On Tue, Mar 14, 2017 at 02:18:48PM +0100, Olaf Hering wrote:
Am Tue, 14 Mar 2017 14:05:37 +0100
schrieb Martin Kletzander <mklet...@redhat.com>:

However without that commit we'll be back to broken code that won't compile on 
some distros.  So unless I missed something (if I did, it should be added to 
the commit message), there should rather be a configure check that defines 
something similar to other code, e.g. HAVE_PERF_CPU_CYCLES and then that should 
be used in the conditional.

There are other places in that and other files which can not be compiled. The 
obvious fix is to just add a #ifndef x;#define x 0 to get libvirt going. The 
question is still unanswered if configure should just error out and refuse to 
go further, or if the few missing pieces should be supplied manually because 
the affected functionality is optional.


if it is optional, it should not error out, but disable that part of the
code (unless it would change the behaviour).

See attached patches for how I deal with it.

Olaf

--- a/src/util/virfile.c
+++ b/src/util/virfile.c
@@ -3623,6 +3623,9 @@ virFileBindMountDevice(const char *src,
    return 0;
}

+#ifndef MS_MOVE
+#define MS_MOVE              8192
+#endif


We need this for qemu's mount namespace.  Functions in this file using
this could error out if MS_MOVE doesn't exist.  The caller can then just
decide to disable namespaces then (it can have different return value
for this kind of error).  I really don't like redefining constants from
somewhere else just to make the code compile.

int
virFileMoveMount(const char *src,
--- a/src/util/virprocess.c
+++ b/src/util/virprocess.c
@@ -1156,6 +1156,13 @@ virProcessRunInMountNamespace(pid_t pid,


#if defined(HAVE_SYS_MOUNT_H) && defined(HAVE_UNSHARE)
+#ifndef MS_REC
+# define MS_REC          16384
+#endif
+
+#ifndef MS_SLAVE
+# define MS_SLAVE                (1<<19)
+#endif

Same here.  I know you that this way you can compile it on older kernel
and then upgrade and it will magically work, but that's not very nice
solution.

int
virProcessSetupPrivateMountNS(void)
{

--- a/src/util/virperf.c
+++ b/src/util/virperf.c
@@ -76,6 +76,12 @@ struct virPerfEventAttr {
    unsigned long long attrConfig;
};

+#ifndef PERF_ATTR_SIZE_VER2
+#define PERF_COUNT_HW_STALLED_CYCLES_FRONTEND   7
+#define PERF_COUNT_HW_STALLED_CYCLES_BACKEND    8
+#define PERF_COUNT_HW_REF_CPU_CYCLES            9
+#endif
+

These, IMHO, should be set to 0s and the code should be fixed to handle
0s as unsupported (as hinted in my previous mail).

static struct virPerfEventAttr attrs[] = {
    {.type = VIR_PERF_EVENT_CMT, .attrType = 0, .attrConfig = 1},
    {.type = VIR_PERF_EVENT_MBMT, .attrType = 0, .attrConfig = 2},

Attachment: signature.asc
Description: Digital signature

--
libvir-list mailing list
libvir-list@redhat.com
https://www.redhat.com/mailman/listinfo/libvir-list

Reply via email to