All of lore.kernel.org
 help / color / mirror / Atom feed
From: Dario Faggioli <dario.faggioli@citrix.com>
To: xen-devel@lists.xen.org
Cc: keir@xen.org, Ian.Campbell@citrix.com, Andrew.Cooper3@citrix.com,
	George.Dunlap@citrix.com, JBeulich@suse.com,
	Ian.Jackson@citrix.com, Wei Liu <wei.liu2@citrix.com>
Subject: [PATCH v8 07/13] libxl: get and set soft affinity
Date: Fri, 13 Jun 2014 15:10:04 +0200	[thread overview]
Message-ID: <20140613131003.4106.88556.stgit@Solace> (raw)
In-Reply-To: <20140613124847.4106.70161.stgit@Solace>

Make space a new cpumap in vcpu_info, called cpumap_soft,
for retrieving soft affinity, and amend the relevant API
accordingly.

libxl_set_vcpuaffinity() now takes two cpumaps, one for hard
and one for soft affinity (LIBXL_API_VERSION is exploited to
retain source level backword compatibility). Either of the
two cpumap can be NULL, in which case, only the affinity
corresponding to the non-NULL cpumap will be affected.

Getting soft affinity happens indirectly (see, e.g.,
`xl vcpu-list'), as it is already for hard affinity).

This commit also introduces some logic to check whether the
affinity which will be used by Xen to schedule the vCPU(s)
does actually match with the cpumaps provided. In fact, we
want to allow every possible combination of hard and soft
affinity to be set, but we warn the user upon particularly
weird situations (e.g., hard and soft being disjoint sets
of pCPUs).

This very change also update the error handling for calls
to libxl_set_vcpuaffinity() in xl, as that can now be any
libxl error code, not just only -1.

Signed-off-by: Dario Faggioli <dario.faggioli@citrix.com>
Reviewed-by: George Dunlap <george.dunlap@eu.citrix.com>
Acked-by: Ian Campbell <ian.campbell@citrix.com>
---
Changes from v7
 * Added 0x40500 to the list of valid libxl versions;
 * killed the cpumap_soft field from libxl_build_info.
   That is not necessary, as subsequent patches will
   introduce and alternative way to set soft affinity.
 * fix a typo.

Changes from v6:
 * fix a typo and update LIBXL_API_VERSION appropriately,
   as requested during review;
 * libxl_bitmap_equal() rewritten in such a way that it now
   automatically deals with bitmaps of different sizes. Bits
   non present in one of the maps are just considered to be
   0, as suggested during review;
 * there is no need for libxl_bitmap_bit_valid() any longer,
   so killed it;
 * inside libxl, retrieve and set both hard and soft affinity
   atomically, in just one call (to xc_vcpu_getaffinity() and
   xc_vcpu_setaffinity(), respectively), as requested during
   review.
 * update the changelog about changing the error handling for
   the call to libxl_set_vcpuaffinity() in xl.

Changes from v4:
 * get rid of inline stubs inside the LIBXL_API_VERSION_XXX
   block and just use define, as suggested during review
 * adapt to the new xc interface;
 * avoid leaking cpumap_soft in libxl_list_vcpu on error, as
   requested during review;
 * fix bogous `return 0' in libxl_set_vcpuaffinity, as
   requested during review;
 * clarify the comment for LIBXL_HAVE_SOFT_AFFINITY, as
   suggested during review;
 * renamed libxl_bitmap_valid() to libxl_bitmap_bit_valid(),
   as suggested uring review.

Changes from v3:
 * only introduce one LIBXL_HAVE_ symbol for soft affinity,
   as requested during review;
 * use LIBXL_API_VERSION instead of having multiple version
   of the same function, as suggested during review;
 * use libxl_get_nr_cpus() rather than libxl_get_cputopology(),
   as suggested during review;
 * use LOGE() instead of LIBXL__LOG_ERRNO(), as requested
   during review;
 * kill the flags and use just one _set_vcpuaffinity()
   function with two cpumaps, allowing either of them to
   be NULL, as suggested during review;
 * avoid overflowing the bitmaps in libxl_bitmap_equal(),
   as suggested during review.

