x86, fix smp_num_siblings calculation and usage
cover x86, fix smp_num_siblings calculation and usage
Paulo, this is what I'm going to send upstream tomorrow. FYI.
P.
----8<---
I have a system on which I have disabled threading in the BIOS, and I am booting the kernel with the option "idle=poll".
The kernel displays
process: WARNING: polling idle and HT enabled, performance may degrade
which is incorrect -- I've already disabled HT.
This warning is issued here:
void select_idle_routine(const struct cpuinfo_x86 *c)
{
if (boot_option_idle_override == IDLE_POLL && smp_num_siblings > 1)
pr_warn_once("WARNING: polling idle and HT enabled, performance may degrade\n");Quoted 1 line
From my understanding of the other areas of kernel that use
smp_num_siblings, the value is supposed to be the the number of threads per core.
The value of smp_num_siblings is incorrect. In theory, it should be 1 but it is reported as 2. When I looked into how smp_num_siblings is calculated I found the following call sequence in the kernel:
start_kernel ->
check_bugs ->
identify_boot_cpu ->
identify_cpu ->
c_init = init_intel
init_intel ->
detect_extended_topology
(sets value)
OR
c_init = init_amd
init_amd -> amd_detect_cmp
-> amd_get_topology
(sets value)
-> detect_ht()
... (sets value)
detect_ht()
(also sets value)ie) it is set three times in some cases and is overwritten by the call to detect_ht() from identify_cpu() in all cases.
It should be noted that nothing in the identify_cpu() path or the cpu_up() path requires smp_num_siblings to be set, prior to the final call to detect_ht().
For x86 boxes, smp_num_siblings is set to a value read in a CPUID call in detect_ht(). This value is the *factory defined* value in all cases; even if HT is disabled in BIOS the value still returns 2 if the CPU supports HT. AMD also reports the factory defined value in all cases.
That is, even with threading disabled,
crash> p smp_num_siblingssmp_num_siblings = $1 = 0x2
on processors that support multi-threading.
smp_num_siblings should be calculated a single time on cpu 0 to determine whether or not the system is multi-threaded or not.
On a system with HT enabled,
crash> p smp_num_siblingssmp_num_siblings = $1 = 0x2
On a system with HT disabled,
crash> p smp_num_siblingssmp_num_siblings = $1 = 0x1
Prarit Bhargava (2):
x86, Clean up smp_num_siblings calculation
x86, Calculate smp_num_siblings once
arch/x86/kernel/cpu/amd.c | 1 - arch/x86/kernel/cpu/common.c | 23 +++++++++++------------ arch/x86/kernel/cpu/topology.c | 2 +- arch/x86/kernel/smpboot.c | 10 +++++++--- 4 files changed, 19 insertions(+), 17 deletions(-)
Signature · 1 line
1.7.9.3
PATCH 1/2 x86, Clean up smp_num_siblings calculation
I have a system on which I have disabled threading in the BIOS, and I am booting the kernel with the option "idle=poll".
The kernel displays
process: WARNING: polling idle and HT enabled, performance may degrade
which is incorrect -- I've already disabled HT.
This warning is issued here:
void select_idle_routine(const struct cpuinfo_x86 *c)
{
if (boot_option_idle_override == IDLE_POLL && smp_num_siblings > 1)
pr_warn_once("WARNING: polling idle and HT enabled, performance may degrade\n");Quoted 1 line
From my understanding of the other ares of kernel that use
smp_num_siblings, the value is supposed to be the the number of threads per core.
The value of smp_num_siblings is incorrect. In theory, it should be 1 but it is reported as 2. When I looked into how smp_num_siblings is calculated I found the following call sequence in the kernel:
start_kernel ->
check_bugs ->
identify_boot_cpu ->
identify_cpu ->
c_init = init_intel
init_intel ->
detect_extended_topology
(sets value)
OR
c_init = init_amd
init_amd -> amd_detect_cmp
-> amd_get_topology
(sets value)
-> detect_ht()
... (sets value)
detect_ht()
(also sets value)ie) it is set three times in some cases and is overwritten by the call to detect_ht() from identify_cpu() in all cases.
It should be noted that nothing in the identify_cpu() path or the cpu_up() path requires smp_num_siblings to be set, prior to the final call to detect_ht().
For x86 boxes, smp_num_siblings is set to a value read in a CPUID call in detect_ht(). This value is the *factory defined* value in all cases; even if HT is disabled in BIOS the value still returns 2 if the CPU supports HT. AMD also reports the factory defined value in all cases.
