x86, therm: Only read the initial value of thermal LVT entry on BSP
PATCH 1/1 x86, therm: Only read the initial value of thermal LVT entry on BSP
Only read the initial value of thermal LVT entry on BSP. The initial value of thermal LVT entries on all APs always reads 0x10000 because APs are woken up by BSP issuing INIT-SIPI-SIPI sequence to them and LVT registers are reset to 0s except for mask bits which are set to 1s when APs receive INIT IPI.
Also restore the value that BIOS has programmed on AP based on BSP's info we saved since BIOS is always setting the same value for all threads/cores.
--- arch/x86/kernel/cpu/mcheck/therm_throt.c | 20 +++++++++++++++++++- 1 file changed, 19 insertions(+), 1 deletion(-)
arch/x86/kernel/cpu/mcheck/therm_throt.c +19 -1
@@ -259,6 +259,7 @@ void intel_init_thermal(struct cpuinfo_x86 *c)unsigned int cpu =smp_processor_id ();int tm2 =0 ;u32 l, h; +static u32 lvtthmr;/* Thermal monitoring depends on ACPI and clock modulation*/ if (!cpu_has (c,X86_FEATURE_ACPI ) || !cpu_has (c,X86_FEATURE_ACC ))
@@ -270,7 +271,24 @@ void intel_init_thermal(struct cpuinfo_x86 *c)* since it might be delivered via SMI already: */ rdmsr (MSR_IA32_MISC_ENABLE , l, h); - h =apic_read (APIC_LVTTHMR ); + +/* +* Only read the initial value of thermal LVT entry on BSP. The +* initial value of thermal LVT entries on all APs always reads +* 0x10000 because APs are woken up by BSP issuing INIT-SIPI-SIPI +* sequence to them and LVT registers are reset to 0s except for +* the mask bits which are set to 1s when APs receive INIT IPI. +* Also restore the value that BIOS has programmed on AP based on +* BSP's info we saved since BIOS is always setting the same value +* for all threads/cores +*/ +if (cpu ==0 ) + lvtthmr =apic_read (APIC_LVTTHMR ); +else +apic_write (APIC_LVTTHMR , lvtthmr); + + h = lvtthmr; +if ((l &MSR_IA32_MISC_ENABLE_TM1 ) && (h &APIC_DM_SMI )) {printk (KERN_DEBUG "CPU%d: Thermal monitoring handled by SMI\n" , cpu);
Quoted 9 lines · Yong Wang
Only read the initial value of thermal LVT entry on BSP. The initial value of thermal LVT entries on all APs always reads
0x10000because APs are woken up by BSP issuing INIT-SIPI-SIPI sequence to them and LVT registers are reset to 0s except for mask bits which are set to 1s when APs receive INIT IPI.Also restore the value that BIOS has programmed on AP based on BSP's info we saved since BIOS is always setting the same value for all threads/cores.
Yong, I have appended a new patch with an enhanced change log and subject. In future, when you modify and post another version of the patch, can you please update the patch version and elaborate what has changed, why etc, so that it will be easier for the reviewers.
Ingo/Peter, please review and queue this patch from Yong. Thanks.
---
From: Yong Wang <[email protected]>
Subject: x86: under bios control, restore AP's APIC_LVTTHMR to the BSP value
On platforms where bios handles the thermal monitor interrupt, APIC_LVTTHMR on each logical CPU is programmed to generate a SMI and OS can't touch it.
Unfortunately AP bringup sequence using INIT-SIPI-SIPI clear all the LVT entries except the mask bit. Essentially this results in all LVT entries including the thermal monitoring interrupt set to masked (clearing the bios programmed value for APIC_LVTTHMR).
And this leads to kernel take over the thermal monitoring interrupt on AP's but not on BSP (leaving the bios programmed value only on BSP).
As a result of this, we have seen system hangs when the thermal monitoring interrupt is generated.
Fix this by reading the initial value of thermal LVT entry on BSP and if bios has taken over the control, then program the same value on all AP's and leave the thermal monitoring interrupt control on all the logical cpu's to the bios.
