x86/resctrl: Support wider MBM counters
Subsystems (2)
Reinette Chatre
maintainercover x86/resctrl: Support wider MBM counters
Memory Bandwidth Monitoring (MBM) is an Intel Resource Director Technology (RDT) feature that tracks Total and Local bandwidth generated which misses the L3 cache.
The original Memory Bandwidth Monitoring (MBM) architectural definition defines counters of up to 62 bits and the first-generation MBM implementation uses 24 bit counters. Software is required to poll at 1 second or faster to ensure that data is retrieved before a counter rollover occurs more than once under worst conditions.
As system bandwidths scale the software requirement is maintained with the introduction of a per-resource enumerable MBM counter width.
This series adds support for the new enumerable MBM counter width.
Details about the feature can be found in Chapter 9 of the most recent Intel ISE available from
https://software.intel.com/sites/default/files/managed/c5/15/architecture-instruction-set-extensions-programming-reference.pdf
Reinette Chatre (2):
x86/resctrl: Maintain MBM counter width per resource x86/resctrl: Support CPUID enumeration of MBM counter width
arch/x86/include/asm/processor.h | 1 + arch/x86/kernel/cpu/common.c | 5 +++++ arch/x86/kernel/cpu/resctrl/ctrlmondata.c | 8 ++++--- arch/x86/kernel/cpu/resctrl/internal.h | 15 ++++++++++--- arch/x86/kernel/cpu/resctrl/monitor.c | 27 ++++++++++++++++------- arch/x86/kernel/cpu/resctrl/rdtgroup.c | 2 +- 6 files changed, 43 insertions(+), 15 deletions(-)
Signature · 1 line
2.21.0
Reinette Chatre
maintainerPATCH 1/2 x86/resctrl: Maintain MBM counter width per resource
The original Memory Bandwidth Monitoring (MBM) architectural definition defines counters of up to 62 bits in the IA32_QM_CTR MSR, and the first-generation MBM implementation uses 24 bit counters. Software is required to poll at 1 second or faster to ensure that data is retrieved before a counter rollover occurs more than once under worst conditions.
As system bandwidths scale the software requirement is maintained with the introduction of a per-resource enumerable MBM counter width.
In preparation for supporting hardware with an enumerable MBM counter width the current globally static MBM counter width is moved to a per-resource MBM counter width. Currently initialized to 24 always to result in no functional change.
In essence there is one function, mbm_overflow_count() that needs to know the counter width to handle rollovers. The static value used within mbm_overflow_count() will be replaced with a value discovered from the hardware. Support for learning the MBM counter width from hardware is added in the change that follows.
--- arch/x86/kernel/cpu/resctrl/ctrlmondata.c | 8 +++++--- arch/x86/kernel/cpu/resctrl/internal.h | 7 +++++-- arch/x86/kernel/cpu/resctrl/monitor.c | 21 +++++++++++++-------- arch/x86/kernel/cpu/resctrl/rdtgroup.c | 2 +- 4 files changed, 24 insertions(+), 14 deletions(-)
arch/x86/kernel/cpu/resctrl/ctrlmondata.c +5 -3
@@ -495,14 +495,16 @@ int rdtgroup_schemata_show(struct kernfs_open_file *of,return ret; } -void mon_event_read (struct rmid_read *rr,struct rdt_domain *d, -struct rdtgroup *rdtgrp,int evtid,int first) +void mon_event_read (struct rmid_read *rr,struct rdt_resource *r, +struct rdt_domain *d,struct rdtgroup *rdtgrp, +int evtid,int first) {/* * setup the parameters to send to the IPI to read the data. */ rr->rgrp = rdtgrp; rr->evtid = evtid; + rr->r = r; rr->d = d; rr->val =0 ; rr->first = first;
@@ -539,7 +541,7 @@ int rdtgroup_mondata_show(struct seq_file *m, void *arg) goto out; } -mon_event_read (&rr, d, rdtgrp, evtid, false); +mon_event_read (&rr, r, d, rdtgrp, evtid, false);if (rr.val &RMID_VAL_ERROR )seq_puts (m,"Error\n" );
arch/x86/kernel/cpu/resctrl/internal.h +5 -2
@@ -87,6 +87,7 @@ union mon_data_bits {struct rmid_read {struct rdtgroup *rgrp; +struct rdt_resource *r;struct rdt_domain *d;int evtid;bool first;
@@ -460,6 +461,7 @@ struct rdt_resource {struct list_head evt_list;int num_rmid;unsigned int mon_scale; +unsigned int mbm_width;unsigned long fflags; };
@@ -587,8 +589,9 @@ void rmdir_mondata_subdir_allrdtgrp(struct rdt_resource *r, unsigned int dom_id);void mkdir_mondata_subdir_allrdtgrp (struct rdt_resource *r,struct rdt_domain *d); -void mon_event_read (struct rmid_read *rr,struct rdt_domain *d, -struct rdtgroup *rdtgrp,int evtid,int first); +void mon_event_read (struct rmid_read *rr,struct rdt_resource *r, +struct rdt_domain *d,struct rdtgroup *rdtgrp, +int evtid,int first);void mbm_setup_overflow_handler (struct rdt_domain *dom,unsigned long delay_ms);void mbm_handle_overflow (struct work_struct *work);
