[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index]

[Xen-devel] [PATCH 3/5] tools: libxl/xl: allow to get/set Credit1's vcpu_migration_delay



Make it possible to get and set a (Credit1) scheduler's
vCPU migration delay via the SCHEDOP sysctl, from both
libxl and xl (no change needed in libxc).

Signed-off-by: Dario Faggioli <dfaggioli@xxxxxxxx>
---
Cc: Ian Jackson <ian.jackson@xxxxxxxxxxxxx>
Cc: Wei Liu <wei.liu2@xxxxxxxxxx>
Cc: George Dunlap <george.dunlap@xxxxxxxxxxxxx>
Cc: Andrew Cooper <andrew.cooper3@xxxxxxxxxx>
---
 docs/man/xl.pod.1.in        |   11 +++++++++++
 tools/libxl/libxl.h         |    7 +++++++
 tools/libxl/libxl_sched.c   |   10 ++++++++++
 tools/libxl/libxl_types.idl |    1 +
 tools/xl/xl_cmdtable.c      |    1 +
 tools/xl/xl_sched.c         |   22 +++++++++++++++-------
 6 files changed, 45 insertions(+), 7 deletions(-)

diff --git a/docs/man/xl.pod.1.in b/docs/man/xl.pod.1.in
index 7fd35c9ae7..48da2052cc 100644
--- a/docs/man/xl.pod.1.in
+++ b/docs/man/xl.pod.1.in
@@ -1046,6 +1046,17 @@ we will allow a higher-priority VM to pre-empt it.  The 
default value
 is 1000 microseconds (1ms).  Valid range is 100 to 500000 (500ms).
 The ratelimit length must be lower than the timeslice length.
 
+=item B<-m DELAY>, B<--migration_delay_us=DELAY>
+
+Migration delay specifies for how long a vCPU, after it stopped running should
+be considered "cache-hot". Basically, if less than DELAY us passed since when
+the vCPU was executing on a CPU, it is likely that most of the vCPU's working
+set is still in the CPU's cache, and therefore the vCPU is not migrated.
+
+Default is 0. Maximum is 100 ms. This can be effective at preventing vCPUs
+to bounce among CPUs too quickly, but, at the same time, the scheduler stops
+being fully work-conserving.
+
 =back
 
 B<COMBINATION>
diff --git a/tools/libxl/libxl.h b/tools/libxl/libxl.h
index eca0ea2c50..edd244278a 100644
--- a/tools/libxl/libxl.h
+++ b/tools/libxl/libxl.h
@@ -299,6 +299,13 @@
  */
 #define LIBXL_HAVE_SCHED_CREDIT2_PARAMS 1
 
+/*
+ * LIBXL_HAVE_SCHED_CREDIT_MIGR_DELAY indicates that there is a field
+ * in libxl_sched_credit_params called vcpu_migr_delay_us which controls
+ * the resistance of the vCPUs of the cpupool to migrations among pCPUs.
+ */
+#define LIBXL_HAVE_SCHED_CREDIT_MIGR_DELAY
+
 /*
  * LIBXL_HAVE_VIRIDIAN_CRASH_CTL indicates that the 'crash_ctl' value
  * is present in the viridian enlightenment enumeration.
diff --git a/tools/libxl/libxl_sched.c b/tools/libxl/libxl_sched.c
index 512788f736..07289079ce 100644
--- a/tools/libxl/libxl_sched.c
+++ b/tools/libxl/libxl_sched.c
@@ -291,6 +291,7 @@ int libxl_sched_credit_params_get(libxl_ctx *ctx, uint32_t 
poolid,
 
     scinfo->tslice_ms = sparam.tslice_ms;
     scinfo->ratelimit_us = sparam.ratelimit_us;
+    scinfo->vcpu_migr_delay_us = sparam.vcpu_migr_delay_us;
 
     rc = 0;
  out:
@@ -321,9 +322,17 @@ int libxl_sched_credit_params_set(libxl_ctx *ctx, uint32_t 
poolid,
         rc = ERROR_INVAL;
         goto out;
     }
+    if (scinfo->vcpu_migr_delay_us
+        && scinfo->vcpu_migr_delay_us > XEN_SYSCTL_CSCHED_MGR_DLY_MAX) {
+        LOG(ERROR, "vcpu migration delay should be > 0 and < %d",
+            XEN_SYSCTL_CSCHED_MGR_DLY_MAX);
+        rc = ERROR_INVAL;
+        goto out;
+    }
 
     sparam.tslice_ms = scinfo->tslice_ms;
     sparam.ratelimit_us = scinfo->ratelimit_us;
+    sparam.vcpu_migr_delay_us = scinfo->vcpu_migr_delay_us;
 
     r = xc_sched_credit_params_set(ctx->xch, poolid, &sparam);
     if ( r < 0 ) {
@@ -334,6 +343,7 @@ int libxl_sched_credit_params_set(libxl_ctx *ctx, uint32_t 
poolid,
 
     scinfo->tslice_ms = sparam.tslice_ms;
     scinfo->ratelimit_us = sparam.ratelimit_us;
+    scinfo->vcpu_migr_delay_us = sparam.vcpu_migr_delay_us;
 
