x86,sched: allow topologies where NUMA nodes share an LLC
PATCH v4 1/1 x86,sched: allow topologies where NUMA nodes share an LLC
From: Alison Schofield <[email protected]>
Intel's Skylake Server CPUs have a different LLC topology than previous generations. When in Sub-NUMA-Clustering (SNC) mode, the package is divided into two "slices", each containing half the cores, half the LLC, and one memory controller and each slice is enumerated to Linux as a NUMA node. This is similar to how the cores and LLC were arranged for the Cluster-On-Die (CoD) feature.
CoD allowed the same cache line to be present in each half of the LLC. But, with SNC, each line is only ever present in *one* slice. This means that the portion of the LLC *available* to a CPU depends on the data being accessed:
Remote socket: entire package LLC is shared
Local socket->local slice: data goes into local slice LLC Local socket->remote slice: data goes into remote-slice LLC. Slightly
higher latency than local slice LLC.
The biggest implication from this is that a process accessing all NUMA-local memory only sees half the LLC capacity.
The CPU describes its cache hierarchy with the CPUID instruction. One of the CPUID leaves enumerates the "logical processors sharing this cache". This information is used for scheduling decisions so that tasks move more freely between CPUs sharing the cache.
But, the CPUID for the SNC configuration discussed above enumerates the LLC as being shared by the entire package. This is not 100% precise because the entire cache is not usable by all accesses. But, it *is* the way the hardware enumerates itself, and this is not likely to change.
The userspace visible impact of all the above is that the sysfs info reports the entire LLC as being available to the entire package. As noted above, this is not true for local socket accesses. This patch does not correct the sysfs info. It is the same, pre and post patch.
This patch continues to allow this SNC topology and it does so without complaint. It eliminates a warning that looks like this:
sched: CPU #3's llc-sibling CPU #0 is not on the same node! [node: 1 != 0]. Ignoring dependency.
The warning is coming from the sane_topology check() in smpboot.c. To fix this, add a vendor and model specific check to never call topology_sane() for these systems. Also, just like "Cluster-on-Die" we throw out the "coregroup" sched_domain_topology_level and use NUMA information from the SRAT alone.
This is OK at least on the hardware we are immediately concerned about because the LLC sharing happens at both the slice and at the package level, which are also NUMA boundaries.
---
Changes in v4:
* Added this to the patch description above:
The userspace visible impact of all the above is that the sysfs info reports the entire LLC as being available to the entire package. As noted above, this is not true for local socket accesses. This patch does not correct the sysfs info. It is the same, pre and post patch.
This patch continues to allow this SNC topology and it does so without complaint. It eliminates a warning that looks like this:
* Changed the code comment per PeterZ/DaveH discussion wrt bypassing that topology_sane() check in match_llc()
/*
* false means 'c' does not share the LLC of 'o'.
* Note: this decision gets reflected all the way
* out to userspace
*/
This message hopes to clarify what happens when we return false. Note that returning false is *not* new to this patch. Without this patch we always returned false - with a warning. This avoids that warning and returns false directly.
* Remove __initconst from snc_cpu[] declaration that I had added in v3. This is not an init time only path.
* Do not deal with the wrong sysfs info. It was wrong before this patch and it will be the exact same 'wrongness' after this patch.
We can address the sysfs reporting separately. Here are some options:
1) Change the way the LLC-size is reported. Enumerate two separate, half-sized LLCs shared only by the slice when SNC mode is on.
2) Do not export the sysfs info that is wrong. Prevents userspace from making bad decisions based on inaccurate info.
Changes in v3:
* Use x86_match_cpu() for vendor & model check and moved related comments to the array define. (Still just one system model)
* Updated the comments surrounding the topology_sane() check.
Changes in v2:
* Add vendor check (Intel) where we only had a model check (Skylake_X). Considered the suggestion of adding a new flag here but thought that to be overkill for this usage.
* Return false, instead of true, from match_llc() per reviewer suggestion. That also cleaned up a topology broken bug message in sched_domain().