arch/x86/kernel/cpu/resctrl/monitor.c +13 -8
@@ -214,9 +214,9 @@ void free_rmid(u32 rmid)list_add_tail (&entry->list, &rmid_free_lru); } -static u64 mbm_overflow_count (u64 prev_msr,u64 cur_msr) +static u64 mbm_overflow_count (u64 prev_msr,u64 cur_msr,unsigned int width) { -u64 shift =64 -MBM_CNTR_WIDTH , chunks; +u64 shift =64 - width, chunks; chunks = (cur_msr << shift) - (prev_msr << shift);return chunks >>= shift;
@@ -256,7 +256,7 @@ static int __mon_event_count(u32 rmid, struct rmid_read *rr)return 0 ; } - chunks =mbm_overflow_count (m->prev_msr, tval); + chunks =mbm_overflow_count (m->prev_msr, tval, rr->r->mbm_width); m->chunks += chunks; m->prev_msr = tval;
@@ -278,7 +278,7 @@ static void mbm_bw_count(u32 rmid, struct rmid_read *rr)if (tval & (RMID_VAL_ERROR |RMID_VAL_UNAVAIL ))return ; - chunks =mbm_overflow_count (m->prev_bw_msr, tval); + chunks =mbm_overflow_count (m->prev_bw_msr, tval, rr->r->mbm_width); m->chunks_bw += chunks; m->chunks = m->chunks_bw; cur_bw = (chunks * r->mon_scale) >>20 ;
@@ -433,11 +433,12 @@ static void update_mba_bw(struct rdtgroup *rgrp, struct rdt_domain *dom_mbm) } } -static void mbm_update (struct rdt_domain *d,int rmid) +static void mbm_update (struct rdt_resource *r,struct rdt_domain *d,int rmid) {struct rmid_read rr; rr.first = false; + rr.r = r; rr.d = d;/*
@@ -510,6 +511,7 @@ void mbm_handle_overflow(struct work_struct *work)struct rdtgroup *prgrp, *crgrp;int cpu =smp_processor_id ();struct list_head *head; +struct rdt_resource *r;struct rdt_domain *d;mutex_lock (&rdtgroup_mutex);
@@ -517,16 +519,18 @@ void mbm_handle_overflow(struct work_struct *work)if (!static_branch_likely (&rdt_mon_enable_key)) goto out_unlock; - d =get_domain_from_cpu (cpu, &rdt_resources_all[RDT_RESOURCE_L3 ]); + r = &rdt_resources_all[RDT_RESOURCE_L3 ]; + + d =get_domain_from_cpu (cpu, r);if (!d) goto out_unlock;list_for_each_entry (prgrp, &rdt_all_groups, rdtgroup_list) { -mbm_update (d, prgrp->mon.rmid); +mbm_update (r, d, prgrp->mon.rmid); head = &prgrp->mon.crdtgrp_list;list_for_each_entry (crgrp, head, mon.crdtgrp_list) -mbm_update (d, crgrp->mon.rmid); +mbm_update (r, d, crgrp->mon.rmid);if (is_mba_sc (NULL ))update_mba_bw (prgrp, d);
@@ -619,6 +623,7 @@ int rdt_get_mon_l3_config(struct rdt_resource *r) r->mon_scale = boot_cpu_data.x86_cache_occ_scale; r->num_rmid = boot_cpu_data.x86_cache_max_rmid +1 ; + r->mbm_width =MBM_CNTR_WIDTH ;/* * A reasonable upper limit on the max threshold is the number
arch/x86/kernel/cpu/resctrl/rdtgroup.c +1 -1
@@ -2459,7 +2459,7 @@ static int mkdir_mondata_subdir(struct kernfs_node *parent_kn, goto out_destroy;if (is_mbm_event (mevt->evtid)) -mon_event_read (&rr, d, prgrp, mevt->evtid, true); +mon_event_read (&rr, r, d, prgrp, mevt->evtid, true); }kernfs_activate (kn);return 0 ;
Signature · 1 line
2.21.0
Reinette Chatre
maintainerPATCH 2/2 x86/resctrl: Support CPUID enumeration of MBM counter width
In reply to Reinette ChatreThe original Memory Bandwidth Monitoring (MBM) architectural definition defines counters of up to 62 bits in the IA32_QM_CTR MSR while the first-generation MBM implementation uses statically defined 24 bit counters.
Expand the MBM CPUID enumeration properties to include the MBM counter width. The previously unused EAX output register contains, in bits [7:0], the MBM counter width encoded as an offset from 24 bits. Enumerating this property is only specified for Intel CPUs. A vendor check is added to ensure original behavior is maintained for AMD CPUs.
While eight bits are available for the counter width offset IA32_QM_CTR MSR only supports 62 bit counters. A sanity check, with warning printed when encountered, is added to ensure counters cannot exceed the 62 bit limit.
---
Dear Maintainers,
Because of the need for a vendor check there seems to be a few ways in which enumerating this new MBM property can be implemented. While this implementation seems to be the simplest it does introduce an explicit vendor check into arch/x86/kernel/cpu/common.c where there does not seem to be a precedent for such a change.
The goal is to support this correctly and I look forward to your guidance on how the enable this feature correctly.
