Implement vcpu soft affinity for credit1
v2 cover Implement vcpu soft affinity for credit1
Hi,
This is take 2 of the per-vcpu NUMA affinity seies... Which is not about per-vcpu NUMA affinity any more! :-)
As agreed with Jan and George, I re-architected the thing and made it more general. So, now it's about allowing each vcpu to have:
- an hard affinity, which they already do, and we usually call pinning. This is the list of pcpus where a vcpu is allowed to run;
- a soft affinity, which this series introduces. This is the list of pcpus where a vcpu *prefers* to run.
Once that is done, per-vcpu NUMA-aware scheduling is easily implemented on top of that, just by instructing libxl to issue the proper call to setup the soft affinity of the domain's vcpus to be equal to its node-affinity. (see, for instance, patch 16).
I'm sorry it took way longer than I expected it to take, but it turned out that just renaming variables and moving hunks around the series is an activity very very prone to stupid mistakes! :-P Also, I had to update, and then double check to have done it correcly, all the changes to the in-tree docs and manpages, to make sure they actually reflect the new architecture.
As it was before, the first part of the series (patches 1-4) carries the updated syntax for specifying cpu (both hard and soft) affinity, which has been discussed and (mostly) acked already, either in v1 or in a previous submission (under a different series name).
The whole series should be pretty straightforward, as it's mostly renaming variables and moving or rewiring existing stuff. Perhaps (and I'm staring at you tool's maintainers :-D) have a thorougher look at patche 11 and 12 (especially 12), where the interface for this new feature is crafted, at the libxc (patch 11) and libxl (patch 12) level.
Release wise, I think this is a nice feature to have, especially now that we've made it more general, and it can find users and use cases outside of the NUMa domain, where it was confined before. Also, I don't think it's that much of a complex series (as I said, it's mosly renaming/reshuffling), so perhaps it can go in. At the same time, the natural consumer for this feature is vNUMA. Thus, since vNUMA is probably going to be a 4.5 thing, having this series slipping wouldn't be too bad.
The series is also available here:
git://xenbits.xen.org/people/dariof/xen.git numa/per-vcpu-affinity-v2
Thanks and Regards,
Dario
---
Dario Faggioli (16):
xl: match output of vcpu-list with pinning syntax
xl: allow for node-wise specification of vcpu pinning
* xl: implement and enable dryrun mode for `xl vcpu-pin'
* xl: test script for the cpumap parser (for vCPU pinning)
xen: fix leaking of v->cpu_affinity_saved
xen: sched: make space for cpu_soft_affinity
xen: sched: rename v->cpu_affinity into v->cpu_hard_affinity
xen: derive NUMA node affinity from hard and soft CPU affinity
xen: sched: DOMCTL_*vcpuaffinity works with hard and soft affinity
xen: sched: use soft-affinity instead of domain's node-affinity
libxc: get and set soft and hard affinity
libxl: get and set soft affinity
xl: show soft affinity in `xl vcpu-list'
xl: enable setting soft affinity
xl: enable for specifying node-affinity in the config file
libxl: automatic NUMA placement affects soft affinity
docs/man/xl.cfg.pod.5 | 66 +++ docs/man/xl.pod.1 | 34 ++ docs/misc/xl-numa-placement.markdown | 164 ++++++--- tools/libxc/xc_domain.c | 153 +++++++- tools/libxc/xenctrl.h | 53 +++ tools/libxl/Makefile | 1 tools/libxl/check-xl-vcpupin-parse | 294 +++++++++++++++ tools/libxl/check-xl-vcpupin-parse.data-example | 53 +++ tools/libxl/libxl.c | 206 ++++++++++- tools/libxl/libxl.h | 23 + tools/libxl/libxl_create.c | 6 tools/libxl/libxl_dom.c | 22 + tools/libxl/libxl_types.idl | 4 tools/libxl/libxl_utils.h | 13 + tools/libxl/xl_cmdimpl.c | 439 ++++++++++++++++------- tools/libxl/xl_cmdtable.c | 8 xen/arch/x86/traps.c | 13 - xen/common/domain.c | 48 ++- xen/common/domctl.c | 38 ++ xen/common/keyhandler.c | 4 xen/common/sched_credit.c | 161 +++----- xen/common/sched_sedf.c | 2 xen/common/schedule.c | 56 ++- xen/common/wait.c | 10 - xen/include/public/domctl.h | 15 + xen/include/xen/sched.h | 14 - 26 files changed, 1515 insertions(+), 385 deletions(-) create mode 100755 tools/libxl/check-xl-vcpupin-parse create mode 100644 tools/libxl/check-xl-vcpupin-parse.data-example
Signature · 1 line
Signature
PATCH v2 1/16 xl: match output of vcpu-list with pinning syntax
in fact, pinning to all the pcpus happens by specifying "all" (either on the command line or in the config file), while `xl vcpu-list' report it as "any cpu".
Change this into something more consistent, by using "all" everywhere.
---
Changes since v1:
* this patch was not there in v1. It is now as using the same syntax for both input and output was requested during review.
--- tools/libxl/xl_cmdimpl.c | 27 +++++++-------------------- 1 file changed, 7 insertions(+), 20 deletions(-)
tools/libxl/xl_cmdimpl.c +7 -20
@@ -3101,8 +3101,7 @@ out: } } -/* If map is not full, prints it and returns 0. Returns 1 otherwise. */ -static int print_bitmap (uint8_t *map,int maplen,FILE *stream) +static void print_bitmap (uint8_t *map,int maplen,FILE *stream) {int i;uint8_t pmap =0 , bitmask =0 ;
@@ -3140,28 +3139,16 @@ static int print_bitmap(uint8_t *map, int maplen, FILE *stream)case 2 :break ;case 1 : -if (firstset ==0 ) -return 1 ; +if (firstset ==0 ) { +fprintf (stream,"all" ); +break ; + }case 3 :fprintf (stream,"%s%d" , state >1 ?"," :"" , firstset);if (i -1 > firstset)fprintf (stream,"-%d" , i -1 );break ; } - -return 0 ; -} - -static void print_cpumap (uint8_t *map,int maplen,FILE *stream) -{ -if (print_bitmap (map, maplen, stream)) -fprintf (stream,"any cpu" ); -} - -static void print_nodemap (uint8_t *map,int maplen,FILE *stream) -{ -if (print_bitmap (map, maplen, stream)) -fprintf (stream,"any node" ); } static void list_domains(int verbose, int context, int claim, int numa,
@@ -3234,7 +3221,7 @@ static void list_domains(int verbose, int context, int claim, int numa,libxl_domain_get_nodeaffinity (ctx, info[i].domid, &nodemap);putchar (' ' ); -print_nodemap (nodemap.map, physinfo.nr_nodes, stdout); +print_bitmap (nodemap.map, physinfo.nr_nodes, stdout); }putchar ('\n' ); }
@@ -4446,7 +4433,7 @@ static void print_vcpuinfo(uint32_t tdomid,/* TIM */ printf ("%9.1f " , ((float )vcpuinfo->vcpu_time /1e9 ));/* CPU AFFINITY */ -print_cpumap (vcpuinfo->cpumap.map, nr_cpus, stdout); +print_bitmap (vcpuinfo->cpumap.map, nr_cpus, stdout);printf ("\n" ); }
PATCH v2 2/16 xl: allow for node-wise specification of vcpu pinning
In reply to Dario FaggioliMaking it possible to use something like the following:
* "nodes:0-3": all pCPUs of nodes 0,1,2,3;
* "nodes:0-3,^node:2": all pCPUS of nodes 0,1,3;
* "1,nodes:1-2,^6": pCPU 1 plus all pCPUs of nodes 1,2 but not pCPU 6;
* ...
In both domain config file and `xl vcpu-pin'.
---
Changes from v1 (of this very series):
* actually checking for both "nodes:" and "node:" as per the doc says;
* using strcmp() (rather than strncmp()) when matching "all", to avoid returning success on any longer string that just begins with "all";
* fixing the handling (well, the rejection, actually) of "^all" and "^nodes:all";
* make some string pointers const.
Changes from v2 (of original series):
* turned a 'return' into 'goto out', consistently with the most of exit patterns;
* harmonized error handling: now parse_range() return a libxl error code, as requested during review;
* dealing with "all" moved inside update_cpumap_range(). It's tricky to move it in parse_range() (as requested during review), since we need the cpumap being modified handy when dealing with it. However, having it in update_cpumap_range() simplifies the code just as much as that;
* explicitly checking for junk after a valid value or range in parse_range(), as requested during review;
* xl exits on parsing failing, so no need to reset the cpumap to something sensible in vcpupin_parse(), as suggested during review;
Changes from v1 (of original series):
* code rearranged in order to look more simple to follow and understand, as requested during review;
* improved docs in xl.cfg.pod.5, as requested during review;
* strtoul() now returns into unsigned long, and the case where it returns ULONG_MAX is now taken into account, as requested during review;
* stuff like "all,^7" now works, as requested during review. Specifying just "^7" does not work either before or after this change
* killed some magic (i.e., ptr += 5 + (ptr[4] == 's') by introducing STR_SKIP_PREFIX() macro, as requested during review.
--- docs/man/xl.cfg.pod.5 | 20 ++++++ tools/libxl/xl_cmdimpl.c | 153 +++++++++++++++++++++++++++++++++------------- 2 files changed, 128 insertions(+), 45 deletions(-)
docs/man/xl.cfg.pod.5 +19 -1
@@ -115,7 +115,25 @@ To allow all the vcpus of the guest to run on all the cpus on the host. =item "0-3,5,^1" -To allow all the vcpus of the guest to run on cpus 0,2,3,5. +To allow all the vcpus of the guest to run on cpus 0,2,3,5. Combining +this with "all" is possible, meaning "all,^7" results in all the vcpus +of the guest running on all the cpus on the host except cpu 7. + +=item "nodes:0-3,node:^2" + +To allow all the vcpus of the guest to run on the cpus from NUMA nodes +0,1,3 of the host. So, if cpus 0-3 belongs to node 0, cpus 4-7 belongs +to node 1 and cpus 8-11 to node 3, the above would mean all the vcpus +of the guest will run on cpus 0-3,8-11. + +Combining this notation with the one above is possible. For instance, +"1,node:2,^6", means all the vcpus of the guest will run on cpu 1 and +on all the cpus of NUMA node 2, but not on cpu 6. Following the same +example as above, that would be cpus 1,4,5,7. + +Combining this with "all" is also possible, meaning "all,^nodes:1" +results in all the vcpus of the guest running on all the cpus on the +host, except for the cpus belonging to the host NUMA node 1. =item ["2", "3"] (or [2, 3])
tools/libxl/xl_cmdimpl.c +109 -44
176 lines not on this page · raw message
PATCH v2 3/16 xl: implement and enable dryrun mode for `xl vcpu-pin'
In reply to Dario FaggioliAs it can be useful to see if the outcome of some complex vCPU pinning bitmap specification looks as expected.
This also allow for the introduction of some automatic testing and verification for the bitmap parsing code, as it happens already in check-xl-disk-parse and check-xl-vif-parse.
In particular, to make the above possible, this commit also changes the implementation of the vcpu-pin command so that, instead of always returning 0, it returns an error if the parsing fails.
---
Changes since v2 (of this very series):
* fixed a typo in the changelog
--- tools/libxl/xl_cmdimpl.c | 49 +++++++++++++++++++++++++++++++++------------ tools/libxl/xl_cmdtable.c | 2 +- 2 files changed, 37 insertions(+), 14 deletions(-)
tools/libxl/xl_cmdimpl.c +36 -13
@@ -4566,30 +4566,53 @@ int main_vcpulist(int argc, char **argv)return 0 ; } -static void vcpupin (uint32_t domid,const char *vcpu,char *cpu) +static int vcpupin (uint32_t domid,const char *vcpu,char *cpu) {libxl_vcpuinfo *vcpuinfo;libxl_bitmap cpumap;uint32_t vcpuid;char *endptr; -int i, nb_vcpu; +int i =0 , nb_vcpu, rc =-1 ; + +libxl_bitmap_init (&cpumap); vcpuid =strtoul (vcpu, &endptr,10 );if (vcpu == endptr) {if (strcmp (vcpu,"all" )) {fprintf (stderr,"Error: Invalid argument.\n" ); -return ; + goto out; } vcpuid =-1 ; } -if (libxl_cpu_bitmap_alloc (ctx, &cpumap,0 )) { - goto vcpupin_out; - } +if (libxl_cpu_bitmap_alloc (ctx, &cpumap,0 )) + goto out;if (vcpupin_parse (cpu, &cpumap)) - goto vcpupin_out1; + goto out; + +if (dryrun_only) { +libxl_cputopology *info =libxl_get_cpu_topology (ctx, &i); + +if (!info) { +fprintf (stderr,"libxl_get_cpu_topology failed.\n" ); + goto out; + } +libxl_cputopology_list_free (info, i); + +fprintf (stdout,"cpumap: " ); +print_bitmap (cpumap.map, i, stdout); +fprintf (stdout,"\n" ); + +if (ferror (stdout) ||fflush (stdout)) { +perror ("stdout" ); +exit (-1 ); + } + + rc =0 ; + goto out; + }if (vcpuid !=-1 ) {if (libxl_set_vcpuaffinity (ctx, domid, vcpuid, &cpumap) ==-1 ) {
@@ -4599,7 +4622,7 @@ static void vcpupin(uint32_t domid, const char *vcpu, char *cpu)else {if (!(vcpuinfo =libxl_list_vcpu (ctx, domid, &nb_vcpu, &i))) {fprintf (stderr,"libxl_list_vcpu failed.\n" ); - goto vcpupin_out1; + goto out; }for (i =0 ; i < nb_vcpu; i++) { if (libxl_set_vcpuaffinity(ctx, domid, vcpuinfo[i].vcpuid,
@@ -4610,10 +4633,11 @@ static void vcpupin(uint32_t domid, const char *vcpu, char *cpu) }libxl_vcpuinfo_list_free (vcpuinfo, nb_vcpu); } - vcpupin_out1: + + rc =0 ; + out:libxl_bitmap_dispose (&cpumap); - vcpupin_out: - ; +return rc; }int main_vcpupin (int argc,char **argv)
@@ -4624,8 +4648,7 @@ int main_vcpupin(int argc, char **argv)/* No options */ } -vcpupin (find_domain (argv[optind]), argv[optind+1 ] , argv[optind+2 ]); -return 0 ; +return vcpupin (find_domain (argv[optind]), argv[optind+1 ] , argv[optind+2 ]); }static void vcpuset (uint32_t domid,const char * nr_vcpus,int check_host)
tools/libxl/xl_cmdtable.c +1 -1
@@ -211,7 +211,7 @@ struct cmd_spec cmd_table[] = {"[Domain, ...]" , }, { "vcpu-pin", - &main_vcpupin,0 ,1 , + &main_vcpupin,1 ,1 ,"Set which CPUs a VCPU can use" , "<Domain> <VCPU|all> <CPUs|all>", },
PATCH v2 4/16 xl: test script for the cpumap parser (for vCPU pinning)
In reply to Dario FaggioliThis commit introduces "check-xl-vcpupin-parse" for helping verifying and debugging the (v)CPU bitmap parsing code in xl.
The script runs "xl -N vcpu-pin 0 all <some strings>" repeatedly, with various input strings, and checks that the output is as expected.
This is what the script can do:
# ./check-xl-vcpupin-parse -h usage: ./check-xl-vcpupin-parse [options]
Tests various vcpu-pinning strings. If run without arguments acts as follows:
- generates some test data and saves them in
check-xl-vcpupin-parse.data;
- tests all the generated configurations (reading them back from check-xl-vcpupin-parse.data).
An example of a test vector file is provided in
check-xl-vcpupin-parse.data-example.
Options:
-h prints this message
-r seed uses seed for initializing the rundom number generator
(default: the script PID)
-s string tries using string as a vcpu pinning configuration and
reports whether that succeeds or not
-o ofile save the test data in ofile
(default: check-xl-vcpupin-parse.data)
-i ifile read test data from ifile
An example test data file (generated on a 2 NUMA nodes, 16 CPUs host) is being provided in check-xl-vcpupin-parse.data-example.
---
Changes from v2 (of original series):
* killed the `sleep 1', as requested during review;
* allow for passing a custom randon seed, and report what is the actual random seed used, as requested during review;
* allow for testing for specific pinning configuration strings, as suggested during review;
* stores the test data in a file, after generating them, and read them back from there for actual testing, as suggested during review;
* allow for reading the test data from an existing test file instead than always generating new ones.
Changes from v1 (of original series):
* this was not there in v1, and adding it has been requested during review.
--- tools/libxl/check-xl-vcpupin-parse | 294 +++++++++++++++++++++++ tools/libxl/check-xl-vcpupin-parse.data-example | 53 ++++ 2 files changed, 347 insertions(+) create mode 100755 tools/libxl/check-xl-vcpupin-parse create mode 100644 tools/libxl/check-xl-vcpupin-parse.data-example
tools/libxl/check-xl-vcpupin-parse new file +294 -0
@@ -0,0 +1,294 @@ +#!/bin/bash + +set -e + +if [ -x ./xl ] ; then + export LD_LIBRARY_PATH=.:../libxc:../xenstore: + XL=./xl +else + XL=xl +fi + +fprefix=tmp.check-xl-vcpupin-parse +outfile=check-xl-vcpupin-parse.data + +usage () { +cat <<END +usage: $0 [options] + +Tests various vcpu-pinning strings. If run without arguments acts +as follows: + - generates some test data and saves them in $outfile; + - tests all the generated configurations (reading them back from + $outfile). + +An example of a test vector file is provided in ${outfile}-example. + +Options: + -h prints this message + -r seed uses seed for initializing the rundom number generator + (default: the script PID) + -s string tries using string as a vcpu pinning configuration and + reports whether that succeeds or not + -o ofile save the test data in ofile (default: $outfile) + -i ifile read test data from ifile +END +} + +expected () { + cat >$fprefix.expected +} + +# by default, re-seed with our PID +seed=$$ +failures=0 + +# Execute one test and check the result against the provided +# rc value and output +one () { + expected_rc=$1; shift + printf "test case %s...\n" "$*" + set +e + ${XL} -N vcpu-pin 0 all "$@" </dev/null >$fprefix.actual 2>/dev/null + actual_rc=$? + if [ $actual_rc != $expected_rc ]; then + diff -u $fprefix.expected $fprefix.actual + echo >&2 "test case \`$*' failed ($actual_rc $diff_rc)" + failures=$(( $failures + 1 )) + fi + set -e +} + +# Write an entry in the test vector file. Format is as follows: +# test-string*expected-rc*expected-output +write () { + printf "$1*$2*$3\n" >> $outfile +} + +complete () { + if [ "$failures" = 0 ]; then + echo all ok.; exit 0 + else + echo "$failures tests failed."; exit 1 + fi +} + +# Test a specific pinning string +string () { + expected_rc=$1; shift + printf "test case %s...\n" "$*" + set +e + ${XL} -N vcpu-pin 0 all "$@" &> /dev/null + actual_rc=$? + set -e + + if [ $actual_rc != $expected_rc ]; then + echo >&2 "test case \`$*' failed ($actual_rc)" + else + echo >&2 "test case \`$*' succeeded" + fi + + exit 0 +} + +# Read a test vector file (provided as $1) line by line and +# test all the entries it contains +run () +{ + while read line + do + if [ ${line:0:1} != '#' ]; then + test_string="`echo $line | cut -f1 -d'*'`" + exp_rc="`echo $line | cut -f2 -d'*'`" + exp_output="`echo $line | cut -f3 -d'*'`" + + expected <<END +$exp_output +END + one $exp_rc "$test_string" + fi + done < $1 + + complete + + exit 0 +} + +while getopts "hr:s:o:i:" option +do + case $option in + h) + usage + exit 0 + ;; + r) + seed=$OPTARG + ;; + s) + string 0 "$OPTARG" + ;; + o) + outfile=$OPTARG + ;; + i) + run $OPTARG + ;; + esac +done + +#---------- test data ---------- +# +nr_cpus=`xl info | grep nr_cpus | cut -f2 -d':'` +nr_nodes=`xl info | grep nr_nodes | cut -f2 -d':'` +nr_cpus_per_node=`xl info -n | sed '/cpu:/,/numa_info/!d' | head -n -1 | \ + awk '{print $4}' | uniq -c | tail -1 | awk '{print $1}'` +cat >$outfile <<END +# WARNING: some of these tests are topology based tests. +# Expect failures if the topology is not detected correctly +# detected topology: $nr_cpus CPUs, $nr_nodes nodes, $nr_cpus_per_node CPUs per node. +# +# seed used for random number generation: seed=${seed}. +# +# Format is as follows: +# test-string*expected-return-code*expected-output +# +END + +# Re-seed the random number generator +RANDOM=$seed + +echo "# Testing a wrong configuration" >> $outfile +write foo 255 "" + +echo "# Testing the 'all' syntax" >> $outfile +write "all" 0 "cpumap: all" +write "nodes:all" 0 "cpumap: all" +write "all,nodes:all" 0 "cpumap: all" +write "all,^nodes:0,all" 0 "cpumap: all" + +echo "# Testing the empty cpumap case" >> $outfile +write "^0" 0 "cpumap: none" + +echo "# A few attempts of pinning to just one random cpu" >> $outfile +if [ $nr_cpus -gt 1 ]; then + for i in `seq 0 3`; do + cpu=$(($RANDOM % nr_cpus)) + write "$cpu" 0 "cpumap: $cpu" + done +fi + +echo "# A few attempts of pinning to all but one random cpu" >> $outfile +if [ $nr_cpus -gt 2 ]; then + for i in `seq 0 3`; do + cpu=$(($RANDOM % nr_cpus)) + if [ $cpu -eq 0 ]; then + expected_range="1-$((nr_cpus - 1))" + elif [ $cpu -eq 1 ]; then + expected_range="0,2-$((nr_cpus - 1))" + elif [ $cpu -eq $((nr_cpus - 2)) ]; then + expected_range="0-$((cpu - 1)),$((nr_cpus - 1))" + elif [ $cpu -eq $((nr_cpus - 1)) ]; then + expected_range="0-$((nr_cpus - 2))" + else + expected_range="0-$((cpu - 1)),$((cpu + 1))-$((nr_cpus - 1))" + fi + write "all,^$cpu" 0 "cpumap: $expected_range" + done +fi + +echo "# A few attempts of pinning to a random range of cpus" >> $outfile +if [ $nr_cpus -gt 2 ]; then + for i in `seq 0 3`; do + cpua=$(($RANDOM % nr_cpus)) + range=$((nr_cpus - cpua)) + cpub=$(($RANDOM % range)) + cpubb=$((cpua + cpub)) + if [ $cpua -eq $cpubb ]; then + expected_range="$cpua" + else + expected_range="$cpua-$cpubb" + fi + write "$expected_range" 0 "cpumap: $expected_range" + done +fi + +echo "# A few attempts of pinning to just one random node" >> $outfile +if [ $nr_nodes -gt 1 ]; then + for i in `seq 0 3`; do + node=$(($RANDOM % nr_nodes)) + # this assumes that the first $nr_cpus_per_node (from cpu + # 0 to cpu $nr_cpus_per_node-1) are assigned to the first node + # (node 0), the second $nr_cpus_per_node (from $nr_cpus_per_node + # to 2*$nr_cpus_per_node-1) are assigned to the second node (node + # 1), etc. Expect failures if that is not the case. + write "nodes:$node" 0 "cpumap: $((nr_cpus_per_node*node))-$((nr_cpus_per_node*(node+1)-1))" + done +fi + +echo "# A few attempts of pinning to all but one random node" >> $outfile +if [ $nr_nodes -gt 1 ]; then + for i in `seq 0 3`; do + node=$(($RANDOM % nr_nodes)) + # this assumes that the first $nr_cpus_per_node (from cpu + # 0 to cpu $nr_cpus_per_node-1) are assigned to the first node + # (node 0), the second $nr_cpus_per_node (from $nr_cpus_per_node + # to 2*$nr_cpus_per_node-1) are assigned to the second node (node + # 1), etc. Expect failures if that is not the case. + if [ $node -eq 0 ]; then + expected_range="$nr_cpus_per_node-$((nr_cpus - 1))" + elif [ $node -eq $((nr_nodes - 1)) ]; then + expected_range="0-$((nr_cpus - nr_cpus_per_node - 1))" + else + expected_range="0-$((nr_cpus_per_node*node-1)),$((nr_cpus_per_node*(node+1)))-$nr_cpus" + fi + write "all,^nodes:$node" 0 "cpumap: $expected_range" + done +fi + +echo "# A few attempts of pinning to a random range of nodes" >> $outfile +if [ $nr_nodes -gt 1 ]; then + for i in `seq 0 3`; do + nodea=$(($RANDOM % nr_nodes)) + range=$((nr_nodes - nodea)) + nodeb=$(($RANDOM % range)) + nodebb=$((nodea + nodeb)) + # this assumes that the first $nr_cpus_per_node (from cpu + # 0 to cpu $nr_cpus_per_node-1) are assigned to the first node + # (node 0), the second $nr_cpus_per_node (from $nr_cpus_per_node + # to 2*$nr_cpus_per_node-1) are assigned to the second node (node + # 1), etc. Expect failures if that is not the case. + if [ $nodea -eq 0 ] && [ $nodebb -eq $((nr_nodes - 1)) ]; then + expected_range="all" + else + expected_range="$((nr_cpus_per_node*nodea))-$((nr_cpus_per_node*(nodebb+1) - 1))" + fi + write "nodes:$nodea-$nodebb" 0 "cpumap: $expected_range" + done +fi + +echo "# A few attempts of pinning to a node but excluding one random cpu" >> $outfile +if [ $nr_nodes -gt 1 ]; then + for i in `seq 0 3`; do + node=$(($RANDOM % nr_nodes)) + # this assumes that the first $nr_cpus_per_node (from cpu + # 0 to cpu $nr_cpus_per_node-1) are assigned to the first node + # (node 0), the second $nr_cpus_per_node (from $nr_cpus_per_node + # to 2*$nr_cpus_per_node-1) are assigned to the second node (node + # 1), etc. Expect failures if that is not the case. + cpu=$(($RANDOM % nr_cpus_per_node + nr_cpus_per_node*node)) + if [ $cpu -eq $((nr_cpus_per_node*node)) ]; then + expected_range="$((nr_cpus_per_node*node + 1))-$((nr_cpus_per_node*(node+1) - 1))" + elif [ $cpu -eq $((nr_cpus_per_node*node + 1)) ]; then + expected_range="$((nr_cpus_per_node*node)),$((nr_cpus_per_node*node + 2))-$((nr_cpus_per_node*(node+1) - 1))" + elif [ $cpu -eq $((nr_cpus_per_node*(node+1) - 2)) ]; then + expected_range="$((nr_cpus_per_node*node))-$((nr_cpus_per_node*(node+1) - 3)),$((nr_cpus_per_node*(node+1) - 1))" + elif [ $cpu -eq $((nr_cpus_per_node*(node+1) - 1)) ]; then + expected_range="$((nr_cpus_per_node*node))-$((nr_cpus_per_node*(node+1) - 2))" + else + expected_range="$((nr_cpus_per_node*node))-$((cpu - 1)),$((cpu + 1))-$((nr_cpus_per_node*(node+1) - 1))" + fi + write "nodes:$node,^$cpu" 0 "cpumap: $expected_range" + done +fi + +run $outfile
tools/libxl/check-xl-vcpupin-parse.data-example new file +53 -0
53 lines not on this page · raw message
PATCH v2 5/16 xen: fix leaking of v->cpu_affinity_saved
In reply to Dario Faggiolion domain destruction.
--- xen/common/domain.c | 1 + 1 file changed, 1 insertion(+)
xen/common/domain.c +1 -0
@@ -736,6 +736,7 @@ static void complete_domain_destroy(struct rcu_head *head) {free_cpumask_var (v->cpu_affinity);free_cpumask_var (v->cpu_affinity_tmp); +free_cpumask_var (v->cpu_affinity_saved);free_cpumask_var (v->vcpu_dirty_cpumask);free_vcpu_struct (v); }
PATCH v2 6/16 xen: sched: make space for cpu_soft_affinity
In reply to Dario FaggioliBefore this change, each vcpu had its own vcpu-affinity (in v->cpu_affinity), representing the set of pcpus where the vcpu is allowed to run. Since when NUMA-aware scheduling was introduced the (credit1 only, for now) scheduler also tries as much as it can to run all the vcpus of a domain on one of the nodes that constitutes the domain's node-affinity.