Other uses of smp_num_siblings involve oprofile (used after boot), and the perf code which is done well after the initial cpus are brought online.
This patch removes dead code and moves the assignment of smp_num_siblings to only the detect_ht() code; it is still always reporting 2. A follow on patch will fix the calculation.
--- arch/x86/kernel/cpu/amd.c | 1 - arch/x86/kernel/cpu/common.c | 23 +++++++++++------------ arch/x86/kernel/cpu/topology.c | 2 +- arch/x86/kernel/smpboot.c | 5 ++--- 4 files changed, 14 insertions(+), 17 deletions(-)
arch/x86/kernel/cpu/amd.c +0 -1
@@ -304,7 +304,6 @@ static void amd_get_topology(struct cpuinfo_x86 *c) node_id = ecx &7 ;/* get compute unit information */ - smp_num_siblings = ((ebx >>8 ) &3 ) +1 ; c->compute_unit_id = ebx &0xff ; cores_per_cu += ((ebx >>8 ) &3 ); }else if (cpu_has (c,X86_FEATURE_NODEID_MSR )) {
arch/x86/kernel/cpu/common.c +11 -12
@@ -507,42 +507,41 @@ void detect_ht(struct cpuinfo_x86 *c)u32 eax, ebx, ecx, edx;int index_msb, core_bits;static bool printed; +int threads_per_core;if (!cpu_has (c,X86_FEATURE_HT ))return ; -if (cpu_has (c,X86_FEATURE_CMP_LEGACY )) +if (cpu_has (c,X86_FEATURE_CMP_LEGACY )) { + threads_per_core =1 ; goto out; + }if (cpu_has (c,X86_FEATURE_XTOPOLOGY ))return ;cpuid (1 , &eax, &ebx, &ecx, &edx); - smp_num_siblings = (ebx &0xff0000 ) >>16 ; + threads_per_core = smp_num_siblings = (ebx &0xff0000 ) >>16 ; -if (smp_num_siblings ==1 ) { -printk_once (KERN_INFO "CPU0: Hyper-Threading is disabled\n" ); +if (threads_per_core <=1 ) { +pr_info_once ("CPU: Hyper-Threading is unsupported on this processor.\n" ); goto out; } -if (smp_num_siblings <=1 ) - goto out; - - index_msb =get_count_order (smp_num_siblings); + index_msb =get_count_order (threads_per_core); c->phys_proc_id = apic->phys_pkg_id (c->initial_apicid, index_msb); - smp_num_siblings = smp_num_siblings / c->x86_max_cores; + threads_per_core = threads_per_core / c->x86_max_cores; - index_msb =get_count_order (smp_num_siblings); + index_msb =get_count_order (threads_per_core); core_bits =get_count_order (c->x86_max_cores); c->cpu_core_id = apic->phys_pkg_id (c->initial_apicid, index_msb) & ((1 << core_bits) -1 ); - out: -if (!printed && (c->x86_max_cores * smp_num_siblings) >1 ) { +if (!printed && (c->x86_max_cores * threads_per_core) >1 ) {printk (KERN_INFO "CPU: Physical Processor ID: %d\n" , c->phys_proc_id); printk(KERN_INFO "CPU: Processor Core ID: %d\n",
arch/x86/kernel/cpu/topology.c +1 -1
@@ -55,7 +55,7 @@ void detect_extended_topology(struct cpuinfo_x86 *c)/* * Populate HT related information from sub-leaf level 0. */ - core_level_siblings = smp_num_siblings =LEVEL_MAX_SIBLINGS (ebx); + core_level_siblings =LEVEL_MAX_SIBLINGS (ebx); core_plus_mask_width = ht_mask_width =BITS_SHIFT_NEXT_LEVEL (eax); sub_index =1 ;
arch/x86/kernel/smpboot.c +2 -3
@@ -351,8 +351,7 @@ static bool match_mc(struct cpuinfo_x86 *c, struct cpuinfo_x86 *o)void set_cpu_sibling_map (int cpu) { -bool has_smt = smp_num_siblings >1 ; -bool has_mp = has_smt || boot_cpu_data.x86_max_cores >1 ; +bool has_mp = boot_cpu_data.x86_max_cores >1 ;struct cpuinfo_x86 *c = &cpu_data (cpu);struct cpuinfo_x86 *o;int i;
@@ -370,7 +369,7 @@ void set_cpu_sibling_map(int cpu)for_each_cpu (i, cpu_sibling_setup_mask) { o = &cpu_data (i); -if ((i == cpu) || (has_smt &&match_smt (c, o))) +if ((i == cpu) ||match_smt (c, o))link_mask (sibling, cpu, i);if ((i == cpu) || (has_mp &&match_llc (c, o)))
Signature · 1 line
1.7.9.3
Re: [PATCH 1/2] x86, Clean up smp_num_siblings calculation
Hi Prarit,
See below,
Quoted 47 lines · Prarit Bhargava
I have a system on which I have disabled threading in the BIOS, and I am booting the kernel with the option "idle=poll".