Regards,
Reinette
arch/x86/include/asm/processor.h | 1 + arch/x86/kernel/cpu/common.c | 5 +++++ arch/x86/kernel/cpu/resctrl/internal.h | 8 +++++++- arch/x86/kernel/cpu/resctrl/monitor.c | 8 +++++++- 4 files changed, 20 insertions(+), 2 deletions(-)
arch/x86/include/asm/processor.h +1 -0
@@ -115,6 +115,7 @@ struct cpuinfo_x86 {/* Cache QoS architectural values: */ int x86_cache_max_rmid;/* max index */ int x86_cache_occ_scale;/* scale to bytes */ +int x86_cache_mbm_width_offset;int x86_power;unsigned long loops_per_jiffy;/* cpuid returned max cores value: */
arch/x86/kernel/cpu/common.c +5 -0
@@ -856,6 +856,8 @@ static void init_speculation_control(struct cpuinfo_x86 *c)static void init_cqm (struct cpuinfo_x86 *c) { + c->x86_cache_mbm_width_offset =-1 ; +if (!cpu_has (c,X86_FEATURE_CQM_LLC )) { c->x86_cache_max_rmid =-1 ; c->x86_cache_occ_scale =-1 ;
@@ -875,6 +877,9 @@ static void init_cqm(struct cpuinfo_x86 *c) c->x86_cache_max_rmid = ecx; c->x86_cache_occ_scale = ebx; +/* EAX contents is only defined for Intel CPUs */ +if (c->x86_vendor ==X86_VENDOR_INTEL ) + c->x86_cache_mbm_width_offset = eax &0xff ; } }
arch/x86/kernel/cpu/resctrl/internal.h +7 -1
@@ -31,7 +31,7 @@#define CQM_LIMBOCHECK_INTERVAL 1000 -#define MBM_CNTR_WIDTH 24 +#define MBM_CNTR_WIDTH_BASE 24#define MBM_OVERFLOW_INTERVAL 1000#define MAX_MBA_BW 100u#define MBA_IS_LINEAR 0x4
@@ -40,6 +40,12 @@#define RMID_VAL_ERROR BIT_ULL(63)#define RMID_VAL_UNAVAIL BIT_ULL(62) +/* +* With the above fields in use 62 bits remain in MSR_IA32_QM_CTR for +* data to be returned. The counter width is discovered from the hardware +* as an offset from MBM_CNTR_WIDTH_BASE. +*/ +#define MBM_CNTR_WIDTH_OFFSET_MAX (62 - MBM_CNTR_WIDTH_BASE)struct rdt_fs_context {
arch/x86/kernel/cpu/resctrl/monitor.c +7 -1
@@ -618,12 +618,18 @@ static void l3_mon_evt_init(struct rdt_resource *r)int rdt_get_mon_l3_config (struct rdt_resource *r) { +unsigned int mbm_offset = boot_cpu_data.x86_cache_mbm_width_offset;unsigned int cl_size = boot_cpu_data.x86_cache_size;int ret; r->mon_scale = boot_cpu_data.x86_cache_occ_scale; r->num_rmid = boot_cpu_data.x86_cache_max_rmid +1 ; - r->mbm_width =MBM_CNTR_WIDTH ; + r->mbm_width =MBM_CNTR_WIDTH_BASE ; + +if (mbm_offset >0 && mbm_offset <=MBM_CNTR_WIDTH_OFFSET_MAX ) + r->mbm_width += mbm_offset; +else if (mbm_offset >MBM_CNTR_WIDTH_OFFSET_MAX ) +pr_warn ("Ignoring impossible MBM counter offset\n" );/* * A reasonable upper limit on the max threshold is the number
Signature · 1 line
2.21.0
Re: [PATCH 2/2] x86/resctrl: Support CPUID enumeration of MBM counter width
Quoted 15 lines · Reinette Chatre
The original Memory Bandwidth Monitoring (MBM) architectural definition defines counters of up to 62 bits in the
IA32_QM_CTRMSR while the first-generation MBM implementation uses statically defined 24 bit counters.arch/x86/kernel/cpu/common.c +2
-0@@ -856,6 +856,8 @@ static void init_speculation_control(struct cpuinfo_x86 *c)static void init_cqm (struct cpuinfo_x86 *c) { + c->x86_cache_mbm_width_offset =-1 ; +if (!cpu_has (c,X86_FEATURE_CQM_LLC )) { c->x86_cache_max_rmid =-1 ; c->x86_cache_occ_scale =-1 ;
arch/x86/kernel/cpu/common.c static void init_cqm(struct cpuinfo_x86 *c) · line 882
Quoted 6 lines
arch/x86/kernel/cpu/common.c +3
-0@@ -875,6 +877,9 @@ static void init_cqm(struct cpuinfo_x86 *c) c->x86_cache_max_rmid = ecx; c->x86_cache_occ_scale = ebx; +/* EAX contents is only defined for Intel CPUs */ +if (c->x86_vendor ==X86_VENDOR_INTEL ) + c->x86_cache_mbm_width_offset = eax &0xff ;
Is it reliable to read eax which is reserved on older platforms that don't support the feature?
Seems the code assumes the reserved eax is 0 on those platforms. Is it reliable?
Quoted 6 lines
arch/x86/kernel/cpu/resctrl/monitor.c +1
-0int rdt_get_mon_l3_config (struct rdt_resource *r) { +unsigned int mbm_offset = boot_cpu_data.x86_cache_mbm_width_offset;unsigned int cl_size = boot_cpu_data.x86_cache_size;int ret;
arch/x86/kernel/cpu/resctrl/monitor.c static void l3_mon_evt_init(struct rdt_resource *r) · line 632
Quoted 10 lines
arch/x86/kernel/cpu/resctrl/monitor.c +6
-1r->mon_scale = boot_cpu_data.x86_cache_occ_scale; r->num_rmid = boot_cpu_data.x86_cache_max_rmid +1 ; - r->mbm_width =MBM_CNTR_WIDTH ; + r->mbm_width =MBM_CNTR_WIDTH_BASE ; + +if (mbm_offset >0 && mbm_offset <=MBM_CNTR_WIDTH_OFFSET_MAX ) + r->mbm_width += mbm_offset; +else if (mbm_offset >MBM_CNTR_WIDTH_OFFSET_MAX ) +pr_warn ("Ignoring impossible MBM counter offset\n" );
Thanks.
-Fenghua
Reinette Chatre
maintainerRe: [PATCH 2/2] x86/resctrl: Support CPUID enumeration of MBM counter width
Hi Fenghua,
Quoted 17 lines · Fenghua Yu
Quoted 16 lines · Reinette Chatre
The original Memory Bandwidth Monitoring (MBM) architectural definition defines counters of up to 62 bits in the
IA32_QM_CTRMSR while the first-generation MBM implementation uses statically defined 24 bit counters.arch/x86/kernel/cpu/common.c +2
-0@@ -856,6 +856,8 @@ static void init_speculation_control(struct cpuinfo_x86 *c)static void init_cqm (struct cpuinfo_x86 *c) { + c->x86_cache_mbm_width_offset =-1 ; +if (!cpu_has (c,X86_FEATURE_CQM_LLC )) { c->x86_cache_max_rmid =-1 ; c->x86_cache_occ_scale =-1 ;@@ -875,6 +877,9 @@ static void init_cqm(struct cpuinfo_x86 *c)
Quoted 8 lines
arch/x86/kernel/cpu/common.c static void init_cqm(struct cpuinfo_x86 *c) · line 882
Quoted 5 lines
arch/x86/kernel/cpu/common.c +3
-0c->x86_cache_max_rmid = ecx; c->x86_cache_occ_scale = ebx; +/* EAX contents is only defined for Intel CPUs */ +if (c->x86_vendor ==X86_VENDOR_INTEL ) + c->x86_cache_mbm_width_offset = eax &0xff ;Is it reliable to read eax which is reserved on older platforms that don't support the feature?