The idea here is making the mechanism more general by:
* allowing for this 'preference' for some pcpus/nodes to be expressed on a per-vcpu basis, instead than for the domain as a whole. That is to say, each vcpu should have its own set of preferred pcpus/nodes, instead than it being the very same for all the vcpus of the domain;
* generalizing the idea of 'preferred pcpus' to not only NUMA awareness and support. That is to say, independently from it being or not (mostly) useful on NUMA systems, it should be possible to specify, for each vcpu, a set of pcpus where it prefers to run (in addition, and possibly unrelated to, the set of pcpus where it is allowed to run).
We will be calling this set of *preferred* pcpus the vcpu's soft affinity, and this change introduces, allocates, frees and initializes the data structure required to host that in struct vcpu (cpu_soft_affinity).
Also, the new field is not used anywhere yet, so no real functional change yet.
---
Changes from v1:
* this patch does something similar to what, in v1, was being done in "5/12 xen: numa-sched: make space for per-vcpu node-affinity"
--- xen/common/domain.c | 3 +++ xen/common/keyhandler.c | 2 ++ xen/common/schedule.c | 2 ++ xen/include/xen/sched.h | 3 +++ 4 files changed, 10 insertions(+)
xen/common/domain.c +3 -0
@@ -128,6 +128,7 @@ struct vcpu *alloc_vcpu(if ( !zalloc_cpumask_var (&v->cpu_affinity) || !zalloc_cpumask_var (&v->cpu_affinity_tmp) || !zalloc_cpumask_var (&v->cpu_affinity_saved) || + !zalloc_cpumask_var (&v->cpu_soft_affinity) || !zalloc_cpumask_var (&v->vcpu_dirty_cpumask) ) goto fail_free;
@@ -159,6 +160,7 @@ struct vcpu *alloc_vcpu(free_cpumask_var (v->cpu_affinity);free_cpumask_var (v->cpu_affinity_tmp);free_cpumask_var (v->cpu_affinity_saved); +free_cpumask_var (v->cpu_soft_affinity);free_cpumask_var (v->vcpu_dirty_cpumask);free_vcpu_struct (v);return NULL ;
@@ -737,6 +739,7 @@ static void complete_domain_destroy(struct rcu_head *head)free_cpumask_var (v->cpu_affinity);free_cpumask_var (v->cpu_affinity_tmp);free_cpumask_var (v->cpu_affinity_saved); +free_cpumask_var (v->cpu_soft_affinity);free_cpumask_var (v->vcpu_dirty_cpumask);free_vcpu_struct (v); }
xen/common/keyhandler.c +2 -0
@@ -298,6 +298,8 @@ static void dump_domains(unsigned char key)printk ("dirty_cpus=%s " , tmpstr);cpuset_print (tmpstr,sizeof (tmpstr), v->cpu_affinity);printk ("cpu_affinity=%s\n" , tmpstr); +cpuset_print (tmpstr,sizeof (tmpstr), v->cpu_soft_affinity); +printk ("cpu_soft_affinity=%s\n" , tmpstr);printk (" pause_count=%d pause_flags=%lx\n" ,atomic_read (&v->pause_count), v->pause_flags);arch_dump_vcpu_info (v);
xen/common/schedule.c +2 -0
@@ -198,6 +198,8 @@ int sched_init_vcpu(struct vcpu *v, unsigned int processor)else cpumask_setall (v->cpu_affinity); +cpumask_setall (v->cpu_soft_affinity); +/* Initialise the per-vcpu timers. */ init_timer (&v->periodic_timer, vcpu_periodic_timer_fn, v, v->processor);
xen/include/xen/sched.h +3 -0
@@ -198,6 +198,9 @@ struct vcpu/* Used to restore affinity across S3. */ cpumask_var_t cpu_affinity_saved; +/* Bitmask of CPUs on which this VCPU prefers to run. */ +cpumask_var_t cpu_soft_affinity; +/* Bitmask of CPUs which are holding onto this VCPU's state. */ cpumask_var_t vcpu_dirty_cpumask;
PATCH v2 7/16 xen: sched: rename v->cpu_affinity into v->cpu_hard_affinity
In reply to Dario Faggioliin order to distinguish it from the cpu_soft_affinity introduced by the previous commit ("xen: sched: make space for cpu_soft_affinity"). This patch does not imply any functional change, it is basically the result of something like the following:
s/cpu_affinity/cpu_hard_affinity/g
s/cpu_affinity_tmp/cpu_hard_affinity_tmp/g
s/cpu_affinity_saved/cpu_hard_affinity_saved/g
--- xen/arch/x86/traps.c | 11 ++++++----- xen/common/domain.c | 22 +++++++++++----------- xen/common/domctl.c | 2 +- xen/common/keyhandler.c | 2 +- xen/common/sched_credit.c | 12 ++++++------ xen/common/sched_sedf.c | 2 +- xen/common/schedule.c | 21 +++++++++++---------- xen/common/wait.c | 4 ++-- xen/include/xen/sched.h | 8 ++++---- 9 files changed, 43 insertions(+), 41 deletions(-)
xen/arch/x86/traps.c +6 -5
@@ -3083,7 +3083,8 @@ static void nmi_mce_softirq(void)/* Set the tmp value unconditionally, so that * the check in the iret hypercall works. */ -cpumask_copy (st->vcpu->cpu_affinity_tmp, st->vcpu->cpu_affinity); +cpumask_copy (st->vcpu->cpu_hard_affinity_tmp, + st->vcpu->cpu_hard_affinity);if ((cpu != st->processor) || (st->processor != st->vcpu->processor))
@@ -3118,11 +3119,11 @@ void async_exception_cleanup(struct vcpu *curr)return ;/* Restore affinity. */ -if ( !cpumask_empty (curr->cpu_affinity_tmp) && - !cpumask_equal (curr->cpu_affinity_tmp, curr->cpu_affinity) ) +if ( !cpumask_empty (curr->cpu_hard_affinity_tmp) && + !cpumask_equal (curr->cpu_hard_affinity_tmp, curr->cpu_hard_affinity) ) { -vcpu_set_affinity (curr, curr->cpu_affinity_tmp); -cpumask_clear (curr->cpu_affinity_tmp); +vcpu_set_affinity (curr, curr->cpu_hard_affinity_tmp); +cpumask_clear (curr->cpu_hard_affinity_tmp); }if ( !(curr->async_exception_mask & (curr->async_exception_mask -1 )) )
xen/common/domain.c +11 -11
@@ -125,9 +125,9 @@ struct vcpu *alloc_vcpu(tasklet_init (&v->continue_hypercall_tasklet,NULL ,0 ); -if ( !zalloc_cpumask_var (&v->cpu_affinity) || - !zalloc_cpumask_var (&v->cpu_affinity_tmp) || - !zalloc_cpumask_var (&v->cpu_affinity_saved) || +if ( !zalloc_cpumask_var (&v->cpu_hard_affinity) || + !zalloc_cpumask_var (&v->cpu_hard_affinity_tmp) || + !zalloc_cpumask_var (&v->cpu_hard_affinity_saved) || !zalloc_cpumask_var (&v->cpu_soft_affinity) || !zalloc_cpumask_var (&v->vcpu_dirty_cpumask) ) goto fail_free;
@@ -157,9 +157,9 @@ struct vcpu *alloc_vcpu( fail_wq:destroy_waitqueue_vcpu (v); fail_free: -free_cpumask_var (v->cpu_affinity); -free_cpumask_var (v->cpu_affinity_tmp); -free_cpumask_var (v->cpu_affinity_saved); +free_cpumask_var (v->cpu_hard_affinity); +free_cpumask_var (v->cpu_hard_affinity_tmp); +free_cpumask_var (v->cpu_hard_affinity_saved);free_cpumask_var (v->cpu_soft_affinity);free_cpumask_var (v->vcpu_dirty_cpumask);free_vcpu_struct (v);
@@ -373,7 +373,7 @@ void domain_update_node_affinity(struct domain *d) for_each_vcpu ( d, v ) { -cpumask_and (online_affinity, v->cpu_affinity, online); +cpumask_and (online_affinity, v->cpu_hard_affinity, online);cpumask_or (cpumask, cpumask, online_affinity); }
@@ -736,9 +736,9 @@ static void complete_domain_destroy(struct rcu_head *head)for ( i = d->max_vcpus -1 ; i >=0 ; i-- )if ( (v = d->vcpu[i]) !=NULL ) { -free_cpumask_var (v->cpu_affinity); -free_cpumask_var (v->cpu_affinity_tmp); -free_cpumask_var (v->cpu_affinity_saved); +free_cpumask_var (v->cpu_hard_affinity); +free_cpumask_var (v->cpu_hard_affinity_tmp); +free_cpumask_var (v->cpu_hard_affinity_saved);free_cpumask_var (v->cpu_soft_affinity);free_cpumask_var (v->vcpu_dirty_cpumask);free_vcpu_struct (v);
@@ -878,7 +878,7 @@ int vcpu_reset(struct vcpu *v) v->async_exception_mask =0 ;memset (v->async_exception_state,0 ,sizeof (v->async_exception_state));#endif -cpumask_clear (v->cpu_affinity_tmp); +cpumask_clear (v->cpu_hard_affinity_tmp);clear_bit (_VPF_blocked, &v->pause_flags);clear_bit (_VPF_in_reset, &v->pause_flags);
xen/common/domctl.c +1 -1
@@ -629,7 +629,7 @@ long do_domctl(XEN_GUEST_HANDLE_PARAM(xen_domctl_t) u_domctl)else { ret =cpumask_to_xenctl_bitmap ( - &op->u.vcpuaffinity.cpumap, v->cpu_affinity); + &op->u.vcpuaffinity.cpumap, v->cpu_hard_affinity); } }break ;
xen/common/keyhandler.c +1 -1
@@ -296,7 +296,7 @@ static void dump_domains(unsigned char key) !vcpu_event_delivery_is_enabled(v));cpuset_print (tmpstr,sizeof (tmpstr), v->vcpu_dirty_cpumask);printk ("dirty_cpus=%s " , tmpstr); -cpuset_print (tmpstr,sizeof (tmpstr), v->cpu_affinity); +cpuset_print (tmpstr,sizeof (tmpstr), v->cpu_hard_affinity);printk ("cpu_affinity=%s\n" , tmpstr);cpuset_print (tmpstr,sizeof (tmpstr), v->cpu_soft_affinity);printk ("cpu_soft_affinity=%s\n" , tmpstr);
xen/common/sched_credit.c +6 -6
@@ -332,13 +332,13 @@ csched_balance_cpumask(const struct vcpu *vc, int step, cpumask_t *mask)if ( step ==CSCHED_BALANCE_NODE_AFFINITY ) {cpumask_and (mask,CSCHED_DOM (vc->domain)->node_affinity_cpumask, - vc->cpu_affinity); + vc->cpu_hard_affinity);if (unlikely (cpumask_empty (mask)) ) -cpumask_copy (mask, vc->cpu_affinity); +cpumask_copy (mask, vc->cpu_hard_affinity); }else /* step == CSCHED_BALANCE_CPU_AFFINITY */ -cpumask_copy (mask, vc->cpu_affinity); +cpumask_copy (mask, vc->cpu_hard_affinity); }static void burn_credits (struct csched_vcpu *svc,s_time_t now)
@@ -407,7 +407,7 @@ __runq_tickle(unsigned int cpu, struct csched_vcpu *new)if ( balance_step ==CSCHED_BALANCE_NODE_AFFINITY && !__vcpu_has_node_affinity (new->vcpu, - new->vcpu->cpu_affinity) ) + new->vcpu->cpu_hard_affinity) )continue ;/* Are there idlers suitable for new (for this balance step)? */
@@ -642,7 +642,7 @@ _csched_cpu_pick(const struct scheduler *ops, struct vcpu *vc, bool_t commit)/* Store in cpus the mask of online cpus on which the domain can run */ online =cpupool_scheduler_cpumask (vc->domain->cpupool); -cpumask_and (&cpus, vc->cpu_affinity, online); +cpumask_and (&cpus, vc->cpu_hard_affinity, online);for_each_csched_balance_step ( balance_step ) {
@@ -1487,7 +1487,7 @@ csched_runq_steal(int peer_cpu, int cpu, int pri, int balance_step)* or counter. */ if ( balance_step ==CSCHED_BALANCE_NODE_AFFINITY - && !__vcpu_has_node_affinity (vc, vc->cpu_affinity) ) + && !__vcpu_has_node_affinity (vc, vc->cpu_hard_affinity) )continue ;csched_balance_cpumask (vc, balance_step, csched_balance_mask);
xen/common/sched_sedf.c +1 -1
@@ -396,7 +396,7 @@ static int sedf_pick_cpu(const struct scheduler *ops, struct vcpu *v)cpumask_t *online; online =cpupool_scheduler_cpumask (v->domain->cpupool); -cpumask_and (&online_affinity, v->cpu_affinity, online); +cpumask_and (&online_affinity, v->cpu_hard_affinity, online);return cpumask_cycle (v->vcpu_id %cpumask_weight (&online_affinity) -1 , &online_affinity); }
xen/common/schedule.c +11 -10
@@ -194,9 +194,9 @@ int sched_init_vcpu(struct vcpu *v, unsigned int processor)*/ v->processor = processor;if (is_idle_domain (d) || d->is_pinned ) -cpumask_copy (v->cpu_affinity,cpumask_of (processor)); +cpumask_copy (v->cpu_hard_affinity,cpumask_of (processor));else -cpumask_setall (v->cpu_affinity); +cpumask_setall (v->cpu_hard_affinity);cpumask_setall (v->cpu_soft_affinity);
@@ -287,7 +287,7 @@ int sched_move_domain(struct domain *d, struct cpupool *c)migrate_timer (&v->singleshot_timer, new_p);migrate_timer (&v->poll_timer, new_p); -cpumask_setall (v->cpu_affinity); +cpumask_setall (v->cpu_hard_affinity); lock =vcpu_schedule_lock_irq (v); v->processor = new_p;
@@ -459,7 +459,7 @@ static void vcpu_migrate(struct vcpu *v)*/ if ( pick_called && (new_lock ==per_cpu (schedule_data, new_cpu).schedule_lock) && -cpumask_test_cpu (new_cpu, v->cpu_affinity) && +cpumask_test_cpu (new_cpu, v->cpu_hard_affinity) &&cpumask_test_cpu (new_cpu, v->domain->cpupool->cpu_valid) )break ;
@@ -563,7 +563,7 @@ void restore_vcpu_affinity(struct domain *d) {printk (XENLOG_DEBUG "Restoring affinity for d%dv%d\n" , d->domain_id, v->vcpu_id); -cpumask_copy (v->cpu_affinity, v->cpu_affinity_saved); +cpumask_copy (v->cpu_hard_affinity, v->cpu_hard_affinity_saved); v->affinity_broken =0 ; }
@@ -606,20 +606,21 @@ int cpu_disable_scheduler(unsigned int cpu)unsigned long flags;spinlock_t *lock =vcpu_schedule_lock_irqsave (v, &flags); -cpumask_and (&online_affinity, v->cpu_affinity, c->cpu_valid); +cpumask_and (&online_affinity, v->cpu_hard_affinity, c->cpu_valid);if (cpumask_empty (&online_affinity) && -cpumask_test_cpu (cpu, v->cpu_affinity) ) +cpumask_test_cpu (cpu, v->cpu_hard_affinity) ) {printk (XENLOG_DEBUG "Breaking affinity for d%dv%d\n" , d->domain_id, v->vcpu_id);if (system_state == SYS_STATE_suspend) { -cpumask_copy (v->cpu_affinity_saved, v->cpu_affinity); +cpumask_copy (v->cpu_hard_affinity_saved, + v->cpu_hard_affinity); v->affinity_broken =1 ; } -cpumask_setall (v->cpu_affinity); +cpumask_setall (v->cpu_hard_affinity); }if ( v->processor == cpu )
@@ -667,7 +668,7 @@ int vcpu_set_affinity(struct vcpu *v, const cpumask_t *affinity) lock =vcpu_schedule_lock_irq (v); -cpumask_copy (v->cpu_affinity, affinity); +cpumask_copy (v->cpu_hard_affinity, affinity);/* Always ask the scheduler to re-evaluate placement * when changing the affinity */
xen/common/wait.c +2 -2
@@ -134,7 +134,7 @@ static void __prepare_to_wait(struct waitqueue_vcpu *wqv)/* Save current VCPU affinity; force wakeup on *this* CPU only. */ wqv->wakeup_cpu =smp_processor_id (); -cpumask_copy (&wqv->saved_affinity, curr->cpu_affinity); +cpumask_copy (&wqv->saved_affinity, curr->cpu_hard_affinity);if (vcpu_set_affinity (curr,cpumask_of (wqv->wakeup_cpu)) ) {gdprintk (XENLOG_ERR ,"Unable to set vcpu affinity\n" );
@@ -183,7 +183,7 @@ void check_wakeup_from_wait(void) {/* Re-set VCPU affinity and re-enter the scheduler. */ struct vcpu *curr = current; -cpumask_copy (&wqv->saved_affinity, curr->cpu_affinity); +cpumask_copy (&wqv->saved_affinity, curr->cpu_hard_affinity);if (vcpu_set_affinity (curr,cpumask_of (wqv->wakeup_cpu)) ) {gdprintk (XENLOG_ERR ,"Unable to set vcpu affinity\n" );
xen/include/xen/sched.h +4 -4
@@ -192,11 +192,11 @@ struct vcpuspinlock_t virq_lock;/* Bitmask of CPUs on which this VCPU may run. */ -cpumask_var_t cpu_affinity; +cpumask_var_t cpu_hard_affinity;/* Used to change affinity temporarily. */ -cpumask_var_t cpu_affinity_tmp; +cpumask_var_t cpu_hard_affinity_tmp;/* Used to restore affinity across S3. */ -cpumask_var_t cpu_affinity_saved; +cpumask_var_t cpu_hard_affinity_saved;/* Bitmask of CPUs on which this VCPU prefers to run. */ cpumask_var_t cpu_soft_affinity;
@@ -795,7 +795,7 @@ void watchdog_domain_destroy(struct domain *d);#define has_hvm_container_domain (d) ((d)->guest_type != guest_type_pv)#define has_hvm_container_vcpu (v) (has_hvm_container_domain((v)->domain))#define is_pinned_vcpu (v) ((v)->domain->is_pinned || \ - cpumask_weight((v)->cpu_affinity) == 1) + cpumask_weight((v)->cpu_hard_affinity) == 1)#ifdef HAS_PASSTHROUGH #define need_iommu (d) ((d)->need_iommu)#else
PATCH v2 8/16 xen: derive NUMA node affinity from hard and soft CPU affinity
In reply to Dario Faggioliif a domain's NUMA node-affinity (which is what controls memory allocations) is provided by the user/toolstack, it just is not touched. However, if the user does not say anything, leaving it all to Xen, let's compute it in the following way:
1. cpupool's cpus & hard-affinity & soft-affinity
2. if (1) is empty: cpupool's cpus & hard-affinity
This guarantees memory to be allocated from the narrowest possible set of NUMA nodes, ad makes it relatively easy to set up NUMA-aware scheduling on top of soft affinity.
--- xen/common/domain.c | 22 +++++++++++++++++++++- 1 file changed, 21 insertions(+), 1 deletion(-)
xen/common/domain.c +21 -1
@@ -353,7 +353,7 @@ struct domain *domain_create(void domain_update_node_affinity (struct domain *d) { -cpumask_var_t cpumask; +cpumask_var_t cpumask, cpumask_soft;cpumask_var_t online_affinity;const cpumask_t *online;struct vcpu *v;
@@ -361,9 +361,15 @@ void domain_update_node_affinity(struct domain *d)if ( !zalloc_cpumask_var (&cpumask) )return ; +if ( !zalloc_cpumask_var (&cpumask_soft) ) + { +free_cpumask_var (cpumask); +return ; + }if ( !alloc_cpumask_var (&online_affinity) ) {free_cpumask_var (cpumask); +free_cpumask_var (cpumask_soft);return ; }
@@ -373,8 +379,12 @@ void domain_update_node_affinity(struct domain *d) for_each_vcpu ( d, v ) { +/* Build up the mask of online pcpus we have hard affinity with */ cpumask_and (online_affinity, v->cpu_hard_affinity, online);cpumask_or (cpumask, cpumask, online_affinity); + +/* As well as the mask of all pcpus we have soft affinity with */ +cpumask_or (cpumask_soft, cpumask_soft, v->cpu_soft_affinity); }/*
@@ -386,6 +396,15 @@ void domain_update_node_affinity(struct domain *d)*/ if ( d->auto_node_affinity ) { +/* +* We're looking for the narower possible set of nodes. So, if +* possible (i.e., if not empty!) let's use the intersection +* between online, hard and soft affinity. If not, just fall back +* to online & hard affinity. +*/ +if (cpumask_intersects (cpumask, cpumask_soft) ) +cpumask_and (cpumask, cpumask, cpumask_soft); +nodes_clear (d->node_affinity);for_each_online_node ( node ) if ( cpumask_intersects(&node_to_cpumask(node), cpumask) )
@@ -397,6 +416,7 @@ void domain_update_node_affinity(struct domain *d)spin_unlock (&d->node_affinity_lock);free_cpumask_var (online_affinity); +free_cpumask_var (cpumask_soft);free_cpumask_var (cpumask); }
PATCH v2 9/16 xen: sched: DOMCTL_*vcpuaffinity works with hard and soft affinity
In reply to Dario Faggioliby adding a flag for the caller to specify which one he cares about.
Add also another cpumap there. This way, in case of DOMCTL_setvcpuaffinity, Xen can return back to the caller the "effective affinity" of the vcpu. We call the effective affinity the intersection between cpupool's cpus, the new hard affinity and (if asking to set soft affinity) the new soft affinity.
The purpose of this is allowing the toolstack to figure out whether or not the requested change in affinity produced sensible results, when combined with the other settings that are already in place.
--- tools/libxc/xc_domain.c | 4 +++- xen/arch/x86/traps.c | 4 ++-- xen/common/domctl.c | 38 ++++++++++++++++++++++++++++++++++---- xen/common/schedule.c | 35 ++++++++++++++++++++++++----------- xen/common/wait.c | 6 +++--- xen/include/public/domctl.h | 15 +++++++++++++-- xen/include/xen/sched.h | 3 ++- 7 files changed, 81 insertions(+), 24 deletions(-)
tools/libxc/xc_domain.c +3 -1
@@ -215,7 +215,9 @@ int xc_vcpu_setaffinity(xc_interface *xch, domctl.cmd = XEN_DOMCTL_setvcpuaffinity; domctl.domain = (domid_t )domid; - domctl.u.vcpuaffinity.vcpu = vcpu; + domctl.u.vcpuaffinity.vcpu = vcpu; +/* Soft affinity is there, but not used anywhere for now, so... */ + domctl.u.vcpuaffinity.flags =XEN_VCPUAFFINITY_HARD ;memcpy (local, cpumap, cpusize);
xen/arch/x86/traps.c +2 -2
@@ -3093,7 +3093,7 @@ static void nmi_mce_softirq(void)* Make sure to wakeup the vcpu on the * specified processor. */ -vcpu_set_affinity (st->vcpu,cpumask_of (st->processor)); +vcpu_set_hard_affinity (st->vcpu,cpumask_of (st->processor));/* Affinity is restored in the iret hypercall. */ }
@@ -3122,7 +3122,7 @@ void async_exception_cleanup(struct vcpu *curr)if ( !cpumask_empty (curr->cpu_hard_affinity_tmp) && !cpumask_equal (curr->cpu_hard_affinity_tmp, curr->cpu_hard_affinity) ) { -vcpu_set_affinity (curr, curr->cpu_hard_affinity_tmp); +vcpu_set_hard_affinity (curr, curr->cpu_hard_affinity_tmp);cpumask_clear (curr->cpu_hard_affinity_tmp); }
xen/common/domctl.c +34 -4
@@ -617,19 +617,49 @@ long do_domctl(XEN_GUEST_HANDLE_PARAM(xen_domctl_t) u_domctl)if ( op->cmd == XEN_DOMCTL_setvcpuaffinity ) {cpumask_var_t new_affinity; +cpumask_t *online; ret =xenctl_bitmap_to_cpumask ( &new_affinity, &op->u.vcpuaffinity.cpumap); -if ( !ret ) +if ( ret ) +break ; + + ret = -EINVAL ; +if ( op->u.vcpuaffinity.flags &XEN_VCPUAFFINITY_HARD ) + ret =vcpu_set_hard_affinity (v, new_affinity); +else if ( op->u.vcpuaffinity.flags &XEN_VCPUAFFINITY_SOFT ) + ret =vcpu_set_soft_affinity (v, new_affinity); + +if ( ret ) { - ret =vcpu_set_affinity (v, new_affinity);free_cpumask_var (new_affinity); +break ; } + +/* +* Report back to the caller what the "effective affinity", that +* is the intersection of cpupool's pcpus, the (new?) hard +* affinity and the (new?) soft-affinity. +*/ + online =cpupool_online_cpumask (v->domain->cpupool); +cpumask_and (new_affinity, online, v->cpu_hard_affinity); +cpumask_and (new_affinity, new_affinity , v->cpu_soft_affinity); + + ret =cpumask_to_xenctl_bitmap ( + &op->u.vcpuaffinity.effective_affinity, new_affinity); + +free_cpumask_var (new_affinity); }else { - ret =cpumask_to_xenctl_bitmap ( - &op->u.vcpuaffinity.cpumap, v->cpu_hard_affinity); +if ( op->u.vcpuaffinity.flags &XEN_VCPUAFFINITY_HARD ) + ret =cpumask_to_xenctl_bitmap ( + &op->u.vcpuaffinity.cpumap, v->cpu_hard_affinity); +else if ( op->u.vcpuaffinity.flags &XEN_VCPUAFFINITY_SOFT ) + ret =cpumask_to_xenctl_bitmap ( + &op->u.vcpuaffinity.cpumap, v->cpu_soft_affinity); +else + ret = -EINVAL ; } }break ;
xen/common/schedule.c +24 -11
@@ -653,22 +653,14 @@ void sched_set_node_affinity(struct domain *d, nodemask_t *mask)SCHED_OP (DOM2OP (d), set_node_affinity, d, mask); } -int vcpu_set_affinity (struct vcpu *v,const cpumask_t *affinity) +static int vcpu_set_affinity ( +struct vcpu *v,const cpumask_t *affinity,cpumask_t **which) { -cpumask_t online_affinity; -cpumask_t *online;spinlock_t *lock; -if ( v->domain->is_pinned ) -return -EINVAL ; - online =VCPU2ONLINE (v); -cpumask_and (&online_affinity, affinity, online); -if (cpumask_empty (&online_affinity) ) -return -EINVAL ; - lock =vcpu_schedule_lock_irq (v); -cpumask_copy (v->cpu_hard_affinity, affinity); +cpumask_copy (*which, affinity);/* Always ask the scheduler to re-evaluate placement * when changing the affinity */
@@ -687,6 +679,27 @@ int vcpu_set_affinity(struct vcpu *v, const cpumask_t *affinity)return 0 ; } +int vcpu_set_hard_affinity (struct vcpu *v,const cpumask_t *affinity) +{ +cpumask_t online_affinity; +cpumask_t *online; + +if ( v->domain->is_pinned ) +return -EINVAL ; + + online =VCPU2ONLINE (v); +cpumask_and (&online_affinity, affinity, online); +if (cpumask_empty (&online_affinity) ) +return -EINVAL ; + +return vcpu_set_affinity (v, affinity, &v->cpu_hard_affinity); +} + +int vcpu_set_soft_affinity (struct vcpu *v,const cpumask_t *affinity) +{ +return vcpu_set_affinity (v, affinity, &v->cpu_soft_affinity); +} +/* Block the currently-executing domain until a pertinent event occurs. */ void vcpu_block (void ) {
xen/common/wait.c +3 -3
@@ -135,7 +135,7 @@ static void __prepare_to_wait(struct waitqueue_vcpu *wqv)/* Save current VCPU affinity; force wakeup on *this* CPU only. */ wqv->wakeup_cpu =smp_processor_id ();cpumask_copy (&wqv->saved_affinity, curr->cpu_hard_affinity); -if (vcpu_set_affinity (curr,cpumask_of (wqv->wakeup_cpu)) ) +if (vcpu_set_hard_affinity (curr,cpumask_of (wqv->wakeup_cpu)) ) {gdprintk (XENLOG_ERR ,"Unable to set vcpu affinity\n" );domain_crash_synchronous ();
@@ -166,7 +166,7 @@ static void __prepare_to_wait(struct waitqueue_vcpu *wqv)static void __finish_wait (struct waitqueue_vcpu *wqv) { wqv->esp =NULL ; - (void )vcpu_set_affinity (current, &wqv->saved_affinity); + (void )vcpu_set_hard_affinity (current, &wqv->saved_affinity); }void check_wakeup_from_wait (void )
@@ -184,7 +184,7 @@ void check_wakeup_from_wait(void)/* Re-set VCPU affinity and re-enter the scheduler. */ struct vcpu *curr = current;cpumask_copy (&wqv->saved_affinity, curr->cpu_hard_affinity); -if (vcpu_set_affinity (curr,cpumask_of (wqv->wakeup_cpu)) ) +if (vcpu_set_hard_affinity (curr,cpumask_of (wqv->wakeup_cpu)) ) {gdprintk (XENLOG_ERR ,"Unable to set vcpu affinity\n" );domain_crash_synchronous ();
xen/include/public/domctl.h +13 -2
@@ -300,8 +300,19 @@ DEFINE_XEN_GUEST_HANDLE(xen_domctl_nodeaffinity_t);/* XEN_DOMCTL_setvcpuaffinity */ /* XEN_DOMCTL_getvcpuaffinity */ struct xen_domctl_vcpuaffinity { -uint32_t vcpu;/* IN */ -struct xenctl_bitmap cpumap;/* IN/OUT */ +/* IN variables. */ +uint32_t vcpu; +/* Set/get the hard affinity for vcpu */ +#define _XEN_VCPUAFFINITY_HARD 0 +#define XEN_VCPUAFFINITY_HARD (1U<<_XEN_VCPUAFFINITY_HARD) +/* Set/get the soft affinity for vcpu */ +#define _XEN_VCPUAFFINITY_SOFT 1 +#define XEN_VCPUAFFINITY_SOFT (1U<<_XEN_VCPUAFFINITY_SOFT) +uint32_t flags; +/* IN/OUT variables. */ +struct xenctl_bitmap cpumap; +/* OUT variables. */ +struct xenctl_bitmap effective_affinity; };typedef struct xen_domctl_vcpuaffinity xen_domctl_vcpuaffinity_t ;DEFINE_XEN_GUEST_HANDLE (xen_domctl_vcpuaffinity_t);
xen/include/xen/sched.h +2 -1
@@ -755,7 +755,8 @@ void scheduler_free(struct scheduler *sched);int schedule_cpu_switch (unsigned int cpu,struct cpupool *c);void vcpu_force_reschedule (struct vcpu *v);int cpu_disable_scheduler (unsigned int cpu); -int vcpu_set_affinity (struct vcpu *v,const cpumask_t *affinity); +int vcpu_set_hard_affinity (struct vcpu *v,const cpumask_t *affinity); +int vcpu_set_soft_affinity (struct vcpu *v,const cpumask_t *affinity);void restore_vcpu_affinity (struct domain *d);void vcpu_runstate_get (struct vcpu *v,struct vcpu_runstate_info *runstate);
PATCH v2 10/16 xen: sched: use soft-affinity instead of domain's node-affinity
In reply to Dario Faggiolinow that we have it, use soft affinity for scheduling, and replace the indirect use of the domain's NUMA node-affinity. This is more general, as soft affinity does not have to be related to NUMA. At the same time it allows to achieve the same results as NUMA-aware scheduling, just by making soft affinity equal to the domain's node affinity, for all the vCPUs (e.g., from the toolstack).
