[libvirt] [PATCH] fix size parameter in calls to numa_node_to_cpus

[Jim Meyering]

Dave Allan <dallan at redhat.com> wrote:
> I was getting the error:
>
> map size mismatch; abort
>
> when running the daemon-conf tests.
>
> I traced it to two places in which we were passing a parameter as a
> number of bits to numa_node_to_cpus which expects a number of bytes
> (see numa_node_to_cpus_compat in numa.h in the numactl source).
>
> The attached patch fixes the problem.
>
> Dave
> diff --git a/src/qemu_conf.c b/src/qemu_conf.c
> index e6c378f..7019c14 100644
> --- a/src/qemu_conf.c
> +++ b/src/qemu_conf.c
> @@ -324,7 +324,7 @@ qemudCapsInitNUMA(virCapsPtr caps)
>
>      for (n = 0 ; n <= numa_max_node() ; n++) {
>
> -        if (numa_node_to_cpus(n, mask, MAX_CPUS_MASK_LEN) < 0)
> +        if (numa_node_to_cpus(n, mask, MAX_CPUS_MASK_LEN / 8) < 0)

Hi Dave,

[Sorry I didn't catch this earlier. ]

With that patch, I am now seeing that out of memory message in
"make check" test failures, e.g.,

    testing with corrupted config: min_workers
    --- expected-err 2008-12-12 15:30:20.000000000 +0100
    +++ err 2008-12-12 15:30:20.000000000 +0100
    @@ -1,2 +1,5 @@
    +map size mismatch; abort
    +: Success
    +umlStartup: out of memory
     remoteReadConfigFile: in23.conf: min_workers: invalid type: got string; expected long

Since we're using NUMA_VERSION1_COMPATIBILITY mode (anyone know why?)

  #if HAVE_NUMACTL
  #define NUMA_VERSION1_COMPATIBILITY 1
  #include <numa.h>
  #endif

numa_node_to_cpus expands to a call to numa_node_to_cpus_compat,
which is defined in numa.h:

    static inline int numa_node_to_cpus_compat(int node, unsigned long *buffer,
                                                            int buffer_len)
    {
            struct bitmask tmp;

            tmp.maskp = (unsigned long *)buffer;
            tmp.size = buffer_len * 8;
            return numa_node_to_cpus(node, &tmp);
    }


So the question is, what units to use in "tmp.size".
To answer that, I looked above, also in numa.h at this:

    struct bitmask {
            unsigned long size; /* number of bits in the map */
            unsigned long *maskp;
    };

And since MAX_CPUS_MASK_LEN is a buffer length, in bytes,

    from qemu_conf.c:
    qemudCapsInitNUMA(virCapsPtr caps)
    {
    ...
        if (VIR_ALLOC_N(mask, MAX_CPUS_MASK_LEN) < 0)
            goto cleanup;

and tmp.size is measured in bits, we have to continue to use that
MAX_CPUS_MASK_LEN parameter, and not divide it by 8:

So I propose this change:
(with it, "make check" does pass once again)

>From ceae19496cd60bd9a7b3527046b42e9f8460ae82 Mon Sep 17 00:00:00 2001
From: Jim Meyering <meyering at redhat.com>
Date: Fri, 12 Dec 2008 16:22:34 +0100
Subject: [PATCH] Revert "fix bits/bytes memory request mismatches"

* src/qemu_conf.c (qemudCapsInitNUMA): Do pass MAX_CPUS_MASK_LEN
as buffer length in call to numa_node_to_cpus.
* src/uml_conf.c (umlCapsInitNUMA): Likewise.
---
 src/qemu_conf.c |    2 +-
 src/uml_conf.c  |    2 +-
 2 files changed, 2 insertions(+), 2 deletions(-)

diff --git a/src/qemu_conf.c b/src/qemu_conf.c
index 59171e7..206fb0b 100644
--- a/src/qemu_conf.c
+++ b/src/qemu_conf.c
@@ -324,7 +324,7 @@ qemudCapsInitNUMA(virCapsPtr caps)

     for (n = 0 ; n <= numa_max_node() ; n++) {

-        if (numa_node_to_cpus(n, mask, MAX_CPUS_MASK_LEN / 8) < 0)
+        if (numa_node_to_cpus(n, mask, MAX_CPUS_MASK_LEN) < 0)
             goto cleanup;

         for (ncpus = 0, i = 0 ; i < MAX_CPUS ; i++)
diff --git a/src/uml_conf.c b/src/uml_conf.c
index 3659c6b..7eb630d 100644
--- a/src/uml_conf.c
+++ b/src/uml_conf.c
@@ -80,7 +80,7 @@ umlCapsInitNUMA(virCapsPtr caps)

     for (n = 0 ; n <= numa_max_node() ; n++) {

-        if (numa_node_to_cpus(n, mask, MAX_CPUS_MASK_LEN / 8) < 0)
+        if (numa_node_to_cpus(n, mask, MAX_CPUS_MASK_LEN) < 0)
             goto cleanup;

         for (ncpus = 0, i = 0 ; i < MAX_CPUS ; i++)
--
1.6.0.4.1044.g77718