The new ISE specification contains an architectural redefinition of EAX.
Quoted 3 lines
Seems the code assumes the reserved eax is 0 on those platforms. Is it reliable?
Testing on BDW, SKX, and CLX confirmed that EAX is 0. This addition thus results in no functional change on these systems with them continuing to use the original MBM width of 24.
Reinette
Reinette Chatre
maintainerHi Thomas and Borislav,
Quoted 23 lines · Reinette Chatre
Memory Bandwidth Monitoring (MBM) is an Intel Resource Director Technology (RDT) feature that tracks Total and Local bandwidth generated which misses the L3 cache.
The original Memory Bandwidth Monitoring (MBM) architectural definition defines counters of up to 62 bits and the first-generation MBM implementation uses 24 bit counters. Software is required to poll at 1 second or faster to ensure that data is retrieved before a counter rollover occurs more than once under worst conditions.
As system bandwidths scale the software requirement is maintained with the introduction of a per-resource enumerable MBM counter width.
This series adds support for the new enumerable MBM counter width.
Details about the feature can be found in Chapter 9 of the most recent Intel ISE available from
https://software.intel.com/sites/default/files/managed/c5/15/architecture-instruction-set-extensions-programming-reference.pdfReinette Chatre (2):
x86/resctrl: Maintain MBM counter width per resource x86/resctrl: Support CPUID enumeration of MBM counter width
Quoted 8 lines
arch/x86/include/asm/processor.h | 1 + arch/x86/kernel/cpu/common.c | 5 +++++ arch/x86/kernel/cpu/resctrl/ctrlmondata.c | 8 ++++--- arch/x86/kernel/cpu/resctrl/internal.h | 15 ++++++++++--- arch/x86/kernel/cpu/resctrl/monitor.c | 27 ++++++++++++++++------- arch/x86/kernel/cpu/resctrl/rdtgroup.c | 2 +- 6 files changed, 43 insertions(+), 15 deletions(-)
Would it be possible to consider this for inclusion in v5.8?
Thank you
Reinette
ps. Upon your consideration I am planning to follow up with same the question for the next resctrl new feature series available at:https://lore.kernel.org/lkml/[email protected]/
You could view outstanding resctrl changes at branch resctrl/next of
https://github.com/rchatre/linux.git
Borislav Petkov
maintainerRe: [PATCH 2/2] x86/resctrl: Support CPUID enumeration of MBM counter width
In reply to Reinette ChatreQuoted 14 lines · Reinette Chatre
arch/x86/kernel/cpu/common.c +2
-0@@ -856,6 +856,8 @@ static void init_speculation_control(struct cpuinfo_x86 *c)static void init_cqm (struct cpuinfo_x86 *c) { + c->x86_cache_mbm_width_offset =-1 ; +if (!cpu_has (c,X86_FEATURE_CQM_LLC )) { c->x86_cache_max_rmid =-1 ; c->x86_cache_occ_scale =-1 ;
arch/x86/kernel/cpu/common.c static void init_cqm(struct cpuinfo_x86 *c) · line 882
Quoted 6 lines
arch/x86/kernel/cpu/common.c +3
-0@@ -875,6 +877,9 @@ static void init_cqm(struct cpuinfo_x86 *c) c->x86_cache_max_rmid = ecx; c->x86_cache_occ_scale = ebx; +/* EAX contents is only defined for Intel CPUs */ +if (c->x86_vendor ==X86_VENDOR_INTEL ) + c->x86_cache_mbm_width_offset = eax &0xff ;
Remind me again pls why is all this RDT stuff replicated per CPU instead of it being properly detected somewhere down in resctrl_late_init()?
Looking at get_rdt_resources(), it kinda wants to have that in there too?
IOW, how about moving all that gunk in init_cqm() to resctrl/ where it truly belongs? Doesn't have to be this patchset but this patchset can start moving it...
Thx.
Signature · 4 lines
Regards/Gruss,
Boris.
https://people.kernel.org/tglx/notes-about-netiquetteReinette Chatre
maintainerRe: [PATCH 2/2] x86/resctrl: Support CPUID enumeration of MBM counter width
Hi Borislav,
Quoted 25 lines · Borislav Petkov
Quoted 14 lines · Reinette Chatre
arch/x86/kernel/cpu/common.c +2
-0@@ -856,6 +856,8 @@ static void init_speculation_control(struct cpuinfo_x86 *c)static void init_cqm (struct cpuinfo_x86 *c) { + c->x86_cache_mbm_width_offset =-1 ; +if (!cpu_has (c,X86_FEATURE_CQM_LLC )) { c->x86_cache_max_rmid =-1 ; c->x86_cache_occ_scale =-1 ;arch/x86/kernel/cpu/common.c static void init_cqm(struct cpuinfo_x86 *c) · line 882
Quoted 6 lines
arch/x86/kernel/cpu/common.c +3
-0@@ -875,6 +877,9 @@ static void init_cqm(struct cpuinfo_x86 *c) c->x86_cache_max_rmid = ecx; c->x86_cache_occ_scale = ebx; +/* EAX contents is only defined for Intel CPUs */ +if (c->x86_vendor ==X86_VENDOR_INTEL ) + c->x86_cache_mbm_width_offset = eax &0xff ;Remind me again pls why is all this RDT stuff replicated per CPU instead of it being properly detected somewhere down in
resctrl_late_init()?
Quoted 6 lines
Looking at
get_rdt_resources(), it kinda wants to have that in there too?IOW, how about moving all that gunk in
init_cqm()to resctrl/ where it truly belongs? Doesn't have to be this patchset but this patchset can start moving it...