--- arch/x86/kernel/cpu/mcheck/therm_throt.c | 20 +++++++++++++++++++- 1 file changed, 19 insertions(+), 1 deletion(-)
arch/x86/kernel/cpu/mcheck/therm_throt.c +19 -1
@@ -259,6 +259,7 @@ void intel_init_thermal(struct cpuinfo_x86 *c)unsigned int cpu =smp_processor_id ();int tm2 =0 ;u32 l, h; +static u32 lvtthmr;/* Thermal monitoring depends on ACPI and clock modulation*/ if (!cpu_has (c,X86_FEATURE_ACPI ) || !cpu_has (c,X86_FEATURE_ACC ))
@@ -270,7 +271,24 @@ void intel_init_thermal(struct cpuinfo_x86 *c)* since it might be delivered via SMI already: */ rdmsr (MSR_IA32_MISC_ENABLE , l, h); - h =apic_read (APIC_LVTTHMR ); + +/* +* Only read the initial value of thermal LVT entry on BSP. The +* initial value of thermal LVT entries on all APs always reads +* 0x10000 because APs are woken up by BSP issuing INIT-SIPI-SIPI +* sequence to them and LVT registers are reset to 0s except for +* the mask bits which are set to 1s when APs receive INIT IPI. +* Also restore the value that BIOS has programmed on AP based on +* BSP's info we saved since BIOS is always setting the same value +* for all threads/cores +*/ +if (cpu ==0 ) + lvtthmr =apic_read (APIC_LVTTHMR ); +else +apic_write (APIC_LVTTHMR , lvtthmr); + + h = lvtthmr; +if ((l &MSR_IA32_MISC_ENABLE_TM1 ) && (h &APIC_DM_SMI )) {printk (KERN_DEBUG "CPU%d: Thermal monitoring handled by SMI\n" , cpu);
Quoted 7 lines · Suresh Siddha
Quoted 6 lines · Yong Wang
Only read the initial value of thermal LVT entry on BSP. The initial value of thermal LVT entries on all APs always reads
0x10000because APs are woken up by BSP issuing INIT-SIPI-SIPI sequence to them and LVT registers are reset to 0s except for mask bits which are set to 1s when APs receive INIT IPI.
Quoted 9 lines
Quoted 3 lines
Also restore the value that BIOS has programmed on AP based on BSP's info we saved since BIOS is always setting the same value for all threads/cores.
Yong, I have appended a new patch with an enhanced change log and subject. In future, when you modify and post another version of the patch, can you please update the patch version and elaborate what has changed, why etc, so that it will be easier for the reviewers.
Thanks a lot for your review and comments. Will follow the conventions going forward.
Quoted 74 lines · Suresh Siddha
Quoted 9 lines · Yong Wang
Only read the initial value of thermal LVT entry on BSP. The initial value of thermal LVT entries on all APs always reads
0x10000because APs are woken up by BSP issuing INIT-SIPI-SIPI sequence to them and LVT registers are reset to 0s except for mask bits which are set to 1s when APs receive INIT IPI.Also restore the value that BIOS has programmed on AP based on BSP's info we saved since BIOS is always setting the same value for all threads/cores.
Yong, I have appended a new patch with an enhanced change log and subject. In future, when you modify and post another version of the patch, can you please update the patch version and elaborate what has changed, why etc, so that it will be easier for the reviewers.
Ingo/Peter, please review and queue this patch from Yong. Thanks.
---From: Yong Wang <[email protected]>
Subject: x86: under bios control, restore AP'sAPIC_LVTTHMRto the BSP valueOn platforms where bios handles the thermal monitor interrupt,
APIC_LVTTHMRon each logical CPU is programmed to generate a SMI and OS can't touch it.Unfortunately AP bringup sequence using INIT-SIPI-SIPI clear all the LVT entries except the mask bit. Essentially this results in all LVT entries including the thermal monitoring interrupt set to masked (clearing the bios programmed value for
APIC_LVTTHMR).And this leads to kernel take over the thermal monitoring interrupt on AP's but not on BSP (leaving the bios programmed value only on BSP).
As a result of this, we have seen system hangs when the thermal monitoring interrupt is generated.
Fix this by reading the initial value of thermal LVT entry on BSP and if bios has taken over the control, then program the same value on all AP's and leave the thermal monitoring interrupt control on all the logical cpu's to the bios.
Signed-off-by Yong Wang <[email protected]>Reviewed-by Suresh Siddha <[email protected]>--- arch/x86/kernel/cpu/mcheck/therm_throt.c | 20 +++++++++++++++++++- 1 file changed, 19 insertions(+), 1 deletion(-)arch/x86/kernel/cpu/mcheck/therm_throt.c +9
-1@@ -259,6 +259,7 @@ void intel_init_thermal(struct cpuinfo_x86 *c)unsigned int cpu =smp_processor_id ();int tm2 =0 ;u32 l, h; +static u32 lvtthmr;/* Thermal monitoring depends on ACPI and clock modulation*/ if (!cpu_has (c,X86_FEATURE_ACPI ) || !cpu_has (c,X86_FEATURE_ACC ))@@ -270,7 +271,24 @@ void intel_init_thermal(struct cpuinfo_x86 *c)* since it might be delivered via SMI already: */ rdmsr (MSR_IA32_MISC_ENABLE , l, h); - h =apic_read (APIC_LVTTHMR ); + +/* +* Only read the initial value of thermal LVT entry on BSP. The +* initial value of thermal LVT entries on all APs always reads +* 0x10000 because APs are woken up by BSP issuing INIT-SIPI-SIPI +* sequence to them and LVT registers are reset to 0s except for +* the mask bits which are set to 1s when APs receive INIT IPI. +* Also restore the value that BIOS has programmed on AP based on
arch/x86/kernel/cpu/mcheck/therm_throt.c void intel_init_thermal(struct cpuinfo_x86 *c) · line 291
Quoted 10 lines
arch/x86/kernel/cpu/mcheck/therm_throt.c +10
-0+* BSP's info we saved since BIOS is always setting the same value +* for all threads/cores +*/ +if (cpu ==0 ) + lvtthmr =apic_read (APIC_LVTTHMR ); +else +apic_write (APIC_LVTTHMR , lvtthmr); + + h = lvtthmr; +
i dont disagree with the fix, but could we please do it a bit cleaner, and initialize a proper file-scope lvtthrm_init value from a different boot-CPU-only function? (not intel_init_thermal)
that makes it cleaner, and also it will work if we dont boot on cpu==0. (should that ever occur)
Thanks,
Ingo
Quoted 5 lines · Ingo Molnar
i dont disagree with the fix, but could we please do it a bit cleaner, and initialize a proper file-scope
lvtthrm_initvalue from a different boot-CPU-only function? (notintel_init_thermal)
Thanks for your comments. Just want to make sure I understand correctly. By 'file-scope', do you want me to define lvtthrm_init as a static variable but not to define it in any function?