Changes from v2:
 * interface completely redesigned, as discussed during
   review.
---
 tools/libxl/libxl.c         |   99 ++++++++++++++++++++++++++++++++++++++-----
 tools/libxl/libxl.h         |   26 ++++++++++-
 tools/libxl/libxl_dom.c     |    3 +
 tools/libxl/libxl_types.idl |    3 +
 tools/libxl/libxl_utils.h   |   25 +++++++++++
 tools/libxl/xl_cmdimpl.c    |    6 +--
 6 files changed, 142 insertions(+), 20 deletions(-)

diff --git a/tools/libxl/libxl.c b/tools/libxl/libxl.c
index 1527283..47a5ed3 100644
--- a/tools/libxl/libxl.c
+++ b/tools/libxl/libxl.c
@@ -4590,13 +4590,17 @@ libxl_vcpuinfo *libxl_list_vcpu(libxl_ctx *ctx, uint32_t domid,
         libxl_bitmap_init(&ptr->cpumap);
         if (libxl_cpu_bitmap_alloc(ctx, &ptr->cpumap, 0))
             goto err;
+        libxl_bitmap_init(&ptr->cpumap_soft);
+        if (libxl_cpu_bitmap_alloc(ctx, &ptr->cpumap_soft, 0))
+            goto err;
         if (xc_vcpu_getinfo(ctx->xch, domid, *nr_vcpus_out, &vcpuinfo) == -1) {
             LOGE(ERROR, "getting vcpu info");
             goto err;
         }
+
         if (xc_vcpu_getaffinity(ctx->xch, domid, *nr_vcpus_out,
-                                ptr->cpumap.map, NULL,
-                                XEN_VCPUAFFINITY_HARD) == -1) {
+                                ptr->cpumap.map, ptr->cpumap_soft.map,
+                                XEN_VCPUAFFINITY_SOFT|XEN_VCPUAFFINITY_HARD) == -1) {
             LOGE(ERROR, "getting vcpu affinity");
             goto err;
         }
@@ -4612,34 +4616,105 @@ libxl_vcpuinfo *libxl_list_vcpu(libxl_ctx *ctx, uint32_t domid,
 
 err:
     libxl_bitmap_dispose(&ptr->cpumap);
+    libxl_bitmap_dispose(&ptr->cpumap_soft);
     free(ret);
     GC_FREE;
     return NULL;
 }
 
 int libxl_set_vcpuaffinity(libxl_ctx *ctx, uint32_t domid, uint32_t vcpuid,
-                           libxl_bitmap *cpumap)
+                           const libxl_bitmap *cpumap_hard,
+                           const libxl_bitmap *cpumap_soft)
 {
-    if (xc_vcpu_setaffinity(ctx->xch, domid, vcpuid, cpumap->map, NULL,
-                            XEN_VCPUAFFINITY_HARD)) {
-        LIBXL__LOG_ERRNO(ctx, LIBXL__LOG_ERROR, "setting vcpu affinity");
-        return ERROR_FAIL;
+    GC_INIT(ctx);
+    libxl_bitmap hard, soft;
+    int rc, flags = 0;
+
+    libxl_bitmap_init(&hard);
+    libxl_bitmap_init(&soft);
+
+    if (!cpumap_hard && !cpumap_soft) {
+        rc = ERROR_INVAL;
+        goto out;
     }
-    return 0;
+
+    /*
+     * Xen wants writable hard and/or soft cpumaps, to put back in them
+     * the effective hard and/or soft affinity that will be used.
+     */
+    if (cpumap_hard) {
+        rc = libxl_cpu_bitmap_alloc(ctx, &hard, 0);
+        if (rc)
+            goto out;
+
+        libxl_bitmap_copy(ctx, &hard, cpumap_hard);
+        flags = XEN_VCPUAFFINITY_HARD;
+    }
+    if (cpumap_soft) {
+        rc = libxl_cpu_bitmap_alloc(ctx, &soft, 0);
+        if (rc)
+            goto out;
+
+        libxl_bitmap_copy(ctx, &soft, cpumap_soft);
+        flags |= XEN_VCPUAFFINITY_SOFT;
+    }
+
+    if (xc_vcpu_setaffinity(ctx->xch, domid, vcpuid,
+                            cpumap_hard ? hard.map : NULL,
+                            cpumap_soft ? soft.map : NULL,
+                            flags)) {
+        LOGE(ERROR, "setting vcpu affinity");
+        rc = ERROR_FAIL;
+        goto out;
+    }
+
+    /*
+     * Let's check the results. Hard affinity will never be empty, but it
+     * is possible that Xen will use something different from what we asked
+     * for various reasons. If that's the case, report it.
+     */
+    if (cpumap_hard &&
+        !libxl_bitmap_equal(cpumap_hard, &hard, 0))
+        LOG(DEBUG, "New hard affinity for vcpu %d has unreachable cpus",
+        vcpuid);
+    /*
+     * Soft affinity can both be different from what asked and empty. Check
+     * for (and report) both.
+     */
+    if (cpumap_soft) {
+        if (!libxl_bitmap_equal(cpumap_soft, &soft, 0))
+            LOG(DEBUG, "New soft affinity for vcpu %d has unreachable cpus",
+                vcpuid);
+        if (libxl_bitmap_is_empty(&soft))
+            LOG(WARN, "all cpus in soft affinity of vcpu %d are unreachable."
+                " Only hard affinity will be considered for scheduling",
+                vcpuid);
+    }
+
+    rc = 0;
+ out:
+    libxl_bitmap_dispose(&hard);
+    libxl_bitmap_dispose(&soft);
+    GC_FREE;
+    return rc;
 }
 
