Workaround for _PPC (BIOS cpufreq limitations)

PATCH In review Thomas Renninger 2 participants 3 messages May 19, 2007 ACPI

lore

Share & tools

lore

Thomas Renninger

author

2007-05-19 02:59 UTC #0 lore raw

PATCH 1/1 Workaround for _PPC (BIOS cpufreq limitations)

Len, can you apply this one, pls.

Workaround for _PPC (BIOS cpufreq limitations)

There have been fixes using _PPC, which seem to unhide a problem on HP nx6125 (double cpufreq switch freezes the machine for several seconds).
This one should provide a workaround for the nx6125 and for possible other machines that show any weird _PPC behaviour.

Signed-off-by Thomas Renninger <[email protected]>

---
 drivers/acpi/processor_perflib.c |   16 +++++++++++++++-
 1 file changed, 15 insertions(+), 1 deletion(-)

drivers/acpi/processor_perflib.c +15 -1

Index: linux-2.6.21/drivers/acpi/processor_perflib.c
===================================================================
@@ -60,6 +60,11 @@ static DEFINE_MUTEX(performance_mutex);
  * policy is adjusted accordingly.
  */
 
+static unsigned int ignore_ppc = 0;
+module_param(ignore_ppc, uint, 0644);
+MODULE_PARM_DESC(ignore_ppc, "If the frequency of your machine gets wrongly" \
+		 "limited by BIOS, this should help");
+
 #define PPC_REGISTERED   1
 #define PPC_IN_USE       2

@@ -72,6 +77,9 @@ static int acpi_processor_ppc_notifier(s
 	struct acpi_processor *pr;
 	unsigned int ppc = 0;
 
+	if (ignore_ppc)
+		return 0;
+
 	mutex_lock(&performance_mutex);
 
 	if (event != CPUFREQ_INCOMPATIBLE)

@@ -130,7 +138,13 @@ static int acpi_processor_get_platform_l
 
 int acpi_processor_ppc_has_changed(struct acpi_processor *pr)
 {
-	int ret = acpi_processor_get_platform_limit(pr);
+	int ret;
+
+	if (ignore_ppc)
+		return 0;
+
+	ret = acpi_processor_get_platform_limit(pr);
+
 	if (ret < 0)
 		return (ret);
 	else

Workaround for _PPC (BIOS cpufreq limitations)

Signed-off-by Thomas Renninger <[email protected]>

---
 drivers/acpi/processor_perflib.c |   16 +++++++++++++++-
 1 file changed, 15 insertions(+), 1 deletion(-)

drivers/acpi/processor_perflib.c +15 -1

Index: linux-2.6.21/drivers/acpi/processor_perflib.c
===================================================================
@@ -60,6 +60,11 @@ static DEFINE_MUTEX(performance_mutex);
  * policy is adjusted accordingly.
  */
 
+static unsigned int ignore_ppc = 0;
+module_param(ignore_ppc, uint, 0644);
+MODULE_PARM_DESC(ignore_ppc, "If the frequency of your machine gets wrongly" \
+		 "limited by BIOS, this should help");
+
 #define PPC_REGISTERED   1
 #define PPC_IN_USE       2

@@ -72,6 +77,9 @@ static int acpi_processor_ppc_notifier(s
 	struct acpi_processor *pr;
 	unsigned int ppc = 0;
 
+	if (ignore_ppc)
+		return 0;
+
 	mutex_lock(&performance_mutex);
 
 	if (event != CPUFREQ_INCOMPATIBLE)

@@ -130,7 +138,13 @@ static int acpi_processor_get_platform_l
 
 int acpi_processor_ppc_has_changed(struct acpi_processor *pr)
 {
-	int ret = acpi_processor_get_platform_limit(pr);
+	int ret;
+
+	if (ignore_ppc)
+		return 0;
+
+	ret = acpi_processor_get_platform_limit(pr);
+
 	if (ret < 0)
 		return (ret);
 	else

Len Brown

reviewer

2007-05-19 04:55 UTC #1 lore raw

Quoted 9 lines · Thomas Renninger

Len, can you apply this one, pls.
Workaround for _PPC (BIOS cpufreq limitations)
There have been fixes using _PPC, which seem to unhide a problem on HP nx6125 (double cpufreq switch freezes the machine for several seconds).
This one should provide a workaround for the nx6125 and for possible other machines that show any weird _PPC behaviour.

I don't understand what the failure is, and why this workaround is effective. Is this a clue here to a real bug
that requires a real fix, rather than a workaround?