Quoted 3 lines
that makes it cleaner, and also it will work if we dont boot on cpu==0. (should that ever occur)
May I know when will this happen?
Thanks
-Yong
Quoted 10 lines · Yong Wang
Quoted 5 lines · Ingo Molnar
i dont disagree with the fix, but could we please do it a bit cleaner, and initialize a proper file-scope
lvtthrm_initvalue from a different boot-CPU-only function? (notintel_init_thermal)Thanks for your comments. Just want to make sure I understand correctly. By 'file-scope', do you want me to define
lvtthrm_initas a static variable but not to define it in any function?
Correct - i'd suggest to put it next to other file-scope variables at the top of the .c file. Maybe make it __read_mostly as well.
Quoted 5 lines
Quoted 3 lines
that makes it cleaner, and also it will work if we dont boot on cpu==0. (should that ever occur)
May I know when will this happen?
It's not really expected - we factorize the CPU IDs (which are logical) so that the boot CPU is 0. But relying on cpu==0 is the boot cpu is not clean and the resulting code is harder to read.
Ingo
[Ingo Molnar - Sun, Nov 08, 2009 at 02:16:55PM +0100] |
...
| | > > that makes it cleaner, and also it will work if we dont boot on | > > cpu==0. (should that ever occur) | > > | > | > May I know when will this happen? | | It's not really expected - we factorize the CPU IDs (which are logical) | so that the boot CPU is 0. But relying on cpu==0 is the boot cpu is not | clean and the resulting code is harder to read.
Perhaps we need some is_bsp_cpu() helper? Though to cover all x86 places we need some efforts to apply :/
| | Ingo |
-- Cyrill
Quoted 10 lines · Ingo Molnar
Quoted 7 lines · Yong Wang
Quoted 5 lines · Ingo Molnar
i dont disagree with the fix, but could we please do it a bit cleaner, and initialize a proper file-scope
lvtthrm_initvalue from a different boot-CPU-only function? (notintel_init_thermal)
Quoted 7 lines
Quoted 3 lines
Thanks for your comments. Just want to make sure I understand correctly. By 'file-scope', do you want me to define
lvtthrm_initas a static variable but not to define it in any function?Correct - i'd suggest to put it next to other file-scope variables at the top of the .c file. Maybe make it
__read_mostlyas well.
OK, will do.
Quoted 10 lines
Quoted 5 lines
Quoted 3 lines
that makes it cleaner, and also it will work if we dont boot on cpu==0. (should that ever occur)
May I know when will this happen?
It's not really expected - we factorize the CPU IDs (which are logical) so that the boot CPU is 0. But relying on cpu==0 is the boot cpu is not clean and the resulting code is harder to read.
There does not seem to be 'boot-CPU-only function' in that .c file. What about changing cpu==0 to cpu==boot_cpu_id? Does that help?
Thanks
-Yong
Quoted 27 lines · Yong Wang
Quoted 10 lines · Ingo Molnar
Quoted 7 lines · Yong Wang
Quoted 5 lines · Ingo Molnar
i dont disagree with the fix, but could we please do it a bit cleaner, and initialize a proper file-scope
lvtthrm_initvalue from a different boot-CPU-only function? (notintel_init_thermal)Quoted 7 lines
Quoted 3 lines
Thanks for your comments. Just want to make sure I understand correctly. By 'file-scope', do you want me to define
lvtthrm_initas a static variable but not to define it in any function?Correct - i'd suggest to put it next to other file-scope variables at the top of the .c file. Maybe make it
__read_mostlyas well.OK, will do.
Quoted 6 lines
Quoted 5 lines
Quoted 3 lines
that makes it cleaner, and also it will work if we dont boot on cpu==0. (should that ever occur)
May I know when will this happen?
Quoted 7 lines
Quoted 4 lines
It's not really expected - we factorize the CPU IDs (which are logical) so that the boot CPU is 0. But relying on cpu==0 is the boot cpu is not clean and the resulting code is harder to read.
There does not seem to be 'boot-CPU-only function' in that .c file. What about changing cpu==0 to cpu==boot_cpu_id? Does that help?
Then create one and call it - it's still cleaner that way.
Ingo