The kernel displays
process: WARNING: polling idle and HT enabled, performance may degrade
which is incorrect -- I've already disabled HT.
This warning is issued here:
void
select_idle_routine(const struct cpuinfo_x86 *c){ if (boot_option_idle_override == IDLE_POLL && smp_num_siblings > 1) pr_warn_once("WARNING: polling idle and HT enabled, performance may degrade\n");Quoted 1 line
From my understanding of the other ares of kernel that use
smp_num_siblings, the value is supposed to be the the number of threads per core.The value of
smp_num_siblingsis incorrect. In theory, it should be 1 but it is reported as 2. When I looked into howsmp_num_siblingsis calculated I found the following call sequence in the kernel:
start_kernel->check_bugs -> identify_boot_cpu -> identify_cpu -> c_init = init_intel init_intel -> detect_extended_topology (sets value) OR c_init = init_amd init_amd -> amd_detect_cmp -> amd_get_topology (sets value) -> detect_ht() ... (sets value) detect_ht() (also sets value)ie) it is set three times in some cases and is overwritten by the call to
detect_ht()fromidentify_cpu()in all cases.
Quoted 8 lines
It should be noted that nothing in the
identify_cpu()path or thecpu_up()path requiressmp_num_siblingsto be set, prior to the final call todetect_ht().For x86 boxes,
smp_num_siblingsis set to a value read in a CPUID call indetect_ht(). This value is the *factory defined* value in all cases; even if HT is disabled in BIOS the value still returns 2 if the CPU supports HT. AMD also reports the factory defined value in all cases.
The above is incorrect in case of X-TOPOLOGY mode. I such a case the information about number of siblings comes from the LEVEL_MAX_SIBLINGS() macro and the X86_FEATURE_XTOPOLOGY flag is set to skip detect_ht() work :
void detect_ht(struct cpuinfo_x86 *c)
...
if (cpu_has(c, X86_FEATURE_XTOPOLOGY))
return;In addition, the information about the number of sibling no longer comes from CPUID(0x1)->ebx but rather from the 0xb leaf of CPUID.
I've marked below the problematic code change.
Thanks,
Oren Twaig
Quoted 105 lines
Other uses of
smp_num_siblingsinvolve oprofile (used after boot), and the perf code which is done well after the initial cpus are brought online.This patch removes dead code and moves the assignment of
smp_num_siblingsto only thedetect_ht()code; it is still always reporting 2. A follow on patch will fix the calculation.Cc Thomas Gleixner <[email protected]>Cc Ingo Molnar <[email protected]>Cc "H. Peter Anvin" <[email protected]>Cc Borislav Petkov <[email protected]>Cc Paul Gortmaker <[email protected]>Cc Andrew Morton <[email protected]>Cc Andi Kleen <[email protected]>Cc Dave Jones <[email protected]>Cc Torsten Kaiser <[email protected]>Cc Jan Beulich <[email protected]>Cc Jan Kiszka <[email protected]>Cc Toshi Kani <[email protected]>Cc Andrew Jones <[email protected]>Signed-off-by Prarit Bhargava <[email protected]>--- arch/x86/kernel/cpu/amd.c | 1 - arch/x86/kernel/cpu/common.c | 23 +++++++++++------------ arch/x86/kernel/cpu/topology.c | 2 +- arch/x86/kernel/smpboot.c | 5 ++--- 4 files changed, 14 insertions(+), 17 deletions(-)arch/x86/kernel/cpu/amd.c +0
-1@@ -304,7 +304,6 @@ static void amd_get_topology(struct cpuinfo_x86 *c) node_id = ecx &7 ;/* get compute unit information */ - smp_num_siblings = ((ebx >>8 ) &3 ) +1 ; c->compute_unit_id = ebx &0xff ; cores_per_cu += ((ebx >>8 ) &3 ); }else if (cpu_has (c,X86_FEATURE_NODEID_MSR )) {arch/x86/kernel/cpu/common.c +11
