Changes

linux: don't leak a DIR on failure to list NUMA nodes
Signed-off-by: Brice Goglin <Brice.Goglin@inria.fr>

bgq: make sure the allowed_set is properly initialized before filling
it
It worked so far because the core allocates it with alloc_full() but it
could break if we ever call the get_allowed_resources() callback again
later (which may happen in the future API).
Signed-off-by: Brice Goglin <Brice.Goglin@inria.fr>

linux: clarify the semantics of read_fd_as_cpulist()
Don't require a full bitmap on input, do it ourself. Specify that
output bitmap is undefined on error.
Signed-off-by: Brice Goglin <Brice.Goglin@inria.fr>

linux: clarify hwloc_admin_disable_set_from_cpuset()
Rename variables for clarity. Abstract-out
hwloc__read_path_as_cpulist(). Add some messages.
Signed-off-by: Brice Goglin <Brice.Goglin@inria.fr>

linux: document why we don't use /sys/devices/system/cpu/kernel_max
Signed-off-by: Brice Goglin <Brice.Goglin@inria.fr>

linux: use /sys/devices/system/cpu/online when available
It's faster than reading one /sys/devices/system/cpu/cpu%d/online file
per PU.
Also it works when PU#0 is offline, which isn't supposed to happen in
Linux
(there's usually no /sys/devices/system/cpu/cpu0/online) but actually
happens in mOS for now.
Signed-off-by: Brice Goglin <Brice.Goglin@inria.fr>

linux: use /sys/devices/system/node/possible for memory binding
Use this mask to initialize the lookup for get_mempolicy mask required
size.
Signed-off-by: Brice Goglin <Brice.Goglin@inria.fr>

linux: use /sys/devices/system/node/possible for listing NUMA nodes
during early discovery
Instead of listing /sys/devices/system/node/ manually first.
By the way, add assertions about the number of nodes.
Signed-off-by: Brice Goglin <Brice.Goglin@inria.fr>

linux: fix some scan-build warnings
Signed-off-by: Brice Goglin <Brice.Goglin@inria.fr>

x86: fix management of failure to reallocate the cache array
Signed-off-by: Brice Goglin <Brice.Goglin@inria.fr>

x86: clarify how we manage non-unique APIC ids
Signed-off-by: Brice Goglin <Brice.Goglin@inria.fr>

xml: set import state.global to NULL when it's not used (diff import)
Signed-off-by: Brice Goglin <Brice.Goglin@inria.fr>

xml/nolibxml: use the right tagname on close
Not strictly required since the close_tag() callback ignores it.
Signed-off-by: Brice Goglin <Brice.Goglin@inria.fr>

xml/nolibxml: better handle parsing failure at the beginning of XML
diffs
Signed-off-by: Brice Goglin <Brice.Goglin@inria.fr>

xml: fix a leak on failure to import a v1 distance
Signed-off-by: Brice Goglin <Brice.Goglin@inria.fr>

xml/nolibxml: fix a leak a failure to pass diff headers after refname
Signed-off-by: Brice Goglin <Brice.Goglin@inria.fr>