 int libxl_set_vcpuaffinity_all(libxl_ctx *ctx, uint32_t domid,
-                               unsigned int max_vcpus, libxl_bitmap *cpumap)
+                               unsigned int max_vcpus,
+                               const libxl_bitmap *cpumap_hard,
+                               const libxl_bitmap *cpumap_soft)
 {
+    GC_INIT(ctx);
     int i, rc = 0;
 
     for (i = 0; i < max_vcpus; i++) {
-        if (libxl_set_vcpuaffinity(ctx, domid, i, cpumap)) {
-            LIBXL__LOG(ctx, LIBXL__LOG_WARNING,
-                       "failed to set affinity for %d", i);
+        if (libxl_set_vcpuaffinity(ctx, domid, i, cpumap_hard, cpumap_soft)) {
+            LOG(WARN, "failed to set affinity for %d", i);
             rc = ERROR_FAIL;
         }
     }
+
+    GC_FREE;
     return rc;
 }
 
diff --git a/tools/libxl/libxl.h b/tools/libxl/libxl.h
index 17b8a7b..c2b143b 100644
--- a/tools/libxl/libxl.h
+++ b/tools/libxl/libxl.h
@@ -104,6 +104,13 @@
 #define LIBXL_HAVE_BUILDINFO_EVENT_CHANNELS 1
 
 /*
+ * LIBXL_HAVE_VCPUINFO_SOFT_AFFINITY indicates that a 'cpumap_soft'
+ * field (of libxl_bitmap type) is present in libxl_vcpuinfo,
+ * containing the soft affinity of a vcpu.
+ */
+#define LIBXL_HAVE_VCPUINFO_SOFT_AFFINITY 1
+
+/*
  * LIBXL_HAVE_DEVICE_DISK_DIRECT_IO_SAFE indicates that a
  * 'direct_io_safe' field (of boolean type) is present in
  * libxl_device_disk.
@@ -325,7 +332,7 @@
 /* API compatibility. */
 #ifdef LIBXL_API_VERSION
 #if LIBXL_API_VERSION != 0x040200 && LIBXL_API_VERSION != 0x040300 && \
-    LIBXL_API_VERSION != 0x040400
+    LIBXL_API_VERSION != 0x040400 && LIBXL_API_VERSION != 0x040500
 #error Unknown LIBXL_API_VERSION
 #endif
 #endif
@@ -1108,9 +1115,22 @@ 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);
+                           const libxl_bitmap *cpumap_hard,
+                           const libxl_bitmap *cpumap_soft);
 int libxl_set_vcpuaffinity_all(libxl_ctx *ctx, uint32_t domid,
-                               unsigned int max_vcpus, libxl_bitmap *cpumap);
+                               unsigned int max_vcpus,
+                               const libxl_bitmap *cpumap_hard,
+                               const libxl_bitmap *cpumap_soft);
+
+#if defined (LIBXL_API_VERSION) && LIBXL_API_VERSION < 0x040500
+
+#define libxl_set_vcpuaffinity(ctx, domid, vcpuid, map) \
+    libxl_set_vcpuaffinity((ctx), (domid), (vcpuid), (map), NULL)
+#define libxl_set_vcpuaffinity_all(ctx, domid, max_vcpus, map) \
+    libxl_set_vcpuaffinity_all((ctx), (domid), (max_vcpus), (map), NULL)
+
+#endif
+
 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,
diff --git a/tools/libxl/libxl_dom.c b/tools/libxl/libxl_dom.c
index 661999c..a90a8d5 100644
--- a/tools/libxl/libxl_dom.c
+++ b/tools/libxl/libxl_dom.c
@@ -261,7 +261,8 @@ int libxl__build_pre(libxl__gc *gc, uint32_t domid,
             return rc;
     }
     libxl_domain_set_nodeaffinity(ctx, domid, &info->nodemap);
-    libxl_set_vcpuaffinity_all(ctx, domid, info->max_vcpus, &info->cpumap);
+    libxl_set_vcpuaffinity_all(ctx, domid, info->max_vcpus,
+                               &info->cpumap, NULL);
 
