support for global CPU list abbreviations
Subsystems (5)
v2 cover support for global CPU list abbreviations
RFC/v1 ---> v2:
commit #1:
leave one line stub behind for !SMP solving build failures. Reported by Randy Dunlap and various build bots.
commit #4
manage to remember '\0' char in strlen from one line to the next. Reported by Colin King.
Original description from v1/RFC below remains unchanged...
---
The basic objective here was to add support for "nohz_full=8-last" and/or "rcu_nocbs="4-last" -- essentially introduce "last" as a portable reference evaluated at boot/runtime for anything using a CPU list.
The thinking behind this, is that people carve off a few early CPUs to support housekeeping tasks, and perhaps dedicate one to a busy I/O peripheral, and then the remaining pool of CPUs out to the end are a part of a commonly configured pool used for the real work the user cares about.
Extend that logic out to a fleet of machines - some new, and some nearing EOL, and you've probably got a wide range of core counts to contend with - even though the early number of cores dedicated to the system overhead probably doesn't vary.
This change would enable sysadmins to have a common bootarg across all such systems, and would also avoid any off-by-one fencepost errors that happen for users who might briefly forget that core counts start at zero.
Looking around before starting, I noticed RCU already had a short-form abbreviation "all" -- but if we want to treat CPU lists in a uniform matter, then tokens shouldn't be implemented at a subsystem level and hence be subsystem specific; each with their own variations.
So I moved "all" to global use - for boot args, and for cgroups. Then I added the inverse "none" and finally, the one I wanted -- "last".
The use of "last" isn't a standalone word like "all" or "none". It will be a part of a complete range specification, possibly with CSV separate ranges, and possibly specified multiple times. So I had to be a bit more careful with string matching - and hence un-inlined the parse function as commit #1 in this series.
But it really is a generic support for "replace token ABC with known at boot value XYZ" - for example, it would be trivial to extend support to add "half" as a dynamic token to be replaced with 1/2 the core count, even though I wouldn't suggest that has a use case like "last" does.
I tested the string matching with a bunch of intentionally badly crafted strings in a user-space harness, and tested bootarg use with nohz_full and rcu_nocbs, and also the post-boot cgroup use case as per below:
root@hackbox:/sys/fs/cgroup/cpuset# mkdir foo root@hackbox:/sys/fs/cgroup/cpuset# cd foo root@hackbox:/sys/fs/cgroup/cpuset/foo# cat cpuset.cpus
root@hackbox:/sys/fs/cgroup/cpuset/foo# /bin/echo 10-last > cpuset.cpus root@hackbox:/sys/fs/cgroup/cpuset/foo# cat cpuset.cpus 10-15 root@hackbox:/sys/fs/cgroup/cpuset/foo# /bin/echo all > cpuset.cpus root@hackbox:/sys/fs/cgroup/cpuset/foo# cat cpuset.cpus 0-15 root@hackbox:/sys/fs/cgroup/cpuset/foo# /bin/echo none > cpuset.cpus root@hackbox:/sys/fs/cgroup/cpuset/foo# cat cpuset.cpus
root@hackbox:/sys/fs/cgroup/cpuset/foo#
This was on a 16 core machine with CONFIG_NR_CPUS=16 in .config file.
Note that the two use cases (boot and runtime) are why you see "early" parameter in the code - I entertained just sticking the string copy on the stack vs. the early alloc dance, but this felt more correct/robust. The cgroup and modular code using cpulist_parse() are runtime cases.
---
Paul Gortmaker (4):
cpumask: un-inline cpulist_parse for SMP; prepare for ascii helpers
cpumask: make "all" alias global and not just RCU
cpumask: add a "none" alias to complement "all"
cpumask: add "last" alias for cpu list specifications
.../admin-guide/kernel-parameters.rst | 20 +++ .../admin-guide/kernel-parameters.txt | 4 +- include/linux/cpumask.h | 8 ++ kernel/rcu/tree_plugin.h | 13 +- lib/cpumask.c | 132 ++++++++++++++++++ 5 files changed, 165 insertions(+), 12 deletions(-)
Signature · 1 line
2.25.1
PATCH v2 1/4 cpumask: un-inline cpulist_parse for SMP; prepare for ascii helpers
In order to support convenience tokens like "all", and "none" and "last" in CPU lists, we'll have to use string operations and expand on what is currently a simple wrapper around the underlying bitmap function call.