-12@@ -507,42 +507,41 @@ void detect_ht(struct cpuinfo_x86 *c)u32 eax, ebx, ecx, edx;int index_msb, core_bits;static bool printed; +int threads_per_core;if (!cpu_has (c,X86_FEATURE_HT ))return ; -if (cpu_has (c,X86_FEATURE_CMP_LEGACY )) +if (cpu_has (c,X86_FEATURE_CMP_LEGACY )) { + threads_per_core =1 ; goto out; + }if (cpu_has (c,X86_FEATURE_XTOPOLOGY ))return ;cpuid (1 , &eax, &ebx, &ecx, &edx); - smp_num_siblings = (ebx &0xff0000 ) >>16 ; + threads_per_core = smp_num_siblings = (ebx &0xff0000 ) >>16 ; -if (smp_num_siblings ==1 ) { -printk_once (KERN_INFO "CPU0: Hyper-Threading is disabled\n" ); +if (threads_per_core <=1 ) { +pr_info_once ("CPU: Hyper-Threading is unsupported on this processor.\n" ); goto out; } -if (smp_num_siblings <=1 ) - goto out; - - index_msb =get_count_order (smp_num_siblings); + index_msb =get_count_order (threads_per_core); c->phys_proc_id = apic->phys_pkg_id (c->initial_apicid, index_msb); - smp_num_siblings = smp_num_siblings / c->x86_max_cores; + threads_per_core = threads_per_core / c->x86_max_cores; - index_msb =get_count_order (smp_num_siblings); + index_msb =get_count_order (threads_per_core); core_bits =get_count_order (c->x86_max_cores); c->cpu_core_id = apic->phys_pkg_id (c->initial_apicid, index_msb) & ((1 << core_bits) -1 ); - out: -if (!printed && (c->x86_max_cores * smp_num_siblings) >1 ) { +if (!printed && (c->x86_max_cores * threads_per_core) >1 ) {printk (KERN_INFO "CPU: Physical Processor ID: %d\n" , c->phys_proc_id); printk(KERN_INFO "CPU: Processor Core ID: %d\n",arch/x86/kernel/cpu/topology.c +0
-0
arch/x86/kernel/cpu/topology.c void detect_extended_topology(struct cpuinfo_x86 *c) · line 58
Quoted 5 lines
arch/x86/kernel/cpu/topology.c +1
-1@@ -55,7 +55,7 @@ void detect_extended_topology(struct cpuinfo_x86 *c)/* * Populate HT related information from sub-leaf level 0. */ - core_level_siblings = smp_num_siblings =LEVEL_MAX_SIBLINGS (ebx); + core_level_siblings =LEVEL_MAX_SIBLINGS (ebx);
The above is the problem which will make smp_num_sibling to be uninitialised in case of X-TOPOLOGY.
Quoted 26 lines
arch/x86/kernel/cpu/topology.c +0
-0core_plus_mask_width = ht_mask_width =BITS_SHIFT_NEXT_LEVEL (eax); sub_index =1 ;arch/x86/kernel/smpboot.c +2
-3@@ -351,8 +351,7 @@ static bool match_mc(struct cpuinfo_x86 *c, struct cpuinfo_x86 *o)void set_cpu_sibling_map (int cpu) { -bool has_smt = smp_num_siblings >1 ; -bool has_mp = has_smt || boot_cpu_data.x86_max_cores >1 ; +bool has_mp = boot_cpu_data.x86_max_cores >1 ;struct cpuinfo_x86 *c = &cpu_data (cpu);struct cpuinfo_x86 *o;int i;@@ -370,7 +369,7 @@ void set_cpu_sibling_map(int cpu)for_each_cpu (i, cpu_sibling_setup_mask) { o = &cpu_data (i); -if ((i == cpu) || (has_smt &&match_smt (c, o))) +if ((i == cpu) ||match_smt (c, o))link_mask (sibling, cpu, i);if ((i == cpu) || (has_mp &&match_llc (c, o)))
Re: [PATCH 1/2] x86, Clean up smp_num_siblings calculation
Quoted 66 lines · Oren Twaig
Hi Prarit,
See below,
Quoted 47 lines · Prarit Bhargava
I have a system on which I have disabled threading in the BIOS, and I am booting the kernel with the option "idle=poll".
The kernel displays
process: WARNING: polling idle and HT enabled, performance may degrade
which is incorrect -- I've already disabled HT.
