|
Programming Interface Information:
The Monitor Records are intended Programming Interfaces that allow the customer to write programs to obtain services of z/VM. |
MRPRCSXL
PrologControl Block Contents
Cross Reference
MRPRCSXL Prolog
DSECT NAME - PRCSXL
FUNCTION - Map a Monitor record.
LOCATED BY -
Through the Monitor Control Area
REFERENCED CONTROL BLOCKS -
MRRECHDR Monitor Record Header
NAME - MRPRCSXL
DESCRIPTIVE NAME - Monitor Sample Record
Domain 5 - Processor Domain
Record 22 - Shared-Exclusive Spinlock Utilization
DESCRIPTION - Distributed shared-exclusive spinlock per-processor
data. These records are generated for any processor
that has been online any time since IPL, even if the
processor is currently offline.
List of locks with data in the record:
Lock ID Lock Description
-------- ----------------
SRMSLOCK: scheduler lock (distributed shared-exclusive)
The name SRMSLOCK is used to be consistent with
the name already used in MRSYTLCK(D0 R23). However,
SRMSLOCK is no longer the Scheduler Lock data structure.
It is now an SXL-style lock with the SXLXL located at
SRMSCHLK.
HCPDSVTL: topology lock (distributed shared-exclusive)
This record consists of two sections:
1) a header at the start of the record
2) an array of entries. One per spinlock listed above.
These contain the statistics for each of the locks
listed above.
The header includes the information needed to locate and
interpret the data in the other section. That is,
for the array, the header has fields that contain the
number of entries in the array, the size of an entry,
and the displacement in bytes from the beginning of this
record to the first entry in the array.
The array has entries for all distributed shared-exclusive
spinlocks. An entry in this array is described by
PRCSXL_CALLKDATA.
SHARED-EXCLUSIVE SPINLOCK STATISTICS:
This Monitor record provides explanatory information
about shared-exclusive spinlock usage and how that
contributes to changes in spinlock contention reported
in MRSYTLCK (D0 R23) for Scheduler and Topology locks.
The purpose and use of this record is similar to
MRSYTLCK (D0 R23), but this record provides more
detailed information.
Spin time for shared requests is reported separately for
standard unconditional requests, as well as for the
LOCKSEQ internal spin used in HCPDSP for the Scheduler
Lock. And that is reported separately for cancelled
spins and successful spins.
Increases in spinlock contention are often related to
the request rates, or the mix of request types. While
shared lock holds can run concurrently with each other,
exclusive holds block all requests regardless of type.
Therefore lock contention is more sensitive to increases
in exclusive request rates than share request rates.
While share requests can be blocked only when exclusive
is held, exclusive requests can be blocked when the lock
is held exclusive or shared, or when a share-pending
condition is present. Counts and spin time for each of
these are reported separately for exclusive requests.
The count of shares for which exclusive requests had to
wait indicates how frequently share scans paused to wait
for a share to be released.
The share-pending statistics provide indications as to
how often share requests relied on the share-pending
setting to halt exclusive requests and allow share
requests to be processed.
The difference between the Diagnose 9C usage data reported
in this record and MRSYTLCK is that this record organizes
the data based on the type of target being considered
which is determined by the condition causing the spin.
SHARED-EXCLUSIVE SPINLOCK REQUEST TYPES:
Each spinlock is described by a set of data that explains
the usage of the lock. Because these locks are shared-
exclusive spinlocks, they may be acquired shared as well
as exclusive. Each acquisition type may be acquired by a
conditional, or an unconditional request method. For the
conditional requests, the lock is acquired only if it is
immediately available. For conditional share requests,
there is support for a LOCKSEQ parameter which is used
in specialized routines that perform their own lock spin
including all of the necessary checks for processor
failures, use of Diagnose x'9C', etc.
Conditional requests may be successful or unsuccessful
while unconditional requests always complete successfully
but may need to spin for a while before they complete.
Conditional requests never end up in the normal spin
routine if the lock isn't available. However, when the
invoker manages its own spin (LOCKSEQ case), then the
conditional request may spin, but not in the normal
spin routine. This special case is the only case where a
spin on a lock may be ended without acquiring the lock.
There are two additional lock request types. These
additional types require that the lock is held in one
mode and switch to the other. These types are EX2SHR,
which is a demote from exclusive to shared, and SHR2EX,
which is a promote from shared to exclusive. The demote
is always successful but promote is successful only if
the an exclusive hold is immediately available. Neither
of these request types ever spin.
LOCK ACQUISITION MECHANISMS:
Unlike the typical semaphore lock design, the shared-
exclusive spinlock design is a distributed lock design.
There is still a global lock word that records whether
the lock is held exclusive and who holds it, but that
word holds nothing about shared holds of the lock.