     if (xc_domain_setmaxmem(ctx->xch, domid, info->target_memkb +
         LIBXL_MAXMEM_CONSTANT) < 0) {
diff --git a/tools/libxl/libxl_types.idl b/tools/libxl/libxl_types.idl
index f0f6e34..2da58e2 100644
--- a/tools/libxl/libxl_types.idl
+++ b/tools/libxl/libxl_types.idl
@@ -514,7 +514,8 @@ 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", [
diff --git a/tools/libxl/libxl_utils.h b/tools/libxl/libxl_utils.h
index 8bfb81b..13b42a1 100644
--- a/tools/libxl/libxl_utils.h
+++ b/tools/libxl/libxl_utils.h
@@ -101,6 +101,31 @@ 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))
 
+/*
+ * Compares two bitmaps bit by bit, up to nr_bits or, if nr_bits is 0, up
+ * to the size of the largest bitmap. If sizes does not match, bits past the
+ * of a bitmap are considered as being 0, which matches with the semantic and
+ * implementation of libxl_bitmap_test I think().
+ *
+ * So, basically, [0,1,0] and [0,1] are considered equal, while [0,1,1] and
+ * [0,1] are different.
+ */
+static inline int libxl_bitmap_equal(const libxl_bitmap *ba,
+                                     const libxl_bitmap *bb,
+                                     int nr_bits)
+{
+    int i;
+
+    if (nr_bits == 0)
+        nr_bits = ba->size > bb->size ? ba->size * 8 : bb->size * 8;
+
+    for (i = 0; i < nr_bits; i++) {
+        if (libxl_bitmap_test(ba, i) != libxl_bitmap_test(bb, i))
+            return 0;
+    }
+    return 1;
+}
+
 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);
diff --git a/tools/libxl/xl_cmdimpl.c b/tools/libxl/xl_cmdimpl.c
index 64a1c77..860ec04 100644
--- a/tools/libxl/xl_cmdimpl.c
+++ b/tools/libxl/xl_cmdimpl.c
@@ -2232,7 +2232,7 @@ start:
             } else {
                 libxl_bitmap_set_any(&vcpu_cpumap);
             }
-            if (libxl_set_vcpuaffinity(ctx, domid, i, &vcpu_cpumap)) {
+            if (libxl_set_vcpuaffinity(ctx, domid, i, &vcpu_cpumap, NULL)) {
                 fprintf(stderr, "setting affinity failed on vcpu `%d'.\n", i);
                 libxl_bitmap_dispose(&vcpu_cpumap);
                 free(vcpu_to_pcpu);
@@ -4663,7 +4663,7 @@ static int vcpupin(uint32_t domid, const char *vcpu, char *cpu)
     }
 
     if (vcpuid != -1) {
-        if (libxl_set_vcpuaffinity(ctx, domid, vcpuid, &cpumap) == -1) {
+        if (libxl_set_vcpuaffinity(ctx, domid, vcpuid, &cpumap, NULL)) {
             fprintf(stderr, "Could not set affinity for vcpu `%u'.\n", vcpuid);
             goto out;
         }
@@ -4675,7 +4675,7 @@ static int vcpupin(uint32_t domid, const char *vcpu, char *cpu)
         }
         for (i = 0; i < nb_vcpu; i++) {
             if (libxl_set_vcpuaffinity(ctx, domid, vcpuinfo[i].vcpuid,
-                                       &cpumap) == -1) {
+                                       &cpumap, NULL)) {
                 fprintf(stderr, "libxl_set_vcpuaffinity failed"
                                 " on vcpu `%u'.\n", vcpuinfo[i].vcpuid);
             }

  parent reply	other threads:[~2014-06-13 13:10 UTC|newest]