This also means renaming most of the NUMA-aware scheduling related functions, in credit1, to something more generic, hinting toward the concept of soft affinity rather than directly to NUMA awareness.
As a side effects, this simplifies the code quit a bit. In fact, prior to this change, we needed to cache the translation of d->node_affinity (which is a nodemask_t) to a cpumask_t, since that is what scheduling decisions require (we used to keep it in node_affinity_cpumask). This, and all the complicated logic required to keep it updated, is not necessary any longer.
The high level description of NUMA placement and scheduling in docs/misc/xl-numa-placement.markdown is being updated too, to match the new architecture.
---
Changes from v1:
* in v1, "7/12 xen: numa-sched: use per-vcpu node-affinity for actual scheduling" was doing something very similar to this patch.
--- docs/misc/xl-numa-placement.markdown | 148 ++++++++++++++++++++++----------- xen/common/domain.c | 2 xen/common/sched_credit.c | 153 +++++++++++++--------------------- 3 files changed, 157 insertions(+), 146 deletions(-)
docs/misc/xl-numa-placement.markdown +97 -51
@@ -12,13 +12,6 @@ is quite more complex and slow. On these machines, a NUMA node is usually defined as a set of processor cores (typically a physical CPU package) and the memory directly attached to the set of cores. -The Xen hypervisor deals with NUMA machines by assigning to each domain -a "node affinity", i.e., a set of NUMA nodes of the host from which they -get their memory allocated. Also, even if the node affinity of a domain -is allowed to change on-line, it is very important to "place" the domain -correctly when it is fist created, as the most of its memory is allocated -at that time and can not (for now) be moved easily. - NUMA awareness becomes very important as soon as many domains start running memory-intensive workloads on a shared host. In fact, the cost of accessing non node-local memory locations is very high, and the
@@ -27,14 +20,37 @@ performance degradation is likely to be noticeable. For more information, have a look at the [Xen NUMA Introduction][numa_intro] page on the Wiki. +## Xen and NUMA machines: the concept of _node-affinity_ ## + +The Xen hypervisor deals with NUMA machines throughout the concept of +_node-affinity_. The node-affinity of a domain is the set of NUMA nodes +of the host where the memory for the domain is being allocated (mostly, +at domain creation time). This is, at least in principle, different and +unrelated with the vCPU (hard and soft, see below) scheduling affinity, +which instead is the set of pCPUs where the vCPU is allowed (or prefers) +to run. + +Of course, despite the fact that they belong to and affect different +subsystems, the domain node-affinity and the vCPUs affinity are not +completely independent. +In fact, if the domain node-affinity is not explicitly specified by the +user, via the proper libxl calls or xl config item, it will be computed +basing on the vCPUs' scheduling affinity. + +Notice that, even if the node affinity of a domain may change on-line, +it is very important to "place" the domain correctly when it is fist +created, as the most of its memory is allocated at that time and can +not (for now) be moved easily. + ### Placing via pinning and cpupools ### -The simplest way of placing a domain on a NUMA node is statically pinning -the domain's vCPUs to the pCPUs of the node. This goes under the name of -CPU affinity and can be set through the "cpus=" option in the config file -(more about this below). Another option is to pool together the pCPUs -spanning the node and put the domain in such a cpupool with the "pool=" -config option (as documented in our [Wiki][cpupools_howto]). +The simplest way of placing a domain on a NUMA node is setting the hard +scheduling affinity of the domain's vCPUs to the pCPUs of the node. This +also goes under the name of vCPU pinning, and can be done through the +"cpus=" option in the config file (more about this below). Another option +is to pool together the pCPUs spanning the node and put the domain in +such a _cpupool_ with the "pool=" config option (as documented in our +[Wiki][cpupools_howto]). In both the above cases, the domain will not be able to execute outside the specified set of pCPUs for any reasons, even if all those pCPUs are
@@ -45,24 +61,45 @@ may come at he cost of some load imbalances. ### NUMA aware scheduling ### -If the credit scheduler is in use, the concept of node affinity defined -above does not only apply to memory. In fact, starting from Xen 4.3, the -scheduler always tries to run the domain's vCPUs on one of the nodes in -its node affinity. Only if that turns out to be impossible, it will just -pick any free pCPU. - -This is, therefore, something more flexible than CPU affinity, as a domain -can still run everywhere, it just prefers some nodes rather than others. -Locality of access is less guaranteed than in the pinning case, but that -comes along with better chances to exploit all the host resources (e.g., -the pCPUs). - -In fact, if all the pCPUs in a domain's node affinity are busy, it is -possible for the domain to run outside of there, but it is very likely that -slower execution (due to remote memory accesses) is still better than no -execution at all, as it would happen with pinning. For this reason, NUMA -aware scheduling has the potential of bringing substantial performances -benefits, although this will depend on the workload. +If using the credit1 scheduler, and starting from Xen 4.3, the scheduler +itself always tries to run the domain's vCPUs on one of the nodes in +its node-affinity. Only if that turns out to be impossible, it will just +pick any free pCPU. Locality of access is less guaranteed than in the +pinning case, but that comes along with better chances to exploit all +the host resources (e.g., the pCPUs). + +Starting from Xen 4.4, credit1 supports two forms of affinity: hard and +soft, both on a per-vCPU basis. This means each vCPU can have its own +soft affinity, stating where such vCPU prefers to execute on. This is +less strict than what it (also starting from 4.4) is called hard affinity, +as the vCPU can potentially run everywhere, it just prefers some pCPUs +rather than others. +In Xen 4.4, therefore, NUMA-aware scheduling is achieved by matching the +soft affinity of the vCPUs of a domain with its node-affinity. + +In fact, as it was for 4.3, if all the pCPUs in a vCPU's soft affinity +are busy, it is possible for the domain to run outside from there. The +idea is that slower execution (due to remote memory accesses) is still +better than no execution at all (as it would happen with pinning). For +this reason, NUMA aware scheduling has the potential of bringing +substantial performances benefits, although this will depend on the +workload. + +Notice that, for each vCPU, the following three scenarios are possbile: + + * a vCPU *is pinned* to some pCPUs and *does not have* any soft affinity + In this case, the vCPU is always scheduled on one of the pCPUs to which + it is pinned, without any specific peference among them. + * a vCPU *has* its own soft affinity and *is not* pinned to any particular + pCPU. In this case, the vCPU can run on every pCPU. Nevertheless, the + scheduler will try to have it running on one of the pCPUs in its soft + affinity; + * a vCPU *has* its own vCPU soft affinity and *is also* pinned to some + pCPUs. In this case, the vCPU is always scheduled on one of the pCPUs + onto which it is pinned, with, among them, a preference for the ones + that also forms its soft affinity. In case pinning and soft affinity + form two disjoint sets of pCPUs, pinning "wins", and the soft affinity + is just ignored. ## Guest placement in xl ##
@@ -71,25 +108,23 @@ both manual or automatic placement of them across the host's NUMA nodes. Note that xm/xend does a very similar thing, the only differences being the details of the heuristics adopted for automatic placement (see below), -and the lack of support (in both xm/xend and the Xen versions where that\ +and the lack of support (in both xm/xend and the Xen versions where that was the default toolstack) for NUMA aware scheduling. ### Placing the guest manually ### Thanks to the "cpus=" option, it is possible to specify where a domain should be created and scheduled on, directly in its config file. This -affects NUMA placement and memory accesses as the hypervisor constructs -the node affinity of a VM basing right on its CPU affinity when it is -created. +affects NUMA placement and memory accesses as, in this case, the +hypervisor constructs the node-affinity of a VM basing right on its +vCPU pinning when it is created. This is very simple and effective, but requires the user/system -administrator to explicitly specify affinities for each and every domain, +administrator to explicitly specify the pinning for each and every domain, or Xen won't be able to guarantee the locality for their memory accesses. -Notice that this also pins the domain's vCPUs to the specified set of -pCPUs, so it not only sets the domain's node affinity (its memory will -come from the nodes to which the pCPUs belong), but at the same time -forces the vCPUs of the domain to be scheduled on those same pCPUs. +That, of course, also mean the vCPUs of the domain will only be able to +execute on those same pCPUs. ### Placing the guest automatically ###
@@ -97,7 +132,9 @@ If no "cpus=" option is specified in the config file, libxl tries to figure out on its own on which node(s) the domain could fit best. If it finds one (some), the domain's node affinity get set to there, and both memory allocations and NUMA aware scheduling (for the credit -scheduler and starting from Xen 4.3) will comply with it. +scheduler and starting from Xen 4.3) will comply with it. Starting from +Xen 4.4, this also means that the mask resulting from this "fitting" +procedure will become the soft affinity of all the vCPUs of the domain. It is worthwhile noting that optimally fitting a set of VMs on the NUMA nodes of an host is an incarnation of the Bin Packing Problem. In fact,
@@ -142,34 +179,43 @@ any placement from happening: libxl_defbool_set(&domain_build_info->numa_placement, false); -Also, if `numa_placement` is set to `true`, the domain must not -have any CPU affinity (i.e., `domain_build_info->cpumap` must -have all its bits set, as it is by default), or domain creation -will fail returning `ERROR_INVAL`. +Also, if `numa_placement` is set to `true`, the domain's vCPUs must +not be pinned (i.e., `domain_build_info->cpumap` must have all its +bits set, as it is by default), or domain creation will fail with +`ERROR_INVAL`. Starting from Xen 4.3, in case automatic placement happens (and is -successful), it will affect the domain's node affinity and _not_ its -CPU affinity. Namely, the domain's vCPUs will not be pinned to any +successful), it will affect the domain's node-affinity and _not_ its +vCPU pinning. Namely, the domain's vCPUs will not be pinned to any pCPU on the host, but the memory from the domain will come from the selected node(s) and the NUMA aware scheduling (if the credit scheduler -is in use) will try to keep the domain there as much as possible. +is in use) will try to keep the domain's vCPUs there as much as possible. Besides than that, looking and/or tweaking the placement algorithm search "Automatic NUMA placement" in libxl\_internal.h. Note this may change in future versions of Xen/libxl. +## Xen < 4.4 ## + +The concept of vCPU soft affinity has been introduced for the first time +in Xen 4.4. In 4.3, it is the domain's node-affinity that drives the +NUMA-aware scheduler. The main difference is soft affinity is per-vCPU, +and so each vCPU can have its own mask of pCPUs, while node-affinity is +per-domain, that is the equivalent of having all the vCPUs with the same +soft affinity. + ## Xen < 4.3 ## As NUMA aware scheduling is a new feature of Xen 4.3, things are a little bit different for earlier version of Xen. If no "cpus=" option is specified and Xen 4.2 is in use, the automatic placement algorithm still runs, but the results is used to _pin_ the vCPUs of the domain to the output node(s). -This is consistent with what was happening with xm/xend, which were also -affecting the domain's CPU affinity. +This is consistent with what was happening with xm/xend. On a version of Xen earlier than 4.2, there is not automatic placement at -all in xl or libxl, and hence no node or CPU affinity being affected. +all in xl or libxl, and hence no node-affinity, vCPU affinity or pinning +being introduced/modified. ## Limitations ##
xen/common/domain.c +0 -2
@@ -411,8 +411,6 @@ void domain_update_node_affinity(struct domain *d)node_set (node, d->node_affinity); } -sched_set_node_affinity (d, &d->node_affinity); -spin_unlock (&d->node_affinity_lock);free_cpumask_var (online_affinity);
xen/common/sched_credit.c +60 -93
282 lines not on this page · raw message
PATCH v2 11/16 libxc: get and set soft and hard affinity
In reply to Dario Faggioliby using the new flag introduced in the parameters of DOMCTL_{get,set}_vcpuaffinity.
This happens in two new xc calls: xc_vcpu_setaffinity_hard() and xc_vcpu_setaffinity_soft() (an in the corresponding getters, of course).
The existing xc_vcpu_{set,get}affinity() call is also retained, with the following behavior:
* xc_vcpu_setaffinity() sets both the hard and soft affinity;
* xc_vcpu_getaffinity() gets the hard affinity.
This is mainly for backward compatibility reasons, i.e., trying not to break existing callers/users.
--- tools/libxc/xc_domain.c | 153 ++++++++++++++++++++++++++++++++++++++++++----- tools/libxc/xenctrl.h | 53 ++++++++++++++++ 2 files changed, 190 insertions(+), 16 deletions(-)
tools/libxc/xc_domain.c +137 -16
@@ -189,13 +189,16 @@ int xc_domain_node_getaffinity(xc_interface *xch,return ret; } -int xc_vcpu_setaffinity (xc_interface *xch, -uint32_t domid, -int vcpu, -xc_cpumap_t cpumap) +static int _vcpu_setaffinity (xc_interface *xch, +uint32_t domid, +int vcpu, +xc_cpumap_t cpumap, +uint32_t flags, +xc_cpumap_t ecpumap) {DECLARE_DOMCTL ; -DECLARE_HYPERCALL_BUFFER (uint8_t, local); +DECLARE_HYPERCALL_BUFFER (uint8_t, cpumap_local); +DECLARE_HYPERCALL_BUFFER (uint8_t, ecpumap_local);int ret =-1 ;int cpusize;
@@ -206,39 +209,119 @@ int xc_vcpu_setaffinity(xc_interface *xch, goto out; } - local =xc_hypercall_buffer_alloc (xch, local, cpusize); -if ( local ==NULL ) + cpumap_local =xc_hypercall_buffer_alloc (xch, cpumap_local, cpusize); +if ( cpumap_local ==NULL ) + { +PERROR ("Could not allocate cpumap_local for DOMCTL_setvcpuaffinity" ); + goto out; + } + ecpumap_local =xc_hypercall_buffer_alloc (xch, ecpumap_local, cpusize); +if ( cpumap_local ==NULL ) { -PERROR ("Could not allocate memory for setvcpuaffinity domctl hypercall" ); +xc_hypercall_buffer_free (xch, cpumap_local); +PERROR ("Could not allocate ecpumap_local for DOMCTL_setvcpuaffinity" ); goto out; } domctl.cmd = XEN_DOMCTL_setvcpuaffinity; domctl.domain = (domid_t )domid; domctl.u.vcpuaffinity.vcpu = vcpu; -/* Soft affinity is there, but not used anywhere for now, so... */ - domctl.u.vcpuaffinity.flags =XEN_VCPUAFFINITY_HARD ; - -memcpy (local, cpumap, cpusize); - -set_xen_guest_handle (domctl.u.vcpuaffinity.cpumap.bitmap, local); + domctl.u.vcpuaffinity.flags = flags; +memcpy (cpumap_local, cpumap, cpusize); +set_xen_guest_handle (domctl.u.vcpuaffinity.cpumap.bitmap, cpumap_local); domctl.u.vcpuaffinity.cpumap.nr_bits = cpusize *8 ; +set_xen_guest_handle (domctl.u.vcpuaffinity.effective_affinity.bitmap, + ecpumap_local); + domctl.u.vcpuaffinity.effective_affinity.nr_bits = cpusize *8 ; + ret =do_domctl (xch, &domctl); -xc_hypercall_buffer_free (xch, local); +memcpy (ecpumap, ecpumap_local, cpusize); + +xc_hypercall_buffer_free (xch, cpumap_local); +xc_hypercall_buffer_free (xch, ecpumap_local); out:return ret; } +int xc_vcpu_setaffinity_soft (xc_interface *xch, +uint32_t domid, +int vcpu, +xc_cpumap_t cpumap, +xc_cpumap_t ecpumap) +{ +return _vcpu_setaffinity (xch, + domid, + vcpu, + cpumap, +XEN_VCPUAFFINITY_SOFT , + ecpumap); +} -int xc_vcpu_getaffinity (xc_interface *xch, +int xc_vcpu_setaffinity_hard (xc_interface *xch, +uint32_t domid, +int vcpu, +xc_cpumap_t cpumap, +xc_cpumap_t ecpumap) +{ +return _vcpu_setaffinity (xch, + domid, + vcpu, + cpumap, +XEN_VCPUAFFINITY_HARD , + ecpumap); +} + +/* Provided for backword compattibility: sets both hard and soft affinity */ +int xc_vcpu_setaffinity (xc_interface *xch,uint32_t domid,int vcpu,xc_cpumap_t cpumap) { +xc_cpumap_t ecpumap; +int ret =-1 ; + + ecpumap =xc_cpumap_alloc (xch); +if (ecpumap ==NULL ) + { +PERROR ("Could not allocate memory for DOMCTL_setvcpuaffinity" ); +return -1 ; + } + + ret =_vcpu_setaffinity (xch, + domid, + vcpu, + cpumap, +XEN_VCPUAFFINITY_SOFT , + ecpumap); + +if ( ret ) + { +PERROR ("Could not set soft affinity via DOMCTL_setvcpuaffinity" ); + goto out; + } + + ret =_vcpu_setaffinity (xch, + domid, + vcpu, + cpumap, +XEN_VCPUAFFINITY_HARD , + ecpumap); + out: +free (ecpumap); +return ret; +} + + +static int _vcpu_getaffinity (xc_interface *xch, +uint32_t domid, +int vcpu, +xc_cpumap_t cpumap, +uint32_t flags) +{DECLARE_DOMCTL ;DECLARE_HYPERCALL_BUFFER (uint8_t, local);int ret =-1 ;
@@ -261,6 +344,7 @@ int xc_vcpu_getaffinity(xc_interface *xch, domctl.cmd = XEN_DOMCTL_getvcpuaffinity; domctl.domain = (domid_t )domid; domctl.u.vcpuaffinity.vcpu = vcpu; + domctl.u.vcpuaffinity.flags = flags;set_xen_guest_handle (domctl.u.vcpuaffinity.cpumap.bitmap, local); domctl.u.vcpuaffinity.cpumap.nr_bits = cpusize *8 ;
@@ -274,6 +358,43 @@ out:return ret; } +int xc_vcpu_getaffinity_soft (xc_interface *xch, +uint32_t domid, +int vcpu, +xc_cpumap_t cpumap) +{ +return _vcpu_getaffinity (xch, + domid, + vcpu, + cpumap, +XEN_VCPUAFFINITY_SOFT ); +} + +int xc_vcpu_getaffinity_hard (xc_interface *xch, +uint32_t domid, +int vcpu, +xc_cpumap_t cpumap) +{ +return _vcpu_getaffinity (xch, + domid, + vcpu, + cpumap, +XEN_VCPUAFFINITY_HARD ); +} + +/* Provided for backward compatibility and wired to hard affinity */ +int xc_vcpu_getaffinity (xc_interface *xch, +uint32_t domid, +int vcpu, +xc_cpumap_t cpumap) +{ +return _vcpu_getaffinity (xch, + domid, + vcpu, + cpumap, +XEN_VCPUAFFINITY_HARD ); +} +int xc_domain_get_guest_width (xc_interface *xch,uint32_t domid,unsigned int *guest_width) {
tools/libxc/xenctrl.h +53 -0
63 lines not on this page · raw message
PATCH v2 12/16 libxl: get and set soft affinity
In reply to Dario Faggioliwhich basically means making space for a new cpumap in both vcpu_info (for getting soft affinity) and build_info (for setting it), along with providing the get/set functions, and wiring them to the proper xc calls. Interface is as follows:
* libxl_{get,set}_vcpuaffinity() deals with hard affinity, as it always has happened;
* libxl_get,set}_vcpuaffinity_soft() deals with soft affinity.
*_set_* functions include some logic for checking whether the affinity that would indeed be used matches the one requested by the caller, and printing some warnings if that is not the case.
That is because, despite what the user asks, for instance, for soft affinity, the scheduler only considers running a vCPU where its hard affinity and cpupool mandate. So, although we want to allow any possible combinations (e.g., we do not want to error out if hard affinity and soft afinity are disjoint), we at very least print some warnings, hoping to help the sysadmin to figure out what is really going on.
Also, as this apparently is the first change being checked in that breaks libxl ABI, bump MAJOR to 4.4.
--- tools/libxl/Makefile | 1 tools/libxl/libxl.c | 206 ++++++++++++++++++++++++++++++++++++++++--- tools/libxl/libxl.h | 23 +++++ tools/libxl/libxl_create.c | 6 + tools/libxl/libxl_types.idl | 4 + tools/libxl/libxl_utils.h | 13 +++ 6 files changed, 239 insertions(+), 14 deletions(-)
tools/libxl/Makefile +1 -0
@@ -5,6 +5,7 @@ XEN_ROOT = $(CURDIR)/../.. include $(XEN_ROOT)/tools/Rules.mk +#MAJOR = 4.4 MAJOR = 4.3 MINOR = 0
tools/libxl/libxl.c +193 -13
235 lines not on this page · raw message
tools/libxl/libxl.h +23 -0
@@ -82,6 +82,20 @@#define LIBXL_HAVE_DOMAIN_NODEAFFINITY 1/* +* LIBXL_HAVE_VCPUINFO_SOFTAFFINITY indicates that a 'cpumap_soft' +* field (of libxl_bitmap type) is present in libxl_vcpuinfo, +* containing the soft affinity for the vcpu. +*/ +#define LIBXL_HAVE_VCPUINFO_SOFTAFFINITY 1 + +/* +* LIBXL_HAVE_BUILDINFO_SOFTAFFINITY indicates that a 'cpumap_soft' +* field (of libxl_bitmap type) is present in libxl_domain_build_info, +* containing the soft affinity for the vcpu. +*/ +#define LIBXL_HAVE_BUILDINFO_SOFTAFFINITY 1 + +/* * LIBXL_HAVE_BUILDINFO_HVM_VENDOR_DEVICE indicates that the * libxl_vendor_device field is present in the hvm sections of * libxl_domain_build_info. This field tells libxl which
@@ -971,8 +985,17 @@ int libxl_userdata_retrieve(libxl_ctx *ctx, uint32_t domid,int libxl_get_physinfo (libxl_ctx *ctx,libxl_physinfo *physinfo);int libxl_set_vcpuaffinity (libxl_ctx *ctx,uint32_t domid,uint32_t vcpuid,libxl_bitmap *cpumap); +int libxl_set_vcpuaffinity_soft (libxl_ctx *ctx,uint32_t domid, +uint32_t vcpuid,libxl_bitmap *cpumap);int libxl_set_vcpuaffinity_all (libxl_ctx *ctx,uint32_t domid,unsigned int max_vcpus,libxl_bitmap *cpumap); +int libxl_set_vcpuaffinity_all_soft (libxl_ctx *ctx,uint32_t domid, +unsigned int max_vcpus, +libxl_bitmap *cpumap); +int libxl_get_vcpuaffinity (libxl_ctx *ctx,uint32_t domid, +uint32_t vcpuid,libxl_bitmap *cpumap); +int libxl_get_vcpuaffinity_soft (libxl_ctx *ctx,uint32_t domid, +uint32_t vcpuid,libxl_bitmap *cpumap);int libxl_domain_set_nodeaffinity (libxl_ctx *ctx,uint32_t domid,libxl_bitmap *nodemap); int libxl_domain_get_nodeaffinity(libxl_ctx *ctx, uint32_t domid,
tools/libxl/libxl_create.c +6 -0
@@ -192,6 +192,12 @@ int libxl__domain_build_info_setdefault(libxl__gc *gc,libxl_bitmap_set_any (&b_info->cpumap); } +if (!b_info->cpumap_soft.size) { +if (libxl_cpu_bitmap_alloc (CTX , &b_info->cpumap_soft,0 )) +return ERROR_FAIL ; +libxl_bitmap_set_any (&b_info->cpumap_soft); + } +libxl_defbool_setdefault (&b_info->numa_placement, true);if (!b_info->nodemap.size) {
tools/libxl/libxl_types.idl +3 -1
@@ -297,6 +297,7 @@ libxl_domain_build_info = Struct("domain_build_info",[ ("max_vcpus", integer), ("avail_vcpus", libxl_bitmap), ("cpumap", libxl_bitmap), + ("cpumap_soft", libxl_bitmap), ("nodemap", libxl_bitmap), ("numa_placement", libxl_defbool), ("tsc_mode", libxl_tsc_mode),
@@ -509,7 +510,8 @@ libxl_vcpuinfo = Struct("vcpuinfo", [ ("blocked", bool), ("running", bool), ("vcpu_time", uint64), # total vcpu time ran (ns) - ("cpumap", libxl_bitmap), # current cpu's affinities + ("cpumap", libxl_bitmap), # current hard cpu affinity + ("cpumap_soft", libxl_bitmap), # current soft cpu affinity ], dir=DIR_OUT) libxl_physinfo = Struct("physinfo", [
tools/libxl/libxl_utils.h +13 -0
@@ -98,6 +98,19 @@ static inline int libxl_bitmap_cpu_valid(libxl_bitmap *bitmap, int bit)#define libxl_for_each_set_bit (v, m) for (v = 0; v < (m).size * 8; v++) \ if (libxl_bitmap_test(&(m), v)) +static inline int libxl_bitmap_equal (const libxl_bitmap *ba, +const libxl_bitmap *bb) +{ +int i; + +libxl_for_each_bit (i, *ba) { +if (libxl_bitmap_test (ba, i) !=libxl_bitmap_test (bb, i)) +return 0 ; + } + +return 1 ; +} +static inline int libxl_cpu_bitmap_alloc (libxl_ctx *ctx,libxl_bitmap *cpumap,int max_cpus) {
PATCH v2 13/16 xl: show soft affinity in `xl vcpu-list'
In reply to Dario Faggioliif the '-s'/'--soft' option is provided. An example of such output is this:
# xl vcpu-list -s Name ID VCPU CPU State Time(s) Hard Affinity / Soft Affinity Domain-0 0 0 11 -b- 5.4 8-15 / all Domain-0 0 1 11 -b- 1.0 8-15 / all Domain-0 0 14 13 -b- 1.4 8-15 / all Domain-0 0 15 8 -b- 1.6 8-15 / all vm-test 3 0 4 -b- 2.5 0-12 / 0-7 vm-test 3 1 0 -b- 3.2 0-12 / 0-7
xl manual page is updated accordingly.