It does contain a flag for share-pending which is
described in the next section. Every processor has its
own processor-local cache line where it records whether
it holds a share of the lock.
A processor acquiring a share starts by marking itself
as holding a share in its local cache line and then
inspecting the global word to see if the lock is held
exclusive. If it isn't, then the share is held. If
exclusive is held and the share request was conditional,
it turns off the share in its local cache line and ends
the share request unsuccessfully. If exclusive was held
and the request was unconditional, the local share flag
is turned off to allow the exclusive request forward
progress, and then spins until exclusive is turned off.
When that happens it repeats the original process of
marking the share in the local cache line and then
checking exclusive again until it has observed exclusive
not held while it has its share set. Share requests
spin only because of exclusive requests.
An exclusive request begins by attempting to turn on
exclusive while observing share-pending and exclusive
are off. Conditional requests fail if this can't be done
on the inital attempt while unconditional requests spin
until successful. Once it sets exclusive it must then
inspect every processor's local memory area to ensure no
shares are held. If it encounters a processor that
holds a share then it waits until the share gets turned
off before checking the next share. The exclusive
acquisition is complete when all of the processors have
been observed as not holding a share. Exclusive
requests could spin when trying to start the exclusive
acquisition process because exclusive is set or because
share-pending is set. For the purposes of statistics
tracking, this spin is considered a spin on exclusive if
exclusive was observed on the last time it spun before a
successful acquisition because that was the last request
that blocked it. However, if it last observed only
share-pending on then it is considered a spin on share-
pending. After exclusive acquisition has started, it
may then spin on one or more shares as it checks all of
the processors. It is considered a spin on shares
unless none were observed held for any processor.
Conditional exclusive requests complete successfully
only if exclusive can be set while exclusive and share-
pending are off and no shares are observed held during
the scan of shares. If the conditional request
encounters a share, it turns off exclusive and
terminates the request.
EX2SHR (demote) requests simply set the share on in the
local memory area and turn off exclusive. The fact that
they held exclusive means no other exclusive request is
held or in SXLBK scan processing, and no shares are
actively held yet. No other checks are needed.
SHR2EX (promote) requests follow the same logic as a
conditional exclusive acquisition while leaving the
share set. If exclusive is acquired successfully,
it turns off the share and the promote is successful. If
the conditional exclusive acquisition was unsuccessful,
the share is left held and the promote is not successful.
SHARE STARVATION PROTECTION:
The shared-exclusive lock manager design gives priority
to exclusive requests in that once they make forward
progress in the exclusive acquisition, shares must back
out and wait for the exclusive to complete. Exclusive
requests wait for existing share holds to complete but
block new share requests. High exclusive usage rates
could potentially starve share requests. To prevent this,
share requests that have been delayed by exclusive
requests that exceed a threshold are allowed to use the
share-pending mechanism to force a window for share
acquisitions by causing new exclusive requests to wait
until that pending share request completes. While the
share-pending state is set, other processors can also
acquire shares allowing the share backlog to drain before
exclusive requests start again. The share-pending window
is closed when the processor that set share-pending
releases its share and turns off share-pending.
LOCAL LOCK SPIN REPORTING:
A deficiency in the old shared-exclusive spinlock
manager was that the time HCPDSP spent in its
internal spin loop for the Scheduler Lock was not
reported in SYTLCK_CALSTIME, SYTSCG_CALSLKTM, or
SYTPRP_PFXSPINT. This resulted in significant under-
reporting of Scheduler Lock contention. Those fields
now include the local spin time in HCPDSP. Further, this
Monitor record breaks out HCPDSP local spin statistics
separately. Because the new shared-exclusive spinlock
manager is much more cache-efficient, the total spin
times are expected to be significantly lower even with
this additional component of spin time included.
------------------------------------------------------------
TABLES OF STATISTICS REPORTED PER LOCK:
- Abbreviation and symbol descriptions follow the tables.
------------------------------------------------------------
REQUEST RATES, SPIN TIMES, SHARE-PENDING USAGE:
------------------------------------------------------------
| | | Shared |Exclusive| | |
------------------------------------------------------------
| | rm-> | SC | SL | SU | XC | XU | SX | XS |
------------------------------------------------------------
| Statistics | ssss | | | | | | | |
------------------------------------------------------------
| Req cts | | | | | | | | |
| Unsucc | CU | Y | Y | | Y | | Y | |
| Succ | | | | | | | | |
| No Spin | CN | Y | Y | Y | Y | Y | Y | Y |
| Spin (any) | CSA | | Y | Y | | Y | | |
| on X | CSX | | Y | Y | | Y | | |
| on S-P | CSP | | | | | Y | | |
| on S | CSS | | | | | Y | | |
------------------------------------------------------------
| Spin time | | | | | | | | |
| Unsucc | TSU | | Y | | | | | |
| Succ | | | | | | | | |
| Spin (any) | | | | | | | | |
| on X | TSX | | Y | Y | | Y | | |
| on S-P | TSP | | | | | Y | | |
| on S | TSS | | | | | Y | | |
------------------------------------------------------------
| Shr-Pnd | | | | | | | | |
| Set | PS | | Y | Y | | | | |
| Reset | PR | | Y | | | | | |
| Obs | PO | Y | Y | Y | | | | |
------------------------------------------------------------
| Shares Obs | SO | | | | | Y | | |
| Excl Obs | XO | | Y | Y | | Y | | |
------------------------------------------------------------
KEY TO ABBREVIATIONS IN THE TABLE ABOVE:
- field names are defined by rm and ssss in this format:
PRCSXL_SXLErmssss.
