/platform/x86/Kconfig +0 -5
@@ -1161,8 +1161,3 @@ config WINMATE_FM07_KEYS that delivers key events when these buttons are pressed. endif # X86_PLATFORM_DEVICES - -config PMC_ATOM - def_bool y - depends on PCI - select COMMON_CLK
drivers/platform/x86/Makefile +1 -1
@@ -126,7 +126,7 @@ obj-$(CONFIG_INTEL_SCU_PCI) += intel_scu_pcidrv.o obj-$(CONFIG_INTEL_SCU_PLATFORM) += intel_scu_pltdrv.o obj-$(CONFIG_INTEL_SCU_WDT) += intel_scu_wdt.o obj-$(CONFIG_INTEL_SCU_IPC_UTIL) += intel_scu_ipcutil.o -obj-$(CONFIG_PMC_ATOM) += pmc_atom.o +obj-$(CONFIG_X86_INTEL_LPSS) += pmc_atom.o # Siemens Simatic Industrial PCs obj-$(CONFIG_SIEMENS_SIMATIC_IPC) += simatic-ipc.o
This just folds the enabling of the pmc_atom code into the same Kconfig option as the one used the the LPSS code, so this actually simplify things; while at the same time making it possible to not build the pmc_atom code by unselecting the CONFIG_X86_INTEL_LPSS option.
Regards,
Hans
Quoted 5 lines · Paul Gortmaker
A few months back I was doing a test build for "defconfig-like" COTS hardware and happened to notice some pmc-atom stuff being built. Fine, I thought - the defconfig isn't minimal - we all know that - what
Kconfigdo I use to turn that off? Well, imagine my surprise when I found out you couldn't turn it [Atom Power Management Controller] code off!
Turning it off on BayTrail and CherryTrail devices will be devastating to the users' experience. And plenty of cheap tablets are exactly made on that SoCs.
Quoted 23 lines
Normally we all agree to not use "default y" unless unavoidable, but somehow this was added as "
def_booly" which is even worse ; you can see theKconfigsetting but there is *no* way you can turn it off.After investigating, I found no reason why the atom code couldn't be opt-out with just minor changes that the original addition overlooked.
And so this series addreses that. I tried to be sensitive to not break any existing configs in the process, but the defconfig will now intentionally be different and exclude this atom specific code.
Using a defconfig on today's linux-next, here is the delta summary:
$ ./scripts/bloat-o-meter -c ../pmc-atom-pre/vmlinux ../pmc-atom-post/vmlinux|grep add/remove add/remove: 0/410 grow/shrink: 0/7 up/down: 0/-47659 (-47659) add/remove: 0/105 grow/shrink: 0/1 up/down: 0/-6848 (-6848) add/remove: 0/56 grow/shrink: 0/1 up/down: 0/-10155 (-10155)$ ./scripts/bloat-o-meter -c ../pmc-atom-pre/vmlinux ../pmc-atom-post/vmlinux|grep Total Total: Before=13626994, After=13579335, chg -0.35% Total: Before=3572137, After=3565289, chg -0.19% Total: Before=1235335, After=1225180, chg -0.82%
Quoted 4 lines
It is hard to reclaim anything against the inevitable growth in footprint, so I'd say we should be glad to take whatever we can.
Boot tested defconfig on today's linux-next on older non-atom COTS.
I believe this needs an extensive test done by Hans who possesses a lot of problematic devices that require that module to be present.
Note to all your patches, please, use --cc option instead of putting noisy lines in the each of the commit messages. For myself I created a "smart" script [1] to avoid bothering with that. Feel free to use, modify, send PRs back to improve.
[1]: https://github.com/andy-shev/home-bin-tools/blob/master/ge2maintainer.sh
Signature · 2 lines
With Best Regards, Andy Shevchenko
Quoted 10 lines · [Re: [PATCH 0/4] platform: allow ATOM PMC code to be optional] On 28/04/2022…
Quoted 5 lines · Paul Gortmaker
A few months back I was doing a test build for "defconfig-like" COTS hardware and happened to notice some pmc-atom stuff being built. Fine, I thought - the defconfig isn't minimal - we all know that - what
Kconfigdo I use to turn that off? Well, imagine my surprise when I found out you couldn't turn it [Atom Power Management Controller] code off!Turning it off on BayTrail and CherryTrail devices will be devastating to the users' experience. And plenty of cheap tablets are exactly made on that SoCs.