I am not familiar with the history to clarify to you why the RDT feature enumeration is not consistent. Even so, there was an effort to consolidate this in [1] but it was found to go against the goal of centralizing the CPUID information [2] and was not pursued further. I would be happy to revisit this if this is the direction that you prefer. This would essentially be resubmitting [1] though. Do you expect that this change would receive a different reception at this time?
Thank you
Reinette
Borislav Petkov
maintainerRe: [PATCH 2/2] x86/resctrl: Support CPUID enumeration of MBM counter width
Quoted 2 lines · Reinette Chatre
This would essentially be resubmitting [1] though. Do you expect that this change would receive a different reception at this time?
Right, Thomas and I talked it over a bit last night. So the proper thing to do is to read all that information *once* and put it in boot_cpu_data. Because that information is replicated the same over CPUID on each core. If there's per-CPU stuff, then it should remain per-CPU but looking at how the RDT code uses boot_cpu_data, I'd say this is global info.
So, it should be parsed once on the BSP during boot and put into boot_cpu_data. And then silly stuff like x86_init_cache_qos() should go away too.
If this info is needed on Intel only, then it should be parsed in cpu/intel.c, in a ->c_bsp_init helper and if it is needed on AMD too, then a function which does this should be called by the respective c_bsp_init helper.
Then all its users can continue reading it out of boot_cpu_data and future RDT hw info can be added there.
Makes sense?
Thx.
Signature · 4 lines
Regards/Gruss,
Boris.
https://people.kernel.org/tglx/notes-about-netiquetteReinette Chatre
maintainerRe: [PATCH 2/2] x86/resctrl: Support CPUID enumeration of MBM counter width
Hi Borislav,
Quoted 11 lines · Borislav Petkov
Quoted 2 lines · Reinette Chatre
This would essentially be resubmitting [1] though. Do you expect that this change would receive a different reception at this time?
Right, Thomas and I talked it over a bit last night. So the proper thing to do is to read all that information *once* and put it in
boot_cpu_data. Because that information is replicated the same over CPUID on each core. If there's per-CPU stuff, then it should remain per-CPU but looking at how the RDT code usesboot_cpu_data, I'd say this is global info.
Quoted 3 lines
So, it should be parsed once on the BSP during boot and put into
boot_cpu_data. And then silly stuff likex86_init_cache_qos()should go away too.
You are correct. I looked through the hardware errata and confirmed with a few people very knowledgeable about the history and technical details of CMT that x86_init_cache_qos() is not necessary. There exists no platform where the CPUs on a platform that support CMT either does not report a RMID or report different RMID.
Also, for the existing implementation moving init_cqm() to early_identify_cpu() makes sense for all the reasons you mention.
I am struggling with what should follow ...
Quoted 5 lines
If this info is needed on Intel only, then it should be parsed in
cpu/intel.c, in a ->c_bsp_init helper and if it is needed on AMD too, then a function which does this should be called by the respectivec_bsp_inithelper.
Using c_bsp_init may be needed to obtain the Intel-only property that the patch that started this originally attempted to do. AMD and Intel support the same CPUID leaf and two sub-leaves ... only differing in the one new register that is defined for Intel but undefined for AMD.
I am concerned with how I am interpreting your suggestion here since my interpretation does end up with duplicate code between the two c_bsp_init helpers. Below is this snippet - is this how you envisioned this change? (Please note that in this snippet you would find init_cqm() moved from early_identify_cpu(), this change is thus done with the assumption that your earlier suggestions have all been applied already.)
(quoted diff) +0 -0
b/arch/x86/include/asm/processor.h
index 3bcf27caf6c9..76f86fdb02af 100644
arch/x86/include/asm/processor.h +1 -0
@@ -116,6 +116,7 @@ struct cpuinfo_x86 {/* Cache QoS architectural values: */ int x86_cache_max_rmid;/* max index */ int x86_cache_occ_scale;/* scale to bytes */ +int x86_cache_mbm_width_offset;int x86_power;unsigned long loops_per_jiffy;/* cpuid returned max cores value: */
arch/x86/kernel/cpu/amd.c +34 -0