ROW HEADINGS FOR TABLE 1
- ROW 1
- Shared = Shared lock requests. These may be
unconditional, conditional with the caller doing a
local spin, or typical single-attempt conditional.
- Exclusive = Exclusive lock requests. These may be
either unconditional or conditional.
- ROW 2 (rm values)
- rm-> = Request type and Modifier values in the row
following this label indicate the rm value used in
the field name PRCSXL_SXLErmssss. The specific
rm values are described below.
- SC = DSPOLICY=RETURN was specified, but LOCKSEQ was
not. This is a conditional share request. A single
attempt is made to acquire the lock. If not
immediately available, the lock is not acquired.
- SL = DSPOLICY=RETURN was specified with LOCKSEQ.
Conditional, but the requestor does its own spin
which might not acquire a share if it cancels the
spin. This is used only for the Scheduler Lock
(SRMSCHLK, previously known as SRMSLOCK) in HCPDSP.
- SU = DSPOLICY=SPIN was specified. This is an
unconditional request for a share.
- XC = DSPOLICY=RETURN was specified. These are
conditional exclusive requests.
- XU = DSPOLICY=SPIN was specified. These are
unconditional exclusive requests.
- SX = SHR2EX request (promote). These are
conditional by definition.
- XS = EX2SHR request (demote). These are
unconditional by definition because they always
succeed, however never need to spin.
COLUMN HEADINGS FOR TABLE 1
- COLUMN 1
- Req cts = Lock Request counts
- Unsucc = Unsuccessful requests. These occur only
for conditional requests. For the
SHARE,DSPOLICY=RETURN,LOCKSEQ case this includes
only cases where the invoking routine cancelled its
local spin before the lock was acquired.
- Succ = Successful requests.
- No Spin = Lock requests that completed successfully
without spinning.
- Spin (any) = requests that spun for any reason.
This count is incremented only once for requests that
spun for multiple different reasons. Spin time does
not use this category because time is allocated to
specific spin reasons. Subcategories follow for the
specific causes.
- on X = requests that spun because eXclusive was held.
- on S-P = requests that spun because a Share-Pending
condition was present.
- on S = requests that spun because a Share was held.
- Shr-Pnd = Share-Pending - a mechanism used by share
requests to prevent share starvation. The statistics
in this category describe behavior of requests that
exploit share-pending.
- Set = share-pending was Set during lock acquisition.
It is set only during share requests that spin.
Although LOCKSEQ is specified with DSPOLICY=RETURN,
the invoker manages its own spin and may use share-
pending. In this case it may also reset it if the
spin is cancelled before the share is acquired.
- Reset = share-pending was turned off (Reset) during
lock acquisition because the requestor stopped
waiting to acquire the lock. It can be reset only by
the same lock request that set it. This occurs only in
cases where the LOCKSEQ spin in HCPDSP was cancelled.
- Obs = share-pending was Observed set at the time the
share was acquired, but was not set by this requestor.
- Shares Obs = The number of Shares that were Observed
held as Exclusive requests scanned processors for
shares held.
- Excl Obs = The number of exclusive requests that were
processed during the spin for other lock requests.
It is updated only for requests that acquire the lock.
- COLUMN 2 (ssss values)
- rm-> = see the description for row 2 above.
- ssss = The values in column 2 after this row are the
ssss values used in the field name PRCSXL_SXLErmssss.
The value indicates the type of statistic this field
contains. The statistic is described above in the
corresponding column 1 entry.
