Skip to contentSkip to conversation

acpi: Fix hed module initialization order when it is built-in

PATCH Superseded Xiaofei Tan 4 participants 7 messages Nov 15, 2024 – Dec 28, 2024 ACPI

v1 of 3 Superseded Latest is v3 →

lore
Share & tools
lore
#0 lore raw

PATCH 1/1 acpi: Fix hed module initialization order when it is built-in

When the module hed is built-in, the init order is determined by Makefile order. That order violates expectations. Because the module hed init is behind evged. RAS records can't be handled in the special time window that evged has initialized while hed not. If the number of such RAS records is more than the APEI HEST error source number, the HEST resources could be occupied all, and then could affect subsequent RAS error reporting.

---
 drivers/acpi/Makefile | 8 +++++++-
 1 file changed, 7 insertions(+), 1 deletion(-)
drivers/acpi/Makefile +7 -1
index 61ca4afe83dc..54f60b7922ad 100644
@@ -15,6 +15,13 @@ endif
 
 obj-$(CONFIG_ACPI)		+= tables.o
 
+#
+# The hed.o needs to be in front of evged.o to avoid the problem that
+# RAS errors cannot be handled in the special time window of startup
+# phase that evged has initialized while hed not.
+#
+obj-$(CONFIG_ACPI_HED)		+= hed.o
+
 #
 # ACPI Core Subsystem (Interpreter)
 #
@@ -95,7 +102,6 @@ obj-$(CONFIG_ACPI_HOTPLUG_IOAPIC) += ioapic.o
 obj-$(CONFIG_ACPI_BATTERY)	+= battery.o
 obj-$(CONFIG_ACPI_SBS)		+= sbshc.o
 obj-$(CONFIG_ACPI_SBS)		+= sbs.o
-obj-$(CONFIG_ACPI_HED)		+= hed.o
 obj-$(CONFIG_ACPI_EC_DEBUGFS)	+= ec_sys.o
 obj-$(CONFIG_ACPI_BGRT)		+= bgrt.o
 obj-$(CONFIG_ACPI_CPPC_LIB)	+= cppc_acpi.o
Signature · 1 line
2.33.0
#1 lore raw
Quoted 3 lines · Xiaofei Tan

When the module hed is built-in, the init order is determined by Makefile order.

Are you sure?

Quoted 6 lines

That order violates expectations. Because the module hed init is behind evged. RAS records can't be handled in the special time window that evged has initialized while hed not. If the number of such RAS records is more than the APEI HEST error source number, the HEST resources could be occupied all, and then could affect subsequent RAS error reporting.

Well, the problem is real, but does the change really prevent it from happening or does it just increase the likelihood of success?

In the latter case, and generally speaking too, it would be better to add explicit synchronization between evged and hed.

Quoted 35 lines
---
 drivers/acpi/Makefile | 8 +++++++-
 1 file changed, 7 insertions(+), 1 deletion(-)
drivers/acpi/Makefile +7 -2
index 61ca4afe83dc..54f60b7922ad 100644
@@ -15,6 +15,13 @@ endif
 
 obj-$(CONFIG_ACPI)             += tables.o
 
+#
+# The hed.o needs to be in front of evged.o to avoid the problem that
+# RAS errors cannot be handled in the special time window of startup
+# phase that evged has initialized while hed not.
+#
+obj-$(CONFIG_ACPI_HED)         += hed.o
+
 #
 # ACPI Core Subsystem (Interpreter)
 #
@@ -95,7 +102,6 @@ obj-$(CONFIG_ACPI_HOTPLUG_IOAPIC) += ioapic.o
 obj-$(CONFIG_ACPI_BATTERY)     += battery.o
 obj-$(CONFIG_ACPI_SBS)         += sbshc.o
 obj-$(CONFIG_ACPI_SBS)         += sbs.o
-obj-$(CONFIG_ACPI_HED)         += hed.o
 obj-$(CONFIG_ACPI_EC_DEBUGFS)  += ec_sys.o
 obj-$(CONFIG_ACPI_BGRT)                += bgrt.o
 obj-$(CONFIG_ACPI_CPPC_LIB)    += cppc_acpi.o
--

2.33.0

#2 lore raw

Hi Rafael,

Quoted 4 lines · Rafael J. Wysocki
Quoted 2 lines · Xiaofei Tan

When the module hed is built-in, the init order is determined by Makefile order.

Are you sure?

yes

Quoted 8 lines
Quoted 6 lines

That order violates expectations. Because the module hed init is behind evged. RAS records can't be handled in the special time window that evged has initialized while hed not. If the number of such RAS records is more than the APEI HEST error source number, the HEST resources could be occupied all, and then could affect subsequent RAS error reporting.