@@ -512,6 +512,38 @@ static void early_init_amd_mc(struct cpuinfo_x86 *c)#endif } +/* +* Significant overlap with the Intel initialization found in +* init_llc_monitoring_intel() since CPUID leaf 0xF subleaf 0x1 +* differ in all but one register (used to initialize +* x86_cache_mbm_width_offset) that is undefined on AMD. +*/ +static void init_llc_monitoring_amd (struct cpuinfo_x86 *c) +{ +if (!cpu_has (c,X86_FEATURE_CQM_LLC )) { + c->x86_cache_max_rmid =-1 ; + c->x86_cache_occ_scale =-1 ; + c->x86_cache_mbm_width_offset =-1 ; +return ; + } + +/* will be overridden if occupancy monitoring exists */ + c->x86_cache_max_rmid =cpuid_ebx (0xf ); + +if (cpu_has (c,X86_FEATURE_CQM_OCCUP_LLC ) || +cpu_has (c,X86_FEATURE_CQM_MBM_TOTAL ) || +cpu_has (c,X86_FEATURE_CQM_MBM_LOCAL )) { +u32 eax, ebx, ecx, edx; + +/* QoS sub-leaf, EAX=0Fh, ECX=1 */ +cpuid_count (0xf ,1 , &eax, &ebx, &ecx, &edx); + + c->x86_cache_max_rmid = ecx; + c->x86_cache_occ_scale = ebx; + c->x86_cache_mbm_width_offset =-1 ; + } +} +static void bsp_init_amd (struct cpuinfo_x86 *c) {
@@ -597,6 +629,8 @@ static void bsp_init_amd(struct cpuinfo_x86 *c) x86_amd_ls_cfg_ssbd_mask =1ULL << bit; } } + +init_llc_monitoring_amd (c); }static void early_detect_mem_encrypt (struct cpuinfo_x86 *c)
arch/x86/kernel/cpu/common.c +0 -0
@@ -854,30 +854,6 @@ static void init_speculation_control(struct
cpuinfo_x86 *c)
}
}
-static void init_cqm(struct cpuinfo_x86 *c)
-{
- if (!cpu_has(c, X86_FEATURE_CQM_LLC)) {
- c->x86_cache_max_rmid = -1;
- c->x86_cache_occ_scale = -1;
- return;
- }
-
- /* will be overridden if occupancy monitoring exists */
- c->x86_cache_max_rmid = cpuid_ebx(0xf);
-
- if (cpu_has(c, X86_FEATURE_CQM_OCCUP_LLC) || - cpu_has(c, X86_FEATURE_CQM_MBM_TOTAL) ||
- cpu_has(c, X86_FEATURE_CQM_MBM_LOCAL)) {
- u32 eax, ebx, ecx, edx;
-
- /* QoS sub-leaf, EAX=0Fh, ECX=1 */
- cpuid_count(0xf, 1, &eax, &ebx, &ecx, &edx);
-
- c->x86_cache_max_rmid = ecx;
- c->x86_cache_occ_scale = ebx;
- }
-}
-
void get_cpu_cap(struct cpuinfo_x86 *c)
{
u32 eax, ebx, ecx, edx;
@@ -1212,7 +1188,6 @@ static void __init early_identify_cpu(struct cpuinfo_x86 *c)
c->cpu_index = 0;
filter_cpuid_features(c, false);
- init_cqm(c);
if (this_cpu->c_bsp_init)
this_cpu->c_bsp_init(c);
arch/x86/kernel/cpu/intel.c +41 -0
53 lines not on this page · raw message
Quoted 3 lines
Then all its users can continue reading it out of
boot_cpu_dataand future RDT hw info can be added there.
Understood.
Thank you very much
Reinette
Borislav Petkov
maintainerRe: [PATCH 2/2] x86/resctrl: Support CPUID enumeration of MBM counter width
Hi,
Quoted 1 line · Reinette Chatre
I am struggling with what should follow ...
Since a diff is better than a thousand words :-) see below.
It builds here with my .config but I'm sure it still needs work but you get the idea. When you do the final version, please split it properly as in:
- one patch does this:
rename from arch/x86/include/asm/resctrl_sched.h
rename to arch/x86/include/asm/resctrl.h
- the other adds resctrl_cpu_detect()
- last one removes init_cqm() and x86_init_cache_qos(), etc.
Thx.
---
arch/x86/include/asm/processor.h +2 -1
@@ -113,9 +113,10 @@ struct cpuinfo_x86 {/* in KB - valid for CPUS which support this call: */ unsigned int x86_cache_size;int x86_cache_alignment;/* In bytes */ -/* Cache QoS architectural values: */ +/* Cache QoS architectural values, valid only on the BSP: */ int x86_cache_max_rmid;/* max index */ int x86_cache_occ_scale;/* scale to bytes */ +int x86_cache_mbm_width_offset;int x86_power;unsigned long loops_per_jiffy;/* cpuid returned max cores value: */
arch/x86/include/asm/resctrl.h renamed from arch/x86/include/asm/resctrl_sched.h +3 -0
@@ -84,9 +84,12 @@ static inline void resctrl_sched_in(void) __resctrl_sched_in(); } +void resctrl_cpu_detect (struct cpuinfo_x86 *c); +#else static inline void resctrl_sched_in (void ) {} +static inline void resctrl_cpu_detect (struct cpuinfo_x86 *c) {}#endif /* CONFIG_X86_CPU_RESCTRL */
arch/x86/kernel/cpu/amd.c +3 -0
@@ -18,6 +18,7 @@#include <asm/pci-direct.h> #include <asm/delay.h> #include <asm/debugreg.h> +#include <asm/resctrl.h> #ifdef CONFIG_X86_64 # include <asm/mmconfig.h>
@@ -597,6 +598,8 @@ static void bsp_init_amd(struct cpuinfo_x86 *c) x86_amd_ls_cfg_ssbd_mask =1ULL << bit; } } + +resctrl_cpu_detect (c); }static void early_detect_mem_encrypt (struct cpuinfo_x86 *c)
arch/x86/kernel/cpu/common.c +0 -40
@@ -877,30 +877,6 @@ static void init_speculation_control(struct cpuinfo_x86 *c) } } -static void init_cqm (struct cpuinfo_x86 *c) -{ -if (!cpu_has (c,X86_FEATURE_CQM_LLC )) { - c->x86_cache_max_rmid =-1 ; - c->x86_cache_occ_scale =-1 ; -return ; - } - -/* will be overridden if occupancy monitoring exists */ - c->x86_cache_max_rmid =cpuid_ebx (0xf ); - -if (cpu_has (c,X86_FEATURE_CQM_OCCUP_LLC ) || -cpu_has (c,X86_FEATURE_CQM_MBM_TOTAL ) || -cpu_has (c,X86_FEATURE_CQM_MBM_LOCAL )) { -u32 eax, ebx, ecx, edx; - -/* QoS sub-leaf, EAX=0Fh, ECX=1 */ -cpuid_count (0xf ,1 , &eax, &ebx, &ecx, &edx); - - c->x86_cache_max_rmid = ecx; - c->x86_cache_occ_scale = ebx; - } -} -void get_cpu_cap (struct cpuinfo_x86 *c) {u32 eax, ebx, ecx, edx;
@@ -968,7 +944,6 @@ void get_cpu_cap(struct cpuinfo_x86 *c)init_scattered_cpuid_features (c);init_speculation_control (c); -init_cqm (c);/* * Clear/Set all flags overridden by options, after probe.