------------------------------------------------------------
DIAGNOSE X'9C' USAGE:
--------------------------------------
| | Shared | Exclusive |
--------------------------------------
| rm-> | SL | SU | XU | XU | XU |
--------------------------------------
| b-> | X | X | X | P | S |
--------------------------------------
| ss | | | | | |
--------------------------------------
| 9T | Y | Y | Y | Y | Y |
| 9S | Y | Y | Y | Y | Y |
| 9F | Y | Y | Y | Y | Y |
| 99 | Y | Y | Y | Y | Y |
| 94 | Y | Y | Y | Y | Y |
--------------------------------------
KEY TO ABBREVIATIONS IN THE TABLE ABOVE:
- field names in the main section of the lock entry
that are mapped by SXLEX are of this format:
PRCSXL_SXLEXrmb
- field names in the SXLEX are of this format:
PRCSXL_SXLEXss
ROW HEADINGS
- ROW 1
- Shared = Shared lock requests. These may be
unconditional, conditional with the caller doing a
local spin, or typical single-attempt conditional.
- Exclusive = Exclusive lock requests. These may be
either unconditional or conditional.
- ROW 2 (rm values)
- SL = DSPOLICY=RETURN was specified with LOCKSEQ.
Conditional, but the requestor does its own spin
which might not acquire a share if it cancels the
spin. This is used only for the Scheduler Lock
(SRMSCHLK, previously known as SRMSLOCK) in HCPDSP.
- SU = DSPOLICY=SPIN was specified. This is an
unconditional request for a share.
- XU = DSPOLICY=SPIN was specified. These are
unconditional exclusive requests.
- ROW 3 (b values - conditions causing the spin)
- X = while spinning due to exclusive being held.
- P = while spinning due to share-pending condition.
- S = while spinning due to shares held.
COLUMN HEADINGS
- COLUMN 1 (ss values)
- 9T = The number of times attempts were made to find
a target for Diag 9C during lock spins.
- 9S = The number of times the state of a potential
Diagnose 9C target was checked during lock spins.
- 9F = The number of times a target was successfully
found during lock spins.
- 99 = The number of times a Diagnose 9C was issued
for a target identified during lock spins.
- 94 = The number of times a Diagnose 44 was issued
during lock spins.
------------------------------------------------------------
Block comments in the record below provide suggestions
for using the individual values to calculate higher level
statistics for evaluating shared-exclusive spinlock usage.
------------------------------------------------------------
NOTES:
The statistics reported in this record are preserved over
periods of time when the processor is varied offline so
the counts will appear monotonically increasing for the
life of the system.
All spin time is measured as elapsed time using the TOD
clock and is reported in TOD clock units.
The Scheduler Lock, which is one of the two locks reported
in this Monitor record, is usually the most frequently
acquired spinlock in the system. These detailed statistics
are defined specifically for the more complicated usage
of the Scheduler lock. Therefore some of the fields will
contain zeroes for other spinlocks. For example, LOCKSEQ
is not used with any other lock.
Definitions of record version numbers.
Version N+1 contains all the sections of data in version
N along with any new sections defined by the version N+1.
In effect the version number determines the structure of the
record header area where variable sections are described.
The version number is placed in the field PRCSXL_CALVERSN
in the header.
Versions:
PRCSXL_CALVERSN = 0
Version 0 includes basic spin lock data as defined by
PRCSXL_CALLKDATA.
MRPRCSXL Control Block Contents
Offsets Dec Hex Type Len Name (Dim) Description 0 0 Structure 68+ PRCSXL Start of monitor record 0 0 Character 0 PRCSXL_CALMRHDR Record header. See MRRECHDR for details. 0 0 Character 20 MRHDR 0 0 Unsigned 2 MRHDRLEN 2 2 Unsigned 2 MRHDRZER 4 4 Unsigned 1 MRHDRDM 5 5 Unsigned 1 * 6 6 Unsigned 2 MRHDRRC 8 8 Character 8 MRHDRTOD 16 10 Character 4 * 20 14 Character 0 MRHDR_END 20 14 Unsigned 4 PRCSXL_CALNMLKS Number of spin locks in this record. That is, the number of PRCSXL_CALLKDATA entries included after this header. 24 18 Unsigned 2 PRCSXL_CALENTSZ The size (in bytes) of each PRCSXL_CALLKDATA entry. 26 1A Unsigned 2 PRCSXL_CALENTDSP The displacement (in bytes) from the beginning of this record to the first PRCSXL_CALLKDATA entry. The entries are consecutive in storage starting with the first at this displacement. 28 1C Unsigned 1 PRCSXL_CALVERSN Record version number: See PRCSXL_VERnnn constant definition in the prolog above. 