-Len

Quoted 49 lines

Signed-off-by Thomas Renninger <[email protected]>

---
 drivers/acpi/processor_perflib.c |   16 +++++++++++++++-
 1 file changed, 15 insertions(+), 1 deletion(-)

Index: linux-2.6.21/drivers/acpi/processor_perflib.c
===================================================================

drivers/acpi/processor_perflib.c +15 -1

@@ -60,6 +60,11 @@ static DEFINE_MUTEX(performance_mutex);
  * policy is adjusted accordingly.
  */
 
+static unsigned int ignore_ppc = 0;
+module_param(ignore_ppc, uint, 0644);
+MODULE_PARM_DESC(ignore_ppc, "If the frequency of your machine gets wrongly" \
+		 "limited by BIOS, this should help");
+
 #define PPC_REGISTERED   1
 #define PPC_IN_USE       2

@@ -72,6 +77,9 @@ static int acpi_processor_ppc_notifier(s
 	struct acpi_processor *pr;
 	unsigned int ppc = 0;
 
+	if (ignore_ppc)
+		return 0;
+
 	mutex_lock(&performance_mutex);
 
 	if (event != CPUFREQ_INCOMPATIBLE)

@@ -130,7 +138,13 @@ static int acpi_processor_get_platform_l
 
 int acpi_processor_ppc_has_changed(struct acpi_processor *pr)
 {
-	int ret = acpi_processor_get_platform_limit(pr);
+	int ret;
+
+	if (ignore_ppc)
+		return 0;
+
+	ret = acpi_processor_get_platform_limit(pr);
+
 	if (ret < 0)
 		return (ret);
 	else

Thomas Renninger

author

2007-05-19 19:41 UTC #2 lore raw

Quoted 11 lines · Len Brown

Quoted 9 lines · Thomas Renninger
Len, can you apply this one, pls.
Workaround for _PPC (BIOS cpufreq limitations)
There have been fixes using _PPC, which seem to unhide a problem on HP nx6125 (double cpufreq switch freezes the machine for several seconds).
This one should provide a workaround for the nx6125 and for possible other machines that show any weird _PPC behaviour.

Quoted 3 lines

I don't understand what the failure is, and why this workaround is effective. Is this a clue here to a real bug
that requires a real fix, rather than a workaround?

I am not sure what the real cause is, the machine hangs because CPU freq is limited and then all freqs are allowed again at the same time.

Also this is a SLES bug, not sure whether this also happens in mainline. (I mentioned that so if someone else with this machine experience that problem he has a pointer).
The reason why I think this should go into mainline is because there were
three patches concerning _PPC in the last months:
- one from Bruno Ducrot (which I expect unhided the other problems)
- one from Ingo Molnar (Do not read _PPC on startup - fixes some
ThinkPads)
- one from me to get highest freq again if booted on battery/limited
freq.

I expect more machines have problems here (or will have in future) and this is a nice
and easy possibility to workaround such problems, without loosing frequency
scaling functionality.

Thomas

Quoted 52 lines

-Len

Quoted 49 lines

Signed-off-by Thomas Renninger <[email protected]>

---
 drivers/acpi/processor_perflib.c |   16 +++++++++++++++-
 1 file changed, 15 insertions(+), 1 deletion(-)

Index: linux-2.6.21/drivers/acpi/processor_perflib.c
===================================================================

drivers/acpi/processor_perflib.c +15 -1

@@ -60,6 +60,11 @@ static DEFINE_MUTEX(performance_mutex);
  * policy is adjusted accordingly.
  */
 
+static unsigned int ignore_ppc = 0;
+module_param(ignore_ppc, uint, 0644);
+MODULE_PARM_DESC(ignore_ppc, "If the frequency of your machine gets wrongly" \
+		 "limited by BIOS, this should help");
+
 #define PPC_REGISTERED   1
 #define PPC_IN_USE       2

@@ -72,6 +77,9 @@ static int acpi_processor_ppc_notifier(s
 	struct acpi_processor *pr;
 	unsigned int ppc = 0;
 
+	if (ignore_ppc)
+		return 0;
+
 	mutex_lock(&performance_mutex);
 
 	if (event != CPUFREQ_INCOMPATIBLE)

@@ -130,7 +138,13 @@ static int acpi_processor_get_platform_l
 
 int acpi_processor_ppc_has_changed(struct acpi_processor *pr)
 {
-	int ret = acpi_processor_get_platform_limit(pr);
+	int ret;
+
+	if (ignore_ppc)
+		return 0;
+
+	ret = acpi_processor_get_platform_limit(pr);
+
 	if (ret < 0)
 		return (ret);
 	else