--- docs/man/xl.pod.1 | 10 ++++++ tools/libxl/xl_cmdimpl.c | 73 ++++++++++++++++++++++++++++----------------- tools/libxl/xl_cmdtable.c | 3 +- 3 files changed, 57 insertions(+), 29 deletions(-)
docs/man/xl.pod.1 +10 -0
@@ -619,6 +619,16 @@ after B<vcpu-set>, go to B<SEE ALSO> section for information. Lists VCPU information for a specific domain. If no domain is specified, VCPU information for all domains will be provided. +B<OPTIONS> + +=over 4 + +=item B<-s>, B<--soft> + +Print the CPU soft affinity of each VCPU too. + +=back + =item B<vcpu-pin> I<domain-id> I<vcpu> I<cpus> Pins the VCPU to only run on the specific CPUs. The keyword
tools/libxl/xl_cmdimpl.c +45 -28
119 lines not on this page · raw message
tools/libxl/xl_cmdtable.c +2 -1
@@ -208,7 +208,8 @@ struct cmd_spec cmd_table[] = { {"vcpu-list" , &main_vcpulist,0 ,0 ,"List the VCPUs for all/some domains" , -"[Domain, ...]" , +"[option] [Domain, ...]" , +"-s, --soft Show CPU soft affinity" , }, { "vcpu-pin", &main_vcpupin, 1, 1,
PATCH v2 14/16 xl: enable setting soft affinity
In reply to Dario FaggioliAlso by adding a '-s'/'--soft' switch to `xl vcpu-pin'.
--- docs/man/xl.pod.1 | 24 ++++++++++++++++- tools/libxl/xl_cmdimpl.c | 65 +++++++++++++++++++++++++++------------------ tools/libxl/xl_cmdtable.c | 3 +- 3 files changed, 64 insertions(+), 28 deletions(-)
docs/man/xl.pod.1 +23 -1
@@ -629,7 +629,7 @@ Print the CPU soft affinity of each VCPU too. =back -=item B<vcpu-pin> I<domain-id> I<vcpu> I<cpus> +=item B<vcpu-pin> [I<OPTIONS>] I<domain-id> I<vcpu> I<cpus> Pins the VCPU to only run on the specific CPUs. The keyword B<all> can be used to apply the I<cpus> list to all VCPUs in the
@@ -640,6 +640,28 @@ different run state is appropriate. Pinning can be used to restrict this, by ensuring certain VCPUs can only run on certain physical CPUs. +B<OPTIONS> + +=over 4 + +=item B<-s>, B<--soft> + +The same as above, but affect I<soft affinity> rather than pinning +(also called I<hard affinity>). + +Normally, VCPUs just wander among the CPUs where it is allowed to +run (either all the CPUs or the ones to which it is pinned, as said +for B<vcpu-list>). Soft affinity offer a mean to specify one or more +I<preferred> CPUs. Basically, among the ones where it can run, the +VCPU the VCPU will greately prefer to execute on one of these CPUs, +whenever that is possible. + +Notice that, in order for soft affinity to actually work, it needs +special support in the scheduler. Right now, only credit1 provides +that. + +=back + =item B<vm-list> Prints information about guests. This list excludes information about
tools/libxl/xl_cmdimpl.c +39 -26
95 lines not on this page · raw message
tools/libxl/xl_cmdtable.c +2 -1
@@ -214,7 +214,8 @@ struct cmd_spec cmd_table[] = { {"vcpu-pin" , &main_vcpupin,1 ,1 ,"Set which CPUs a VCPU can use" , -"<Domain> <VCPU|all> <CPUs|all>" , +"[option] <Domain> <VCPU|all> <CPUs|all>" , +"-s, --soft Deal with soft affinity" , }, { "vcpu-set", &main_vcpuset, 0, 1,
PATCH v2 15/16 xl: enable for specifying node-affinity in the config file
In reply to Dario Faggioliin a similar way to how it is possible to specify vcpu-affinity.
--- docs/man/xl.cfg.pod.5 | 27 ++++++++++++++-- tools/libxl/xl_cmdimpl.c | 78 +++++++++++++++++++++++++++++++++++++++++++++- 2 files changed, 100 insertions(+), 5 deletions(-)
docs/man/xl.cfg.pod.5 +24 -3
@@ -144,19 +144,40 @@ run on cpu #3 of the host. =back If this option is not specified, no vcpu to cpu pinning is established, -and the vcpus of the guest can run on all the cpus of the host. +and the vcpus of the guest can run on all the cpus of the host. If this +option is specified, the intersection of the vcpu pinning mask, provided +here, and the soft affinity mask, provided via B<cpus\_soft=> (if any), +is utilized to compute the domain node-affinity, for driving memory +allocations. If we are on a NUMA machine (i.e., if the host has more than one NUMA node) and this option is not specified, libxl automatically tries to place the guest on the least possible number of nodes. That, however, will not affect vcpu pinning, so the guest will still be able to run on -all the cpus, it will just prefer the ones from the node it has been -placed on. A heuristic approach is used for choosing the best node (or +all the cpus. A heuristic approach is used for choosing the best node (or set of nodes), with the goals of maximizing performance for the guest and, at the same time, achieving efficient utilization of host cpus and memory. See F<docs/misc/xl-numa-placement.markdown> for more details. +=item B<cpus_soft="CPU-LIST"> + +Exactly as B<cpus=>, but specifies soft affinity, rather than pinning +(also called hard affinity). Starting from Xen 4.4, and if the credit +scheduler is used, this means the vcpus of the domain prefers to run +these pcpus. Default is either all pcpus or xl (via libxl) guesses +(depending on what other options are present). + +A C<CPU-LIST> is specified exactly as above, for B<cpus=>. + +If this option is not specified, the vcpus of the guest will not have +any preference regarding on what cpu to run, and the scheduler will +treat all the cpus where a vcpu can execute (if B<cpus=> is specified), +or all the host cpus (if not), the same. If this option is specified, +the intersection of the soft affinity mask, provided here, and the vcpu +pinning, provided via B<cpus=> (if any), is utilized to compute the +domain node-affinity, for driving memory allocations. + =back =head3 CPU Scheduling
tools/libxl/xl_cmdimpl.c +76 -2
103 lines not on this page · raw message
PATCH v2 16/16 libxl: automatic NUMA placement affects soft affinity
In reply to Dario FaggiolivCPU soft affinity and NUMA-aware scheduling does not have to be related. However, soft affinity is how NUMA-aware scheduling is actually implemented, and therefore, by default, the results of automatic NUMA placement (at VM creation time) are also used to set the soft affinity of all the vCPUs of the domain.
Of course, this only happens if automatic NUMA placement is enabled and actually takes place (for instance, if the user does not specify any hard and soft affiniy in the xl config file).
This also takes care of the vice-versa, i.e., don't trigger automatic placement if the config file specifies either an hard (the check for which was already there) or a soft (the check for which is introduced by this commit) affinity.
--- docs/man/xl.cfg.pod.5 | 21 +++++++++++---------- docs/misc/xl-numa-placement.markdown | 16 ++++++++++++++-- tools/libxl/libxl_dom.c | 22 ++++++++++++++++++++-- 3 files changed, 45 insertions(+), 14 deletions(-)
docs/man/xl.cfg.pod.5 +11 -10
@@ -150,16 +150,6 @@ here, and the soft affinity mask, provided via B<cpus\_soft=> (if any), is utilized to compute the domain node-affinity, for driving memory allocations. -If we are on a NUMA machine (i.e., if the host has more than one NUMA -node) and this option is not specified, libxl automatically tries to -place the guest on the least possible number of nodes. That, however, -will not affect vcpu pinning, so the guest will still be able to run on -all the cpus. A heuristic approach is used for choosing the best node (or -set of nodes), with the goals of maximizing performance for the guest -and, at the same time, achieving efficient utilization of host cpus -and memory. See F<docs/misc/xl-numa-placement.markdown> for more -details. - =item B<cpus_soft="CPU-LIST"> Exactly as B<cpus=>, but specifies soft affinity, rather than pinning
@@ -178,6 +168,17 @@ the intersection of the soft affinity mask, provided here, and the vcpu pinning, provided via B<cpus=> (if any), is utilized to compute the domain node-affinity, for driving memory allocations. +If this option is not specified (and B<cpus=> is not specified either), +libxl automatically tries to place the guest on the least possible +number of nodes. A heuristic approach is used for choosing the best +node (or set of nodes), with the goal of maximizing performance for +the guest and, at the same time, achieving efficient utilization of +host cpus and memory. In that case, the soft affinity of all the vcpus +of the domain will be set to the pcpus belonging to the NUMA nodes +chosen during placement. + +For more details, see F<docs/misc/xl-numa-placement.markdown>. + =back =head3 CPU Scheduling
docs/misc/xl-numa-placement.markdown +14 -2
@@ -126,10 +126,22 @@ or Xen won't be able to guarantee the locality for their memory accesses. That, of course, also mean the vCPUs of the domain will only be able to execute on those same pCPUs. +Starting from 4.4, is is also possible to specify a "cpus\_soft=" option +in the xl config file. This, independently from whether or not "cpus=" is +specified too, affect the NUMA placement in a way very similar to what +is described above. In fact, the hypervisor will build up the node-affinity +of the VM basing right on it or, if both pinning (via "cpus=") and soft +affinity (via "cpus\_soft=") are present, basing on their intersection. + +Besides that, "cpus\_soft=" also means, of course, that the vCPUs of the +domain will prefer to execute on, among the pCPUs where they can run, +those particular pCPUs. + + ### Placing the guest automatically ### -If no "cpus=" option is specified in the config file, libxl tries -to figure out on its own on which node(s) the domain could fit best. +If neither "cpus=" nor "cpus\_soft=" are present in the config file, libxl +tries to figure out on its own on which node(s) the domain could fit best. If it finds one (some), the domain's node affinity get set to there, and both memory allocations and NUMA aware scheduling (for the credit scheduler and starting from Xen 4.3) will comply with it. Starting from
tools/libxl/libxl_dom.c +20 -2
@@ -222,21 +222,39 @@ int libxl__build_pre(libxl__gc *gc, uint32_t domid,* some weird error manifests) the subsequent call to * libxl_domain_set_nodeaffinity() will do the actual placement, * whatever that turns out to be. +* +* As far as scheduling is concerned, we achieve NUMA-aware scheduling +* by having the results of placement affect the soft affinity of all +* the vcpus of the domain. Of course, we want that iff placement is +* enabled and actually happens, so we only change info->cpumap_soft to +* reflect the placement result if that is the case */ if (libxl_defbool_val (info->numa_placement)) { -if (!libxl_bitmap_is_full (&info->cpumap)) { +/* We require both hard and soft affinity not to be set */ +if (!libxl_bitmap_is_full (&info->cpumap) || + !libxl_bitmap_is_full (&info->cpumap_soft)) {LOG (ERROR ,"Can run NUMA placement only if no vcpu " -"affinity is specified" ); +"(hard or soft) affinity is specified" );return ERROR_INVAL ; } rc =numa_place_domain (gc, domid, info);if (rc)return rc; + +/* +* We change the soft affinity in domain_build_info here, of course +* after converting the result of placement from nodes to cpus. the +* following call to libxl_set_vcpuaffinity_all_soft() will do the +* actual updating of the domain's vcpus' soft affinity. +*/ +libxl_nodemap_to_cpumap (ctx, &info->nodemap, &info->cpumap_soft); }libxl_domain_set_nodeaffinity (ctx, domid, &info->nodemap);libxl_set_vcpuaffinity_all (ctx, domid, info->max_vcpus, &info->cpumap); +libxl_set_vcpuaffinity_all_soft (ctx, domid, info->max_vcpus, + &info->cpumap_soft);xc_domain_setmaxmem (ctx->xch, domid, info->target_memkb +LIBXL_MAXMEM_CONSTANT ); xs_domid =xs_read (ctx->xsh,XBT_NULL ,"/tool/xenstored/domid" ,NULL );
Re: [PATCH v2 12/16] libxl: get and set soft affinity
In reply to Dario FaggioliQuoted 19 lines · Dario Faggioli
Also, as this apparently is the first change being checked in that breaks libxl ABI, bump MAJOR to 4.4.
Signed-off-by Dario Faggioli <[email protected]>--- tools/libxl/Makefile | 1 tools/libxl/libxl.c | 206 ++++++++++++++++++++++++++++++++++++++++--- tools/libxl/libxl.h | 23 +++++ tools/libxl/libxl_create.c | 6 + tools/libxl/libxl_types.idl | 4 + tools/libxl/libxl_utils.h | 13 +++ 6 files changed, 239 insertions(+), 14 deletions(-)tools/libxl/Makefile +0
-0
tools/libxl/Makefile line 11
Quoted 7 lines
tools/libxl/Makefile +1
-0@@ -5,6 +5,7 @@ XEN_ROOT = $(CURDIR)/../.. include $(XEN_ROOT)/tools/Rules.mk +#MAJOR = 4.4 MAJOR = 4.3 MINOR = 0
And, of course, this is quite wrong! :-(
Sorry for that. Correct version with this line _not_ commented out attached.
Regards,
Dario
Signature · 4 lines
<<This happens because I choose it to happen!>> (Raistlin Majere) ----------------------------------------------------------------- Dario Faggioli, Ph.D, http://about.me/dario.faggioli Senior Software Engineer, Citrix Systems R&D Ltd., Cambridge (UK)
commit af862ae01c5399de37d7728d9bc254e2314562cc
Author: Dario Faggioli <[email protected]> Date: Fri Nov 8 15:48:47 2013 +0100
libxl: get and set soft affinity
which basically means making space for a new cpumap in both vcpu_info (for getting soft affinity) and build_info (for setting it), along with providing the get/set functions, and wiring them to the proper xc calls. Interface is as follows:
* libxl_{get,set}_vcpuaffinity() deals with hard affinity, as it always has happened;
* libxl_get,set}_vcpuaffinity_soft() deals with soft affinity.
*_set_* functions include some logic for checking whether the affinity that would indeed be used matches the one requested by the caller, and printing some warnings if that is not the case.
That is because, despite what the user asks, for instance, for soft affinity, the scheduler only considers running a vCPU where its hard affinity and cpupool mandate. So, although we want to allow any possible combinations (e.g., we do not want to error out if hard affinity and soft afinity are disjoint), we at very least print some warnings, hoping to help the sysadmin to figure out what is really going on.
Also, as this apparently is the first change being checked in that breaks libxl ABI, bump MAJOR to 4.4.
tools/libxl/Makefile +1 -1
@@ -5,7 +5,7 @@ XEN_ROOT = $(CURDIR)/../.. include $(XEN_ROOT)/tools/Rules.mk -MAJOR = 4.3 +MAJOR = 4.4 MINOR = 0 XLUMAJOR = 4.3
tools/libxl/libxl.c +193 -13
235 lines not on this page · raw message
tools/libxl/libxl.h +23 -0
@@ -82,6 +82,20 @@#define LIBXL_HAVE_DOMAIN_NODEAFFINITY 1/* +* LIBXL_HAVE_VCPUINFO_SOFTAFFINITY indicates that a 'cpumap_soft' +* field (of libxl_bitmap type) is present in libxl_vcpuinfo, +* containing the soft affinity for the vcpu. +*/ +#define LIBXL_HAVE_VCPUINFO_SOFTAFFINITY 1 + +/* +* LIBXL_HAVE_BUILDINFO_SOFTAFFINITY indicates that a 'cpumap_soft' +* field (of libxl_bitmap type) is present in libxl_domain_build_info, +* containing the soft affinity for the vcpu. +*/ +#define LIBXL_HAVE_BUILDINFO_SOFTAFFINITY 1 + +/* * LIBXL_HAVE_BUILDINFO_HVM_VENDOR_DEVICE indicates that the * libxl_vendor_device field is present in the hvm sections of * libxl_domain_build_info. This field tells libxl which
@@ -971,8 +985,17 @@ int libxl_userdata_retrieve(libxl_ctx *ctx, uint32_t domid,int libxl_get_physinfo (libxl_ctx *ctx,libxl_physinfo *physinfo);int libxl_set_vcpuaffinity (libxl_ctx *ctx,uint32_t domid,uint32_t vcpuid,libxl_bitmap *cpumap); +int libxl_set_vcpuaffinity_soft (libxl_ctx *ctx,uint32_t domid, +uint32_t vcpuid,libxl_bitmap *cpumap);int libxl_set_vcpuaffinity_all (libxl_ctx *ctx,uint32_t domid,unsigned int max_vcpus,libxl_bitmap *cpumap); +int libxl_set_vcpuaffinity_all_soft (libxl_ctx *ctx,uint32_t domid, +unsigned int max_vcpus, +libxl_bitmap *cpumap); +int libxl_get_vcpuaffinity (libxl_ctx *ctx,uint32_t domid, +uint32_t vcpuid,libxl_bitmap *cpumap); +int libxl_get_vcpuaffinity_soft (libxl_ctx *ctx,uint32_t domid, +uint32_t vcpuid,libxl_bitmap *cpumap);int libxl_domain_set_nodeaffinity (libxl_ctx *ctx,uint32_t domid,libxl_bitmap *nodemap); int libxl_domain_get_nodeaffinity(libxl_ctx *ctx, uint32_t domid,
tools/libxl/libxl_create.c +6 -0
@@ -192,6 +192,12 @@ int libxl__domain_build_info_setdefault(libxl__gc *gc,libxl_bitmap_set_any (&b_info->cpumap); } +if (!b_info->cpumap_soft.size) { +if (libxl_cpu_bitmap_alloc (CTX , &b_info->cpumap_soft,0 )) +return ERROR_FAIL ; +libxl_bitmap_set_any (&b_info->cpumap_soft); + } +libxl_defbool_setdefault (&b_info->numa_placement, true);if (!b_info->nodemap.size) {
tools/libxl/libxl_types.idl +3 -1
@@ -297,6 +297,7 @@ libxl_domain_build_info = Struct("domain_build_info",[ ("max_vcpus", integer), ("avail_vcpus", libxl_bitmap), ("cpumap", libxl_bitmap), + ("cpumap_soft", libxl_bitmap), ("nodemap", libxl_bitmap), ("numa_placement", libxl_defbool), ("tsc_mode", libxl_tsc_mode),
@@ -509,7 +510,8 @@ libxl_vcpuinfo = Struct("vcpuinfo", [ ("blocked", bool), ("running", bool), ("vcpu_time", uint64), # total vcpu time ran (ns) - ("cpumap", libxl_bitmap), # current cpu's affinities + ("cpumap", libxl_bitmap), # current hard cpu affinity + ("cpumap_soft", libxl_bitmap), # current soft cpu affinity ], dir=DIR_OUT) libxl_physinfo = Struct("physinfo", [
tools/libxl/libxl_utils.h +13 -0
@@ -98,6 +98,19 @@ static inline int libxl_bitmap_cpu_valid(libxl_bitmap *bitmap, int bit)#define libxl_for_each_set_bit (v, m) for (v = 0; v < (m).size * 8; v++) \ if (libxl_bitmap_test(&(m), v)) +static inline int libxl_bitmap_equal (const libxl_bitmap *ba, +const libxl_bitmap *bb) +{ +int i; + +libxl_for_each_bit (i, *ba) { +if (libxl_bitmap_test (ba, i) !=libxl_bitmap_test (bb, i)) +return 0 ; + } + +return 1 ; +} +static inline int libxl_cpu_bitmap_alloc (libxl_ctx *ctx,libxl_bitmap *cpumap,int max_cpus) {
_______________________________________________
Xen-devel mailing list
[email protected]
http://lists.xen.org/xen-devel
Re: [PATCH v2 01/16] xl: match output of vcpu-list with pinning syntax
In reply to Dario FaggioliQuoted 6 lines · Dario Faggioli
in fact, pinning to all the pcpus happens by specifying "all" (either on the command line or in the config file), while `xl vcpu-list' report it as "any cpu".
Change this into something more consistent, by using "all" everywhere.
While I can see a certain logic to having the specification and the output match, there is a part of me that really prefers the "any" syntax in the output.
This is just expressing a preference, not a nack; if others think it would be better to be consistent, I won't argue the case.
-George
Re: [PATCH v2 02/16] xl: allow for node-wise specification of vcpu pinning
In reply to Dario FaggioliQuoted 10 lines · Dario Faggioli
Making it possible to use something like the following:
* "nodes:0-3": all pCPUs of nodes 0,1,2,3;
* "nodes:0-3,^node:2": all pCPUS of nodes 0,1,3;
* "1,nodes:1-2,^6": pCPU 1 plus all pCPUs of nodes 1,2but not pCPU 6;* ...
In both domain config file and `xl vcpu-pin'.
Signed-off-by Dario Faggioli <[email protected]>
Quoted 261 lines
---
Changes from v1 (of this very series):
* actually checking for both "nodes:" and "node:" as perthe doc says; * using strcmp() (rather than strncmp()) when matching "all", to avoid returning success on any longer string that just begins with "all"; * fixing the handling (well, the rejection, actually) of "^all" and "^nodes:all";* make some string pointers const.
Changes from v2 (of original series):
* turned a 'return' into 'goto out', consistently with themost of exit patterns; * harmonized error handling: now parse_range() return a libxl error code, as requested during review; * dealing with "all" moved inside update_cpumap_range(). It's tricky to move it in parse_range() (as requested during review), since we need the cpumap being modified handy when dealing with it. However, having it in update_cpumap_range() simplifies the code just as much as that; * explicitly checking for junk after a valid value or range in parse_range(), as requested during review;* xl exits on parsing failing, so no need to reset the cpumap to something sensible in
vcpupin_parse(), as suggested during review;Changes from v1 (of original series):
* code rearranged in order to look more simple to followand understand, as requested during review; * improved docs in xl.cfg.pod.5, as requested during review; * strtoul() now returns into unsigned long, and the case where it returns ULONG_MAX is now taken into account, as requested during review; * stuff like "all,^7" now works, as requested during review. Specifying just "^7" does not work either before or after this change * killed some magic (i.e., `ptr += 5 + (ptr[4] == 's'`) by introducing STR_SKIP_PREFIX() macro, as requested during review. --- docs/man/xl.cfg.pod.5 | 20 ++++++ tools/libxl/xl_cmdimpl.c | 153 +++++++++++++++++++++++++++++++++------------- 2 files changed, 128 insertions(+), 45 deletions(-)docs/man/xl.cfg.pod.5 +19
-1@@ -115,7 +115,25 @@ To allow all the vcpus of the guest to run on all the cpus on the host. =item "0-3,5,^1" -To allow all the vcpus of the guest to run on cpus 0,2,3,5. +To allow all the vcpus of the guest to run on cpus 0,2,3,5. Combining +this with "all" is possible, meaning "all,^7" results in all the vcpus +of the guest running on all the cpus on the host except cpu 7. + +=item "nodes:0-3,node:^2" + +To allow all the vcpus of the guest to run on the cpus from NUMA nodes +0,1,3 of the host. So, if cpus 0-3 belongs to node 0, cpus 4-7 belongs +to node 1 and cpus 8-11 to node 3, the above would mean all the vcpus +of the guest will run on cpus 0-3,8-11. + +Combining this notation with the one above is possible. For instance, +"1,node:2,^6", means all the vcpus of the guest will run on cpu 1 and +on all the cpus of NUMA node 2, but not on cpu 6. Following the same +example as above, that would be cpus 1,4,5,7. + +Combining this with "all" is also possible, meaning "all,^nodes:1" +results in all the vcpus of the guest running on all the cpus on the +host, except for the cpus belonging to the host NUMA node 1. =item ["2", "3"] (or [2, 3])tools/libxl/xl_cmdimpl.c +109
-44@@ -59,6 +59,11 @@ } \ }) +#define STR_HAS_PREFIX ( a, b ) \ + ( strncmp(a, b, strlen(b)) == 0 ) +#define STR_SKIP_PREFIX ( a, b ) \ + ( STR_HAS_PREFIX(a, b) ? ((a) += strlen(b), 1) : 0 ) +int logfile =2 ;@@ -513,61 +518,121 @@ static void split_string_into_string_list(const char *str,free (s); } -static int vcpupin_parse (char *cpu,libxl_bitmap *cpumap) +static int parse_range (const char *str,unsigned long *a,unsigned long *b) { -libxl_bitmap exclude_cpumap; -uint32_t cpuida, cpuidb; -char *endptr, *toka, *tokb, *saveptr =NULL ; -int i, rc =0 , rmcpu; +const char *nstr; +char *endptr; -if (!strcmp (cpu,"all" )) { -libxl_bitmap_set_any (cpumap); -return 0 ; + *a = *b =strtoul (str, &endptr,10 ); +if (endptr == str || *a ==ULONG_MAX ) +return ERROR_INVAL ; + +if (*endptr =='-' ) { + nstr = endptr +1 ; + + *b =strtoul (nstr, &endptr,10 ); +if (endptr == nstr || *b ==ULONG_MAX || *b < *a) +return ERROR_INVAL ; + } + +/* Valid value or range so far, but we also don't want junk after that */ +if (*endptr !='\0' ) +return ERROR_INVAL ; + +return 0 ; +} + +/* +* Add or removes a specific set of cpus (specified in str, either as +* single cpus or as entire NUMA nodes) to/from cpumap. +*/ +static int update_cpumap_range (const char *str,libxl_bitmap *cpumap) +{ +unsigned long ida, idb; +libxl_bitmap node_cpumap; +bool is_not = false, is_nodes = false; +int rc =0 ; + +libxl_bitmap_init (&node_cpumap); + + rc =libxl_node_bitmap_alloc (ctx, &node_cpumap,0 ); +if (rc) { +fprintf (stderr,"libxl_node_bitmap_alloc failed.\n" ); + goto out; } -if (libxl_cpu_bitmap_alloc (ctx, &exclude_cpumap,0 )) { -fprintf (stderr,"Error: Failed to allocate cpumap.\n" ); -return ENOMEM ; +/* Are we adding or removing cpus/nodes? */ +if (STR_SKIP_PREFIX (str,"^" )) { + is_not = true; } -for (toka =strtok_r (cpu,"," , &saveptr); toka; - toka =strtok_r (NULL ,"," , &saveptr)) { - rmcpu =0 ; -if (*toka =='^' ) { -/* This (These) Cpu(s) will be removed from the map */ - toka++; - rmcpu =1 ; - } -/* Extract a valid (range of) cpu(s) */ - cpuida = cpuidb =strtoul (toka, &endptr,10 ); -if (endptr == toka) { -fprintf (stderr,"Error: Invalid argument.\n" ); - rc =EINVAL ; - goto vcpp_out; - } -if (*endptr =='-' ) { - tokb = endptr +1 ; - cpuidb =strtoul (tokb, &endptr,10 ); -if (endptr == tokb || cpuida > cpuidb) { -fprintf (stderr,"Error: Invalid argument.\n" ); - rc =EINVAL ; - goto vcpp_out; +/* Are we dealing with cpus or full nodes? */ +if (STR_SKIP_PREFIX (str,"node:" ) ||STR_SKIP_PREFIX (str,"nodes:" )) { + is_nodes = true; + } + +if (strcmp (str,"all" ) ==0 ) { +/* We do not accept "^all" or "^nodes:all" */ +if (is_not) { +fprintf (stderr,"Can't combine \"^\" and \"all\".\n" ); + rc =ERROR_INVAL ; + }else +libxl_bitmap_set_any (cpumap); + goto out; + } + + rc =parse_range (str, &ida, &idb); +if (rc) { +fprintf (stderr,"Invalid pcpu range: %s.\n" , str); + goto out; + } + +/* Add or remove the specified cpus in the range */ +while (ida <= idb) { +if (is_nodes) { +/* Add/Remove all the cpus of a NUMA node */ +int i; + + rc =libxl_node_to_cpumap (ctx, ida, &node_cpumap); +if (rc) { +fprintf (stderr,"libxl_node_to_cpumap failed.\n" ); + goto out; } + +/* Add/Remove all the cpus in the node cpumap */ +libxl_for_each_set_bit (i, node_cpumap) { + is_not ?libxl_bitmap_reset (cpumap, i) : +libxl_bitmap_set (cpumap, i); + } + }else { +/* Add/Remove this cpu */ + is_not ?libxl_bitmap_reset (cpumap, ida) : +libxl_bitmap_set (cpumap, ida); } -while (cpuida <= cpuidb) { - rmcpu ==0 ?libxl_bitmap_set (cpumap, cpuida) : -libxl_bitmap_set (&exclude_cpumap, cpuida); - cpuida++; - } + ida++; } -/* Clear all the cpus from the removal list */ -libxl_for_each_set_bit (i, exclude_cpumap) { -libxl_bitmap_reset (cpumap, i); - } + out: +libxl_bitmap_dispose (&node_cpumap); +return rc; +} -vcpp_out: -libxl_bitmap_dispose (&exclude_cpumap); +/* +* Takes a string representing a set of cpus (specified either as +* single cpus or as eintire NUMA nodes) and turns it into the +* corresponding libxl_bitmap (in cpumap). +*/ +static int vcpupin_parse (char *cpu,libxl_bitmap *cpumap) +{ +char *ptr, *saveptr =NULL ; +int rc =0 ; + +for (ptr =strtok_r (cpu,"," , &saveptr); ptr; + ptr =strtok_r (NULL ,"," , &saveptr)) { + rc =update_cpumap_range (ptr, cpumap); +if (rc) +break ; + }return rc; }
Re: [PATCH v2 05/16] xen: fix leaking of v->cpu_affinity_saved
In reply to Dario FaggioliQuoted 3 lines · Dario Faggioli
on domain destruction.