This warning is issued here:
void
select_idle_routine(const struct cpuinfo_x86 *c){ if (boot_option_idle_override == IDLE_POLL && smp_num_siblings > 1) pr_warn_once("WARNING: polling idle and HT enabled, performance may degrade\n");Quoted 1 line
From my understanding of the other ares of kernel that use
smp_num_siblings, the value is supposed to be the the number of threads per core.The value of
smp_num_siblingsis incorrect. In theory, it should be 1 but it is reported as 2. When I looked into howsmp_num_siblingsis calculated I found the following call sequence in the kernel:
start_kernel->check_bugs -> identify_boot_cpu -> identify_cpu -> c_init = init_intel init_intel -> detect_extended_topology (sets value) OR c_init = init_amd init_amd -> amd_detect_cmp -> amd_get_topology (sets value) -> detect_ht() ... (sets value) detect_ht() (also sets value)ie) it is set three times in some cases and is overwritten by the call to
detect_ht()fromidentify_cpu()in all cases.Quoted 8 lines
It should be noted that nothing in the
identify_cpu()path or thecpu_up()path requiressmp_num_siblingsto be set, prior to the final call todetect_ht().For x86 boxes,
smp_num_siblingsis set to a value read in a CPUID call indetect_ht(). This value is the *factory defined* value in all cases; even if HT is disabled in BIOS the value still returns 2 if the CPU supports HT. AMD also reports the factory defined value in all cases.The above is incorrect in case of X-TOPOLOGY mode. I such a case the information about number of siblings comes from the
LEVEL_MAX_SIBLINGS()macro and theX86_FEATURE_XTOPOLOGYflag is set to skipdetect_ht()work :
voiddetect_ht(struct cpuinfo_x86 *c)
...
Quoted 7 lines
if (cpu_has(c, X86_FEATURE_XTOPOLOGY)) return;In addition, the information about the number of sibling no longer comes from CPUID(0x1)->ebx but rather from the
0xbleaf of CPUID.I've marked below the problematic code change.
I will do a [v2] of the patchset that omits this change
Quoted 2 lines
arch/x86/kernel/cpu/topology.c void detect_extended_topology(struct cpuinfo_x86 *c) · line 58
Quoted 2 lines
arch/x86/kernel/cpu/topology.c +1
-1- core_level_siblings = smp_num_siblings =LEVEL_MAX_SIBLINGS (ebx); + core_level_siblings =LEVEL_MAX_SIBLINGS (ebx);
and then removes the setting of smp_num_siblings in 2/2.
Thanks,
P.
PATCH 2/2 x86, Calculate smp_num_siblings once
In reply to Prarit BhargavaFor x86 boxes, smp_num_siblings is set to a value read in a CPUID call in detect_ht(). This value is the *factory defined* value in all cases; even if HT is disabled in BIOS the value still returns 2 if the CPU supports HT. AMD also reports the factory defined value in all cases.
That is, even with threading disabled,
crash> p smp_num_siblingssmp_num_siblings = $1 = 0x2
on processors that support multi-threading.
smp_num_siblings should be calculated a single time on cpu 0 to determine whether or not the system is multi-threaded or not.
On a system with HT enabled,
crash> p smp_num_siblingssmp_num_siblings = $1 = 0x2
On a system with HT disabled,
crash> p smp_num_siblingssmp_num_siblings = $1 = 0x1
--- arch/x86/kernel/cpu/common.c | 2 +- arch/x86/kernel/smpboot.c | 5 +++++ 2 files changed, 6 insertions(+), 1 deletion(-)
arch/x86/kernel/cpu/common.c +1 -1
@@ -522,7 +522,7 @@ void detect_ht(struct cpuinfo_x86 *c)cpuid (1 , &eax, &ebx, &ecx, &edx); - threads_per_core = smp_num_siblings = (ebx &0xff0000 ) >>16 ; + threads_per_core = (ebx &0xff0000 ) >>16 ;if (threads_per_core <=1 ) {pr_info_once ("CPU: Hyper-Threading is unsupported on this processor.\n" );
arch/x86/kernel/smpboot.c +5 -0
@@ -363,6 +363,7 @@ void set_cpu_sibling_map(int cpu)cpumask_set_cpu (cpu,cpu_llc_shared_mask (cpu));cpumask_set_cpu (cpu,cpu_core_mask (cpu)); c->booted_cores =1 ; + smp_num_siblings =1 ;return ; }
@@ -407,6 +408,10 @@ void set_cpu_sibling_map(int cpu) c->booted_cores =cpu_data (i).booted_cores; } } + +/* Only need to check this on the boot cpu, o/w it is disabled */ +if (cpu ==0 ) + smp_num_siblings =cpumask_weight (cpu_sibling_mask (cpu)); }/* maps the cpu to the sched domain representing multi-core */
Signature · 1 line
1.7.9.3