* Updated the comments around the match_llc() return value, continuing to note that the return value doesn't actually matter because we are throwing away coregroups for scheduling.
* Changed submitter. I'm following up on this patch originally submitted by Dave Hansen.
arch/x86/kernel/smpboot.c | 67 ++++++++++++++++++++++++++++++++++++++++------- 1 file changed, 57 insertions(+), 10 deletions(-)
arch/x86/kernel/smpboot.c +57 -10
@@ -77,6 +77,8 @@#include <asm/i8259.h> #include <asm/misc.h> #include <asm/qspinlock.h> +#include <asm/intel-family.h> +#include <asm/cpu_device_id.h> /* Number of siblings per CPU package */ int smp_num_siblings =1 ;
@@ -390,15 +392,66 @@ static bool match_smt(struct cpuinfo_x86 *c, struct cpuinfo_x86 *o)return false; } +/* +* Set if a package/die has multiple NUMA nodes inside. +* AMD Magny-Cours, Intel Cluster-on-Die, and Intel +* Sub-NUMA Clustering have this. +*/ + +static bool x86_has_numa_in_package; + +/* +* Define SNC (Sub-NUMA Cluster) CPUs. +* +* These are Intel CPUs that enumerate an LLC that is shared by +* multiple NUMA nodes. The LLC on these systems is shared for +* off-package data access but private to the NUMA node (half +* of the package) for on-package access. +* +* CPUID (the source of the information about the LLC) can only +* enumerate the cache as being shared *or* unshared, but not +* this particular configuration. The CPU in this case enumerates +* the cache to be shared across the entire package (spanning both +* NUMA nodes). +*/ + +static const struct x86_cpu_id snc_cpu[] = { + {X86_VENDOR_INTEL ,6 ,INTEL_FAM6_SKYLAKE_X }, + {} +}; +static bool match_llc (struct cpuinfo_x86 *c,struct cpuinfo_x86 *o) {int cpu1 = c->cpu_index, cpu2 = o->cpu_index; -if (per_cpu (cpu_llc_id, cpu1) !=BAD_APICID && -per_cpu (cpu_llc_id, cpu1) ==per_cpu (cpu_llc_id, cpu2)) -return topology_sane (c, o,"llc" ); +/* Do not match if we do not have a valid APICID for cpu: */ +if (per_cpu (cpu_llc_id, cpu1) ==BAD_APICID ) +return false; -return false; +/* Do not match if LLC id does not match: */ +if (per_cpu (cpu_llc_id, cpu1) !=per_cpu (cpu_llc_id, cpu2)) +return false; + +/* +* topology_sane() considers LLCs that span NUMA nodes to be +* insane and will display a warning message. Bypass the call +* to topology_sane() for snc_cpu's to avoid that warning. +*/ + +if (!topology_same_node (c, o) &&x86_match_cpu (snc_cpu)) { +/* Indicate that package has NUMA nodes inside: */ + x86_has_numa_in_package = true; + +/* +* false means 'c' does not share the LLC of 'o'. +* Note: this decision gets reflected all the way +* out to userspace. +*/ + +return false; + } + +return topology_sane (c, o,"llc" ); }/*
@@ -454,12 +507,6 @@ static struct sched_domain_topology_level x86_topology[] = { { NULL, }, }; -/* -* Set if a package/die has multiple NUMA nodes inside. -* AMD Magny-Cours and Intel Cluster-on-Die have this. -*/ -static bool x86_has_numa_in_package; -void set_cpu_sibling_map (int cpu) {bool has_smt = smp_num_siblings >1 ;
Signature · 1 line
2.7.4
arch/x86/kernel/smpboot.c static bool match_smt(struct cpuinfo_x86 *c, struct cpuinfo_x86 *o) · line 443
Quoted 10 lines · Alison Schofield
arch/x86/kernel/smpboot.c +10
-0+ +/* +* topology_sane() considers LLCs that span NUMA nodes to be +* insane and will display a warning message. Bypass the call +* to topology_sane() for snc_cpu's to avoid that warning. +*/ + +if (!topology_same_node (c, o) &&x86_match_cpu (snc_cpu)) { +/* Indicate that package has NUMA nodes inside: */ + x86_has_numa_in_package = true;
Why does the x86_has_numa_in_package has to be set here when it would have been done later in set_cpu_sibling_map?
