On Alpha and Integrity server systems, allows modification of the
CPU affinity set for a specified kernel thread.
This service accepts 64-bit addresses.
Format
SYS$PROCESS_AFFINITY [pidadr], [prcnam], [select_mask],
[modify_mask], [prev_mask], [flags]
[,[mask_length]]
C Prototype
int sys$process_affinity (unsigned int *pidadr, void *prcnam,
struct _generic_64 *select_mask,
struct _generic_64 *modify_mask,
struct _generic_64 *prev_mask, struct
_generic_64 *flags,...);
1 – Arguments
pidadr
OpenVMS usage:process_id
type: longword (unsigned)
access: read only
mechanism: by 32- or 64-bit reference
Process identification (PID) of a kernel thread whose affinity
mask is to be modified or returned. The pidadr argument is the
32- or 64-bit address of a longword that contains the PID.
Process selection is made through a combination of the pidadr
and prcnam arguments. If neither are specified or if both have a
zero value, the service operations are made to the user affinity
mask of the current kernel thread of the calling process. The
pidadr argument takes precedence over the prcnam argument in any
circumstances where both are supplied in the service call.
prcnam
OpenVMS usage:process_name
type: character-coded text string
access: read only
mechanism: by 32- or 64-bit descriptor-fixed-length string
descriptor
Process name of the process whose affinity mask is to be modified
or returned. The prcnam argument is the 32- or 64-bit address
of a character string descriptor pointing to the process name
string. A process can be identified with a 1- to 15-character
string. The service operations are made to the user affinity mask
of the initial thread of the specified process.
If pidadr and prcnam are both specified, then pidadr is modified
or returned and prcnam is ignored. If neither argument is
specified, then the context of the current kernel thread of the
calling process is modified or returned.
select_mask
OpenVMS usage:bitmap
type: quadword bitmap
access: read only
mechanism: by 32- or 64-bit reference
The select-mask argument specifies which bits of the specified
process's affinity mask are to be modified. The select_mask
argument is the 32- or 64-bit address of a quadword bit vector
wherein a bit, when set, specifies that the corresponding CPU
position in the mask is to be modified.
modify_mask
OpenVMS usage:bitmap
type: quadword bitmap
access: read only
mechanism: by 32- or 64-bit reference
Mask specifying the settings for those explicit affinities
selected in the select_mask argument. The modify_mask argument is
the 32- or 64-bit address of a quadword bit vector wherein a bit,
when set, specifies that the corresponding CPU is to be added to
the specified process affinity set; when clear, the corresponding
CPU is to be removed from the specified process affinity set.
To add a specific CPU to the affinity mask set, that bit position
must be set in both select_mask and modify_mask. To remove a
specific CPU from the affinity mask set, that bit position must
be set in select_mask and clear in modify_mask.
The constant CAP$K_ALL_CPU_ADD, when specified in modify_mask,
indicates that all CPUs specified in select_mask are to be added
to the affinity mask set. The constant CAP$K_ALL_CPU_REMOVE
indicates that all CPUs in select_mask are to be removed from
the affinity mask set.
prev_mask
OpenVMS usage:bitmap
type: quadword bitmap
access: write only
mechanism: by 32- or 64-bit reference
Previous CPU affinity mask for the specified kernel thread before
execution of this call to $PROCESS_AFFINITY. The prev_mask
argument is the 32- or 64-bit address of a quadword into which
$PROCESS_AFFINITY writes the previous explicit affinity bitmap.
flags
OpenVMS usage:mask_quadword
type: quadword (unsigned)
access: read only
mechanism: by 32- or 64-bit reference
Options selected for affinity modification. The flags argument
is a quadword bit vector wherein a bit corresponds to an option.
Only the bits specified below are used; the remainder of the
quadword bits are reserved and must be 0.
Each option (bit) has a symbolic name, which the $CAPDEF macro
defines. The flags argument is constructed by performing a
logical OR operation using the symbolic names of each desired
option.
The following table describes the symbolic name of each option:
Symbolic Name Description
CAP$M_FLAG_PERMANENT Indicates whether to modify the permanent
process affinities in addition to the
current image copy. If CAP$M_FLAG_
PERMANENT is set, then both the permanent
and current affinities are modified.
If the flag bit is clear or flags is
unspecified, then just the current image
process affinities are modified.
This bit also determines which of the
affinity masks are returned in prev_
mask. If set, the permanent mask, used
to reinitialize the current set at image
rundown, is returned. If the bit is clear
or the flags argument is not specified,
the current running mask is returned.
CAP$M_FLAG_CHECK_CPU Determines whether the kernel thread
can be left in a nonrunnable state under
some circumstances. No operation of this
service will allow a transition from a
runnable to blocked state; however, if
the kernel thread is already at a blocked
state, this bit determines whether the
result of the operation must leave it
runnable. If CAP$M_FLAG_CHECK_CPU is set
or flags is unspecified, the kernel thread
will be checked to ensure it can safely
run on one of the CPUs in the active set;
otherwise, any valid state operations on
kernel threads already in a blocked state
will be allowed.
CAP$M_FLAG_CHECK_CPU_ Indicates whether a check is made to
ACTIVE verify that all CPUs in the select mask
that are about to be selected for affinity
binding are in the active set. This does
not apply to CPUs that are about to be
cleared from the current affinity set.
Unlike CAP$M_FLAG_CHECK_CPU where only
a single CPU has to be valid for the
condition to pass, CAP$M_FLAG_CHECK_
CPU_ACTIVE requires that all CPUs in the
selected set must pass the criteria.
CAP$M_PURGE_WS_IF_ Causes the working set of the process
NEW_RAD to be purged if the choice of affinity
results in a change to the home RAD of the
process.
mask_length
OpenVMS usage:bitmap
type: quadword bitmap
access: read only
mechanism: by 32- or 64-bit reference
The mask_length specifies the length in bytes of each of the
three bitmaps: select_mask, modify_mask, prev_mask. If mask_
length is not supplied or specified as zero, a length of 8 bytes
is used.
The correct value for mask_length is determined by the number of
supported CPUs on the system. You can compute the number of bytes
needed for the bitmap as follows: Use the $GETSYI system service
with an item code of SYI$_MAX_CPUS to find the minimum number of
bits needed, round this number up to a multiple of 64, and divide
the result by 8.