Sure, but I could say the same thing for DRM_I915 and millions of desktop PC users - yet we still give all the other non i915 users the option to be able to turn it off. Pick any other Kconfig value you like and the same thing holds true.
Just so we are on the same page - I want to give the option to let people opt out, and at the same time not break existing users. If you think the defconfig default of being off is too risky, then I am OK with that and we can just start by exposing the option with "default y".
So, to that end - are you OK with exposing the Kconfig so people can opt out, or are you 100% against exposing the Kconfig at all? That obviously has the most impact on what I do or don't do next.
Quoted 23 lines
Quoted 23 lines
Normally we all agree to not use "default y" unless unavoidable, but somehow this was added as "
def_booly" which is even worse ; you can see theKconfigsetting but there is *no* way you can turn it off.After investigating, I found no reason why the atom code couldn't be opt-out with just minor changes that the original addition overlooked.
And so this series addreses that. I tried to be sensitive to not break any existing configs in the process, but the defconfig will now intentionally be different and exclude this atom specific code.
Using a defconfig on today's linux-next, here is the delta summary:
$ ./scripts/bloat-o-meter -c ../pmc-atom-pre/vmlinux ../pmc-atom-post/vmlinux|grep add/remove add/remove: 0/410 grow/shrink: 0/7 up/down: 0/-47659 (-47659) add/remove: 0/105 grow/shrink: 0/1 up/down: 0/-6848 (-6848) add/remove: 0/56 grow/shrink: 0/1 up/down: 0/-10155 (-10155)$ ./scripts/bloat-o-meter -c ../pmc-atom-pre/vmlinux ../pmc-atom-post/vmlinux|grep Total Total: Before=13626994, After=13579335, chg -0.35% Total: Before=3572137, After=3565289, chg -0.19% Total: Before=1235335, After=1225180, chg -0.82%
Quoted 7 lines
Quoted 4 lines
It is hard to reclaim anything against the inevitable growth in footprint, so I'd say we should be glad to take whatever we can.
Boot tested defconfig on today's linux-next on older non-atom COTS.
I believe this needs an extensive test done by Hans who possesses a lot of problematic devices that require that module to be present.
Input from Hans is 100% welcome - but maybe again if we just consider using "default y" even though it isn't typical - then your concerns are not as extensive?
Quoted 4 lines
Note to all your patches, please, use --cc option instead of putting noisy lines in the each of the commit messages. For myself I created a "smart" script [1] to avoid bothering with that. Feel free to use, modify, send PRs back to improve.
Thanks - I'm used to explicitly capturing who was looped into the discussion. But I hadn't considered how things have evolved so that in certain circumstances that might be considered just noise with the wider audience we see in the typical patch sets of today.
Assuming you are OK with exposing the option as a choice, I'll make things "default y" in v2 to ensure we've minimized risk to existing users who rely on it, but wait a while for others like Hans to put in their input. I'll probably follow up in the individual patches to ask for clarification if I'm not clear on what you had in mind.
Thanks,
Paul.
--
Quoted 8 lines
[1]: https://github.com/andy-shev/home-bin-tools/blob/master/ge2maintainer.sh
Signature · 2 lines
With Best Regards, Andy Shevchenko
Quoted 71 lines · Paul Gortmaker
Quoted 10 lines · [Re: [PATCH 0/4] platform: allow ATOM PMC code to be optional] On 28/04/2022…
Quoted 5 lines · Paul Gortmaker
A few months back I was doing a test build for "defconfig-like" COTS hardware and happened to notice some pmc-atom stuff being built. Fine, I thought - the defconfig isn't minimal - we all know that - what
Kconfigdo I use to turn that off? Well, imagine my surprise when I found out you couldn't turn it [Atom Power Management Controller] code off!Turning it off on BayTrail and CherryTrail devices will be devastating to the users' experience. And plenty of cheap tablets are exactly made on that SoCs.
Sure, but I could say the same thing for
DRM_I915and millions of desktop PC users - yet we still give all the other non i915 users the option to be able to turn it off. Pick any otherKconfigvalue you like and the same thing holds true.Just so we are on the same page - I want to give the option to let people opt out, and at the same time not break existing users. If you think the defconfig default of being off is too risky, then I am OK with that and we can just start by exposing the option with "default y".
