mld(3)mld(3)NAME
mld_create, mld_destroy, process_mldlink, process_cpulink, numa_acreate -
memory locality domain operations
SYNOPSIS
#include <sys/types.h>
#include <sys/pmo.h>
pmo_handle_t mld_create(int radius, long size)
int mld_destroy(pmo_handle_t mld_handle)
int process_mldlink(pid_t pid, pmo_handle_t mld_handle, rqmode_t rqmode)
int process_cpulink(pid_t pid, pmo_handle_t mld_handle, cpuid_t lcpuid, rqmode_t rqmode)
void* numa_acreate(pmo_handle_t mld, size_t arena_size, size_t page_size)
DESCRIPTION
mld_create creates a memory locality domain. The argument radius is not
currently used and is normally zero. The argument size is a hint
specifying approximately how much physical memory in bytes will be
required for this MLD. If this argument is set to MLD_DEFAULTSIZE then at
the time of mld placement the kernel will make an estimate of the memory
to be used based on the current total address space in use by the process
and the number of mlds.
mld_destroy destroys a memory locality domain.
process_mldlink attaches a process to an MLD. When a process is attached
to an MLD, the process scheduler tries to activate the process on a CPU
in the node where the MLD has been placed. The argument rqmode is
currently not used and is normally set to RQMODE_ADVISORY. If this
process forks or sprocs a child pid, the child pid does not inherit the
affinity to the node where the MLD is placed. Instead the current default
policy module for the MEM_DATA address space determines the affinity of
the child pid to a node.
process_cpulink attaches a process to a specific CPU within the node that
the MLD is placed on. The CPU number is relative to the node, for example
on an Origin 3000 with a fully populated node the CPU numbering is 0-3.
On an Origin 2000 with a fully populated node, the CPU numbering is 0-1.
If a CPU is disabled on that node, then the CPU numbering is re-ordered
at boot time to always count up consecutively from 0. For example if an
Origin 3000 had a node with only 3 CPUs, the CPU numbering regardless of
which CPU is disabled would be 0,1,2. This operation forces a process to
always run on the same CPU using `MP_MUSTRUN' (see sysmp(2)). The
argument rqmode is currently not used and is normally set to
RQMODE_ADVISORY. If this process forks or sprocs a child pid, the child
pid will inherit the affinity characteristics of the parent pid, per the
actions of the `MP_MUSTRUN' sysmp call.
Page 1
mld(3)mld(3)
numa_acreate creates an arena for memory allocation on a specific MLD.
mld_create will fail and a MLD will not be created if one or more of the
following are true:
EFAULT Arguments could not be copied into kernel space.
ENOMEM No memory or kernel table space is available to create and
hold the MLD.
ENOTSUP Memory locality domains are not supported on this system
architecture.
mld_destroy will fail and no MLD will be destroyed if one or more of the
following are true:
EINVAL The mld_handle is invalid.
ENOTSUP Memory locality domains are not supported on this system
architecture.
process_mldlink will fail and the process will not be attached to an MLD
if one or more of the following are true:
EDEADLK The operation would cause a deadlock to occur.
EINVAL The mld_handle is invalid or does not exist.
ENOTSUP Memory locality domains are not supported on this system
architecture.
ESRCH The specified pid does not exist.
EXDEV The MLD has not been placed. (Call mldset_place(3c)
first.)
process_cpulink will fail and the process will not be attached to an MLD
if one or more of the following are true:
EDEADLK The operation would cause a deadlock to occur.
EINVAL The specified mld_handle is invalid or does not exist.
EINVAL The cpulink is attempted on a CPU which was not present.
EINVAL The pid is part of a Pthreaded application, which is not
supported by MP_MUSTRUN.
EINVAL The process is already resides on a 'mustrun' CPU and is
not allowed to move to another CPU.
Page 2
mld(3)mld(3)
ENOTSUP Memory locality domains are not supported on this system
architecture.
EPERM The user does not have adequate permission to execute the
sysmp(2) system call MP_MUSTRUN command.
EPERM The cpuset does not allow the process to run on the
selected MLD and CPU.
ESRCH The specified pid does not exist.
EXDEV The MLD has not been placed. (Call mldset_place(3c)
first.)
numa_acreate will fail and no memory arena will be created (and set
errno) under the same circumstances that acreate(2) will fail. It will
also fail and no memory arena will be created if one or more of the
following are true:
EFAULT Arguments could not be copied into kernel space.
EINVAL The specified MLD is invalid.
ENOTSUP Memory locality domains are not supported on this system
architecture.
ESRCH The specified MLD does not exist.
SEE ALSOacreate(2), numa(5), mmci(5), mldset(3c), pm(3c), amalloc(3p), sysmp(2),
migration(3c), pminfo(3c).
DIAGNOSTICS
Upon successful completion, mld_create returns a value of type
pmo_handle_t. Otherwise, a value of -1 is returned and errno is set to
indicate the error.
Upon successful completion, mld_destroy, process_mldlink, and
process_cpulink return 0. Otherwise, a value of -1 is returned and errno
is set to indicate the error.
Upon successful completion, numa_acreate returns a pointer to the memory
arena which has been created. Otherwise NULL will be returned and errno
is set to indicate the error.
Page 3