Signed-off-by Dario Faggioli <[email protected]>
Ugh, nasty. Good catch.
Looks like this was introduced in 41e71c26 back in April, so this fix should be backported to 4.3.
Re: [PATCH v2 01/16] xl: match output of vcpu-list with pinning syntax
In reply to George DunlapQuoted 5 lines · George Dunlap
Quoted 4 lines · Dario Faggioli
in fact, pinning to all the pcpus happens by specifying "all" (either on the command line or in the config file), while `xl vcpu-list' report it as "any cpu".
Quoted 7 lines
Quoted 2 lines
Change this into something more consistent, by using "all" everywhere.
While I can see a certain logic to having the specification and the output match, there is a part of me that really prefers the "any" syntax in the output.
I see.
Quoted 3 lines
This is just expressing a preference, not a nack; if others think it would be better to be consistent, I won't argue the case.
I'm fine with both. As I tried (and probably failed) to say here:
"
Changes since v1:
* this patch was not there in v1. It is now as using the same syntax for both input and output was requested during review.
"
It was a request (from Ian Jackson) during review of v1:
Quoted 7 lines
-George
_______________________________________________
Xen-devel mailing list
[email protected]
http://lists.xen.org/xen-devel
Signature · 4 lines
<<This happens because I choose it to happen!>> (Raistlin Majere) ----------------------------------------------------------------- Dario Faggioli, Ph.D, http://about.me/dario.faggioli Senior Software Engineer, Citrix Systems R&D Ltd., Cambridge (UK)
_______________________________________________
Xen-devel mailing list
[email protected]
http://lists.xen.org/xen-devel
Re: [PATCH v2 01/16] xl: match output of vcpu-list with pinning syntax
In reply to George DunlapQuoted 5 lines · George Dunlap
Quoted 4 lines · Dario Faggioli
in fact, pinning to all the pcpus happens by specifying "all" (either on the command line or in the config file), while `xl vcpu-list' report it as "any cpu".
Quoted 7 lines
Quoted 2 lines
Change this into something more consistent, by using "all" everywhere.
While I can see a certain logic to having the specification and the output match, there is a part of me that really prefers the "any" syntax in the output.
I see.
Quoted 3 lines
This is just expressing a preference, not a nack; if others think it would be better to be consistent, I won't argue the case.
I'm fine with both. As I tried (and probably failed) to say here:
"
Changes since v1:
* this patch was not there in v1. It is now as using the same syntax for both input and output was requested during review.
"
It was a request (from Ian Jackson) during last round:
http://bugs.xenproject.org/xen/mid/%[email protected]%3E
Thanks and Regards,
Dario
Signature · 4 lines
<<This happens because I choose it to happen!>> (Raistlin Majere) ----------------------------------------------------------------- Dario Faggioli, Ph.D, http://about.me/dario.faggioli Senior Software Engineer, Citrix Systems R&D Ltd., Cambridge (UK)
_______________________________________________
Xen-devel mailing list
[email protected]
http://lists.xen.org/xen-devel
George Dunlap
Acked-byRe: [PATCH v2 01/16] xl: match output of vcpu-list with pinning syntax
In reply to Dario FaggioliQuoted 19 lines · Dario Faggioli
Quoted 11 lines · George Dunlap
Quoted 6 lines · Dario Faggioli
in fact, pinning to all the pcpus happens by specifying "all" (either on the command line or in the config file), while `xl vcpu-list' report it as "any cpu".
Change this into something more consistent, by using "all" everywhere.
While I can see a certain logic to having the specification and the output match, there is a part of me that really prefers the "any" syntax in the output.
I see.
Quoted 3 lines
This is just expressing a preference, not a nack; if others think it would be better to be consistent, I won't argue the case.
I'm fine with both. As I tried (and probably failed) to say here:
Quoted 8 lines
"
Changes since v1:
* this patch was not there in v1. It is now as using the same syntax for both input and output was requested during review.
"It was a request (from Ian Jackson) during review of v1:
Oh, right -- sorry, missed that comment (and the discussion w/ IanJ). In that case,
-George
Re: [PATCH v2 05/16] xen: fix leaking of v->cpu_affinity_saved
In reply to George DunlapQuoted 7 lines · George Dunlap
Quoted 3 lines · Dario Faggioli
on domain destruction.
Signed-off-by Dario Faggioli <[email protected]>Ugh, nasty. Good catch.
:-)
Quoted 3 lines
Looks like this was introduced in
41e71c26back in April, so this fix should be backported to 4.3.
Oh, right... I really forgot to try figuring out what introduced this and what that was. I'll try to remember at least to file a proper backport request, as soon as this hits the tree.
Thanks and regards,
Dario
Signature · 4 lines
<<This happens because I choose it to happen!>> (Raistlin Majere) ----------------------------------------------------------------- Dario Faggioli, Ph.D, http://about.me/dario.faggioli Senior Software Engineer, Citrix Systems R&D Ltd., Cambridge (UK)
_______________________________________________
Xen-devel mailing list
[email protected]
http://lists.xen.org/xen-devel
Re: [PATCH v2 01/16] xl: match output of vcpu-list with pinning syntax
In reply to George DunlapGeorge Dunlap writes ("Re: [PATCH v2 01/16] xl: match output of vcpu-list with pinning syntax"):
Quoted 11 lines · Dario Faggioli
Quoted 6 lines
in fact, pinning to all the pcpus happens by specifying "all" (either on the command line or in the config file), while `xl vcpu-list' report it as "any cpu".
Change this into something more consistent, by using "all" everywhere.
While I can see a certain logic to having the specification and the output match, there is a part of me that really prefers the "any" syntax in the output.
I definitely think they should be consistent. I don't mind whether it's "any" or "all".
(I see Dario has explained that he did this at my request.)
Thanks,
Ian.
Re: [PATCH v2 07/16] xen: sched: rename v->cpu_affinity into v->cpu_hard_affinity
In reply to Dario FaggioliQuoted 10 lines · Dario Faggioli
in order to distinguish it from the
cpu_soft_affinityintroduced by the previous commit ("xen: sched: make space forcpu_soft_affinity"). This patch does not imply any functional change, it is basically the result of something like the following:s/cpu_affinity/cpu_hard_affinity/g
s/cpu_affinity_tmp/cpu_hard_affinity_tmp/g
s/cpu_affinity_saved/cpu_hard_affinity_saved/gSigned-off-by Dario Faggioli <[email protected]>
Quoted 319 lines
--- xen/arch/x86/traps.c | 11 ++++++----- xen/common/domain.c | 22 +++++++++++----------- xen/common/domctl.c | 2 +- xen/common/keyhandler.c | 2 +- xen/common/sched_credit.c | 12 ++++++------ xen/common/sched_sedf.c | 2 +- xen/common/schedule.c | 21 +++++++++++---------- xen/common/wait.c | 4 ++-- xen/include/xen/sched.h | 8 ++++---- 9 files changed, 43 insertions(+), 41 deletions(-)xen/arch/x86/traps.c +6
-5@@ -3083,7 +3083,8 @@ static void nmi_mce_softirq(void)/* Set the tmp value unconditionally, so that * the check in the iret hypercall works. */ -cpumask_copy (st->vcpu->cpu_affinity_tmp, st->vcpu->cpu_affinity); +cpumask_copy (st->vcpu->cpu_hard_affinity_tmp, + st->vcpu->cpu_hard_affinity);if ((cpu != st->processor) || (st->processor != st->vcpu->processor))@@ -3118,11 +3119,11 @@ void async_exception_cleanup(struct vcpu *curr)return ;/* Restore affinity. */ -if ( !cpumask_empty (curr->cpu_affinity_tmp) && - !cpumask_equal (curr->cpu_affinity_tmp, curr->cpu_affinity) ) +if ( !cpumask_empty (curr->cpu_hard_affinity_tmp) && + !cpumask_equal (curr->cpu_hard_affinity_tmp, curr->cpu_hard_affinity) ) { -vcpu_set_affinity (curr, curr->cpu_affinity_tmp); -cpumask_clear (curr->cpu_affinity_tmp); +vcpu_set_affinity (curr, curr->cpu_hard_affinity_tmp); +cpumask_clear (curr->cpu_hard_affinity_tmp); }if ( !(curr->async_exception_mask & (curr->async_exception_mask -1 )) )xen/common/domain.c +11
-11@@ -125,9 +125,9 @@ struct vcpu *alloc_vcpu(tasklet_init (&v->continue_hypercall_tasklet,NULL ,0 ); -if ( !zalloc_cpumask_var (&v->cpu_affinity) || - !zalloc_cpumask_var (&v->cpu_affinity_tmp) || - !zalloc_cpumask_var (&v->cpu_affinity_saved) || +if ( !zalloc_cpumask_var (&v->cpu_hard_affinity) || + !zalloc_cpumask_var (&v->cpu_hard_affinity_tmp) || + !zalloc_cpumask_var (&v->cpu_hard_affinity_saved) || !zalloc_cpumask_var (&v->cpu_soft_affinity) || !zalloc_cpumask_var (&v->vcpu_dirty_cpumask) ) goto fail_free;@@ -157,9 +157,9 @@ struct vcpu *alloc_vcpu( fail_wq:destroy_waitqueue_vcpu (v); fail_free: -free_cpumask_var (v->cpu_affinity); -free_cpumask_var (v->cpu_affinity_tmp); -free_cpumask_var (v->cpu_affinity_saved); +free_cpumask_var (v->cpu_hard_affinity); +free_cpumask_var (v->cpu_hard_affinity_tmp); +free_cpumask_var (v->cpu_hard_affinity_saved);free_cpumask_var (v->cpu_soft_affinity);free_cpumask_var (v->vcpu_dirty_cpumask);free_vcpu_struct (v);@@ -373,7 +373,7 @@ void domain_update_node_affinity(struct domain *d) for_each_vcpu ( d, v ) { -cpumask_and (online_affinity, v->cpu_affinity, online); +cpumask_and (online_affinity, v->cpu_hard_affinity, online);cpumask_or (cpumask, cpumask, online_affinity); }@@ -736,9 +736,9 @@ static void complete_domain_destroy(struct rcu_head *head)for ( i = d->max_vcpus -1 ; i >=0 ; i-- )if ( (v = d->vcpu[i]) !=NULL ) { -free_cpumask_var (v->cpu_affinity); -free_cpumask_var (v->cpu_affinity_tmp); -free_cpumask_var (v->cpu_affinity_saved); +free_cpumask_var (v->cpu_hard_affinity); +free_cpumask_var (v->cpu_hard_affinity_tmp); +free_cpumask_var (v->cpu_hard_affinity_saved);free_cpumask_var (v->cpu_soft_affinity);free_cpumask_var (v->vcpu_dirty_cpumask);free_vcpu_struct (v);@@ -878,7 +878,7 @@ int vcpu_reset(struct vcpu *v) v->async_exception_mask =0 ;memset (v->async_exception_state,0 ,sizeof (v->async_exception_state));#endif -cpumask_clear (v->cpu_affinity_tmp); +cpumask_clear (v->cpu_hard_affinity_tmp);clear_bit (_VPF_blocked, &v->pause_flags);clear_bit (_VPF_in_reset, &v->pause_flags);xen/common/domctl.c +1
-1@@ -629,7 +629,7 @@ long do_domctl(XEN_GUEST_HANDLE_PARAM(xen_domctl_t) u_domctl)else { ret =cpumask_to_xenctl_bitmap ( - &op->u.vcpuaffinity.cpumap, v->cpu_affinity); + &op->u.vcpuaffinity.cpumap, v->cpu_hard_affinity); } }break ;xen/common/keyhandler.c +1
-1@@ -296,7 +296,7 @@ static void dump_domains(unsigned char key) !vcpu_event_delivery_is_enabled(v));cpuset_print (tmpstr,sizeof (tmpstr), v->vcpu_dirty_cpumask);printk ("dirty_cpus=%s " , tmpstr); -cpuset_print (tmpstr,sizeof (tmpstr), v->cpu_affinity); +cpuset_print (tmpstr,sizeof (tmpstr), v->cpu_hard_affinity);printk ("cpu_affinity=%s\n" , tmpstr);cpuset_print (tmpstr,sizeof (tmpstr), v->cpu_soft_affinity);printk ("cpu_soft_affinity=%s\n" , tmpstr);xen/common/sched_credit.c +6
-6@@ -332,13 +332,13 @@ csched_balance_cpumask(const struct vcpu *vc, int step, cpumask_t *mask)if ( step ==CSCHED_BALANCE_NODE_AFFINITY ) {cpumask_and (mask,CSCHED_DOM (vc->domain)->node_affinity_cpumask, - vc->cpu_affinity); + vc->cpu_hard_affinity);if (unlikely (cpumask_empty (mask)) ) -cpumask_copy (mask, vc->cpu_affinity); +cpumask_copy (mask, vc->cpu_hard_affinity); }else /* step == CSCHED_BALANCE_CPU_AFFINITY */ -cpumask_copy (mask, vc->cpu_affinity); +cpumask_copy (mask, vc->cpu_hard_affinity); }static void burn_credits (struct csched_vcpu *svc,s_time_t now)@@ -407,7 +407,7 @@ __runq_tickle(unsigned int cpu, struct csched_vcpu *new)if ( balance_step ==CSCHED_BALANCE_NODE_AFFINITY && !__vcpu_has_node_affinity (new->vcpu, - new->vcpu->cpu_affinity) ) + new->vcpu->cpu_hard_affinity) )continue ;/* Are there idlers suitable for new (for this balance step)? */ @@ -642,7 +642,7 @@ _csched_cpu_pick(const struct scheduler *ops, struct vcpu *vc, bool_t commit)/* Store in cpus the mask of online cpus on which the domain can run */ online =cpupool_scheduler_cpumask (vc->domain->cpupool); -cpumask_and (&cpus, vc->cpu_affinity, online); +cpumask_and (&cpus, vc->cpu_hard_affinity, online);for_each_csched_balance_step ( balance_step ) {@@ -1487,7 +1487,7 @@ csched_runq_steal(int peer_cpu, int cpu, int pri, int balance_step)* or counter. */ if ( balance_step ==CSCHED_BALANCE_NODE_AFFINITY - && !__vcpu_has_node_affinity (vc, vc->cpu_affinity) ) + && !__vcpu_has_node_affinity (vc, vc->cpu_hard_affinity) )continue ;csched_balance_cpumask (vc, balance_step, csched_balance_mask);xen/common/sched_sedf.c +1
-1@@ -396,7 +396,7 @@ static int sedf_pick_cpu(const struct scheduler *ops, struct vcpu *v)cpumask_t *online; online =cpupool_scheduler_cpumask (v->domain->cpupool); -cpumask_and (&online_affinity, v->cpu_affinity, online); +cpumask_and (&online_affinity, v->cpu_hard_affinity, online);return cpumask_cycle (v->vcpu_id %cpumask_weight (&online_affinity) -1 , &online_affinity); }xen/common/schedule.c +11
-10@@ -194,9 +194,9 @@ int sched_init_vcpu(struct vcpu *v, unsigned int processor)*/ v->processor = processor;if (is_idle_domain (d) || d->is_pinned ) -cpumask_copy (v->cpu_affinity,cpumask_of (processor)); +cpumask_copy (v->cpu_hard_affinity,cpumask_of (processor));else -cpumask_setall (v->cpu_affinity); +cpumask_setall (v->cpu_hard_affinity);cpumask_setall (v->cpu_soft_affinity);@@ -287,7 +287,7 @@ int sched_move_domain(struct domain *d, struct cpupool *c)migrate_timer (&v->singleshot_timer, new_p);migrate_timer (&v->poll_timer, new_p); -cpumask_setall (v->cpu_affinity); +cpumask_setall (v->cpu_hard_affinity); lock =vcpu_schedule_lock_irq (v); v->processor = new_p;@@ -459,7 +459,7 @@ static void vcpu_migrate(struct vcpu *v)*/ if ( pick_called && (new_lock ==per_cpu (schedule_data, new_cpu).schedule_lock) && -cpumask_test_cpu (new_cpu, v->cpu_affinity) && +cpumask_test_cpu (new_cpu, v->cpu_hard_affinity) &&cpumask_test_cpu (new_cpu, v->domain->cpupool->cpu_valid) )break ;@@ -563,7 +563,7 @@ void restore_vcpu_affinity(struct domain *d) {printk (XENLOG_DEBUG "Restoring affinity for d%dv%d\n" , d->domain_id, v->vcpu_id); -cpumask_copy (v->cpu_affinity, v->cpu_affinity_saved); +cpumask_copy (v->cpu_hard_affinity, v->cpu_hard_affinity_saved); v->affinity_broken =0 ; }@@ -606,20 +606,21 @@ int cpu_disable_scheduler(unsigned int cpu)unsigned long flags;spinlock_t *lock =vcpu_schedule_lock_irqsave (v, &flags); -cpumask_and (&online_affinity, v->cpu_affinity, c->cpu_valid); +cpumask_and (&online_affinity, v->cpu_hard_affinity, c->cpu_valid);if (cpumask_empty (&online_affinity) && -cpumask_test_cpu (cpu, v->cpu_affinity) ) +cpumask_test_cpu (cpu, v->cpu_hard_affinity) ) {printk (XENLOG_DEBUG "Breaking affinity for d%dv%d\n" , d->domain_id, v->vcpu_id);if (system_state == SYS_STATE_suspend) { -cpumask_copy (v->cpu_affinity_saved, v->cpu_affinity); +cpumask_copy (v->cpu_hard_affinity_saved, + v->cpu_hard_affinity); v->affinity_broken =1 ; } -cpumask_setall (v->cpu_affinity); +cpumask_setall (v->cpu_hard_affinity); }if ( v->processor == cpu )@@ -667,7 +668,7 @@ int vcpu_set_affinity(struct vcpu *v, const cpumask_t *affinity) lock =vcpu_schedule_lock_irq (v); -cpumask_copy (v->cpu_affinity, affinity); +cpumask_copy (v->cpu_hard_affinity, affinity);/* Always ask the scheduler to re-evaluate placement * when changing the affinity */ xen/common/wait.c +2
-2@@ -134,7 +134,7 @@ static void __prepare_to_wait(struct waitqueue_vcpu *wqv)/* Save current VCPU affinity; force wakeup on *this* CPU only. */ wqv->wakeup_cpu =smp_processor_id (); -cpumask_copy (&wqv->saved_affinity, curr->cpu_affinity); +cpumask_copy (&wqv->saved_affinity, curr->cpu_hard_affinity);if (vcpu_set_affinity (curr,cpumask_of (wqv->wakeup_cpu)) ) {gdprintk (XENLOG_ERR ,"Unable to set vcpu affinity\n" );@@ -183,7 +183,7 @@ void check_wakeup_from_wait(void) {/* Re-set VCPU affinity and re-enter the scheduler. */ struct vcpu *curr = current; -cpumask_copy (&wqv->saved_affinity, curr->cpu_affinity); +cpumask_copy (&wqv->saved_affinity, curr->cpu_hard_affinity);if (vcpu_set_affinity (curr,cpumask_of (wqv->wakeup_cpu)) ) {gdprintk (XENLOG_ERR ,"Unable to set vcpu affinity\n" );xen/include/xen/sched.h +4
-4@@ -192,11 +192,11 @@ struct vcpuspinlock_t virq_lock;/* Bitmask of CPUs on which this VCPU may run. */ -cpumask_var_t cpu_affinity; +cpumask_var_t cpu_hard_affinity;/* Used to change affinity temporarily. */ -cpumask_var_t cpu_affinity_tmp; +cpumask_var_t cpu_hard_affinity_tmp;/* Used to restore affinity across S3. */ -cpumask_var_t cpu_affinity_saved; +cpumask_var_t cpu_hard_affinity_saved;/* Bitmask of CPUs on which this VCPU prefers to run. */ cpumask_var_t cpu_soft_affinity;@@ -795,7 +795,7 @@ void watchdog_domain_destroy(struct domain *d);#define has_hvm_container_domain (d) ((d)->guest_type != guest_type_pv)#define has_hvm_container_vcpu (v) (has_hvm_container_domain((v)->domain))#define is_pinned_vcpu (v) ((v)->domain->is_pinned || \ - cpumask_weight((v)->cpu_affinity) == 1) + cpumask_weight((v)->cpu_hard_affinity) == 1)#ifdef HAS_PASSTHROUGH #define need_iommu (d) ((d)->need_iommu)#else
Re: [PATCH v2 01/16] xl: match output of vcpu-list with pinning syntax
In reply to Dario FaggioliDario Faggioli writes ("[PATCH v2 01/16] xl: match output of vcpu-list with pinning syntax"):
Quoted 6 lines
in fact, pinning to all the pcpus happens by specifying "all" (either on the command line or in the config file), while `xl vcpu-list' report it as "any cpu".
Change this into something more consistent, by using "all" everywhere.
Re: [PATCH v2 02/16] xl: allow for node-wise specification of vcpu pinning
In reply to Dario FaggioliDario Faggioli writes ("[PATCH v2 02/16] xl: allow for node-wise specification of vcpu pinning"):
Quoted 5 lines
Making it possible to use something like the following:
* "nodes:0-3": all pCPUs of nodes 0,1,2,3;
* "nodes:0-3,^node:2": all pCPUS of nodes 0,1,3;
* "1,nodes:1-2,^6": pCPU 1 plus all pCPUs of nodes 1,2 but not pCPU 6;
...
Quoted 2 lines
* fixing the handling (well, the rejection, actually) of "^all" and "^nodes:all";
If it were me I would have made ^all and ^nodes:all work, rather than rejecting them. That makes it possible for semi-automaticallyy-generated configs to specify
"blah,blah,^all,1-3"
to override the first part.
But that's not really important.
Re: [PATCH v2 05/16] xen: fix leaking of v->cpu_affinity_saved
In reply to Dario FaggioliDario Faggioli writes ("[PATCH v2 05/16] xen: fix leaking of v->cpu_affinity_saved"):
Quoted 3 lines
on domain destruction.
Signed-off-by Dario Faggioli <[email protected]>
Re: [PATCH v2 02/16] xl: allow for node-wise specification of vcpu pinning
In reply to Ian JacksonQuoted 7 lines · Ian Jackson
Dario Faggioli writes ("[PATCH v2 02/16] xl: allow for node-wise specification of vcpu pinning"):
Quoted 5 lines
Making it possible to use something like the following:
* "nodes:0-3": all pCPUs of nodes 0,1,2,3;
* "nodes:0-3,^node:2": all pCPUS of nodes 0,1,3;
* "1,nodes:1-2,^6": pCPU 1 plus all pCPUs of nodes 1,2 but not pCPU 6;...
Quoted 9 lines
Quoted 2 lines
* fixing the handling (well, the rejection, actually) of "^all" and "^nodes:all";
If it were me I would have made ^all and ^nodes:all work, rather than rejecting them. That makes it possible for semi-automaticallyy-generated configs to specify
"blah,blah,^all,1-3"
to override the first part.
Oh, I see... No, sorry, I completely failed at spotting the possibility of enabling such an use-case. Now that you mention it, I see it exists and could be useful, although, probably, not very common.
Quoted 2 lines
But that's not really important.
Yeah, I think so. Anyway, it's nothing that can't be done with a follow-up patch, once this series is in (/me adds it to my my TODO list. :-) )
Thanks and Regards,
Dario
Signature · 4 lines
<<This happens because I choose it to happen!>> (Raistlin Majere) ----------------------------------------------------------------- Dario Faggioli, Ph.D, http://about.me/dario.faggioli Senior Software Engineer, Citrix Systems R&D Ltd., Cambridge (UK)
_______________________________________________
Xen-devel mailing list
[email protected]
http://lists.xen.org/xen-devel
Re: [PATCH v2 09/16] xen: sched: DOMCTL_*vcpuaffinity works with hard and soft affinity
In reply to Dario FaggioliQuoted 71 lines · Dario Faggioli
by adding a flag for the caller to specify which one he cares about.
Add also another cpumap there. This way, in case of
DOMCTL_setvcpuaffinity, Xen can return back to the caller the "effective affinity" of the vcpu. We call the effective affinity the intersection between cpupool's cpus, the new hard affinity and (if asking to set soft affinity) the new soft affinity.The purpose of this is allowing the toolstack to figure out whether or not the requested change in affinity produced sensible results, when combined with the other settings that are already in place.
Signed-off-by Dario Faggioli <[email protected]>--- tools/libxc/xc_domain.c | 4 +++- xen/arch/x86/traps.c | 4 ++-- xen/common/domctl.c | 38 ++++++++++++++++++++++++++++++++++---- xen/common/schedule.c | 35 ++++++++++++++++++++++++----------- xen/common/wait.c | 6 +++--- xen/include/public/domctl.h | 15 +++++++++++++-- xen/include/xen/sched.h | 3 ++- 7 files changed, 81 insertions(+), 24 deletions(-)tools/libxc/xc_domain.c +3
-1@@ -215,7 +215,9 @@ int xc_vcpu_setaffinity(xc_interface *xch, domctl.cmd = XEN_DOMCTL_setvcpuaffinity; domctl.domain = (domid_t )domid; - domctl.u.vcpuaffinity.vcpu = vcpu; + domctl.u.vcpuaffinity.vcpu = vcpu; +/* Soft affinity is there, but not used anywhere for now, so... */ + domctl.u.vcpuaffinity.flags =XEN_VCPUAFFINITY_HARD ;memcpy (local, cpumap, cpusize);xen/arch/x86/traps.c +2
-2@@ -3093,7 +3093,7 @@ static void nmi_mce_softirq(void)* Make sure to wakeup the vcpu on the * specified processor. */ -vcpu_set_affinity (st->vcpu,cpumask_of (st->processor)); +vcpu_set_hard_affinity (st->vcpu,cpumask_of (st->processor));/* Affinity is restored in the iret hypercall. */ }@@ -3122,7 +3122,7 @@ void async_exception_cleanup(struct vcpu *curr)if ( !cpumask_empty (curr->cpu_hard_affinity_tmp) && !cpumask_equal (curr->cpu_hard_affinity_tmp, curr->cpu_hard_affinity) ) { -vcpu_set_affinity (curr, curr->cpu_hard_affinity_tmp); +vcpu_set_hard_affinity (curr, curr->cpu_hard_affinity_tmp);cpumask_clear (curr->cpu_hard_affinity_tmp); }xen/common/domctl.c +1
-0@@ -617,19 +617,49 @@ long do_domctl(XEN_GUEST_HANDLE_PARAM(xen_domctl_t) u_domctl)if ( op->cmd == XEN_DOMCTL_setvcpuaffinity ) {cpumask_var_t new_affinity; +cpumask_t *online; ret = xenctl_bitmap_to_cpumask(
xen/common/domctl.c long do_domctl(XEN_GUEST_HANDLE_PARAM(xen_domctl_t) u_domctl) · line 631
Quoted 10 lines
xen/common/domctl.c +8
-1&new_affinity, &op->u.vcpuaffinity.cpumap); -if ( !ret ) +if ( ret ) +break ; + + ret = -EINVAL ; +if ( op->u.vcpuaffinity.flags &XEN_VCPUAFFINITY_HARD ) + ret =vcpu_set_hard_affinity (v, new_affinity); +else if ( op->u.vcpuaffinity.flags &XEN_VCPUAFFINITY_SOFT ) + ret =vcpu_set_soft_affinity (v, new_affinity);
Wait, why are we making this a bitmask of flags, if we can only set one at a time? Shouldn't it instead be a simple enum?