@@ -1400,20 +1375,6 @@ static void generic_identify(struct cpuinfo_x86 *c)#endif } -static void x86_init_cache_qos (struct cpuinfo_x86 *c) -{ -/* -* The heavy lifting of max_rmid and cache_occ_scale are handled -* in get_cpu_cap(). Here we just set the max_rmid for the boot_cpu -* in case CQM bits really aren't there in this CPU. -*/ -if (c != &boot_cpu_data) { - boot_cpu_data.x86_cache_max_rmid = -min (boot_cpu_data.x86_cache_max_rmid, - c->x86_cache_max_rmid); - } -} -/* * Validate that ACPI/mptables have the same information about the * effective APIC id and update the package map.
@@ -1526,7 +1487,6 @@ static void identify_cpu(struct cpuinfo_x86 *c)#endif x86_init_rdrand (c); -x86_init_cache_qos (c);setup_pku (c);/*
arch/x86/kernel/cpu/intel.c +7 -0
@@ -22,6 +22,7 @@#include <asm/cpu_device_id.h> #include <asm/cmdline.h> #include <asm/traps.h> +#include <asm/resctrl.h> #ifdef CONFIG_X86_64 #include <linux/topology.h>
@@ -322,6 +323,11 @@ static void early_init_intel(struct cpuinfo_x86 *c)detect_ht_early (c); } +static void bsp_init_intel (struct cpuinfo_x86 *c) +{ +resctrl_cpu_detect (c); +} +#ifdef CONFIG_X86_32 /* * Early probe support logic for ppro memory erratum #50
@@ -961,6 +967,7 @@ static const struct cpu_dev intel_cpu_dev = {#endif .c_detect_tlb = intel_detect_tlb, .c_early_init = early_init_intel, + .c_bsp_init = bsp_init_intel, .c_init = init_intel, .c_x86_vendor = X86_VENDOR_INTEL, };
arch/x86/kernel/cpu/resctrl/core.c +30 -1
@@ -22,7 +22,7 @@#include <linux/cpuhotplug.h> #include <asm/intel-family.h> -#include <asm/resctrl_sched.h> +#include <asm/resctrl.h> #include "internal.h" /* Mutex to protect rdtgroup access. */
@@ -958,6 +958,35 @@ static __init void rdt_init_res_defs(void)static enum cpuhp_state rdt_online; +/* Runs once on the BSP during boot. */ +void resctrl_cpu_detect (struct cpuinfo_x86 *c) +{ +if (!cpu_has (c,X86_FEATURE_CQM_LLC )) { + c->x86_cache_max_rmid =-1 ; + c->x86_cache_occ_scale =-1 ; + c->x86_cache_mbm_width_offset =-1 ; +return ; + } + +/* will be overridden if occupancy monitoring exists */ + c->x86_cache_max_rmid =cpuid_ebx (0xf ); + +if (cpu_has (c,X86_FEATURE_CQM_OCCUP_LLC ) || +cpu_has (c,X86_FEATURE_CQM_MBM_TOTAL ) || +cpu_has (c,X86_FEATURE_CQM_MBM_LOCAL )) { +u32 eax, ebx, ecx, edx; + +/* QoS sub-leaf, EAX=0Fh, ECX=1 */ +cpuid_count (0xf ,1 , &eax, &ebx, &ecx, &edx); + + c->x86_cache_max_rmid = ecx; + c->x86_cache_occ_scale = ebx; + +if (c->x86_vendor ==X86_VENDOR_INTEL ) + c->x86_cache_mbm_width_offset = eax &0xff ; + } +} +static int __init resctrl_late_init (void ) {struct rdt_resource *r;
arch/x86/kernel/cpu/resctrl/pseudo_lock.c +1 -1
@@ -24,7 +24,7 @@#include <asm/cacheflush.h> #include <asm/intel-family.h> -#include <asm/resctrl_sched.h> +#include <asm/resctrl.h> #include <asm/perf_event.h> #include "../../events/perf_event.h" /* For X86_CONFIG() */
arch/x86/kernel/cpu/resctrl/rdtgroup.c +1 -1
@@ -29,7 +29,7 @@#include <uapi/linux/magic.h> -#include <asm/resctrl_sched.h> +#include <asm/resctrl.h> #include "internal.h" DEFINE_STATIC_KEY_FALSE (rdt_enable_key);
arch/x86/kernel/process_64.c +1 -1
@@ -52,7 +52,7 @@#include <asm/switch_to.h> #include <asm/xen/hypervisor.h> #include <asm/vdso.h> -#include <asm/resctrl_sched.h> +#include <asm/resctrl.h> #include <asm/unistd.h> #include <asm/fsgsbase.h> #ifdef CONFIG_IA32_EMULATION
Signature · 4 lines
Regards/Gruss,
Boris.
https://people.kernel.org/tglx/notes-about-netiquetteReinette Chatre
maintainerRe: [PATCH 2/2] x86/resctrl: Support CPUID enumeration of MBM counter width
Hi Borislav,
Quoted 7 lines · Borislav Petkov
Hi,
Quoted 1 line · Reinette Chatre
I am struggling with what should follow ...
Since a diff is better than a thousand words :-) see below.
Thank you so much for providing the details. Your explanation is clear to me but I do have one clarification question ...
Quoted 8 lines
arch/x86/kernel/cpu/amd.c +2
-0@@ -597,6 +598,8 @@ static void bsp_init_amd(struct cpuinfo_x86 *c) x86_amd_ls_cfg_ssbd_mask =1ULL << bit; } } + +resctrl_cpu_detect (c); }
...