29 1D Unsigned 1 * Reserved for IBM use. 30 1E Unsigned 2 PRCSXL_SXLCPUAD Processor's CPU address. 32 20 Unsigned 1 PRCSXL_SXLCPUTY CPU Type. Valid values: X'00' - General Purpose (CP) X'02' - zSeries Application Assist (zAAP) X'03' - Integrated Facility for Linux (IFL) X'04' - Internal Coupling Facility (ICF) X'05' - zSeries Integrated Information (zIIP) 33 21 Unsigned 1 PRCSXL_SXLFLAGS Flags: X'80' - processor is offline. X'40' - processor was online at some point since IPL. 34 22 Unsigned 2 * Reserved for IBM use. 36 24 Character 8 PRCSXL_SXLTIME TOD value at the time of the transition to the current state which is either online or offline. 44 2C Character 8 PRCSXL_PFXPRBTM Total raw CPU time spent in emulation mode on this processor for all users. This field starts at x'7FFFFFFFF_00000000' and advances down (in the negative direction). 52 34 Character 8 PRCSXL_PFXUTIME Total raw CPU time spent on this processor which was charged to users. It includes both time spent running the guest and time that CP spends on behalf of a user doing such things as instruction simulation or page translation. This is in raw CPU time, not elapsed (or wall clock) time. This field starts at x'7FFFFFFFF_00000000' and advances down (in the negative direction). 60 3C Character 8 PRCSXL_PFXTMSYS Total raw CPU time spent on this processor which has been charged to the system instead of the user. In general, this is processing that is inappropriate to charge to any one particular user. Examples include portions of the following functions: scheduling decisions, handling interruptions, monitoring, and page migration. This field starts at x'7FFFFFFFF_00000000' and advances down (in the negative direction). 68 44 Character 0 PRCSXL_CALENTS Start of the array of lock data which contains the general spin lock data (PRCSXL_CALLKDATA) entries, one for each formal spin lock arranged one after the other at the end of this record. There is an ID field in each entry that should be used to determine which lock the entry represents. See the list of LOCKIDs above to determine which ID represents which lock.Mapping of an array entry for a shared-exclusive spinlock containing statistics for that spinlock. (See the list of locks in the prologue.)
Offsets Dec Hex Type Len Name (Dim) Description 0 0 Structure 424 PRCSXL_CALLKDATA Formal spin lock data entry. 0 0 Character 8 PRCSXL_CALLCKID IDentifier for this lock. 8 8 Bitstring 1 PRCSXL_SXLECTRL Control settings. 1... .... PRCSXL_SXLASSST Assist allow usage when 1 .1.. .... PRCSXL_SXLNIAI2 NIAI-2 allow usage when 1 ..1. .... PRCSXL_SXLNIAI4 NIAI-4 allow usage when 1 ...1 .... PRCSXL_SXLNIAI8 NIAI-8 allow usage when 1 .... 1... PRCSXL_SXLCAD CAD allow usage when 1. Only set if SXLPFD6 is also set. .... .1.. PRCSXL_SXLPFD6 PFD-6/7 used when 1. .... ..1. PRCSXL_SXLPPA14 PPA-14 used when 1. .... ...1 PRCSXL_SXLPPA15 PPA-15 used when 1. 9 9 Bitstring 1 PRCSXL_SXLEFLAGS Lock flags. 1... .... PRCSXL_SXLLKSEQ LOCKSEQ is exploited. .1.. .... * ..1. .... * ...1 .... * .... 1... * .... .1.. * .... ..1. * .... ...1 * 10 A Unsigned 2 * Reserved for IBM use. 12 C Unsigned 2 PRCSXL_SXLXLSQT Sequence number increase threshold for setting share-pending (SXLXLSPD) for unconditional share requests. 14 E Unsigned 2 PRCSXL_SXLXLSQTL Sequence number increase threshold for setting share-pending (SXLXLSPD) for LOCKSEQ invokers with local spin lock routines. 16 10 Character 8 * Reserved for IBM use.
REQUEST COUNTS:
Request counts are reported in mutually exclusive counts for each request type so that the sum of the individual counts indicates the overall request rate for that request type and method. For an rm value, the sum of the CU, CN and CSA values provided for that request type and method provides the total request count. The CSX, CSP and CSS values provide details for the reasons for the spin, but the sum of those is likely to be greater than the CSA value because an exclusive request might wait because of an exclusive hold, or share-pending, but then also wait because of shares. The total share requests is the sum of the total for the rm values SC, SL and SU. For exclusive, sum XC and XU. Various subtotals are meaningful and referred to below: Ctot - total requests. Ltot - total lock acquisitions (CN + CSA). Stot - total lock spins. Normally CSA, but for rm = SL, Stot = CU + CSA because the CU value represents the number of spins started, but cancelled before the lock was acquired. Comparing the relative percentages of the Ctot values across rm values provides indications of which request types and modifiers are most common.