arch/x86/kernel/smpboot.c static bool match_smt(struct cpuinfo_x86 *c, struct cpuinfo_x86 *o) · line 451
Quoted 8 lines
arch/x86/kernel/smpboot.c +8
-0+ +/* +* false means 'c' does not share the LLC of 'o'. +* Note: this decision gets reflected all the way +* out to userspace. +*/ + +return false;
Thanks.
Tim
Quoted 15 lines · Tim Chen
arch/x86/kernel/smpboot.c static bool match_smt(struct cpuinfo_x86 *c, struct cpuinfo_x86 *o) · line 443
Quoted 10 lines · Alison Schofield
arch/x86/kernel/smpboot.c +10
-0+ +/* +* topology_sane() considers LLCs that span NUMA nodes to be +* insane and will display a warning message. Bypass the call +* to topology_sane() for snc_cpu's to avoid that warning. +*/ + +if (!topology_same_node (c, o) &&x86_match_cpu (snc_cpu)) { +/* Indicate that package has NUMA nodes inside: */ + x86_has_numa_in_package = true;Why does the
x86_has_numa_in_packagehas to be set here when it would have been done later inset_cpu_sibling_map?
Tim,
I had that same thought when you commented on it previously. After discussing w DaveH, decided that match_llc() and match_die(c,0) could be different and chose to be (cautiously) redundant. alisons
Quoted 13 lines
arch/x86/kernel/smpboot.c static bool match_smt(struct cpuinfo_x86 *c, struct cpuinfo_x86 *o) · line 451
Quoted 9 lines
arch/x86/kernel/smpboot.c +8
-0+ +/* +* false means 'c' does not share the LLC of 'o'. +* Note: this decision gets reflected all the way +* out to userspace. +*/ + +return false;Thanks.
Tim
Quoted 17 lines · Alison Schofield
Quoted 15 lines · Tim Chen
arch/x86/kernel/smpboot.c static bool match_smt(struct cpuinfo_x86 *c, struct cpuinfo_x86 *o) · line 443
Quoted 10 lines · Alison Schofield
arch/x86/kernel/smpboot.c +10
-0+ +/* +* topology_sane() considers LLCs that span NUMA nodes to be +* insane and will display a warning message. Bypass the call +* to topology_sane() for snc_cpu's to avoid that warning. +*/ + +if (!topology_same_node (c, o) &&x86_match_cpu (snc_cpu)) { +/* Indicate that package has NUMA nodes inside: */ + x86_has_numa_in_package = true;Why does the
x86_has_numa_in_packagehas to be set here when it would have been done later inset_cpu_sibling_map?
Quoted 5 lines
Tim,
I had that same thought when you commented on it previously. After discussing w DaveH, decided thatmatch_llc()andmatch_die(c,0)could be different and chose to be (cautiously) redundant. alisons
If it is redundant, I suggest it be removed, and only added if there is truly a case where the current logic
if (match_die(c, o) && !topology_same_node(c, o))
x86_has_numa_in_package = true;fails. And also the modification of this logic should be at the same place for easy code maintenance.
Tim
Quoted 16 lines
Quoted 16 lines
arch/x86/kernel/smpboot.c static bool match_smt(struct cpuinfo_x86 *c, struct cpuinfo_x86 *o) · line 451
Quoted 12 lines
arch/x86/kernel/smpboot.c +8
-0+ +/* +* false means 'c' does not share the LLC of 'o'. +* Note: this decision gets reflected all the way +* out to userspace. +*/ + +return false;Thanks.