Quoted 56 lines
arch/x86/kernel/cpu/intel.c +6
-0@@ -322,6 +323,11 @@ static void early_init_intel(struct cpuinfo_x86 *c)detect_ht_early (c); } +static void bsp_init_intel (struct cpuinfo_x86 *c) +{ +resctrl_cpu_detect (c); +} +#ifdef CONFIG_X86_32 /* * Early probe support logic for ppro memory erratum #50 @@ -961,6 +967,7 @@ static const struct cpu_dev intel_cpu_dev = {#endif .c_detect_tlb = intel_detect_tlb, .c_early_init = early_init_intel, + .c_bsp_init = bsp_init_intel, .c_init = init_intel, .c_x86_vendor = X86_VENDOR_INTEL, };arch/x86/kernel/cpu/resctrl/core.c +20
-1@@ -22,7 +22,7 @@#include <linux/cpuhotplug.h> #include <asm/intel-family.h> -#include <asm/resctrl_sched.h> +#include <asm/resctrl.h> #include "internal.h" /* Mutex to protect rdtgroup access. */ @@ -958,6 +958,35 @@ static __init void rdt_init_res_defs(void)static enum cpuhp_state rdt_online; +/* Runs once on the BSP during boot. */ +void resctrl_cpu_detect (struct cpuinfo_x86 *c) +{ +if (!cpu_has (c,X86_FEATURE_CQM_LLC )) { + c->x86_cache_max_rmid =-1 ; + c->x86_cache_occ_scale =-1 ; + c->x86_cache_mbm_width_offset =-1 ; +return ; + } + +/* will be overridden if occupancy monitoring exists */ + c->x86_cache_max_rmid =cpuid_ebx (0xf ); + +if (cpu_has (c,X86_FEATURE_CQM_OCCUP_LLC ) || +cpu_has (c,X86_FEATURE_CQM_MBM_TOTAL ) || +cpu_has (c,X86_FEATURE_CQM_MBM_LOCAL )) { +u32 eax, ebx, ecx, edx; + +/* QoS sub-leaf, EAX=0Fh, ECX=1 */
arch/x86/kernel/cpu/resctrl/core.c static __init void rdt_init_res_defs(void) · line 989
Quoted 10 lines
arch/x86/kernel/cpu/resctrl/core.c +10
-0+cpuid_count (0xf ,1 , &eax, &ebx, &ecx, &edx); + + c->x86_cache_max_rmid = ecx; + c->x86_cache_occ_scale = ebx; + +if (c->x86_vendor ==X86_VENDOR_INTEL ) + c->x86_cache_mbm_width_offset = eax &0xff ; + } +} +
resctrl_cpu_detect() is now identical among vendors. Do we still need the c_bsp_init helpers? Could we not perhaps call resctrl_cpu_detect() directly from early_identify_cpu()?
Thank you
Reinette
Reinette Chatre
maintainerRe: [PATCH 2/2] x86/resctrl: Support CPUID enumeration of MBM counter width
Hi Borislav,
Quoted 86 lines · Reinette Chatre
Quoted 7 lines · Borislav Petkov
Hi,
Quoted 1 line · Reinette Chatre
I am struggling with what should follow ...
Since a diff is better than a thousand words :-) see below.
Thank you so much for providing the details. Your explanation is clear to me but I do have one clarification question ...
Quoted 8 lines
arch/x86/kernel/cpu/amd.c +2
-0@@ -597,6 +598,8 @@ static void bsp_init_amd(struct cpuinfo_x86 *c) x86_amd_ls_cfg_ssbd_mask =1ULL << bit; } } + +resctrl_cpu_detect (c); }...
Quoted 62 lines
arch/x86/kernel/cpu/intel.c +6
-0@@ -322,6 +323,11 @@ static void early_init_intel(struct cpuinfo_x86 *c)detect_ht_early (c); } +static void bsp_init_intel (struct cpuinfo_x86 *c) +{ +resctrl_cpu_detect (c); +} +#ifdef CONFIG_X86_32 /* * Early probe support logic for ppro memory erratum #50 @@ -961,6 +967,7 @@ static const struct cpu_dev intel_cpu_dev = {#endif .c_detect_tlb = intel_detect_tlb, .c_early_init = early_init_intel, + .c_bsp_init = bsp_init_intel, .c_init = init_intel, .c_x86_vendor = X86_VENDOR_INTEL, };arch/x86/kernel/cpu/resctrl/core.c +26
-1@@ -22,7 +22,7 @@#include <linux/cpuhotplug.h> #include <asm/intel-family.h> -#include <asm/resctrl_sched.h> +#include <asm/resctrl.h> #include "internal.h" /* Mutex to protect rdtgroup access. */ @@ -958,6 +958,35 @@ static __init void rdt_init_res_defs(void)static enum cpuhp_state rdt_online; +/* Runs once on the BSP during boot. */ +void resctrl_cpu_detect (struct cpuinfo_x86 *c) +{ +if (!cpu_has (c,X86_FEATURE_CQM_LLC )) { + c->x86_cache_max_rmid =-1 ; + c->x86_cache_occ_scale =-1 ; + c->x86_cache_mbm_width_offset =-1 ; +return ; + } + +/* will be overridden if occupancy monitoring exists */ + c->x86_cache_max_rmid =cpuid_ebx (0xf ); + +if (cpu_has (c,X86_FEATURE_CQM_OCCUP_LLC ) || +cpu_has (c,X86_FEATURE_CQM_MBM_TOTAL ) || +cpu_has (c,X86_FEATURE_CQM_MBM_LOCAL )) { +u32 eax, ebx, ecx, edx; + +/* QoS sub-leaf, EAX=0Fh, ECX=1 */ +cpuid_count (0xf ,1 , &eax, &ebx, &ecx, &edx); + + c->x86_cache_max_rmid = ecx; + c->x86_cache_occ_scale = ebx; + +if (c->x86_vendor ==X86_VENDOR_INTEL )
Quoted 9 lines
arch/x86/kernel/cpu/resctrl/core.c static __init void rdt_init_res_defs(void) · line 989
Quoted 4 lines
arch/x86/kernel/cpu/resctrl/core.c +4
-0+ c->x86_cache_mbm_width_offset = eax &0xff ; + } +} +
resctrl_cpu_detect()is now identical among vendors. Do we still need thec_bsp_inithelpers? Could we not perhaps callresctrl_cpu_detect()directly fromearly_identify_cpu()?
I should have given this more thought ... even though the function will be identical between the two vendors it does contain vendor-specific code, thus needing to be called from the c_bsp_init helpers? I will resubmit the new series that intends to follow all your suggestions shortly.
Thank you very much for your feedback
Reinette