Counts of unsuccessful conditional lock requests. These values are used to determine the failure rates of conditional lock requests. - for rm = SC, XC, and SX: %failure = 100*CU/Ctot - for rm = SL (internal spin loop might have been cancelled) %failure = 100*CU/Ctot
Offsets Dec Hex Type Len Name (Dim) Description 24 18 Character 8 PRCSXL_SXLESCCU Count of unsuccessful share requests with DSPOLICY=RETURN but LOCKSEQ was not specified. 32 20 Character 8 PRCSXL_SXLESLCU Count of unsuccessful share requests with DSPOLICY=RETURN and LOCKSEQ but the invoker cancelled the spin before the share was acquired. 40 28 Character 8 PRCSXL_SXLEXCCU Count of unsuccessful exclusive requests with DSPOLICY=RETURN. 48 30 Character 8 PRCSXL_SXLESXCU Count of unsuccessful SHR2EX requests (promote).
Counts of successful lock acquisitions not requiring a spin. Ltot(rm = SC, XC, XS and SX) = CN because those request types don't spin. These values are used with the spin counts to determine Ltot and collision rates for rm = SL, SU and XU which are request types that can spin. See the next section.
Offsets Dec Hex Type Len Name (Dim) Description 56 38 Character 8 PRCSXL_SXLESCCN Count of shares acquired by requests with DSPOLICY=RETURN but LOCKSEQ was not specified. 64 40 Character 8 PRCSXL_SXLESLCN Count of shares acquired by requests with DSPOLICY=RETURN and LOCKSEQ that did not require a spin in the invoker. 72 48 Character 8 PRCSXL_SXLESUCN Count of shares acquired by requests with DSPOLICY=RETURN that did not require a spin. 80 50 Character 8 PRCSXL_SXLEXCCN Count of exclusive acquisitions by requests with DSPOLICY=RETURN. 88 58 Character 8 PRCSXL_SXLEXUCN Count of exclusive acquisitions by requests with DSPOLICY=SPIN that did not require a spin. 96 60 Character 8 PRCSXL_SXLESXCN Count of SHR2EX requests that successfully promoted from a share to exclusive. 104 68 Character 8 PRCSXL_SXLEXSCN Count of EX2SHR requests that successfully demoted from a exclusive to share.
Counts of lock requests that spun trying to acquire the lock for any reason. Incremented once per lock acquisition that spins for one or multiple reasons. Ltot(rm = SL, SU, and XU) = CN + CSA because those request types spin. Lock collision rates as a percentage of total acquisitions 100*CSA/Ltot Lock collision rates as a percentage of total requests 100*Stot/Ctot
Offsets Dec Hex Type Len Name (Dim) Description 112 70 Character 8 PRCSXL_SXLESLCSA Count of shares acquired by requests with DSPOLICY=RETURN and LOCKSEQ that required a spin in the invoker for any reason. 120 78 Character 8 PRCSXL_SXLESUCSA Count of shares acquired by requests with DSPOLICY=SPIN that required a spin for any reason. 128 80 Character 8 PRCSXL_SXLEXUCSA Count of exclusive acquisitions by requests with DSPOLICY=SPIN that required a spin for any reason.
Counts of lock requests that spun for a specific reason during lock acquisition. Incremented once per request that spins for that specific reason. Collision rates for specific reasons can be calculated using these values in place of the CSA values.
Offsets Dec Hex Type Len Name (Dim) Description 136 88 Character 8 PRCSXL_SXLESLCSX Count of shares acquired by requests with DSPOLICY=RETURN and LOCKSEQ that required a spin in the invoker to wait for exclusive to be released. 144 90 Character 8 PRCSXL_SXLESUCSX Count of shares acquired by requests with DSPOLICY=SPIN that required a spin to wait for exclusive to be released. 152 98 Character 8 PRCSXL_SXLEXUCSX Count of exclusive acquisitions by requests with DSPOLICY=SPIN that required a spin to wait for exclusive to be released. 160 A0 Character 8 PRCSXL_SXLEXUCSP Count of exclusive acquisitions by requests with DSPOLICY=SPIN that required a spin to wait because share-pending was set. 168 A8 Character 8 PRCSXL_SXLEXUCSS Count of exclusive acquisitions by requests with DSPOLICY=SPIN that required a spin to wait for one or more shares to be released.
Spin times when the lock was not immediately available. These values accumulate time only when the request is spinning for the specified condition. STIMtot = Total spin time for all share requests is the sum of the TSX and TSU values for SL and SU rm values because it should include spin time for cancelled spins. STIMtot(rm = SL) = TSX + TSU STIMtot(rm = SU) = TSX STIMtot = Total spin time for exclusive acquisitions is SUM of the TSX, TSP and TSS values for the XU rm value. STIMtot(rm = XU) = TSX + TSP + TSS Average spin time = STIMtot/Stot. Spin % busy can be calculated from the STIMtot values. Average spin time and spin % busy can also be calculated for the individual spin conditions for exclusive.