Well, the problem is real, but does the change really prevent it from happening or does it just increase the likelihood of success?

It can be completely solved if the driver used as built-in way. If build HED as a module, it not solved.

Quoted 40 lines

In the latter case, and generally speaking too, it would be better to add explicit synchronization between evged and hed.

Quoted 33 lines
---
  drivers/acpi/Makefile | 8 +++++++-
  1 file changed, 7 insertions(+), 1 deletion(-)
drivers/acpi/Makefile +7 -2
index 61ca4afe83dc..54f60b7922ad 100644
@@ -15,6 +15,13 @@ endif
 
  obj-$(CONFIG_ACPI)             += tables.o
 
+#
+# The hed.o needs to be in front of evged.o to avoid the problem that
+# RAS errors cannot be handled in the special time window of startup
+# phase that evged has initialized while hed not.
+#
+obj-$(CONFIG_ACPI_HED)         += hed.o
+
  #
  # ACPI Core Subsystem (Interpreter)
  #
@@ -95,7 +102,6 @@ obj-$(CONFIG_ACPI_HOTPLUG_IOAPIC) += ioapic.o
  obj-$(CONFIG_ACPI_BATTERY)     += battery.o
  obj-$(CONFIG_ACPI_SBS)         += sbshc.o
  obj-$(CONFIG_ACPI_SBS)         += sbs.o
-obj-$(CONFIG_ACPI_HED)         += hed.o
  obj-$(CONFIG_ACPI_EC_DEBUGFS)  += ec_sys.o
  obj-$(CONFIG_ACPI_BGRT)                += bgrt.o
  obj-$(CONFIG_ACPI_CPPC_LIB)    += cppc_acpi.o
--
Quoted 2 lines

2.33.0

.

#3 lore raw
Quoted 9 lines · On Mon, 23 Dec 2024 17:31:08 +0800 Xiaofei Tan <[email protected]> wrote:

Hi Rafael,

Quoted 4 lines · Rafael J. Wysocki
Quoted 2 lines · Xiaofei Tan

When the module hed is built-in, the init order is determined by Makefile order.

Are you sure?

yes

We had a similar fix in CXL recently (which is why I suggested this approach internally when tanxiaofei mentioned the problem).

https://git.kernel.org/pub/scm/linux/kernel/git/torvalds/linux.git/commit/drivers/cxl?id=6575b268157f37929948a8d1f3bafb3d7c055bc1

The related discussion for the CXL patch was the first time I'd come across solution to load order for built in cases.

Quoted 7 lines
Quoted 7 lines
Quoted 7 lines

That order violates expectations. Because the module hed init is behind evged. RAS records can't be handled in the special time window that evged has initialized while hed not. If the number of such RAS records is more than the APEI HEST error source number, the HEST resources could be occupied all, and then could affect subsequent RAS error reporting.

Quoted 5 lines
Quoted 2 lines

Well, the problem is real, but does the change really prevent it from happening or does it just increase the likelihood of success?

It can be completely solved if the driver used as built-in way. If build HED as a module, it not solved.

Can we enforce that condition not happening with appropriate Kconfig? It's annoying to restrict build options, but if needed to make it work then better than not working!

Jonathan

Quoted 41 lines
Quoted 41 lines

In the latter case, and generally speaking too, it would be better to add explicit synchronization between evged and hed.

Quoted 33 lines
---
  drivers/acpi/Makefile | 8 +++++++-
  1 file changed, 7 insertions(+), 1 deletion(-)
drivers/acpi/Makefile +7 -2
index 61ca4afe83dc..54f60b7922ad 100644
@@ -15,6 +15,13 @@ endif
 
  obj-$(CONFIG_ACPI)             += tables.o
 
+#
+# The hed.o needs to be in front of evged.o to avoid the problem that
+# RAS errors cannot be handled in the special time window of startup
+# phase that evged has initialized while hed not.
+#
+obj-$(CONFIG_ACPI_HED)         += hed.o
+
  #
  # ACPI Core Subsystem (Interpreter)
  #
@@ -95,7 +102,6 @@ obj-$(CONFIG_ACPI_HOTPLUG_IOAPIC) += ioapic.o
  obj-$(CONFIG_ACPI_BATTERY)     += battery.o
  obj-$(CONFIG_ACPI_SBS)         += sbshc.o
  obj-$(CONFIG_ACPI_SBS)         += sbs.o
-obj-$(CONFIG_ACPI_HED)         += hed.o
  obj-$(CONFIG_ACPI_EC_DEBUGFS)  += ec_sys.o
  obj-$(CONFIG_ACPI_BGRT)                += bgrt.o
  obj-$(CONFIG_ACPI_CPPC_LIB)    += cppc_acpi.o
--
Quoted 2 lines

2.33.0

.