Tim
Quoted 24 lines · Tim Chen
Quoted 17 lines · Alison Schofield
Quoted 15 lines · Tim Chen
arch/x86/kernel/smpboot.c static bool match_smt(struct cpuinfo_x86 *c, struct cpuinfo_x86 *o) · line 443
Quoted 10 lines · Alison Schofield
arch/x86/kernel/smpboot.c +10
-0+ +/* +* topology_sane() considers LLCs that span NUMA nodes to be +* insane and will display a warning message. Bypass the call +* to topology_sane() for snc_cpu's to avoid that warning. +*/ + +if (!topology_same_node (c, o) &&x86_match_cpu (snc_cpu)) { +/* Indicate that package has NUMA nodes inside: */ + x86_has_numa_in_package = true;Why does the
x86_has_numa_in_packagehas to be set here when it would have been done later inset_cpu_sibling_map?Quoted 5 lines
Tim,
I had that same thought when you commented on it previously. After discussing w DaveH, decided thatmatch_llc()andmatch_die(c,0)could be different and chose to be (cautiously) redundant. alisons
Quoted 8 lines
If it is redundant, I suggest it be removed, and only added if there is truly a case where the current logic
if (match_die(c, o) && !topology_same_node(c, o)) x86_has_numa_in_package = true;fails. And also the modification of this logic should be at the same place for easy code maintenance.
That makes good sense. I'll look to define the difference or remove the redundancy.
alisons
Quoted 20 lines
Tim
Quoted 16 lines
Quoted 16 lines
arch/x86/kernel/smpboot.c static bool match_smt(struct cpuinfo_x86 *c, struct cpuinfo_x86 *o) · line 451
Quoted 12 lines
arch/x86/kernel/smpboot.c +8
-0+ +/* +* false means 'c' does not share the LLC of 'o'. +* Note: this decision gets reflected all the way +* out to userspace. +*/ + +return false;Thanks.
Tim
Quoted 31 lines · Alison Schofield
Quoted 30 lines · Tim Chen
Quoted 17 lines · Alison Schofield
Quoted 15 lines · Tim Chen
arch/x86/kernel/smpboot.c static bool match_smt(struct cpuinfo_x86 *c, struct cpuinfo_x86 *o) · line 443
Quoted 10 lines · Alison Schofield
arch/x86/kernel/smpboot.c +10
-0+ +/* +* topology_sane() considers LLCs that span NUMA nodes to be +* insane and will display a warning message. Bypass the call +* to topology_sane() for snc_cpu's to avoid that warning. +*/ + +if (!topology_same_node (c, o) &&x86_match_cpu (snc_cpu)) { +/* Indicate that package has NUMA nodes inside: */ + x86_has_numa_in_package = true;Why does the
x86_has_numa_in_packagehas to be set here when it would have been done later inset_cpu_sibling_map?Quoted 5 lines
Tim,
I had that same thought when you commented on it previously. After discussing w DaveH, decided thatmatch_llc()andmatch_die(c,0)could be different and chose to be (cautiously) redundant. alisonsIf it is redundant, I suggest it be removed, and only added if there is truly a case where the current logic
if (match_die(c, o) && !topology_same_node(c, o)) x86_has_numa_in_package = true;
Quoted 7 lines
Quoted 2 lines
fails. And also the modification of this logic should be at the same place for easy code maintenance.
That makes good sense. I'll look to define the difference or remove the redundancy.
alisons
I found not reason for the redundancy via experimentation w my Skylake, nor through code examination. I've removed it in v5. I'll see if anyone claims theoretical case.
alisons
Quoted 21 lines
Quoted 21 lines
Tim
Quoted 16 lines
Quoted 16 lines
arch/x86/kernel/smpboot.c static bool match_smt(struct cpuinfo_x86 *c, struct cpuinfo_x86 *o) · line 451
Quoted 12 lines
arch/x86/kernel/smpboot.c +8
-0+ +/* +* false means 'c' does not share the LLC of 'o'. +* Note: this decision gets reflected all the way +* out to userspace. +*/ + +return false;Thanks.
Tim