So, to that end - are you OK with exposing the
Kconfigso people can opt out, or are you 100% against exposing theKconfigat all? That obviously has the most impact on what I do or don't do next.Quoted 23 lines
Quoted 23 lines
Normally we all agree to not use "default y" unless unavoidable, but somehow this was added as "
def_booly" which is even worse ; you can see theKconfigsetting but there is *no* way you can turn it off.After investigating, I found no reason why the atom code couldn't be opt-out with just minor changes that the original addition overlooked.
And so this series addreses that. I tried to be sensitive to not break any existing configs in the process, but the defconfig will now intentionally be different and exclude this atom specific code.
Using a defconfig on today's linux-next, here is the delta summary:
$ ./scripts/bloat-o-meter -c ../pmc-atom-pre/vmlinux ../pmc-atom-post/vmlinux|grep add/remove add/remove: 0/410 grow/shrink: 0/7 up/down: 0/-47659 (-47659) add/remove: 0/105 grow/shrink: 0/1 up/down: 0/-6848 (-6848) add/remove: 0/56 grow/shrink: 0/1 up/down: 0/-10155 (-10155)$ ./scripts/bloat-o-meter -c ../pmc-atom-pre/vmlinux ../pmc-atom-post/vmlinux|grep Total Total: Before=13626994, After=13579335, chg -0.35% Total: Before=3572137, After=3565289, chg -0.19% Total: Before=1235335, After=1225180, chg -0.82%Quoted 7 lines
Quoted 4 lines
It is hard to reclaim anything against the inevitable growth in footprint, so I'd say we should be glad to take whatever we can.
Boot tested defconfig on today's linux-next on older non-atom COTS.
I believe this needs an extensive test done by Hans who possesses a lot of problematic devices that require that module to be present.
Input from Hans is 100% welcome - but maybe again if we just consider using "default y" even though it isn't typical - then your concerns are not as extensive?
Quoted 4 lines
Note to all your patches, please, use --cc option instead of putting noisy lines in the each of the commit messages. For myself I created a "smart" script [1] to avoid bothering with that. Feel free to use, modify, send PRs back to improve.
Thanks - I'm used to explicitly capturing who was looped into the discussion. But I hadn't considered how things have evolved so that in certain circumstances that might be considered just noise with the wider audience we see in the typical patch sets of today.
Quoted 5 lines
Assuming you are OK with exposing the option as a choice, I'll make things "default y" in v2 to ensure we've minimized risk to existing users who rely on it, but wait a while for others like Hans to put in their input. I'll probably follow up in the individual patches to ask for clarification if I'm not clear on what you had in mind.
My main concern is the existing users' experience. Opting-out the option is a good to have, I'm not against it.
Signature · 2 lines
With Best Regards, Andy Shevchenko
Hi,
Quoted 53 lines · Paul Gortmaker
Quoted 10 lines · [Re: [PATCH 0/4] platform: allow ATOM PMC code to be optional] On 28/04/2022…
Quoted 5 lines · Paul Gortmaker
A few months back I was doing a test build for "defconfig-like" COTS hardware and happened to notice some pmc-atom stuff being built. Fine, I thought - the defconfig isn't minimal - we all know that - what
Kconfigdo I use to turn that off? Well, imagine my surprise when I found out you couldn't turn it [Atom Power Management Controller] code off!Turning it off on BayTrail and CherryTrail devices will be devastating to the users' experience. And plenty of cheap tablets are exactly made on that SoCs.
Sure, but I could say the same thing for
DRM_I915and millions of desktop PC users - yet we still give all the other non i915 users the option to be able to turn it off. Pick any otherKconfigvalue you like and the same thing holds true.Just so we are on the same page - I want to give the option to let people opt out, and at the same time not break existing users. If you think the defconfig default of being off is too risky, then I am OK with that and we can just start by exposing the option with "default y".