#4 lore raw

Hi Jonathan,

Quoted 10 lines · Jonathan Cameron
Quoted 7 lines · On Mon, 23 Dec 2024 17:31:08 +0800 Xiaofei Tan <[email protected]> wrote:

Hi Rafael,

Quoted 4 lines · Rafael J. Wysocki
Quoted 2 lines · Xiaofei Tan

When the module hed is built-in, the init order is determined by Makefile order.

Are you sure?

Quoted 9 lines
Quoted 1 line

yes

We had a similar fix in CXL recently (which is why I suggested this approach internally when tanxiaofei mentioned the problem).

https://git.kernel.org/pub/scm/linux/kernel/git/torvalds/linux.git/commit/drivers/cxl?id=6575b268157f37929948a8d1f3bafb3d7c055bc1

The related discussion for the CXL patch was the first time I'd come across solution to load order for built in cases.

Yes :)

Quoted 6 lines
Quoted 6 lines
Quoted 6 lines
Quoted 6 lines

That order violates expectations. Because the module hed init is behind evged. RAS records can't be handled in the special time window that evged has initialized while hed not. If the number of such RAS records is more than the APEI HEST error source number, the HEST resources could be occupied all, and then could affect subsequent RAS error reporting.

Quoted 7 lines
Quoted 4 lines
Quoted 2 lines

Well, the problem is real, but does the change really prevent it from happening or does it just increase the likelihood of success?

It can be completely solved if the driver used as built-in way. If build HED as a module, it not solved.

Can we enforce that condition not happening with appropriate Kconfig? It's annoying to restrict build options, but if needed to make it work then better than not working!

Agree,  i will change ACPI_HED from tristate to bool if there are no other comments, thanks.

Quoted 44 lines

Jonathan

Quoted 31 lines
Quoted 31 lines

In the latter case, and generally speaking too, it would be better to add explicit synchronization between evged and hed.

Quoted 28 lines
---
   drivers/acpi/Makefile | 8 +++++++-
   1 file changed, 7 insertions(+), 1 deletion(-)
drivers/acpi/Makefile +7 -0
index 61ca4afe83dc..54f60b7922ad 100644
@@ -15,6 +15,13 @@ endif
 
   obj-$(CONFIG_ACPI)             += tables.o
 
+#
+# The hed.o needs to be in front of evged.o to avoid the problem that
+# RAS errors cannot be handled in the special time window of startup
+# phase that evged has initialized while hed not.
+#
+obj-$(CONFIG_ACPI_HED)         += hed.o
+
   #
   # ACPI Core Subsystem (Interpreter)
   #
@@ -95,7 +102,6 @@ obj-$(CONFIG_ACPI_HOTPLUG_IOAPIC) += ioapic.o
   obj-$(CONFIG_ACPI_BATTERY)     += battery.o
   obj-$(CONFIG_ACPI_SBS)         += sbshc.o
   obj-$(CONFIG_ACPI_SBS)         += sbs.o
Quoted 8 lines
Quoted 8 lines
Quoted 7 lines
drivers/acpi/Makefile +0 -2
-obj-$(CONFIG_ACPI_HED)         += hed.o
   obj-$(CONFIG_ACPI_EC_DEBUGFS)  += ec_sys.o
   obj-$(CONFIG_ACPI_BGRT)                += bgrt.o
   obj-$(CONFIG_ACPI_CPPC_LIB)    += cppc_acpi.o
--

2.33.0

.

.

#5 lore raw
In reply to Xiaofei Tan

Em Fri, 15 Nov 2024 11:50:14 +0800
Xiaofei Tan <[email protected]> escreveu:

Please always copy my @kernel.org address for upstream work.

Quoted 7 lines

When the module hed is built-in, the init order is determined by Makefile order. That order violates expectations. Because the module hed init is behind evged. RAS records can't be handled in the special time window that evged has initialized while hed not. If the number of such RAS records is more than the APEI HEST error source number, the HEST resources could be occupied all, and then could affect subsequent RAS error reporting.

IMO, it is a lot better to use a late init call. Please see:
include/linux/init.h

This would be done by, for instance, using late_initcall().

Now, what we have is:

	acpi-y                          += evged.o
	obj-$(CONFIG_ACPI_HED)          += hed.o

Where ACPI_HED being a tri-state.