Offsets Dec Hex Type Len Name (Dim) Description 176 B0 Character 8 PRCSXL_SXLESLTSU Time spent spinning on share requests with DSPOLICY=RETURN and LOCKSEQ after beginning a spin in the invoker but ending the spin before the share was acquired. 184 B8 Character 8 PRCSXL_SXLESLTSX Time spent spinning on share requests with DSPOLICY=RETURN and LOCKSEQ that required a spin in the invoker to wait for exclusive to be released. 192 C0 Character 8 PRCSXL_SXLESUTSX Time spent spinning on share requests with DSPOLICY=SPIN that required a spin to wait for exclusive to be released. 200 C8 Character 8 PRCSXL_SXLEXUTSX Portion of time acquiring exclusive with DSPOLICY=SPIN spent spinning waiting for an exclusive hold to be released. 208 D0 Character 8 PRCSXL_SXLEXUTSP Portion of time acquiring exclusive with DSPOLICY=SPIN spent spinning waiting for share-pending to be removed. 216 D8 Character 8 PRCSXL_SXLEXUTSS Portion of time acquiring exclusive with DSPOLICY=SPIN spent spinning waiting for shares to be released.
Share-pending usage statistics. In general, the higher the ratio of PO/PS, the better because it represents efficient usage. A ratio value near or below the number of processors is reasonable. When calculating the ratio, use the sum of the PO and sum of the PS values.
Offsets Dec Hex Type Len Name (Dim) Description 224 E0 Character 8 PRCSXL_SXLESUPS Count of times share-pending was set to acquire a share excluding LOCKSEQ spins. 232 E8 Character 8 PRCSXL_SXLESLPS Count of times share-pending was set to acquire a share during LOCKSEQ spin. 240 F0 Character 8 PRCSXL_SXLESLPR Count of times share-pending was removed in a LOCKSEQ spin after previously setting it without acquiring a share. 248 F8 Character 8 PRCSXL_SXLESCPO Count of times share-pending was set when acquiring a conditional share acquisition excluding LOCKSEQ spin. 256 100 Character 8 PRCSXL_SXLESLPO Count of times share-pending was observed when getting a share during a LOCKSEQ spin. 264 108 Character 8 PRCSXL_SXLESUPO Count of times share-pending was observed during an unconditional share acquisition.
Shares observed during exclusive acquisition. The ratio SO/CSS is the average number of shares for which an exclusive request had to wait while spinning on shares. The value will be greater than 1, but is not expected to be a lot higher because while the exclusive request is waiting for the first share, other processors are in the process of finishing up their holds as well.
Offsets Dec Hex Type Len Name (Dim) Description 272 110 Character 8 PRCSXL_SXLEXUSO Number of CPUs observed holding shares while attempting to acquire the lock exclusive.
Exclusive requests observed during lock acquisition. The ratio XO/CSX for an rm value is the average number of exclusive acquisitions by other processors that occur while the rm type requests are blocked by exclusive holds. The higher the ratios, the more likely the ratio CSX/Ltot is high. A high CSX/Ltot ratio would be closer to 1, while a high XO/CSX ratio would be higher than 2. These ratios can be calculated for rm = SL, SU and XU.
Offsets Dec Hex Type Len Name (Dim) Description 280 118 Character 8 PRCSXL_SXLESLXO The number of exclusive lock acquisitions that occurred while other requests spun getting a share in HCPDSP using its local spin. 288 120 Character 8 PRCSXL_SXLESUXO The number of exclusive lock acquisitions that occurred while other requests spun getting a share excluding HCPDSP local spin. 296 128 Character 8 PRCSXL_SXLEXUXO The number of exclusive lock acquisitions that occurred while other requests spun getting exclusive.
Diagnose x'9C' and Diag x'44' usage statistics.
Offsets Dec Hex Type Len Name (Dim) Description 304 130 Character 24 PRCSXL_SXLEXSLX(LENGTH(SXLEX)) LOCKSEQ share requests blocked by exclusive. 328 148 Character 24 PRCSXL_SXLEXSUX(LENGTH(SXLEX)) Unconditional Share requests block by exclusive. 352 160 Character 24 PRCSXL_SXLEXXUX(LENGTH(SXLEX)) Exclusive requests blocked by exclusive. 376 178 Character 24 PRCSXL_SXLEXXUP(LENGTH(SXLEX)) Exclusive requests blocked by share-pending. 400 190 Character 24 PRCSXL_SXLEXXUS(LENGTH(SXLEX)) Exclusive requests blocked by CPUs holding a share. 424 198 Character 0 * End of PRCSXL_CALLKDATA.
Mapping for each of the above elements. That is, for: PRCSXL_SXLEXSLX, PRCSXL_SXLEXSUX, PRCSXL_SXLEXXUX, PRCSXL_SXLEXXUP and PRCSXL_SXLEXXUS.