So, to that end - are you OK with exposing the
Kconfigso people can opt out, or are you 100% against exposing theKconfigat all? That obviously has the most impact on what I do or don't do next.Quoted 26 lines
Quoted 26 lines
Normally we all agree to not use "default y" unless unavoidable, but somehow this was added as "
def_booly" which is even worse ; you can see theKconfigsetting but there is *no* way you can turn it off.After investigating, I found no reason why the atom code couldn't be opt-out with just minor changes that the original addition overlooked.
And so this series addreses that. I tried to be sensitive to not break any existing configs in the process, but the defconfig will now intentionally be different and exclude this atom specific code.
Using a defconfig on today's linux-next, here is the delta summary:
$ ./scripts/bloat-o-meter -c ../pmc-atom-pre/vmlinux ../pmc-atom-post/vmlinux|grep add/remove add/remove: 0/410 grow/shrink: 0/7 up/down: 0/-47659 (-47659) add/remove: 0/105 grow/shrink: 0/1 up/down: 0/-6848 (-6848) add/remove: 0/56 grow/shrink: 0/1 up/down: 0/-10155 (-10155)$ ./scripts/bloat-o-meter -c ../pmc-atom-pre/vmlinux ../pmc-atom-post/vmlinux|grep Total Total: Before=13626994, After=13579335, chg -0.35% Total: Before=3572137, After=3565289, chg -0.19% Total: Before=1235335, After=1225180, chg -0.82%It is hard to reclaim anything against the inevitable growth in footprint, so I'd say we should be glad to take whatever we can.
Quoted 8 lines
Quoted 4 lines
Quoted 1 line
Boot tested defconfig on today's linux-next on older non-atom COTS.
I believe this needs an extensive test done by Hans who possesses a lot of problematic devices that require that module to be present.
Input from Hans is 100% welcome - but maybe again if we just consider using "default y" even though it isn't typical - then your concerns are not as extensive?
I have no objection against allowing disabling the PMC_ATOM Kconfig option.
As for users breaking support for BYT/CHT setups because they forget to enable this, without X86_INTEL_LPSS being enabled BYT/CHT are pretty much broken anyways and since patch 2/4 adds a "select PMC_ATOM" to the X86_INTEL_LPSS Kconfig option I'm not really worried about that.
I'm afraid this patch-set might break some randomconfig builds though, but I cannot see anything obviously causing such breakage here, so I think it would be fine to just merge this series as is and then see if we get any breakage reports.
Andy, are you ok with me moving ahead and merging this series as is?
Regards,
Hans
Quoted 1 line · Hans de Goede
On 4/28/22 20:11, Paul Gortmaker wrote:
...
Quoted 11 lines
As for users breaking support for BYT/CHT setups because they forget to enable this, without
X86_INTEL_LPSSbeing enabled BYT/CHT are pretty much broken anyways and since patch 2/4 adds a "selectPMC_ATOM" to theX86_INTEL_LPSSKconfigoption I'm not really worried about that.I'm afraid this patch-set might break some randomconfig builds though, but I cannot see anything obviously causing such breakage here, so I think it would be fine to just merge this series as is and then see if we get any breakage reports.
Andy, are you ok with me moving ahead and merging this series as is?
It seems as is can't be fulfilled due to your own comment, but in general I'm not objecting the idea. So, go ahead if you feel it's ready.
Signature · 2 lines
With Best Regards, Andy Shevchenko
Hi,
Quoted 15 lines · Andy Shevchenko
Quoted 1 line · Hans de Goede
On 4/28/22 20:11, Paul Gortmaker wrote:
...
Quoted 10 lines
As for users breaking support for BYT/CHT setups because they forget to enable this, without
X86_INTEL_LPSSbeing enabled BYT/CHT are pretty much broken anyways and since patch 2/4 adds a "selectPMC_ATOM" to theX86_INTEL_LPSSKconfigoption I'm not really worried about that.I'm afraid this patch-set might break some randomconfig builds though, but I cannot see anything obviously causing such breakage here, so I think it would be fine to just merge this series as is and then see if we get any breakage reports.
Quoted 4 lines
Quoted 1 line
Andy, are you ok with me moving ahead and merging this series as is?
It seems as is can't be fulfilled due to your own comment, but in general I'm not objecting the idea. So, go ahead if you feel it's ready.
Right, my later comment to just replace PMC_ATOM with X86_INTEL_LPSS supersedes this.
I'll send out a patch with that approach so that this can get some comments / review.
Regards,
Hans