Thread overview: 25+ messages / expand[flat|nested]  mbox.gz  Atom feed  top
2014-06-13 13:09 [PATCH v8 00/13] Implement vcpu soft affinity for credit1 Dario Faggioli
2014-06-13 13:09 ` [PATCH v8 01/13] xen: sched: rename v->cpu_affinity into v->cpu_hard_affinity Dario Faggioli
2014-06-13 13:09 ` [PATCH v8 02/13] xen: sched: introduce soft-affinity and use it instead d->node-affinity Dario Faggioli
2014-06-13 13:09 ` [PATCH v8 03/13] xen: derive NUMA node affinity from hard and soft CPU affinity Dario Faggioli
2014-06-13 13:09 ` [PATCH v8 04/13] xen/libxc: sched: DOMCTL_*vcpuaffinity works with hard and soft affinity Dario Faggioli
2014-06-13 13:09 ` [PATCH v8 05/13] libxc/libxl: bump library SONAMEs Dario Faggioli
2014-06-13 13:09 ` [PATCH v8 06/13] libxc: get and set soft and hard affinity Dario Faggioli
2014-06-13 13:10 ` Dario Faggioli [this message]
2014-06-13 13:10 ` [PATCH v8 08/13] xl: enable getting and setting soft affinity Dario Faggioli
2014-06-13 13:10 ` [PATCH v8 09/13] libxl/xl: push VCPU affinity pinning down to libxl Dario Faggioli
2014-06-13 13:25   ` Wei Liu
2014-06-17 10:09     ` Dario Faggioli
2014-06-13 13:10 ` [PATCH v8 10/13] libxl/xl: deprecate the build_info->cpumap field Dario Faggioli
2014-06-13 13:34   ` Wei Liu
2014-06-13 13:38     ` Ian Campbell
2014-06-17 10:11       ` Dario Faggioli
2014-06-13 13:49     ` Wei Liu
2014-06-17  9:59       ` Dario Faggioli
2014-06-17 10:16   ` Wei Liu
2014-06-17 10:29     ` Dario Faggioli
2014-06-17 10:45       ` Wei Liu
2014-06-13 13:10 ` [PATCH v8 11/13] xl: move the vcpu affinity parsing in a function Dario Faggioli
2014-06-13 13:10 ` [PATCH v8 12/13] xl: enable for specifying soft-affinity in the config file Dario Faggioli
2014-06-17 10:34   ` Wei Liu
2014-06-13 13:10 ` [PATCH v8 13/13] libxl: automatic NUMA placement affects soft affinity Dario Faggioli

Reply instructions:

You may reply publicly to this message via plain-text email
using any one of the following methods:

* Save the following mbox file, import it into your mail client,
  and reply-to-all from there: mbox

  Avoid top-posting and favor interleaved quoting:
  https://en.wikipedia.org/wiki/Posting_style#Interleaved_style

* Reply using the --to, --cc, and --in-reply-to
  switches of git-send-email(1):

  git send-email \
    --in-reply-to=20140613131003.4106.88556.stgit@Solace \
    --to=dario.faggioli@citrix.com \
    --cc=Andrew.Cooper3@citrix.com \
    --cc=George.Dunlap@citrix.com \
    --cc=Ian.Campbell@citrix.com \
    --cc=Ian.Jackson@citrix.com \
    --cc=JBeulich@suse.com \
    --cc=keir@xen.org \
    --cc=wei.liu2@citrix.com \
    --cc=xen-devel@lists.xen.org \
    /path/to/YOUR_REPLY

  https://kernel.org/pub/software/scm/git/docs/git-send-email.html

* If your mail client supports setting the In-Reply-To header
  via mailto: links, try the mailto: link
Be sure your reply has a Subject: header at the top and a blank line before the message body.
This is an external index of several public inboxes,
see mirroring instructions on how to clone and mirror
all data and code used by this external index.