It sounds to me, that even, with your patch, if you build HED as a module, you'll still have a problem.

Shouldn't be ACPI_HED be changed from tristate to bool?

Regards,
Mauro

Quoted 33 lines
---
 drivers/acpi/Makefile | 8 +++++++-
 1 file changed, 7 insertions(+), 1 deletion(-)
drivers/acpi/Makefile +7 -1
index 61ca4afe83dc..54f60b7922ad 100644
@@ -15,6 +15,13 @@ endif
 
 obj-$(CONFIG_ACPI)		+= tables.o
 
+#
+# The hed.o needs to be in front of evged.o to avoid the problem that
+# RAS errors cannot be handled in the special time window of startup
+# phase that evged has initialized while hed not.
+#
+obj-$(CONFIG_ACPI_HED)		+= hed.o
+
 #
 # ACPI Core Subsystem (Interpreter)
 #
@@ -95,7 +102,6 @@ obj-$(CONFIG_ACPI_HOTPLUG_IOAPIC) += ioapic.o
 obj-$(CONFIG_ACPI_BATTERY)	+= battery.o
 obj-$(CONFIG_ACPI_SBS)		+= sbshc.o
 obj-$(CONFIG_ACPI_SBS)		+= sbs.o
-obj-$(CONFIG_ACPI_HED)		+= hed.o
 obj-$(CONFIG_ACPI_EC_DEBUGFS)	+= ec_sys.o
 obj-$(CONFIG_ACPI_BGRT)		+= bgrt.o
 obj-$(CONFIG_ACPI_CPPC_LIB)	+= cppc_acpi.o
#6 lore raw

Hi Mauro,

Quoted 4 lines · Mauro Carvalho Chehab

Em Fri, 15 Nov 2024 11:50:14 +0800
Xiaofei Tan <[email protected]> escreveu:

Please always copy my @kernel.org address for upstream work.

OK

Quoted 14 lines
Quoted 7 lines

When the module hed is built-in, the init order is determined by Makefile order. That order violates expectations. Because the module hed init is behind evged. RAS records can't be handled in the special time window that evged has initialized while hed not. If the number of such RAS records is more than the APEI HEST error source number, the HEST resources could be occupied all, and then could affect subsequent RAS error reporting.

IMO, it is a lot better to use a late init call. Please see:
include/linux/init.h

This would be done by, for instance, using late_initcall().

Now, what we have is:

Quoted 7 lines
	acpi-y                          += evged.o
	obj-$(CONFIG_ACPI_HED)          += hed.o

Where ACPI_HED being a tri-state.

It sounds to me, that even, with your patch, if you build HED as a module, you'll still have a problem.

Yes, and it is also  affected by loading sequence of HED and GHES. Anyway, the risk remains.

Quoted 2 lines

Shouldn't be ACPI_HED be changed from tristate to bool?

agree,

@Rafael

Hi Rafael, Please help check if we can do this change, thanks.

Quoted 37 lines

Regards,
Mauro

Quoted 25 lines
---
  drivers/acpi/Makefile | 8 +++++++-
  1 file changed, 7 insertions(+), 1 deletion(-)
drivers/acpi/Makefile +7 -0
index 61ca4afe83dc..54f60b7922ad 100644
@@ -15,6 +15,13 @@ endif
 
  obj-$(CONFIG_ACPI)		+= tables.o
 
+#
+# The hed.o needs to be in front of evged.o to avoid the problem that
+# RAS errors cannot be handled in the special time window of startup
+# phase that evged has initialized while hed not.
+#
+obj-$(CONFIG_ACPI_HED)		+= hed.o
+
  #
  # ACPI Core Subsystem (Interpreter)
  #

drivers/acpi/Makefile obj-$(CONFIG_ACPI_HOTPLUG_IOAPIC) += ioapic.o · line 107

Quoted 7 lines
drivers/acpi/Makefile +0 -1
@@ -95,7 +102,6 @@ obj-$(CONFIG_ACPI_HOTPLUG_IOAPIC) += ioapic.o
  obj-$(CONFIG_ACPI_BATTERY)	+= battery.o
  obj-$(CONFIG_ACPI_SBS)		+= sbshc.o
  obj-$(CONFIG_ACPI_SBS)		+= sbs.o
-obj-$(CONFIG_ACPI_HED)		+= hed.o
  obj-$(CONFIG_ACPI_EC_DEBUGFS)	+= ec_sys.o
  obj-$(CONFIG_ACPI_BGRT)		+= bgrt.o
  obj-$(CONFIG_ACPI_CPPC_LIB)	+= cppc_acpi.o

.