Or alternately (which is what I was expecting when I saw it was 'flags'), shouldn't it allow you to set both at the same time? (i.e., take away the 'else' here?)
Quoted 24 lines
xen/common/domctl.c +17
-1+ +if ( ret ) { - ret =vcpu_set_affinity (v, new_affinity);free_cpumask_var (new_affinity); +break ; } + +/* +* Report back to the caller what the "effective affinity", that +* is the intersection of cpupool's pcpus, the (new?) hard +* affinity and the (new?) soft-affinity. +*/ + online =cpupool_online_cpumask (v->domain->cpupool); +cpumask_and (new_affinity, online, v->cpu_hard_affinity); +cpumask_and (new_affinity, new_affinity , v->cpu_soft_affinity); + + ret =cpumask_to_xenctl_bitmap ( + &op->u.vcpuaffinity.effective_affinity, new_affinity); + +free_cpumask_var (new_affinity); }else {
xen/common/domctl.c long do_domctl(XEN_GUEST_HANDLE_PARAM(xen_domctl_t) u_domctl) · line 662
Quoted 10 lines
xen/common/domctl.c +8
-2- ret =cpumask_to_xenctl_bitmap ( - &op->u.vcpuaffinity.cpumap, v->cpu_hard_affinity); +if ( op->u.vcpuaffinity.flags &XEN_VCPUAFFINITY_HARD ) + ret =cpumask_to_xenctl_bitmap ( + &op->u.vcpuaffinity.cpumap, v->cpu_hard_affinity); +else if ( op->u.vcpuaffinity.flags &XEN_VCPUAFFINITY_SOFT ) + ret =cpumask_to_xenctl_bitmap ( + &op->u.vcpuaffinity.cpumap, v->cpu_soft_affinity); +else + ret = -EINVAL ;
Asking for both the hard and soft affinities at the same time shouldn't return just the hard affinity. It should either return an error, or do something interesting like return the intersection of the two.
Other than that, I think this looks good.
Quoted 133 lines
xen/common/domctl.c +0
-0} }break ;xen/common/schedule.c +24
-11@@ -653,22 +653,14 @@ void sched_set_node_affinity(struct domain *d, nodemask_t *mask)SCHED_OP (DOM2OP (d), set_node_affinity, d, mask); } -int vcpu_set_affinity (struct vcpu *v,const cpumask_t *affinity) +static int vcpu_set_affinity ( +struct vcpu *v,const cpumask_t *affinity,cpumask_t **which) { -cpumask_t online_affinity; -cpumask_t *online;spinlock_t *lock; -if ( v->domain->is_pinned ) -return -EINVAL ; - online =VCPU2ONLINE (v); -cpumask_and (&online_affinity, affinity, online); -if (cpumask_empty (&online_affinity) ) -return -EINVAL ; - lock =vcpu_schedule_lock_irq (v); -cpumask_copy (v->cpu_hard_affinity, affinity); +cpumask_copy (*which, affinity);/* Always ask the scheduler to re-evaluate placement * when changing the affinity */ @@ -687,6 +679,27 @@ int vcpu_set_affinity(struct vcpu *v, const cpumask_t *affinity)return 0 ; } +int vcpu_set_hard_affinity (struct vcpu *v,const cpumask_t *affinity) +{ +cpumask_t online_affinity; +cpumask_t *online; + +if ( v->domain->is_pinned ) +return -EINVAL ; + + online =VCPU2ONLINE (v); +cpumask_and (&online_affinity, affinity, online); +if (cpumask_empty (&online_affinity) ) +return -EINVAL ; + +return vcpu_set_affinity (v, affinity, &v->cpu_hard_affinity); +} + +int vcpu_set_soft_affinity (struct vcpu *v,const cpumask_t *affinity) +{ +return vcpu_set_affinity (v, affinity, &v->cpu_soft_affinity); +} +/* Block the currently-executing domain until a pertinent event occurs. */ void vcpu_block (void ) {xen/common/wait.c +3
-3@@ -135,7 +135,7 @@ static void __prepare_to_wait(struct waitqueue_vcpu *wqv)/* Save current VCPU affinity; force wakeup on *this* CPU only. */ wqv->wakeup_cpu =smp_processor_id ();cpumask_copy (&wqv->saved_affinity, curr->cpu_hard_affinity); -if (vcpu_set_affinity (curr,cpumask_of (wqv->wakeup_cpu)) ) +if (vcpu_set_hard_affinity (curr,cpumask_of (wqv->wakeup_cpu)) ) {gdprintk (XENLOG_ERR ,"Unable to set vcpu affinity\n" );domain_crash_synchronous ();@@ -166,7 +166,7 @@ static void __prepare_to_wait(struct waitqueue_vcpu *wqv)static void __finish_wait (struct waitqueue_vcpu *wqv) { wqv->esp =NULL ; - (void )vcpu_set_affinity (current, &wqv->saved_affinity); + (void )vcpu_set_hard_affinity (current, &wqv->saved_affinity); }void check_wakeup_from_wait (void )@@ -184,7 +184,7 @@ void check_wakeup_from_wait(void)/* Re-set VCPU affinity and re-enter the scheduler. */ struct vcpu *curr = current;cpumask_copy (&wqv->saved_affinity, curr->cpu_hard_affinity); -if (vcpu_set_affinity (curr,cpumask_of (wqv->wakeup_cpu)) ) +if (vcpu_set_hard_affinity (curr,cpumask_of (wqv->wakeup_cpu)) ) {gdprintk (XENLOG_ERR ,"Unable to set vcpu affinity\n" );domain_crash_synchronous ();xen/include/public/domctl.h +13
-2@@ -300,8 +300,19 @@ DEFINE_XEN_GUEST_HANDLE(xen_domctl_nodeaffinity_t);/* XEN_DOMCTL_setvcpuaffinity */ /* XEN_DOMCTL_getvcpuaffinity */ struct xen_domctl_vcpuaffinity { -uint32_t vcpu;/* IN */ -struct xenctl_bitmap cpumap;/* IN/OUT */ +/* IN variables. */ +uint32_t vcpu; +/* Set/get the hard affinity for vcpu */ +#define _XEN_VCPUAFFINITY_HARD 0 +#define XEN_VCPUAFFINITY_HARD (1U<<_XEN_VCPUAFFINITY_HARD) +/* Set/get the soft affinity for vcpu */ +#define _XEN_VCPUAFFINITY_SOFT 1 +#define XEN_VCPUAFFINITY_SOFT (1U<<_XEN_VCPUAFFINITY_SOFT) +uint32_t flags; +/* IN/OUT variables. */ +struct xenctl_bitmap cpumap; +/* OUT variables. */ +struct xenctl_bitmap effective_affinity; };typedef struct xen_domctl_vcpuaffinity xen_domctl_vcpuaffinity_t ;DEFINE_XEN_GUEST_HANDLE (xen_domctl_vcpuaffinity_t);xen/include/xen/sched.h +2
-1@@ -755,7 +755,8 @@ void scheduler_free(struct scheduler *sched);int schedule_cpu_switch (unsigned int cpu,struct cpupool *c);void vcpu_force_reschedule (struct vcpu *v);int cpu_disable_scheduler (unsigned int cpu); -int vcpu_set_affinity (struct vcpu *v,const cpumask_t *affinity); +int vcpu_set_hard_affinity (struct vcpu *v,const cpumask_t *affinity); +int vcpu_set_soft_affinity (struct vcpu *v,const cpumask_t *affinity);void restore_vcpu_affinity (struct domain *d);void vcpu_runstate_get (struct vcpu *v,struct vcpu_runstate_info *runstate);
Re: [PATCH v2 11/16] libxc: get and set soft and hard affinity
In reply to Dario FaggioliDario Faggioli writes ("[PATCH v2 11/16] libxc: get and set soft and hard affinity"):
Quoted 2 lines
by using the new flag introduced in the parameters of DOMCTL_{get,set}_vcpuaffinity.
...
tools/libxc/xc_domain.c int xc_domain_node_getaffinity(xc_interface *xch, · line 197
Quoted 10 lines
tools/libxc/xc_domain.c +6
-4-int xc_vcpu_setaffinity (xc_interface *xch, -uint32_t domid, -int vcpu, -xc_cpumap_t cpumap) +static int _vcpu_setaffinity (xc_interface *xch, +uint32_t domid, +int vcpu, +xc_cpumap_t cpumap, +uint32_t flags, +xc_cpumap_t ecpumap)
It would be clearer if the ecpumap parameter was clearly marked as an out parameter.
Even better if the in parameter was const-correct. However, defining the xc_cpumap_t to be a uint8_t* has made that difficult. We don't want to introduce a new xc_cpumap_const_t. Perhaps we should have typedef uint8_t xc_cpumap_element.
(NB that identifiers ending *_t are reserved for the C implementation. libxc gets this wrong all the time :-/.)
tools/libxc/xc_domain.c int xc_vcpu_setaffinity(xc_interface *xch, · line 212
Quoted 7 lines
tools/libxc/xc_domain.c +1
-2@@ -206,39 +209,119 @@ int xc_vcpu_setaffinity(xc_interface *xch, goto out; } - local =xc_hypercall_buffer_alloc (xch, local, cpusize); -if ( local ==NULL ) + cpumap_local =xc_hypercall_buffer_alloc (xch, cpumap_local, cpusize);
Is xc_hypercall_buffer_free idempotent, and is there a way to init a hypercall buffer to an unallocated state ? If so this function could be a lot simpler, and it could in particular more clearly not leak anything, by using the "goto out" cleanup style.
Quoted 10 lines
tools/libxc/xenctrl.h +10
-0+/** +* This functions specify the scheduling affinity for a vcpu. Soft +* affinity is on what pcpus a vcpu prefers to run. Hard affinity is +* on what pcpus a vcpu is allowed to run. When set independently (by +* the respective _soft and _hard calls) the effective affinity is +* also returned. What we call the effective affinity it the intersection +* of soft affinity, hard affinity and the set of the cpus of the cpupool +* the domain belongs to. It's basically what the Xen scheduler will +* actually use. Returning it to the caller allows him to check if that +* matches with, or at least is good enough for, his purposes.
tools/libxc/xenctrl.h int xc_domain_node_getaffinity(xc_interface *xch, · line 601
Quoted 10 lines
tools/libxc/xenctrl.h +10
-0+* +* A xc_vcpu_setaffinity() call is provided, mainly for backward +* compatibility reasons, and what it does is setting both hard and +* soft affinity for the vcpu. +* +* @param xch a handle to an open hypervisor interface. +* @param domid the id of the domain to which the vcpu belongs +* @param vcpu the vcpu id wihin the domain +* @param cpumap the (hard, soft, both) new affinity map one wants to set +* @param ecpumap the effective affinity for the vcpu
Either the doc comment, or the parameter name, should make it clear that ecpumap is an out parameter.
Ian.
Re: [PATCH v2 06/16] xen: sched: make space for cpu_soft_affinity
In reply to Dario FaggioliQuoted 26 lines · Dario Faggioli
Before this change, each vcpu had its own vcpu-affinity (in
v->cpu_affinity), representing the set of pcpus where the vcpu is allowed to run. Since when NUMA-aware scheduling was introduced the (credit1 only, for now) scheduler also tries as much as it can to run all the vcpus of a domain on one of the nodes that constitutes the domain's node-affinity.The idea here is making the mechanism more general by:
* allowing for this 'preference' for some pcpus/nodes to be expressed on a per-vcpu basis, instead than for the domain as a whole. That is to say, each vcpu should have its own set of preferred pcpus/nodes, instead than it being the very same for all the vcpus of the domain;
* generalizing the idea of 'preferred pcpus' to not only NUMA awareness and support. That is to say, independently from it being or not (mostly) useful on NUMA systems, it should be possible to specify, for each vcpu, a set of pcpus where it prefers to run (in addition, and possibly unrelated to, the set of pcpus where it is allowed to run).We will be calling this set of *preferred* pcpus the vcpu's soft affinity, and this change introduces, allocates, frees and initializes the data structure required to host that in struct vcpu (
cpu_soft_affinity).
Quoted 4 lines
Also, the new field is not used anywhere yet, so no real functional change yet.
Signed-off-by Dario Faggioli <[email protected]>
The breakdown of this in the series doesn't make much sense to me -- I would have folded this one and patch 10 (use soft affinity instead of node affinity) together, and put it in after patch 07 (s/affinity/hard_affinity/g;).
But the code itself is fine, and time is short, so:
Quoted 81 lines
---
Changes from v1:
* this patch does something similar to what, in v1, was being done in "5/12 xen: numa-sched: make space for per-vcpu node-affinity"--- xen/common/domain.c | 3 +++ xen/common/keyhandler.c | 2 ++ xen/common/schedule.c | 2 ++ xen/include/xen/sched.h | 3 +++ 4 files changed, 10 insertions(+)xen/common/domain.c +3
-0@@ -128,6 +128,7 @@ struct vcpu *alloc_vcpu(if ( !zalloc_cpumask_var (&v->cpu_affinity) || !zalloc_cpumask_var (&v->cpu_affinity_tmp) || !zalloc_cpumask_var (&v->cpu_affinity_saved) || + !zalloc_cpumask_var (&v->cpu_soft_affinity) || !zalloc_cpumask_var (&v->vcpu_dirty_cpumask) ) goto fail_free;@@ -159,6 +160,7 @@ struct vcpu *alloc_vcpu(free_cpumask_var (v->cpu_affinity);free_cpumask_var (v->cpu_affinity_tmp);free_cpumask_var (v->cpu_affinity_saved); +free_cpumask_var (v->cpu_soft_affinity);free_cpumask_var (v->vcpu_dirty_cpumask);free_vcpu_struct (v);return NULL ;@@ -737,6 +739,7 @@ static void complete_domain_destroy(struct rcu_head *head)free_cpumask_var (v->cpu_affinity);free_cpumask_var (v->cpu_affinity_tmp);free_cpumask_var (v->cpu_affinity_saved); +free_cpumask_var (v->cpu_soft_affinity);free_cpumask_var (v->vcpu_dirty_cpumask);free_vcpu_struct (v); }xen/common/keyhandler.c +2
-0@@ -298,6 +298,8 @@ static void dump_domains(unsigned char key)printk ("dirty_cpus=%s " , tmpstr);cpuset_print (tmpstr,sizeof (tmpstr), v->cpu_affinity);printk ("cpu_affinity=%s\n" , tmpstr); +cpuset_print (tmpstr,sizeof (tmpstr), v->cpu_soft_affinity); +printk ("cpu_soft_affinity=%s\n" , tmpstr);printk (" pause_count=%d pause_flags=%lx\n" ,atomic_read (&v->pause_count), v->pause_flags);arch_dump_vcpu_info (v);xen/common/schedule.c +2
-0@@ -198,6 +198,8 @@ int sched_init_vcpu(struct vcpu *v, unsigned int processor)else cpumask_setall (v->cpu_affinity); +cpumask_setall (v->cpu_soft_affinity); +/* Initialise the per-vcpu timers. */ init_timer (&v->periodic_timer, vcpu_periodic_timer_fn, v, v->processor);xen/include/xen/sched.h +3
-0@@ -198,6 +198,9 @@ struct vcpu/* Used to restore affinity across S3. */ cpumask_var_t cpu_affinity_saved; +/* Bitmask of CPUs on which this VCPU prefers to run. */ +cpumask_var_t cpu_soft_affinity; +/* Bitmask of CPUs which are holding onto this VCPU's state. */ cpumask_var_t vcpu_dirty_cpumask;
Re: [PATCH v2 12/16] libxl: get and set soft affinity
In reply to Dario FaggioliDario Faggioli writes ("[PATCH v2 12/16] libxl: get and set soft affinity"):
Quoted 8 lines
which basically means making space for a new cpumap in both
vcpu_info(for getting soft affinity) andbuild_info(for setting it), along with providing the get/set functions, and wiring them to the proper xc calls. Interface is as follows:* libxl_{get,set}_vcpuaffinity() deals with hard affinity, as it always has happened;
* libxl_get,set}_vcpuaffinity_soft() deals with soft affinity.
In practice, doesn't this API plus these warnings mean that a toolstack which wants to migrate a domain to a new set of vcpus (or, worse, a new cpupool) will find it difficult to avoid warnings from libxl ?
Because the toolstack will want to change both the hard and soft affinities to new maps, perhaps entirely disjoint from the old ones, but can only do one at a time. So the system will definitely go through "silly" states.
This would be solved with an API that permitted setting both sets of affinities in a single call, even if the underlying libxc and hypercalls are separate, because libxl would do the check only on the overall final state.
So perhaps we should have a singe function that can change the cpupool, hard affinity, and soft affinity, all at once ?
What if the application makes a call to change the cpupool, without touching the affinities ? Should the affinities be reset automatically ?
tools/libxl/Makefile line 10
Quoted 11 lines
tools/libxl/Makefile +1
-0@@ -5,6 +5,7 @@ XEN_ROOT = $(CURDIR)/../.. include $(XEN_ROOT)/tools/Rules.mk +#MAJOR = 4.4 MAJOR = 4.3 MINOR = 0
Commented out ?
tools/libxl/Makefile
Quoted 4 lines
tools/libxl/Makefile +4
-0+ if (libxl_cpu_bitmap_alloc(ctx, &ptr->cpumap_soft, 0)) { + LIBXL__LOG_ERRNO(ctx, LIBXL__LOG_ERROR, "allocating cpumap_soft"); + return NULL; + }
I went and looked at the error handling. Normally in libxl allocators can't fail but this one wants to do a hypercall to look up the max cpus.
But, pulling on the thread, I wonder if it would be better to do this error logging in libxl_cpu_bitmap_alloc rather than all the call sites. (If we did that in principle we should introduce a new #define which allows applications to #ifdef out their own logging for this, but in practice the double-logging isn't likely to be a problem.)
And looking at libxl_cpu_bitmap_alloc, it calls libxl_get_max_cpus and does something very odd with the return value: libxl_get_max_nodes ought to return a positive number or a libxl error code.
So I went to look at libxl_get_max_nodes and it just returns whatever it got from libxc, which is definitely wrong!
Would you mind fixing this as part of this series ?
Thanks,
Ian.
Re: [PATCH v2 13/16] xl: show soft affinity in `xl vcpu-list'
In reply to Dario FaggioliDario Faggioli writes ("[PATCH v2 13/16] xl: show soft affinity in `xl vcpu-list'"):
Quoted 2 lines
if the '-s'/'--soft' option is provided. An example of such output is this:
Do we want to use "-s" for this ? (What does xm do with it?)
Perhaps we should have one more general option that displays cpu affinity and cpupool (ie, basic numa and scheduling info).
Ian.
Re: [PATCH v2 15/16] xl: enable for specifying node-affinity in the config file
In reply to Dario FaggioliDario Faggioli writes ("[PATCH v2 15/16] xl: enable for specifying node-affinity in the config file"):
Quoted 1 line
in a similar way to how it is possible to specify vcpu-affinity.
I think this config file interface is fine, but if the libxl API is going to change the implementaion here will have to too.
So I'll write:
If this patch gets reworked, please take my ack off but mention that I previously acked it :-).
Thanks,
Ian.
Re: [PATCH v2 16/16] libxl: automatic NUMA placement affects soft affinity
In reply to Dario FaggioliDario Faggioli writes ("[PATCH v2 16/16] libxl: automatic NUMA placement affects soft affinity"):
Quoted 6 lines
vCPU soft affinity and NUMA-aware scheduling does not have to be related. However, soft affinity is how NUMA-aware scheduling is actually implemented, and therefore, by default, the results of automatic NUMA placement (at VM creation time) are also used to set the soft affinity of all the vCPUs of the domain.
....
Quoted 6 lines
tools/libxl/libxl_dom.c +0
-0@@ -222,21 +222,39 @@ int libxl__build_pre(libxl__gc *gc, uint32_t domid,* some weird error manifests) the subsequent call to
tools/libxl/libxl_dom.c int libxl__build_pre(libxl__gc *gc, uint32_t domid, · line 232
Quoted 10 lines
tools/libxl/libxl_dom.c +6
-0* libxl_domain_set_nodeaffinity() will do the actual placement, * whatever that turns out to be. +* +* As far as scheduling is concerned, we achieve NUMA-aware scheduling +* by having the results of placement affect the soft affinity of all +* the vcpus of the domain. Of course, we want that iff placement is +* enabled and actually happens, so we only change info->cpumap_soft to +* reflect the placement result if that is the case */ if (libxl_defbool_val (info->numa_placement)) {
But isn't the default for this true ?
tools/libxl/libxl_dom.c int libxl__build_pre(libxl__gc *gc, uint32_t domid, · line 239
Quoted 8 lines
tools/libxl/libxl_dom.c +4
-2-if (!libxl_bitmap_is_full (&info->cpumap)) { +/* We require both hard and soft affinity not to be set */ +if (!libxl_bitmap_is_full (&info->cpumap) || + !libxl_bitmap_is_full (&info->cpumap_soft)) {LOG (ERROR ,"Can run NUMA placement only if no vcpu " -"affinity is specified" ); +"(hard or soft) affinity is specified" );return ERROR_INVAL ;
So the result will be that any attempt to set cpumaps in an otherwise-naive config file will cause errors, rather than just disabling the numa placement ?
Thanks,
Ian.
Re: [PATCH v2 08/16] xen: derive NUMA node affinity from hard and soft CPU affinity
In reply to Dario FaggioliQuoted 48 lines · Dario Faggioli
if a domain's NUMA node-affinity (which is what controls memory allocations) is provided by the user/toolstack, it just is not touched. However, if the user does not say anything, leaving it all to Xen, let's compute it in the following way:
1. cpupool's cpus & hard-affinity & soft-affinity
2. if (1) is empty: cpupool's cpus & hard-affinityThis guarantees memory to be allocated from the narrowest possible set of NUMA nodes, ad makes it relatively easy to set up NUMA-aware scheduling on top of soft affinity.
Signed-off-by Dario Faggioli <[email protected]>--- xen/common/domain.c | 22 +++++++++++++++++++++- 1 file changed, 21 insertions(+), 1 deletion(-)xen/common/domain.c +7
-1@@ -353,7 +353,7 @@ struct domain *domain_create(void domain_update_node_affinity (struct domain *d) { -cpumask_var_t cpumask; +cpumask_var_t cpumask, cpumask_soft;cpumask_var_t online_affinity;const cpumask_t *online;struct vcpu *v;@@ -361,9 +361,15 @@ void domain_update_node_affinity(struct domain *d)if ( !zalloc_cpumask_var (&cpumask) )return ; +if ( !zalloc_cpumask_var (&cpumask_soft) ) + { +free_cpumask_var (cpumask); +return ; + }if ( !alloc_cpumask_var (&online_affinity) ) {free_cpumask_var (cpumask); +free_cpumask_var (cpumask_soft);return ; }
xen/common/domain.c void domain_update_node_affinity(struct domain *d) · line 387
Quoted 9 lines
xen/common/domain.c +4
-0@@ -373,8 +379,12 @@ void domain_update_node_affinity(struct domain *d) for_each_vcpu ( d, v ) { +/* Build up the mask of online pcpus we have hard affinity with */ cpumask_and (online_affinity, v->cpu_hard_affinity, online);cpumask_or (cpumask, cpumask, online_affinity); + +/* As well as the mask of all pcpus we have soft affinity with */ +cpumask_or (cpumask_soft, cpumask_soft, v->cpu_soft_affinity);
Is this really the most efficient way to do this? I would have thought or'ing cpumask and cpumask_soft, and then if it's not empty, then use it; maybe use a pointer so you don't have to copy one into the other one?
Also, all of the above computation happens unconditionally, but is only used if d->auto_node_affinity is true. It seems like it would be better to move this calculation inside the conditional.
Leaving the allocation outside might make sense to reduce the code covered by the lock -- not sure about that; I don't think this is a heavily-contended lock, is it?
Quoted 31 lines
xen/common/domain.c +10
-0}/* @@ -386,6 +396,15 @@ void domain_update_node_affinity(struct domain *d)*/ if ( d->auto_node_affinity ) { +/* +* We're looking for the narower possible set of nodes. So, if +* possible (i.e., if not empty!) let's use the intersection +* between online, hard and soft affinity. If not, just fall back +* to online & hard affinity. +*/ +if (cpumask_intersects (cpumask, cpumask_soft) ) +cpumask_and (cpumask, cpumask, cpumask_soft); +nodes_clear (d->node_affinity);for_each_online_node ( node ) if ( cpumask_intersects(&node_to_cpumask(node), cpumask) )@@ -397,6 +416,7 @@ void domain_update_node_affinity(struct domain *d)spin_unlock (&d->node_affinity_lock);free_cpumask_var (online_affinity); +free_cpumask_var (cpumask_soft);free_cpumask_var(cpumask);}
Re: [PATCH v2 10/16] xen: sched: use soft-affinity instead of domain's node-affinity
In reply to Dario FaggioliQuoted 18 lines · Dario Faggioli
now that we have it, use soft affinity for scheduling, and replace the indirect use of the domain's NUMA node-affinity. This is more general, as soft affinity does not have to be related to NUMA. At the same time it allows to achieve the same results as NUMA-aware scheduling, just by making soft affinity equal to the domain's node affinity, for all the vCPUs (e.g., from the toolstack).
This also means renaming most of the NUMA-aware scheduling related functions, in credit1, to something more generic, hinting toward the concept of soft affinity rather than directly to NUMA awareness.
As a side effects, this simplifies the code quit a bit. In fact, prior to this change, we needed to cache the translation of
d->node_affinity(which is anodemask_t) to acpumask_t, since that is what scheduling decisions require (we used to keep it innode_affinity_cpumask). This, and all the complicated logic required to keep it updated, is not necessary any longer.
Quoted 5 lines
The high level description of NUMA placement and scheduling in docs/misc/xl-numa-placement.markdown is being updated too, to match the new architecture.
Signed-off-by Dario Faggioli <[email protected]>
Just a few things to note below...
xen/common/domain.c void domain_update_node_affinity(struct domain *d) · line 414
Quoted 11 lines
xen/common/domain.c +0
-2@@ -411,8 +411,6 @@ void domain_update_node_affinity(struct domain *d)node_set (node, d->node_affinity); } -sched_set_node_affinity (d, &d->node_affinity); -spin_unlock (&d->node_affinity_lock);
At this point, the only thing inside the spinlock is contingent on d->auto_node_affinity.
Quoted 4 lines
xen/common/sched_credit.c +0
-0
...
xen/common/sched_credit.c __runq_remove(struct csched_vcpu *svc) · line 297
Quoted 13 lines
xen/common/sched_credit.c +3
-7-static inline int __vcpu_has_node_affinity (const struct vcpu *vc, +static inline int __vcpu_has_soft_affinity (const struct vcpu *vc,const cpumask_t *mask) { -const struct domain *d = vc->domain; -const struct csched_dom *sdom =CSCHED_DOM (d); - -if ( d->auto_node_affinity - ||cpumask_full (sdom->node_affinity_cpumask) - || !cpumask_intersects (sdom->node_affinity_cpumask, mask) ) +if (cpumask_full (vc->cpu_soft_affinity) + || !cpumask_intersects (vc->cpu_soft_affinity, mask) )return 0 ;
At this point we've lost a way to make this check potentially much faster (being able to check auto_node_affinity). This isn't a super-hot path but it does happen fairly frequently -- will the "cpumask_full()" check take a significant amount of time on, say, a 4096-core system? If so, we might think about "caching" the results of cpumask_full() at some point.
Re: [PATCH v2 11/16] libxc: get and set soft and hard affinity
In reply to Dario FaggioliQuoted 6 lines · Dario Faggioli
by using the new flag introduced in the parameters of DOMCTL_{get,set}_vcpuaffinity.
This happens in two new xc calls:
xc_vcpu_setaffinity_hard()andxc_vcpu_setaffinity_soft()(an in the corresponding getters, of course).
Personally I think:
* You should be able to set both HARD and SOFT flags, in which case it will set both hard and soft affinities
* You should expose the "flags" to the xc caller, so that the caller can set either one or both.
Quoted 9 lines
The existing xc_vcpu_{set,get}affinity() call is also retained, with the following behavior:
*
xc_vcpu_setaffinity()sets both the hard and soft affinity;
*xc_vcpu_getaffinity()gets the hard affinity.This is mainly for backward compatibility reasons, i.e., trying not to break existing callers/users.
