Skip to contentSkip to conversation

platform: allow ATOM PMC code to be optional

PATCH In review Paul Gortmaker 4 participants 18 messages Apr 28, 2022 – May 6, 2022

Subsystems (4)
lore
Share & tools
lore
#0 lore raw

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.

Cc Andy Shevchenko <[email protected]>
Cc Aubrey Li <[email protected]>
Cc Hans de Goede <[email protected]>
Cc Len Brown <[email protected]>
Cc Mark Gross <[email protected]>
Cc "Rafael J. Wysocki" <[email protected]>

---

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
#1 lore raw

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.

Cc Andy Shevchenko <[email protected]>
Cc Aubrey Li <[email protected]>
Cc Hans de Goede <[email protected]>
Cc Mark Gross <[email protected]>
---
 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
index a40fae6edc84..31cf25d25d66 100644
@@ -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
index 022bcea9edec..6807839c718b 100644
@@ -144,6 +144,5 @@
 #define	SLEEP_ENABLE		0x2000
 
 extern 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
#2 lore raw

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
#3 lore raw

Re: [PATCH 1/4] platform/x86: pmc_atom: remove unused pmc_atom_write()

In reply to Paul Gortmaker

Hi,

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]>

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
index a40fae6edc84..31cf25d25d66 100644
@@ -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
index 022bcea9edec..6807839c718b 100644
@@ -144,6 +144,5 @@
 #define	SLEEP_ENABLE		0x2000
 
 extern int pmc_atom_read(int offset, u32 *value);
-extern int pmc_atom_write(int offset, u32 value);
 
 #endif /* PMC_ATOM_H */
#4 lore raw

PATCH 2/4 ACPI: LPSS: make the Kconfig dependency on PMC_ATOM explicit

In reply to Paul Gortmaker

The 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.

Cc Andy Shevchenko <[email protected]>
Cc Aubrey Li <[email protected]>
Cc "Rafael J. Wysocki" <[email protected]>
Cc Len Brown <[email protected]>
---
 arch/x86/Kconfig | 1 +
 1 file changed, 1 insertion(+)
arch/x86/Kconfig +1 -0
index 2ee26f10a814..163c198ec8ec 100644
@@ -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
#5 lore raw

Re: [PATCH 2/4] ACPI: LPSS: make the Kconfig dependency on PMC_ATOM explicit

Quoted 10 lines · Paul Gortmaker

The 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.

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
maintainer Acked-by on v1 2/4
#6 lore raw

Re: [PATCH 2/4] ACPI: LPSS: make the Kconfig dependency on PMC_ATOM explicit

In reply to Paul Gortmaker
Quoted 11 lines · Paul Gortmaker

The 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.

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]>
---
 arch/x86/Kconfig | 1 +
 1 file changed, 1 insertion(+)
arch/x86/Kconfig +1 -1
index 2ee26f10a814..163c198ec8ec 100644
@@ -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

#7 lore raw

PATCH 3/4 platform/x86: pmc_atom: dont export pmc_atom_read - no modular users

In reply to Paul Gortmaker

There 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.

Cc Andy Shevchenko <[email protected]>
Cc Aubrey Li <[email protected]>
Cc Hans de Goede <[email protected]>
Cc Mark Gross <[email protected]>
---
 drivers/platform/x86/pmc_atom.c | 1 -
 1 file changed, 1 deletion(-)
drivers/platform/x86/pmc_atom.c +0 -1
index 31cf25d25d66..b8b1ed1406de 100644
@@ -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
#8 lore raw

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_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.

Quoted 6 lines
Cc Andy Shevchenko <[email protected]>
Cc Aubrey Li <[email protected]>
Cc Hans de Goede <[email protected]>
Cc Mark Gross <[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
index 31cf25d25d66..b8b1ed1406de 100644
@@ -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)
 {
#9 lore raw

PATCH 4/4 platform/x86: pmc_atom: make the PMC driver actually unselectable

In reply to Paul Gortmaker

This 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.

Cc Aubrey Li <[email protected]>
Cc Andy Shevchenko <[email protected]>
Cc Hans de Goede <[email protected]>
Cc Mark Gross <[email protected]>
---
 drivers/platform/x86/Kconfig | 11 ++++++++---
 1 file changed, 8 insertions(+), 3 deletions(-)
drivers/platform/x86/Kconfig +8 -3
index f08ad85683cb..86459e99d831 100644
@@ -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
#10 lore raw

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_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.

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
#11 lore raw

Re: [PATCH 4/4] platform/x86: pmc_atom: make the PMC driver actually unselectable

In reply to Paul Gortmaker

Hi,

Quoted 34 lines · Paul Gortmaker

This 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.

Cc Aubrey Li <[email protected]>
Cc Andy Shevchenko <[email protected]>
Cc Hans de Goede <[email protected]>
Cc Mark Gross <[email protected]>
---
 drivers/platform/x86/Kconfig | 11 ++++++++---
 1 file changed, 8 insertions(+), 3 deletions(-)
drivers/platform/x86/Kconfig +0 -0
index f08ad85683cb..86459e99d831 100644

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
index 1244c4e568ff..c2088b3c4081 100644
@@ -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
index f08ad85683cb..85c396a43048 100644
@@ -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
index 4a59f47a46e2..cc2a74713313 100644
@@ -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

#12 lore raw
In reply to Paul Gortmaker
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 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!

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_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%
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
#13 lore raw
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 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!

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_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%
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
#14 lore raw
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 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!

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_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%
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
#15 lore raw
In reply to Paul Gortmaker

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 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!

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 26 lines
Quoted 26 lines

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.

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

#16 lore raw
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_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?

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
#17 lore raw

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_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.

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