Offsets Dec Hex Type Len Name (Dim) Description 0 0 Character 24 PRCSXL_SXLEX 0 0 Unsigned 4 PRCSXL_SXLEX9T Count of attempts to identify a diagnose x'9C' target. 4 4 Unsigned 4 PRCSXL_SXLEX9S Count of potential targets considered (for which a SIGP SRS was performed). 8 8 Unsigned 4 PRCSXL_SXLEX9F Count of times a target was found. 12 C Unsigned 4 PRCSXL_SXLEX9C Count of Diag x'9C's issued. 16 10 Unsigned 4 PRCSXL_SXLEX44 Count of Diag x'44's issued. 20 14 Unsigned 4 * Reserved for IBM use. 24 18 Character 0 * End of PRCSXL_SXLEX.
MRPRCSXL Cross Reference
Hex Hex
Name Offset Length Value
MRHDR 0 20
MRHDR_END 14 0
MRHDRDM 4 1
MRHDRLEN 0 2
MRHDRRC 6 2
MRHDRTOD 8 8
MRHDRZER 2 2
PRCSXL 0 69+
PRCSXL_CALENTDSP 1A 2
PRCSXL_CALENTS 2C 0
PRCSXL_CALENTSZ 18 2
PRCSXL_CALLCKID 0 8
PRCSXL_CALLKDATA 0 424
PRCSXL_CALMRHDR 0 0
PRCSXL_CALNMLKS 14 4
PRCSXL_CALVERSN 1C 1
PRCSXL_MRHDR 0 0
PRCSXL_PFXPRBTM 2C 8
PRCSXL_PFXTMSYS 3C 8
PRCSXL_PFXUTIME 34 8
PRCSXL_SXLASSST 8 80
PRCSXL_SXLCAD 8 08
PRCSXL_SXLCUPAD 1E 2
PRCSXL_SXLCPUTY 20 1
PRCSXL_SXLECTRL 8 1
PRCSXL_SXLEFLAGS 9 1
PRCSXL_SXLESCCN 38 8
PRCSXL_SXLESCCU 18 8
PRCSXL_SXLESCPO F8 8
PRCSXL_SXLESLCN 40 8
PRCSXL_SXLESLCSA 70 8
PRCSXL_SXLESLCSX 88 8
PRCSXL_SXLESLCU 20 8
PRCSXL_SXLESLPO 100 8
PRCSXL_SXLESLPR F0 8
PRCSXL_SXLESLPS E8 8
PRCSXL_SXLESLTSU B0 8
PRCSXL_SXLESLTSX B8 8
PRCSXL_SXLESLXO 118 8
PRCSXL_SXLESUCN 48 8
PRCSXL_SXLESUCSA 78 8
PRCSXL_SXLESUCSX 90 8
PRCSXL_SXLESUPO 108 8
PRCSXL_SXLESUPS E0 8
PRCSXL_SXLESUTSX C0 8
PRCSXL_SXLESUXO 120 8
PRCSXL_SXLESXCN 60 8
PRCSXL_SXLESXCU 30 8
PRCSXL_SXLEX 0 24
PRCSXL_SXLEX44 10 4
PRCSXL_SXLEX9C C 4
PRCSXL_SXLEX9F 8 4
PRCSXL_SXLEX9S 4 4
PRCSXL_SXLEX9T 0 4
PRCSXL_SXLEXCCN 50 8
PRCSXL_SXLEXCCU 28 8
PRCSXL_SXLEXSCN 68 8
PRCSXL_SXLEXSLX 130 24
PRCSXL_SXLEXSUX 148 24
PRCSXL_SXLEXUCN 58 8
PRCSXL_SXLEXUCSA 80 8
PRCSXL_SXLEXUCSP A0 8
PRCSXL_SXLEXUCSS A8 8
PRCSXL_SXLEXUCSX 98 8
PRCSXL_SXLEXUSO 110 8
PRCSXL_SXLEXUTSP D0 8
PRCSXL_SXLEXUTSS D8 8
PRCSXL_SXLEXUTSX C8 8
PRCSXL_SXLEXUXO 128 8
PRCSXL_SXLEXXUP 178 24
PRCSXL_SXLEXXUS 190 24
PRCSXL_SXLEXXUX 160 24
PRCSXL_SXLFLAGS 21 1
PRCSXL_SXLLKSEQ 9 80
PRCSXL_SXLNIAI2 8 40
PRCSXL_SXLNIAI4 8 20
PRCSXL_SXLNIAI8 8 10
PRCSXL_SXLPFD6 8 04
PRCSXL_SXLPPA14 8 02
PRCSXL_SXLPPA15 8 01
PRCSXL_SXLTIME 24 8
PRCSXL_SXLXLSQT C 2
PRCSXL_SXLXLSQTL E 2
This information is based on z/VM V6R4.0.
Last updated on 23 Aug 2017 at 13:23:43.
Copyright IBM Corporation, 1990, 2017