The xc interface isn't stable, right? Couldn't we just change the callers (presumably just xend and libxl)?
-George
Re: [PATCH v2 12/16] libxl: get and set soft affinity
In reply to Ian JacksonQuoted 19 lines · Ian Jackson
Dario Faggioli writes ("[PATCH v2 12/16] libxl: get and set soft affinity"):
Quoted 8 lines
which basically means making space for a new cpumap in both
vcpu_info(for getting soft affinity) andbuild_info(for setting it), along with providing the get/set functions, and wiring them to the proper xc calls. Interface is as follows:* libxl_{get,set}_vcpuaffinity() deals with hard affinity, as it
always has happened;* libxl_get,set}_vcpuaffinity_soft() deals with soft affinity.
In practice, doesn't this API plus these warnings mean that a toolstack which wants to migrate a domain to a new set of vcpus (or, worse, a new cpupool) will find it difficult to avoid warnings from libxl ?
Because the toolstack will want to change both the hard and soft affinities to new maps, perhaps entirely disjoint from the old ones, but can only do one at a time. So the system will definitely go through "silly" states.
Quoted 7 lines
This would be solved with an API that permitted setting both sets of affinities in a single call, even if the underlying libxc and hypercalls are separate, because libxl would do the check only on the overall final state.
So perhaps we should have a singe function that can change the cpupool, hard affinity, and soft affinity, all at once ?
I think this is probably a good idea. Would it make sense to basically have libxl_[gs]et_vcpuaffinity2(), which takes a parameter that can specify that the mask is for either hard, soft, or both?
Quoted 4 lines
What if the application makes a call to change the cpupool, without touching the affinities ? Should the affinities be reset automatically ?
I think whatever happens for hard affinities right now should be carried on.
-George
Re: [PATCH v2 16/16] libxl: automatic NUMA placement affects soft affinity
In reply to Dario FaggioliQuoted 12 lines · Dario Faggioli
vCPU soft affinity and NUMA-aware scheduling does not have to be related. However, soft affinity is how NUMA-aware scheduling is actually implemented, and therefore, by default, the results of automatic NUMA placement (at VM creation time) are also used to set the soft affinity of all the vCPUs of the domain.
Of course, this only happens if automatic NUMA placement is enabled and actually takes place (for instance, if the user does not specify any hard and soft affiniy in the xl config file).
Quoted 4 lines
This also takes care of the vice-versa, i.e., don't trigger automatic placement if the config file specifies either an hard (the check for which was already there) or a soft (the check for which is introduced by this commit) affinity.
It looks like with this patch you set *both* hard and soft affinities when doing auto-numa placement. Would it make more sense to change it to setting only the soft affinity, and leaving the hard affinity to "any"?
(My brain is running low, so forgive me if I've mis-read it...)
-George
Quoted 123 lines
Signed-off-by Dario Faggioli <[email protected]>--- docs/man/xl.cfg.pod.5 | 21 +++++++++++---------- docs/misc/xl-numa-placement.markdown | 16 ++++++++++++++-- tools/libxl/libxl_dom.c | 22 ++++++++++++++++++++-- 3 files changed, 45 insertions(+), 14 deletions(-)docs/man/xl.cfg.pod.5 +11
-10@@ -150,16 +150,6 @@ here, and the soft affinity mask, provided via B<cpus\_soft=> (if any), is utilized to compute the domain node-affinity, for driving memory allocations. -If we are on a NUMA machine (i.e., if the host has more than one NUMA -node) and this option is not specified, libxl automatically tries to -place the guest on the least possible number of nodes. That, however, -will not affect vcpu pinning, so the guest will still be able to run on -all the cpus. A heuristic approach is used for choosing the best node (or -set of nodes), with the goals of maximizing performance for the guest -and, at the same time, achieving efficient utilization of host cpus -and memory. See F<docs/misc/xl-numa-placement.markdown> for more -details. - =item B<cpus_soft="CPU-LIST"> Exactly as B<cpus=>, but specifies soft affinity, rather than pinning@@ -178,6 +168,17 @@ the intersection of the soft affinity mask, provided here, and the vcpu pinning, provided via B<cpus=> (if any), is utilized to compute the domain node-affinity, for driving memory allocations. +If this option is not specified (and B<cpus=> is not specified either), +libxl automatically tries to place the guest on the least possible +number of nodes. A heuristic approach is used for choosing the best +node (or set of nodes), with the goal of maximizing performance for +the guest and, at the same time, achieving efficient utilization of +host cpus and memory. In that case, the soft affinity of all the vcpus +of the domain will be set to the pcpus belonging to the NUMA nodes +chosen during placement. + +For more details, see F<docs/misc/xl-numa-placement.markdown>. + =back =head3 CPU Schedulingdocs/misc/xl-numa-placement.markdown +14
-2@@ -126,10 +126,22 @@ or Xen won't be able to guarantee the locality for their memory accesses. That, of course, also mean the vCPUs of the domain will only be able to execute on those same pCPUs. +Starting from 4.4, is is also possible to specify a "cpus\_soft=" option +in the xl config file. This, independently from whether or not "cpus=" is +specified too, affect the NUMA placement in a way very similar to what +is described above. In fact, the hypervisor will build up the node-affinity +of the VM basing right on it or, if both pinning (via "cpus=") and soft +affinity (via "cpus\_soft=") are present, basing on their intersection. + +Besides that, "cpus\_soft=" also means, of course, that the vCPUs of the +domain will prefer to execute on, among the pCPUs where they can run, +those particular pCPUs. + + ### Placing the guest automatically ### -If no "cpus=" option is specified in the config file, libxl tries -to figure out on its own on which node(s) the domain could fit best. +If neither "cpus=" nor "cpus\_soft=" are present in the config file, libxl +tries to figure out on its own on which node(s) the domain could fit best. If it finds one (some), the domain's node affinity get set to there, and both memory allocations and NUMA aware scheduling (for the credit scheduler and starting from Xen 4.3) will comply with it. Starting fromtools/libxl/libxl_dom.c +20
-2@@ -222,21 +222,39 @@ int libxl__build_pre(libxl__gc *gc, uint32_t domid,* some weird error manifests) the subsequent call to * libxl_domain_set_nodeaffinity() will do the actual placement, * whatever that turns out to be. +* +* As far as scheduling is concerned, we achieve NUMA-aware scheduling +* by having the results of placement affect the soft affinity of all +* the vcpus of the domain. Of course, we want that iff placement is +* enabled and actually happens, so we only change info->cpumap_soft to +* reflect the placement result if that is the case */ if (libxl_defbool_val (info->numa_placement)) { -if (!libxl_bitmap_is_full (&info->cpumap)) { +/* We require both hard and soft affinity not to be set */ +if (!libxl_bitmap_is_full (&info->cpumap) || + !libxl_bitmap_is_full (&info->cpumap_soft)) {LOG (ERROR ,"Can run NUMA placement only if no vcpu " -"affinity is specified" ); +"(hard or soft) affinity is specified" );return ERROR_INVAL ; } rc =numa_place_domain (gc, domid, info);if (rc)return rc; + +/* +* We change the soft affinity in domain_build_info here, of course +* after converting the result of placement from nodes to cpus. the +* following call to libxl_set_vcpuaffinity_all_soft() will do the +* actual updating of the domain's vcpus' soft affinity. +*/ +libxl_nodemap_to_cpumap (ctx, &info->nodemap, &info->cpumap_soft); }libxl_domain_set_nodeaffinity (ctx, domid, &info->nodemap);libxl_set_vcpuaffinity_all (ctx, domid, info->max_vcpus, &info->cpumap); +libxl_set_vcpuaffinity_all_soft (ctx, domid, info->max_vcpus, + &info->cpumap_soft);xc_domain_setmaxmem (ctx->xch, domid, info->target_memkb +LIBXL_MAXMEM_CONSTANT ); xs_domid =xs_read (ctx->xsh,XBT_NULL ,"/tool/xenstored/domid" ,NULL );
Re: [PATCH v2 16/16] libxl: automatic NUMA placement affects soft affinity
In reply to Ian JacksonQuoted 12 lines · Ian Jackson
Dario Faggioli writes ("[PATCH v2 16/16] libxl: automatic NUMA placement affects soft affinity"):
Quoted 10 lines
tools/libxl/libxl_dom.c +2
-0@@ -222,21 +222,39 @@ int libxl__build_pre(libxl__gc *gc, uint32_t domid,* some weird error manifests) the subsequent call to * libxl_domain_set_nodeaffinity() will do the actual placement, * whatever that turns out to be. +* +* As far as scheduling is concerned, we achieve NUMA-aware scheduling
Quoted 9 lines
tools/libxl/libxl_dom.c int libxl__build_pre(libxl__gc *gc, uint32_t domid, · line 232
Quoted 6 lines
tools/libxl/libxl_dom.c +4
-0+* by having the results of placement affect the soft affinity of all +* the vcpus of the domain. Of course, we want that iff placement is +* enabled and actually happens, so we only change info->cpumap_soft to +* reflect the placement result if that is the case */ if (libxl_defbool_val (info->numa_placement)) {But isn't the default for this true ?
It is.
Quoted 9 lines
tools/libxl/libxl_dom.c int libxl__build_pre(libxl__gc *gc, uint32_t domid, · line 239
Quoted 8 lines
tools/libxl/libxl_dom.c +4
-2-if (!libxl_bitmap_is_full (&info->cpumap)) { +/* We require both hard and soft affinity not to be set */ +if (!libxl_bitmap_is_full (&info->cpumap) || + !libxl_bitmap_is_full (&info->cpumap_soft)) {LOG (ERROR ,"Can run NUMA placement only if no vcpu " -"affinity is specified" ); +"(hard or soft) affinity is specified" );return ERROR_INVAL ;
Quoted 4 lines
So the result will be that any attempt to set cpumaps in an otherwise-naive config file will cause errors, rather than just disabling the numa placement ?
Nope, because, as it discussed already not more than a couple of days back, what xl does when finding a "cpus=" option (and from this patch on, a "cpus_soft=" option) is:
1. set numa_placement to false
2. set cpumap (or cpumap_soft) as requested.
:-)
So, as far as xl is concerned, this is fine. It is true that every other consumer of libxl needs to do the same (i.e., setting numa_placement to false), or it will hit the error, and that may or may not be obvious. For sure, they'll figure out if the check out xl (which is meant to serve as a 'reference implementation' too, right?), and, it is all documented, at least in docs/misc (I can double check whether it is also stressed enough in libxl.h somewhere, and put it there if not). For one, I probably should improve the comment above the if(), to avoid this sort of confusion to happen again.
Anyway, like it or not, this is the kind of interface we came up with when designing, refining, and checking in that bit:
http://lists.xen.org/archives/html/xen-devel/2012-07/msg01077.html
If we don't like it, I think there is some room for amending, since, despite libxl API being stable, this looks enough as an implementation detail to me.... It's just a matter of agreeing on what we actually want. :-)
Thanks and Regards,
Dario
Signature · 4 lines
<<This happens because I choose it to happen!>> (Raistlin Majere) ----------------------------------------------------------------- Dario Faggioli, Ph.D, http://about.me/dario.faggioli Senior Software Engineer, Citrix Systems R&D Ltd., Cambridge (UK)
_______________________________________________
Xen-devel mailing list
[email protected]
http://lists.xen.org/xen-devel
Re: [PATCH v2 15/16] xl: enable for specifying node-affinity in the config file
In reply to Ian JacksonQuoted 10 lines · Ian Jackson
Dario Faggioli writes ("[PATCH v2 15/16] xl: enable for specifying node-affinity in the config file"):
Quoted 1 line
in a similar way to how it is possible to specify vcpu-affinity.
I think this config file interface is fine, but if the libxl API is going to change the implementaion here will have to too.
So I'll write:
Acked-by Ian Jackson <[email protected]>
Thanks.
Quoted 3 lines
If this patch gets reworked, please take my ack off but mention that I previously acked it :-).
Ok, will do.
Thanks and Regards,
Dario
Signature · 4 lines
<<This happens because I choose it to happen!>> (Raistlin Majere) ----------------------------------------------------------------- Dario Faggioli, Ph.D, http://about.me/dario.faggioli Senior Software Engineer, Citrix Systems R&D Ltd., Cambridge (UK)
_______________________________________________
Xen-devel mailing list
[email protected]
http://lists.xen.org/xen-devel
Re: [PATCH v2 06/16] xen: sched: make space for cpu_soft_affinity
In reply to George DunlapQuoted 5 lines · George Dunlap
The breakdown of this in the series doesn't make much sense to me -- I would have folded this one and patch 10 (use soft affinity instead of node affinity) together, and put it in after patch 07 (s/affinity/hard_affinity/g;).
I see. Well, looks like I'm respinning it. While at it, I'll see if I can put it that way.
If I do (and provided that does not imply any code change in those patches), can I retain your Reviewed-by tag in the result?
Quoted 4 lines
But the code itself is fine, and time is short, so:
Reviewed-by George Dunlap <[email protected]>
Thanks and Regards,
Dario
Signature · 4 lines
<<This happens because I choose it to happen!>> (Raistlin Majere) ----------------------------------------------------------------- Dario Faggioli, Ph.D, http://about.me/dario.faggioli Senior Software Engineer, Citrix Systems R&D Ltd., Cambridge (UK)
_______________________________________________
Xen-devel mailing list
[email protected]
http://lists.xen.org/xen-devel
Re: [PATCH v2 11/16] libxc: get and set soft and hard affinity
In reply to Ian JacksonQuoted 6 lines · Ian Jackson
Dario Faggioli writes ("[PATCH v2 11/16] libxc: get and set soft and hard affinity"):
Quoted 4 lines
tools/libxc/xc_domain.c +0
-0@@ -206,39 +209,119 @@ int xc_vcpu_setaffinity(xc_interface *xch, goto out; }
Quoted 9 lines
tools/libxc/xc_domain.c int xc_vcpu_setaffinity(xc_interface *xch, · line 212
Quoted 3 lines
tools/libxc/xc_domain.c +1
-2- local =xc_hypercall_buffer_alloc (xch, local, cpusize); -if ( local ==NULL ) + cpumap_local =xc_hypercall_buffer_alloc (xch, cpumap_local, cpusize);Is
xc_hypercall_buffer_freeidempotent, and is there a way to init a hypercall buffer to an unallocated state ? If so this function could be a lot simpler, and it could in particular more clearly not leak anything, by using the "goto out" cleanup style.
When I put this together, I followed suit from similar cases (actually, this is mostly renaming the function and adding a new parameter). This to say that I don't think it can be done much differently, but yes, I will check.
Quoted 21 lines
Quoted 10 lines
tools/libxc/xenctrl.h +10
-0+/** +* This functions specify the scheduling affinity for a vcpu. Soft +* affinity is on what pcpus a vcpu prefers to run. Hard affinity is +* on what pcpus a vcpu is allowed to run. When set independently (by +* the respective _soft and _hard calls) the effective affinity is +* also returned. What we call the effective affinity it the intersection +* of soft affinity, hard affinity and the set of the cpus of the cpupool +* the domain belongs to. It's basically what the Xen scheduler will +* actually use. Returning it to the caller allows him to check if that +* matches with, or at least is good enough for, his purposes. tools/libxc/xenctrl.h int xc_domain_node_getaffinity(xc_interface *xch, · line 601
Quoted 10 lines
tools/libxc/xenctrl.h +10
-0+* +* A xc_vcpu_setaffinity() call is provided, mainly for backward +* compatibility reasons, and what it does is setting both hard and +* soft affinity for the vcpu. +* +* @param xch a handle to an open hypervisor interface. +* @param domid the id of the domain to which the vcpu belongs +* @param vcpu the vcpu id wihin the domain +* @param cpumap the (hard, soft, both) new affinity map one wants to set +* @param ecpumap the effective affinity for the vcpu
Quoted 3 lines
Either the doc comment, or the parameter name, should make it clear that ecpumap is an out parameter.
Ok, will do.
Thanks and Regards,
Dario
Signature · 4 lines
<<This happens because I choose it to happen!>> (Raistlin Majere) ----------------------------------------------------------------- Dario Faggioli, Ph.D, http://about.me/dario.faggioli Senior Software Engineer, Citrix Systems R&D Ltd., Cambridge (UK)
_______________________________________________
Xen-devel mailing list
[email protected]
http://lists.xen.org/xen-devel
Re: [PATCH v2 09/16] xen: sched: DOMCTL_*vcpuaffinity works with hard and soft affinity
In reply to George DunlapQuoted 7 lines · George Dunlap
Quoted 6 lines · Dario Faggioli
xen/common/domctl.c +1
-0@@ -617,19 +617,49 @@ long do_domctl(XEN_GUEST_HANDLE_PARAM(xen_domctl_t) u_domctl)if ( op->cmd == XEN_DOMCTL_setvcpuaffinity ) {cpumask_var_t new_affinity; +cpumask_t *online;
Quoted 15 lines
xen/common/domctl.c long do_domctl(XEN_GUEST_HANDLE_PARAM(xen_domctl_t) u_domctl) · line 631
Quoted 11 lines
xen/common/domctl.c +8
-1ret =xenctl_bitmap_to_cpumask ( &new_affinity, &op->u.vcpuaffinity.cpumap); -if ( !ret ) +if ( ret ) +break ; + + ret = -EINVAL ; +if ( op->u.vcpuaffinity.flags &XEN_VCPUAFFINITY_HARD ) + ret =vcpu_set_hard_affinity (v, new_affinity); +else if ( op->u.vcpuaffinity.flags &XEN_VCPUAFFINITY_SOFT ) + ret =vcpu_set_soft_affinity (v, new_affinity);Wait, why are we making this a bitmask of flags, if we can only set one at a time? Shouldn't it instead be a simple enum?
Right. I had a split mind about which one way to go between the ones you outline here and, apparently, ended up with the worst possible combination, i.e., something in the middle! :-P
Quoted 4 lines
Or alternately (which is what I was expecting when I saw it was 'flags'), shouldn't it allow you to set both at the same time? (i.e., take away the 'else' here?)
Indeed. I'll take this path.
Quoted 13 lines
xen/common/domctl.c long do_domctl(XEN_GUEST_HANDLE_PARAM(xen_domctl_t) u_domctl) · line 662
Quoted 12 lines
xen/common/domctl.c +8
-2else { - ret =cpumask_to_xenctl_bitmap ( - &op->u.vcpuaffinity.cpumap, v->cpu_hard_affinity); +if ( op->u.vcpuaffinity.flags &XEN_VCPUAFFINITY_HARD ) + ret =cpumask_to_xenctl_bitmap ( + &op->u.vcpuaffinity.cpumap, v->cpu_hard_affinity); +else if ( op->u.vcpuaffinity.flags &XEN_VCPUAFFINITY_SOFT ) + ret =cpumask_to_xenctl_bitmap ( + &op->u.vcpuaffinity.cpumap, v->cpu_soft_affinity); +else + ret = -EINVAL ;
Quoted 4 lines
Asking for both the hard and soft affinities at the same time shouldn't return just the hard affinity. It should either return an error, or do something interesting like return the intersection of the two.
Right again. Given what we said above, I think I'll go for the intersection.
Quoted 2 lines
Other than that, I think this looks good.
Cool, I'll respin with these changes ASAP.
Thanks and Regards,
Dario
Signature · 4 lines
<<This happens because I choose it to happen!>> (Raistlin Majere) ----------------------------------------------------------------- Dario Faggioli, Ph.D, http://about.me/dario.faggioli Senior Software Engineer, Citrix Systems R&D Ltd., Cambridge (UK)
_______________________________________________
Xen-devel mailing list
[email protected]
http://lists.xen.org/xen-devel
Re: [PATCH v2 12/16] libxl: get and set soft affinity
In reply to George DunlapGeorge Dunlap writes ("Re: [PATCH v2 12/16] libxl: get and set soft affinity"):
Quoted 7 lines · Ian Jackson
Quoted 2 lines
So perhaps we should have a singe function that can change the cpupool, hard affinity, and soft affinity, all at once ?
I think this is probably a good idea. Would it make sense to basically have libxl_[gs]et_vcpuaffinity2(), which takes a parameter that can specify that the mask is for either hard, soft, or both?
No, it needs to take at least two separate parameters for the hard and soft masks, because the toolstack might want to set the hard affinity and soft affinity to different values, without having to think carefully about ordering these calls to avoid generating spurious warnings and incompatible arrangements.
I think this probably applies to cpupool changes too. So perhaps the function should look a bit like setresuid or something: it should take an optional new cpupool (is there a sentinel value that can be used to mean "no change"?), an optional new soft mask, and an optional new hard mask.
Quoted 5 lines
Quoted 3 lines
What if the application makes a call to change the cpupool, without touching the affinities ? Should the affinities be reset automatically ?
I think whatever happens for hard affinities right now should be carried on.
Maybe it is a bug that it doesn't do anything. I think it depends how we expect people to use this. If a caller sets the hard affinities and then changes the cpupool, are they supposed to always then set the hard affinities again to a new suitable value ?
Ian.
Re: [PATCH v2 08/16] xen: derive NUMA node affinity from hard and soft CPU affinity
In reply to George DunlapQuoted 4 lines · George Dunlap
Quoted 4 lines
Signed-off-by Dario Faggioli <[email protected]>xen/common/domain.c +0
-0@@ -373,8 +379,12 @@ void domain_update_node_affinity(struct domain *d)
Quoted 13 lines
xen/common/domain.c void domain_update_node_affinity(struct domain *d) · line 387
Quoted 8 lines
xen/common/domain.c +4
-0for_each_vcpu ( d, v ) { +/* Build up the mask of online pcpus we have hard affinity with */ cpumask_and (online_affinity, v->cpu_hard_affinity, online);cpumask_or (cpumask, cpumask, online_affinity); + +/* As well as the mask of all pcpus we have soft affinity with */ +cpumask_or (cpumask_soft, cpumask_soft, v->cpu_soft_affinity);Is this really the most efficient way to do this? I would have thought or'ing cpumask and
cpumask_soft, and then if it's not empty, then use it; maybe use a pointer so you don't have to copy one into the other one?
I'm not sure I fully get what you mean... I cannot afford neglecting online_affinity, independently from how cpumask and cpumask_soft look like, because that's what's needed to account for cpupools. Anyway, I'll think more about it and see if I can make it better.
Quoted 4 lines
Also, all of the above computation happens unconditionally, but is only used if
d->auto_node_affinityis true. It seems like it would be better to move this calculation inside the conditional.
That is true, and I thought it too. I didn't do this here because it seems an unrelated change (the calculation of cpumask|=hard_affinity&online was outside of the if already).
Anyway, again, I agree with this, so I'll either do it here or in a pre-patch.
Quoted 4 lines
Leaving the allocation outside might make sense to reduce the code covered by the lock -- not sure about that; I don't think this is a heavily-contended lock, is it?
Not really, nor this is an hot-path, but I still think it's worth doing what you suggest.
Thanks and Regards,
Dario
Signature · 4 lines
<<This happens because I choose it to happen!>> (Raistlin Majere) ----------------------------------------------------------------- Dario Faggioli, Ph.D, http://about.me/dario.faggioli Senior Software Engineer, Citrix Systems R&D Ltd., Cambridge (UK)
_______________________________________________
Xen-devel mailing list
[email protected]
http://lists.xen.org/xen-devel
Re: [PATCH v2 11/16] libxc: get and set soft and hard affinity
In reply to George DunlapQuoted 4 lines · George Dunlap
Quoted 3 lines · Dario Faggioli
by using the new flag introduced in the parameters of DOMCTL_{get,set}_vcpuaffinity.
Quoted 8 lines
Quoted 3 lines
This happens in two new xc calls:
xc_vcpu_setaffinity_hard()andxc_vcpu_setaffinity_soft()(an in the corresponding getters, of course).Personally I think:
* You should be able to set both HARD and SOFT flags, in which case it will set both hard and soft affinities
Ok. Just to be sure, you mean that, if I specify both flags, both hard and soft affinity will be set to the same cpumap, right?
Quoted 3 lines
* You should expose the "flags" to the xc caller, so that the caller can set either one or both.
Ok.
Quoted 12 lines
Quoted 8 lines
The existing xc_vcpu_{set,get}affinity() call is also retained, with the following behavior:
*
xc_vcpu_setaffinity()sets both the hard and soft affinity;
*xc_vcpu_getaffinity()gets the hard affinity.This is mainly for backward compatibility reasons, i.e., trying not to break existing callers/users.
The xc interface isn't stable, right? Couldn't we just change the callers (presumably just xend and libxl)?
In tree, yes, those ones plus something intools/ocaml/libs/xc/xenctrl_stubs.c.
I'll do it that way.
Dario
Signature · 4 lines
<<This happens because I choose it to happen!>> (Raistlin Majere) ----------------------------------------------------------------- Dario Faggioli, Ph.D, http://about.me/dario.faggioli Senior Software Engineer, Citrix Systems R&D Ltd., Cambridge (UK)
_______________________________________________
Xen-devel mailing list
[email protected]
http://lists.xen.org/xen-devel
Re: [PATCH v2 16/16] libxl: automatic NUMA placement affects soft affinity
In reply to George DunlapQuoted 13 lines · George Dunlap
Quoted 12 lines · Dario Faggioli
vCPU soft affinity and NUMA-aware scheduling does not have to be related. However, soft affinity is how NUMA-aware scheduling is actually implemented, and therefore, by default, the results of automatic NUMA placement (at VM creation time) are also used to set the soft affinity of all the vCPUs of the domain.
Of course, this only happens if automatic NUMA placement is enabled and actually takes place (for instance, if the user does not specify any hard and soft affiniy in the xl config file).
Quoted 9 lines
Quoted 4 lines
This also takes care of the vice-versa, i.e., don't trigger automatic placement if the config file specifies either an hard (the check for which was already there) or a soft (the check for which is introduced by this commit) affinity.
It looks like with this patch you set *both* hard and soft affinities when doing auto-numa placement. Would it make more sense to change it to setting only the soft affinity, and leaving the hard affinity to "any"?
Nope, it indeed sets only soft affinity after automatic placement, hard affinity is left untouched.
Quoted 2 lines
(My brain is running low, so forgive me if I've mis-read it...)
:-)
This is the spot:
Quoted 26 lines
Quoted 26 lines
Signed-off-by Dario Faggioli <[email protected]>tools/libxl/libxl_dom.c +10
-2@@ -222,21 +222,39 @@ int libxl__build_pre(libxl__gc *gc, uint32_t domid,* some weird error manifests) the subsequent call to * libxl_domain_set_nodeaffinity() will do the actual placement, * whatever that turns out to be. +* +* As far as scheduling is concerned, we achieve NUMA-aware scheduling +* by having the results of placement affect the soft affinity of all +* the vcpus of the domain. Of course, we want that iff placement is +* enabled and actually happens, so we only change info->cpumap_soft to +* reflect the placement result if that is the case */ if (libxl_defbool_val (info->numa_placement)) { -if (!libxl_bitmap_is_full (&info->cpumap)) { +/* We require both hard and soft affinity not to be set */ +if (!libxl_bitmap_is_full (&info->cpumap) || + !libxl_bitmap_is_full (&info->cpumap_soft)) {LOG (ERROR ,"Can run NUMA placement only if no vcpu " -"affinity is specified" ); +"(hard or soft) affinity is specified" );return ERROR_INVAL ; }
Quoted 12 lines
tools/libxl/libxl_dom.c int libxl__build_pre(libxl__gc *gc, uint32_t domid, · line 252
Quoted 11 lines
tools/libxl/libxl_dom.c +8
-0rc =numa_place_domain (gc, domid, info);if (rc)return rc; + +/* +* We change the soft affinity in domain_build_info here, of course +* after converting the result of placement from nodes to cpus. the +* following call to libxl_set_vcpuaffinity_all_soft() will do the +* actual updating of the domain's vcpus' soft affinity. +*/ +libxl_nodemap_to_cpumap (ctx, &info->nodemap, &info->cpumap_soft);^
|
Here: -----------------------------------------------------------/
I only copy the result of placement into info->cpumap_soft, without touching info->cpumap, which is "all" (or we won't be at this point) and stays that way.