     rc = 0;
  out:
diff --git a/tools/libxl/libxl_types.idl b/tools/libxl/libxl_types.idl
index 35038120ca..dbb287d6fe 100644
--- a/tools/libxl/libxl_types.idl
+++ b/tools/libxl/libxl_types.idl
@@ -973,6 +973,7 @@ libxl_pcitopology = Struct("pcitopology", [
 libxl_sched_credit_params = Struct("sched_credit_params", [
     ("tslice_ms", integer),
     ("ratelimit_us", integer),
+    ("vcpu_migr_delay_us", integer),
     ], dispose_fn=None)
 
 libxl_sched_credit2_params = Struct("sched_credit2_params", [
diff --git a/tools/xl/xl_cmdtable.c b/tools/xl/xl_cmdtable.c
index 6d894394ca..bf2ced8140 100644
--- a/tools/xl/xl_cmdtable.c
+++ b/tools/xl/xl_cmdtable.c
@@ -257,6 +257,7 @@ struct cmd_spec cmd_table[] = {
       "-s         --schedparam           Query / modify scheduler parameters\n"
       "-t TSLICE, --tslice_ms=TSLICE     Set the timeslice, in milliseconds\n"
       "-r RLIMIT, --ratelimit_us=RLIMIT  Set the scheduling rate limit, in 
microseconds\n"
+      "-m DLY, --migration_delay_us=DLY  Set the migration delay, in 
microseconds\n"
       "-p CPUPOOL, --cpupool=CPUPOOL     Restrict output to CPUPOOL"
     },
     { "sched-credit2",
diff --git a/tools/xl/xl_sched.c b/tools/xl/xl_sched.c
index 7965ccbca0..1f49560376 100644
--- a/tools/xl/xl_sched.c
+++ b/tools/xl/xl_sched.c
@@ -172,10 +172,11 @@ static int sched_credit_pool_output(uint32_t poolid)
         printf("Cpupool %s: [sched params unavailable]\n",
                poolname);
     } else {
-        printf("Cpupool %s: tslice=%dms ratelimit=%dus\n",
+        printf("Cpupool %s: tslice=%dms ratelimit=%dus migration-delay=%dus\n",
                poolname,
                scparam.tslice_ms,
-               scparam.ratelimit_us);
+               scparam.ratelimit_us,
+               scparam.vcpu_migr_delay_us);
     }
     free(poolname);
     return 0;
@@ -469,10 +470,10 @@ int main_sched_credit(int argc, char **argv)
     const char *dom = NULL;
     const char *cpupool = NULL;
     int weight = 256, cap = 0;
-    int tslice = 0, ratelimit = 0;
+    int tslice = 0, ratelimit = 0, migrdelay = 0;
     bool opt_w = false, opt_c = false;
     bool opt_t = false, opt_r = false;
-    bool opt_s = false;
+    bool opt_s = false, opt_m = false;
     int opt, rc;
     static struct option opts[] = {
         {"domain", 1, 0, 'd'},
@@ -481,11 +482,12 @@ int main_sched_credit(int argc, char **argv)
         {"schedparam", 0, 0, 's'},
         {"tslice_ms", 1, 0, 't'},
         {"ratelimit_us", 1, 0, 'r'},
+        {"migration_delay_us", 1, 0, 'm'},
         {"cpupool", 1, 0, 'p'},
         COMMON_LONG_OPTS
     };
 
-    SWITCH_FOREACH_OPT(opt, "d:w:c:p:t:r:s", opts, "sched-credit", 0) {
+    SWITCH_FOREACH_OPT(opt, "d:w:c:p:t:r:m:s", opts, "sched-credit", 0) {
     case 'd':
         dom = optarg;
         break;
@@ -505,6 +507,9 @@ int main_sched_credit(int argc, char **argv)
         ratelimit = strtol(optarg, NULL, 10);
         opt_r = true;
         break;
+    case 'm':
+        migrdelay = strtol(optarg, NULL, 10);
+        opt_m = true;
     case 's':
         opt_s = true;
         break;
@@ -522,7 +527,7 @@ int main_sched_credit(int argc, char **argv)
         fprintf(stderr, "Must specify a domain.\n");
         return EXIT_FAILURE;
     }
-    if (!opt_s && (opt_t || opt_r)) {
+    if (!opt_s && (opt_t || opt_r || opt_w)) {
         fprintf(stderr, "Must specify schedparam to set schedule "
                 "parameter values.\n");
         return EXIT_FAILURE;
@@ -541,7 +546,7 @@ int main_sched_credit(int argc, char **argv)
             }
         }
 
-        if (!opt_t && !opt_r) { /* Output scheduling parameters */
+        if (!opt_t && !opt_r && !opt_m) { /* Output scheduling parameters */
             if (sched_credit_pool_output(poolid))
                 return EXIT_FAILURE;
         } else { /* Set scheduling parameters*/
@@ -554,6 +559,9 @@ int main_sched_credit(int argc, char **argv)
             if (opt_r)
                 scparam.ratelimit_us = ratelimit;
 
+            if (opt_m)
+                scparam.vcpu_migr_delay_us = migrdelay;
+
             if (sched_credit_params_set(poolid, &scparam))
                 return EXIT_FAILURE;
         }


_______________________________________________
Xen-devel mailing list
Xen-devel@xxxxxxxxxxxxxxxxxxxx
https://lists.xenproject.org/mailman/listinfo/xen-devel

 


Rackspace

Lists.xenproject.org is hosted with RackSpace, monitoring our
servers 24x7x365 and backed by RackSpace's Fanatical Support®.