Quoted 6 lines
tools/libxl/libxl_dom.c int libxl__build_pre(libxl__gc *gc, uint32_t domid, · line 258
Quoted 6 lines
tools/libxl/libxl_dom.c +2
-0}libxl_domain_set_nodeaffinity (ctx, domid, &info->nodemap);libxl_set_vcpuaffinity_all (ctx, domid, info->max_vcpus, &info->cpumap); +libxl_set_vcpuaffinity_all_soft (ctx, domid, info->max_vcpus, + &info->cpumap_soft);
Thanks and Regards,
Dario
Signature · 4 lines
<<This happens because I choose it to happen!>> (Raistlin Majere) ----------------------------------------------------------------- Dario Faggioli, Ph.D, http://about.me/dario.faggioli Senior Software Engineer, Citrix Systems R&D Ltd., Cambridge (UK)
_______________________________________________
Xen-devel mailing list
[email protected]
http://lists.xen.org/xen-devel
George Dunlap
Acked-byRe: [PATCH v2 16/16] libxl: automatic NUMA placement affects soft affinity
Quoted 19 lines · Dario Faggioli
Quoted 18 lines · George Dunlap
Quoted 16 lines · Dario Faggioli
vCPU soft affinity and NUMA-aware scheduling does not have to be related. However, soft affinity is how NUMA-aware scheduling is actually implemented, and therefore, by default, the results of automatic NUMA placement (at VM creation time) are also used to set the soft affinity of all the vCPUs of the domain.
Of course, this only happens if automatic NUMA placement is enabled and actually takes place (for instance, if the user does not specify any hard and soft affiniy in the xl config file).
This also takes care of the vice-versa, i.e., don't trigger automatic placement if the config file specifies either an hard (the check for which was already there) or a soft (the check for which is introduced by this commit) affinity.
Quoted 6 lines
Quoted 4 lines
It looks like with this patch you set *both* hard and soft affinities when doing auto-numa placement. Would it make more sense to change it to setting only the soft affinity, and leaving the hard affinity to "any"?
Nope, it indeed sets only soft affinity after automatic placement, hard affinity is left untouched.
Oh, right -- dur, I had it in my head that you were setting hard affinities before this patch (and since I didn't see them removed I assumed they were still there.) But that's what we did sometime in ancient history, before the scheduler paid attention to the domain numa affinity. :-)
That being the case,
(If I have time I may come by and do a closer review later.)
-George
Re: [PATCH v2 10/16] xen: sched: use soft-affinity instead of domain's node-affinity
In reply to George DunlapQuoted 10 lines · George Dunlap
Quoted 6 lines · Dario Faggioli
[..]
The high level description of NUMA placement and scheduling in docs/misc/xl-numa-placement.markdown is being updated too, to match the new architecture.Signed-off-by Dario Faggioli <[email protected]>Reviewed-by George Dunlap <[email protected]>
Cool, thanks.
Quoted 2 lines
Just a few things to note below...
Ok.
Quoted 5 lines
Quoted 5 lines
xen/common/domain.c +0
-0@@ -411,8 +411,6 @@ void domain_update_node_affinity(struct domain *d)node_set (node, d->node_affinity); }
Quoted 7 lines
xen/common/domain.c void domain_update_node_affinity(struct domain *d) · line 414
Quoted 3 lines
xen/common/domain.c +0
-2-sched_set_node_affinity (d, &d->node_affinity); -spin_unlock (&d->node_affinity_lock);At this point, the only thing inside the spinlock is contingent on
d->auto_node_affinity.
Mmm... Sorry, but I'm not geting what you mean here. :-(
Quoted 9 lines
Quoted 9 lines
xen/common/sched_credit.c +1
-5-static inline int __vcpu_has_node_affinity (const struct vcpu *vc, +static inline int __vcpu_has_soft_affinity (const struct vcpu *vc,const cpumask_t *mask) { -const struct domain *d = vc->domain; -const struct csched_dom *sdom =CSCHED_DOM (d); - - if ( d->auto_node_affinity
Quoted 9 lines
xen/common/sched_credit.c __runq_remove(struct csched_vcpu *svc) · line 297
Quoted 5 lines
xen/common/sched_credit.c +2
-2- || cpumask_full(sdom->node_affinity_cpumask) - || !cpumask_intersects(sdom->node_affinity_cpumask, mask) ) +if (cpumask_full (vc->cpu_soft_affinity) + || !cpumask_intersects (vc->cpu_soft_affinity, mask) )return 0 ;At this point we've lost a way to make this check potentially much faster (being able to check
auto_node_affinity).
Right.
Quoted 3 lines
This isn't a super-hot
path but it does happen fairly frequently --
Quite frequently indeed.
Quoted 5 lines
will the "
cpumask_full()"
check take a significant amount of time on, say, a 4096-core system? If so, we might think about "caching" the results ofcpumask_full()at some point.
Yes, I think cpumask_* operation could be heavy when the number of pcpus is high. However, this is not really a problem introduced by this series. Consider that the default behavior (for libxl and xl) is to go through initial domain placement, which would set a node-affinity for the domain explicitly, which means d->auto_node_affinity is false.
In fact, every domain that does not manually pin its vcpus at creation time --which is what we want, because that way NUMA placement can do its magic-- will have to go through the (cpumask_full || !cpumask_intrscts) anyway. Basically, I'm saying that having d->auto_node_affinity there may look like a speedup, but it really is only for a minority of cases.
So, yes, I think we should aim at optimizing this, but that is something completely orthogonal to this series. That is to say: (a) we should do it anyway, whether or not this series goes in; (b) for that same reason, that shouldn't prevent this series from going in.
If you think this can be an issue for 4.4, I'm fine creating a bug for it and putting it among the blockers. At that point, I'll start looking for a solution, and will commit to post a fix ASAP, but again, that's pretty independent from this very series, at least AFAICT.
Then, the fact that you provided your Reviewed-by above probably means that you are aware and ok with this all, but I felt like it was worth pointing it out anyway. :-)
Thanks and Regards,
Dario
Signature · 4 lines
<<This happens because I choose it to happen!>> (Raistlin Majere) ----------------------------------------------------------------- Dario Faggioli, Ph.D, http://about.me/dario.faggioli Senior Software Engineer, Citrix Systems R&D Ltd., Cambridge (UK)
_______________________________________________
Xen-devel mailing list
[email protected]
http://lists.xen.org/xen-devel
Re: [PATCH v2 12/16] libxl: get and set soft affinity
In reply to Ian JacksonQuoted 9 lines · Ian Jackson
Review
Quoted 4 lines
(quoted diff) +4
-0+ if (libxl_cpu_bitmap_alloc(ctx, &ptr->cpumap_soft, 0)) { + LIBXL__LOG_ERRNO(ctx, LIBXL__LOG_ERROR, "allocating cpumap_soft"); + return NULL; + }I went and looked at the error handling. Normally in libxl allocators can't fail but this one wants to do a hypercall to look up the max cpus.
Yep.
Quoted 4 lines
But, pulling on the thread, I wonder if it would be better to do this error logging in
libxl_cpu_bitmap_allocrather than all the call sites.
Indeed. And the main problem I see here is that the logging mostly says (and mine above is no exception, shame on me! :-P) "failed allocating ...", which is not true. If something failed, it was the retrieval of max cpus/nodes!
Quoted 4 lines
(If we did that in principle we should introduce a new #define which allows applications to #ifdef out their own logging for this, but in practice the double-logging isn't likely to be a problem.)
Do we? Right now we don't have anything like that... Looks to me that logging inside rather than outside the function is going to result in the same amount of logging we have already, with the notable difference that the "new" logging will be much more accurate.
Quoted 7 lines
And looking at
libxl_cpu_bitmap_alloc, it callslibxl_get_max_cpusand does something very odd with the return value:libxl_get_max_nodesought to return a positive number or a libxl error code.So I went to look at
libxl_get_max_nodesand it just returns whatever it got from libxc, which is definitely wrong!
I think I see what you mean. It's a weird path, since those xc calls (xc_get_max_cpus() and xc_get_max_nodes()) either return a positive number (if successful) or zero (if failing).
That's why, I think, libxl ended up checking for zero, as an indication of failure, rather than "< 0" / libxl error code.
Quoted 2 lines
Would you mind fixing this as part of this series ?
I don't, but I'm not sure I'm 100% getting how you think this should be fixed. What about the following (compile tested only) patch?
Basically, I tried to modify libxl_get_max_{cpus,nodes} as you suggeste above (making them return either a positive number or a libxl error code). At the same time, I made them a bit more "robust" against what could come from xc_get_max_{cpus,nodes}() (by checking for <= 0). I also added the proper logging --stating correctly what is it that is failing-- inside libxl_get_max_{cpus,nodes}() and removed it from the callsites that had it (wrong).
If you're fine with that, I've got no problem in folding it in the series.
Thanks and Regards,
Dario
---
tools/libxl/libxl.c +2 -6
@@ -616,10 +616,8 @@ static int cpupool_info(libxl__gc *gc, info->n_dom = xcinfo->n_dom; rc =libxl_cpu_bitmap_alloc (CTX , &info->cpumap,0 );if (rc) - { -LOG (ERROR ,"unable to allocate cpumap %d\n" , rc); goto out; - } +memcpy (info->cpumap.map, xcinfo->cpumap, info->cpumap.size); rc =0 ;
@@ -4204,10 +4202,8 @@ libxl_vcpuinfo *libxl_list_vcpu(libxl_ctx *ctx, uint32_t domid, }for (*nb_vcpu =0 ; *nb_vcpu <= domaininfo.max_vcpu_id; ++*nb_vcpu, ++ptr) { -if (libxl_cpu_bitmap_alloc (ctx, &ptr->cpumap,0 )) { -LIBXL__LOG_ERRNO (ctx,LIBXL__LOG_ERROR ,"allocating cpumap" ); +if (libxl_cpu_bitmap_alloc (ctx, &ptr->cpumap,0 ))return NULL ; - }if (xc_vcpu_getinfo (ctx->xch, domid, *nb_vcpu, &vcpuinfo) ==-1 ) {LIBXL__LOG_ERRNO (ctx,LIBXL__LOG_ERROR ,"getting vcpu info" );return NULL ;
tools/libxl/libxl_utils.c +47 -2
65 lines not on this page · raw message
tools/libxl/libxl_utils.h +6 -26
@@ -98,32 +98,12 @@ static inline int libxl_bitmap_cpu_valid(libxl_bitmap *bitmap, int bit)#define libxl_for_each_set_bit (v, m) for (v = 0; v < (m).size * 8; v++) \ if (libxl_bitmap_test(&(m), v)) -static inline int libxl_cpu_bitmap_alloc (libxl_ctx *ctx,libxl_bitmap *cpumap, -int max_cpus) -{ -if (max_cpus <0 ) -return ERROR_INVAL ; -if (max_cpus ==0 ) - max_cpus =libxl_get_max_cpus (ctx); -if (max_cpus ==0 ) -return ERROR_FAIL ; - -return libxl_bitmap_alloc (ctx, cpumap, max_cpus); -} - -static inline int libxl_node_bitmap_alloc (libxl_ctx *ctx, -libxl_bitmap *nodemap, -int max_nodes) -{ -if (max_nodes <0 ) -return ERROR_INVAL ; -if (max_nodes ==0 ) - max_nodes =libxl_get_max_nodes (ctx); -if (max_nodes ==0 ) -return ERROR_FAIL ; - -return libxl_bitmap_alloc (ctx, nodemap, max_nodes); -} +int libxl_cpu_bitmap_alloc (libxl_ctx *ctx, +libxl_bitmap *cpumap, +int max_cpus); +int libxl_node_bitmap_alloc (libxl_ctx *ctx, +libxl_bitmap *nodemap, +int max_nodes);/* Populate cpumap with the cpus spanned by the nodes in nodemap */ int libxl_nodemap_to_cpumap(libxl_ctx *ctx,
Signature · 4 lines
<<This happens because I choose it to happen!>> (Raistlin Majere) ----------------------------------------------------------------- Dario Faggioli, Ph.D, http://about.me/dario.faggioli Senior Software Engineer, Citrix Systems R&D Ltd., Cambridge (UK)
_______________________________________________
Xen-devel mailing list
[email protected]
http://lists.xen.org/xen-devel
Re: [PATCH v2 12/16] libxl: get and set soft affinity
In reply to Ian JacksonQuoted 9 lines · Ian Jackson
George Dunlap writes ("Re: [PATCH v2 12/16] libxl: get and set soft affinity"):
Quoted 7 lines · Ian Jackson
Quoted 2 lines
So perhaps we should have a singe function that can change the cpupool, hard affinity, and soft affinity, all at once ?
I think this is probably a good idea. Would it make sense to basically have libxl_[gs]et_vcpuaffinity2(), which takes a parameter that can specify that the mask is for either hard, soft, or both?
Quoted 6 lines
No, it needs to take at least two separate parameters for the hard and soft masks, because the toolstack might want to set the hard affinity and soft affinity to different values, without having to think carefully about ordering these calls to avoid generating spurious warnings and incompatible arrangements.
Ok, what about if I provide both?
Something like this:
int libxl_set_vcpuaffinity2(libxl_ctx *ctx, uint32_t domid, uint32_t vcpuid,
const libxl_bitmap *cpumap, int flags)
{
flags & HARD
<set_hard_affinity>(domid, vcpuid, cpumap)
flags & SOFT
<set_soft_affinity>(domid, vcpuid, cpumap)
<check_ecpumap_out_and_warn>
}
int libxl_set_vcpuaffinity3(libxl_ctx *ctx, uint32_t domid, uint32_t vcpuid,
const libxl_bitmap *cpumap_hard, const libxl_bitmap *cpumap_soft)
{
<set_hard_affinity>(domid, vcpuid, cpumap_hard)
<set_soft_affinity>(domid, vcpuid, cpumap_soft)
<check_ecpumap_out_and_warn>
}
This ought to allow both the use case:
1. when one is only want to change one (or both to the same value), and is ok with being WARN'ed if that result in an inconsistent state
2. when one wants to land directly in a specific state, which requires changing both, each one to its own value, and only want to be WARN'ed _iff_ that final state is inconsistent
I've already an half backed updated series doing right this, but I'll be working far from my testbox in the morning, so I'll complete the testing and send it in the afternoon, ok?
Quoted 6 lines
I think this probably applies to cpupool changes too. So perhaps the function should look a bit like setresuid or something: it should take an optional new cpupool (is there a sentinel value that can be used to mean "no change"?), an optional new soft mask, and an optional new hard mask.
I see what you mean, but cpupools are really a different thing, with its own interface, etc. I'd rather try to implement what I described above and being consistent with the current cpupool behavior, copying what it happens already for hard affinity to how soft affinity is handled too (as George was saying), at least for this series.
Quoted 8 lines
Quoted 5 lines
Quoted 3 lines
What if the application makes a call to change the cpupool, without touching the affinities ? Should the affinities be reset automatically ?
I think whatever happens for hard affinities right now should be carried on.
Maybe it is a bug that it doesn't do anything.
Maybe, but that just mean that we can (should) fix that even outside from this series, right?
Quoted 3 lines
I think it depends how
we expect people to use this.
Indeed. :-D
Quoted 4 lines
If a caller sets the hard affinities
and then changes the cpupool, are they supposed to always then set the hard affinities again to a new suitable value ?
Yes and no. :-) I'll provide some examples of what happens to try clarifying the situation as soon as I get the chance.
Thanks for the review and the nice discussion. :-) Dario
Signature · 4 lines
<<This happens because I choose it to happen!>> (Raistlin Majere) ----------------------------------------------------------------- Dario Faggioli, Ph.D, http://about.me/dario.faggioli Senior Software Engineer, Citrix Systems R&D Ltd., Cambridge (UK)
_______________________________________________
Xen-devel mailing list
[email protected]
http://lists.xen.org/xen-devel
Re: [PATCH v2 06/16] xen: sched: make space for cpu_soft_affinity
In reply to Dario FaggioliQuoted 11 lines · Dario Faggioli
Quoted 5 lines · George Dunlap
The breakdown of this in the series doesn't make much sense to me -- I would have folded this one and patch 10 (use soft affinity instead of node affinity) together, and put it in after patch 07 (s/affinity/hard_affinity/g;).
I see. Well, looks like I'm respinning it. While at it, I'll see if I can put it that way.
If I do (and provided that does not imply any code change in those patches), can I retain your Reviewed-by tag in the result?
Yes, if you do a plain merge of this with patch 10 and don't make any changes, go ahead and keep it.
-George
Re: [PATCH v2 08/16] xen: derive NUMA node affinity from hard and soft CPU affinity
In reply to Dario FaggioliQuoted 17 lines · Dario Faggioli
Quoted 4 lines · George Dunlap
Quoted 4 lines
Signed-off-by Dario Faggioli <[email protected]>xen/common/domain.c +0
-0@@ -373,8 +379,12 @@ void domain_update_node_affinity(struct domain *d)Quoted 12 lines
xen/common/domain.c void domain_update_node_affinity(struct domain *d) · line 387
Quoted 8 lines
xen/common/domain.c +4
-0for_each_vcpu ( d, v ) { +/* Build up the mask of online pcpus we have hard affinity with */ cpumask_and (online_affinity, v->cpu_hard_affinity, online);cpumask_or (cpumask, cpumask, online_affinity); + +/* As well as the mask of all pcpus we have soft affinity with */ +cpumask_or (cpumask_soft, cpumask_soft, v->cpu_soft_affinity);Is this really the most efficient way to do this? I would have thought or'ing cpumask and
cpumask_soft, and then if it's not empty, then use it; maybe use a pointer so you don't have to copy one into the other one?
Quoted 4 lines
I'm not sure I fully get what you mean... I cannot afford neglecting
online_affinity, independently from how cpumask andcpumask_softlook like, because that's what's needed to account for cpupools. Anyway, I'll think more about it and see if I can make it better.
So what you have here is (in pseudocode):
online_affinity = hard_affinity & online;
cpumask |= online_affinity;
cpumask_soft |= soft_affinity;
if ( intersects(cpumask, cpumask_soft) )
cpumask &= cpumask_soft;
So at least four full bitwise operations, plus "intersects" which will be the equivalent of a full bitwise operation if it turns out to be false.
How about something like the following:
cpumask_hard = hard_affinity & online;cpumask_soft = cpumask_hard & soft_affinity;
cpumask_p = is_empty(cpumask_soft) ? &cpumask_hard : &cpumask_soft;
So only two full bitwise operations, plus "is_empty", which I think should be faster than "intersects" even if it turns out to be true, since we can (I think) check full words at a time.
Then use cpumask_p in the subsequent node_affinity update. (Or just do a full copy, since doing the nodewise check for intersection is going to be fairly slow anyway.)
-George
Re: [PATCH v2 10/16] xen: sched: use soft-affinity instead of domain's node-affinity
In reply to Dario FaggioliQuoted 21 lines · Dario Faggioli
Quoted 9 lines · George Dunlap
Quoted 6 lines · Dario Faggioli
[..]
The high level description of NUMA placement and scheduling in docs/misc/xl-numa-placement.markdown is being updated too, to match the new architecture.Signed-off-by Dario Faggioli <[email protected]>Reviewed-by George Dunlap <[email protected]>Cool, thanks.
Quoted 2 lines
Just a few things to note below...
Ok.
Quoted 5 lines
Quoted 5 lines
xen/common/domain.c +0
-0@@ -411,8 +411,6 @@ void domain_update_node_affinity(struct domain *d)node_set (node, d->node_affinity); }
Quoted 7 lines
Quoted 6 lines
xen/common/domain.c void domain_update_node_affinity(struct domain *d) · line 414
Quoted 3 lines
xen/common/domain.c +0
-2-sched_set_node_affinity (d, &d->node_affinity); -spin_unlock (&d->node_affinity_lock);At this point, the only thing inside the spinlock is contingent on
d->auto_node_affinity.Mmm... Sorry, but I'm not geting what you mean here. :-(
I mean just what I said -- if d->auto_node_affinity is false, nothing inside the critical region here needs to be done. I'm just pointing it out. :-) (This is sort of related to my comment on the other patch, about not needing to do the work of calculating intersections.)
Quoted 47 lines
Quoted 18 lines
Quoted 5 lines
xen/common/sched_credit.c +1
-1-static inline int __vcpu_has_node_affinity (const struct vcpu *vc, +static inline int __vcpu_has_soft_affinity (const struct vcpu *vc,const cpumask_t *mask)xen/common/sched_credit.c __runq_remove(struct csched_vcpu *svc) · line 297
Quoted 10 lines
xen/common/sched_credit.c +2
-6{ -const struct domain *d = vc->domain; -const struct csched_dom *sdom =CSCHED_DOM (d); - -if ( d->auto_node_affinity - ||cpumask_full (sdom->node_affinity_cpumask) - || !cpumask_intersects (sdom->node_affinity_cpumask, mask) ) +if (cpumask_full (vc->cpu_soft_affinity) + || !cpumask_intersects (vc->cpu_soft_affinity, mask) )return 0 ;At this point we've lost a way to make this check potentially much faster (being able to check
auto_node_affinity).Right.
Quoted 3 lines
This isn't a super-hot
path but it does happen fairly frequently --Quite frequently indeed.
Quoted 5 lines
will the "
cpumask_full()"
check take a significant amount of time on, say, a 4096-core system? If so, we might think about "caching" the results ofcpumask_full()at some point.Yes, I think cpumask_* operation could be heavy when the number of pcpus is high. However, this is not really a problem introduced by this series. Consider that the default behavior (for libxl and xl) is to go through initial domain placement, which would set a node-affinity for the domain explicitly, which means
d->auto_node_affinityis false.In fact, every domain that does not manually pin its vcpus at creation time --which is what we want, because that way NUMA placement can do its magic-- will have to go through the (
cpumask_full|| !cpumask_intrscts) anyway. Basically, I'm saying that havingd->auto_node_affinitythere may look like a speedup, but it really is only for a minority of cases.So, yes, I think we should aim at optimizing this, but that is something completely orthogonal to this series. That is to say: (a) we should do it anyway, whether or not this series goes in; (b) for that same reason, that shouldn't prevent this series from going in.
Quoted 8 lines
If you think this can be an issue for 4.4, I'm fine creating a bug for it and putting it among the blockers. At that point, I'll start looking for a solution, and will commit to post a fix ASAP, but again, that's pretty independent from this very series, at least AFAICT.
Then, the fact that you provided your Reviewed-by above probably means that you are aware and ok with this all, but I felt like it was worth pointing it out anyway. :-)
Yes, the "at some point" was intended to imply that I didn't think this had to be done right away, as was "things to note", which means, "I just want to point this out, they're not something which needs to be acted on right away."
-George
Re: [PATCH v2 12/16] libxl: get and set soft affinity
In reply to Ian JacksonQuoted 19 lines · Ian Jackson
George Dunlap writes ("Re: [PATCH v2 12/16] libxl: get and set soft affinity"):
Quoted 6 lines · Ian Jackson
Quoted 2 lines
So perhaps we should have a singe function that can change the cpupool, hard affinity, and soft affinity, all at once ?
I think this is probably a good idea. Would it make sense to basically have libxl_[gs]et_vcpuaffinity2(), which takes a parameter that can specify that the mask is for either hard, soft, or both?
No, it needs to take at least two separate parameters for the hard and soft masks, because the toolstack might want to set the hard affinity and soft affinity to different values, without having to think carefully about ordering these calls to avoid generating spurious warnings and incompatible arrangements.
I think this probably applies to cpupool changes too. So perhaps the function should look a bit like setresuid or something: it should take an optional new cpupool (is there a sentinel value that can be used to mean "no change"?), an optional new soft mask, and an optional new hard mask.
Quoted 8 lines
Quoted 4 lines
Quoted 3 lines
What if the application makes a call to change the cpupool, without touching the affinities ? Should the affinities be reset automatically ?
I think whatever happens for hard affinities right now should be carried on.
Maybe it is a bug that it doesn't do anything. I think it depends how we expect people to use this. If a caller sets the hard affinities and then changes the cpupool, are they supposed to always then set the hard affinities again to a new suitable value ?
Well in fact, as far as I can tell, it *does* do something. When moving a vcpu to a new pool, it unconditionally calls cpumask_setall(v->cpu_affinity) for each vcpu, which will effectively erase the hard affinity. (xen/common/schedule.c:sched_move_domain()).
And, when unplugging cpus, if it unplugs the last cpu a vcpu can run on, it also resets the affinity to "all".
-George
Re: [PATCH v2 08/16] xen: derive NUMA node affinity from hard and soft CPU affinity
In reply to George DunlapQuoted 13 lines · George Dunlap
Quoted 4 lines · Dario Faggioli
I'm not sure I fully get what you mean... I cannot afford neglecting
online_affinity, independently from how cpumask andcpumask_softlook like, because that's what's needed to account for cpupools. Anyway, I'll think more about it and see if I can make it better.So what you have here is (in pseudocode):
online_affinity = hard_affinity & online; cpumask |= online_affinity; cpumask_soft |= soft_affinity;
Quoted 6 lines
if (
intersects(cpumask, cpumask_soft))
cpumask &=cpumask_soft;So at least four full bitwise operations, plus "intersects" which will be the equivalent of a full bitwise operation if it turns out to be false.
Well, that's not exactly what I have. Fact is, _every_ vcpu has a soft_affinity and a hard_affinity. That's why I have the cpumask_or()-s in the loop, to build a hard haffinity and a soft affinity mask of the _domain_, by ||-ing all the hard (soft) affinity of all the vcpus.
From the pseudo code above (and even from the one below), it's not clear to me whether you mean vcpu hard (and soft) affinity or domain ones.
In fact...
Quoted 7 lines
How about something like the following:
cpumask_hard=hard_affinity& online;cpumask_soft=cpumask_hard&soft_affinity;
cpumask_p=is_empty(cpumask_soft)?&cpumask_hard:&cpumask_soft;
... From reading this last pseudo-statement I think you actually mean the affinity masks for the whole domain, which is exactly what I also want to consider... But I have to construct them somehow... That's, again, form where the cpumask_or()-s come.
So, thanking this per-vcpu-ness into account, what I have is:
for_each_vcpu(i)
dom_cpumask_hard |= online & cpumask_hard(i)
dom_cpumask_soft |= cpumask_soft(i) if ( dom_cpumask_hard & dom_cpumask_soft )
dom_cpumask = dom_cpumask_hard & dom_cpumask_soft
else
<use dom_cpumask_hard>
Perhaps I can rename the variable with something like this 'dom_' prefix in the real code too, to make things more clear.
However, it is indeed true that online is per domain, so yes, perhaps I can ddo something like the following:
for_each_vcpu(i)
dom_cpumask_hard |= cpumask_hard(i)
dom_cpumask_soft |= cpumask_soft(i) dom_cpumask_hard &= online
dom_cpumask_soft &= dom_cpumask_hard
dom_cpumask_p = is_empty(dom_cpumask_soft) ?
dom_cpumask_hard : dom_cpumask_soft;
I think it's both more efficient and more clear than what I have now, and I give you that it's close enough to what you were suggesting... But I don't think I can ditch the '|=' in the loop for building up the two domain wide cpumasks.
How do you like it?
Dario
Signature · 4 lines
<<This happens because I choose it to happen!>> (Raistlin Majere) ----------------------------------------------------------------- Dario Faggioli, Ph.D, http://about.me/dario.faggioli Senior Software Engineer, Citrix Systems R&D Ltd., Cambridge (UK)
_______________________________________________
Xen-devel mailing list
[email protected]
http://lists.xen.org/xen-devel
Re: [PATCH v2 12/16] libxl: get and set soft affinity
In reply to George DunlapQuoted 6 lines · George Dunlap
Quoted 4 lines · Ian Jackson
Maybe it is a bug that it doesn't do anything. I think it depends how we expect people to use this. If a caller sets the hard affinities and then changes the cpupool, are they supposed to always then set the hard affinities again to a new suitable value ?
Quoted 8 lines
Well in fact, as far as I can tell, it *does* do something. When moving a vcpu to a new pool, it unconditionally calls
cpumask_setall(v->cpu_affinity)for each vcpu, which will effectively erase the hard affinity. (xen/common/schedule.c:sched_move_domain()).And, when unplugging cpus, if it unplugs the last cpu a vcpu can run on, it also resets the affinity to "all".
Right. But this all happens in hypervisor level and, personally, I think it's just fine.
The point here is how we should behave and what kind of interface we should have/add at the libxl level. My opinion here is that, while it is ok to have calls that deals with both hard and soft affinity together, we should leave cpupool alone, as it is too different of both a concept and of an interface.
Thanks and Regards,
Dario
Signature · 4 lines
<<This happens because I choose it to happen!>> (Raistlin Majere) ----------------------------------------------------------------- Dario Faggioli, Ph.D, http://about.me/dario.faggioli Senior Software Engineer, Citrix Systems R&D Ltd., Cambridge (UK)
_______________________________________________
Xen-devel mailing list
[email protected]
http://lists.xen.org/xen-devel