Rather than add header dependencies to cpumask.h and code more complex operations not really appropriate for a header file, we prepare by simply un-inlining it here and move it to the lib dir alongside the other more complex cpumask functions.
Since lib/cpumask.c is built conditionally on CONFIG_SMP, and there are non-SMP callers, we leave the one-line stub behind for that case. If they want to check "0-0" is a valid range, they can still do it. In the meantime, we can add the ascii helpers for CONFIG_SMP users. The use of NR_CPUS vs. CONFIG_SMP is consistent with the existing file.
Aside from an additional exported symbol in the SMP case, no functional changes are anticipated with this move.
--- include/linux/cpumask.h | 8 ++++++++ lib/cpumask.c | 13 +++++++++++++ 2 files changed, 21 insertions(+)
include/linux/cpumask.h +8 -0
@@ -679,11 +679,19 @@ static inline int cpumask_parse(const char *buf, struct cpumask *dstp)* @dstp: the cpumask to set. * * Returns -errno, or 0 for success. +* +* There are instances of non-SMP callers of this, and the easiest way +* to remain 100% runtime compatible is to let them continue to have the +* one-line stub, while the SMP version in lib/cpumask.c gets improved. */ +#if NR_CPUS ==1 static inline int cpulist_parse (const char *buf,struct cpumask *dstp) {return bitmap_parselist (buf,cpumask_bits (dstp), nr_cpumask_bits); } +#else +int cpulist_parse (const char *buf,struct cpumask *dstp); +#endif /** * cpumask_size - size to allocate for a 'struct cpumask' in bytes
lib/cpumask.c +13 -0
@@ -95,6 +95,19 @@ int cpumask_next_wrap(int n, const struct cpumask *mask, int start, bool wrap) }EXPORT_SYMBOL (cpumask_next_wrap); +/** +* cpulist_parse - extract a cpumask from a user string of ranges +* @buf: the buffer to extract from +* @dstp: the cpumask to set. +* +* Returns -errno, or 0 for success. +*/ +int cpulist_parse (const char *buf,struct cpumask *dstp) +{ +return bitmap_parselist (buf,cpumask_bits (dstp), nr_cpumask_bits); +} +EXPORT_SYMBOL (cpulist_parse); +/* These are not inline because of header tangles. */ #ifdef CONFIG_CPUMASK_OFFSTACK /**
Signature · 1 line
2.25.1
PATCH v2 2/4 cpumask: make "all" alias global and not just RCU
In reply to Paul GortmakerIt is probably better that we don't have subsystem specific abbreviations or aliases for generic CPU list specifications.
Hence we move the "all" from RCU out to lib/ so that it can be used in any instance where CPU lists are being parsed.
--- Documentation/admin-guide/kernel-parameters.rst | 7 +++++++ Documentation/admin-guide/kernel-parameters.txt | 4 +--- kernel/rcu/tree_plugin.h | 13 ++++--------- lib/cpumask.c | 6 ++++++ 4 files changed, 18 insertions(+), 12 deletions(-)
Documentation/admin-guide/kernel-parameters.rst +7 -0
@@ -68,6 +68,13 @@ For example one can add to the command line following parameter: where the final item represents CPUs 100,101,125,126,150,151,... +The following convenience aliases are also accepted and used: + + foo_cpus=all + +is equivalent to "foo_cpus=0-N" -- where "N" is the numerically last CPU on +the system, thus avoiding looking up the value in "/sys/devices/system/cpu" +in advance on each deployed system. This document may not be entirely up to date and comprehensive. The command
Documentation/admin-guide/kernel-parameters.txt +1 -3
@@ -4023,9 +4023,7 @@ see CONFIG_RAS_CEC help text. rcu_nocbs= [KNL] - The argument is a cpu list, as described above, - except that the string "all" can be used to - specify every CPU on the system. + The argument is a cpu list, as described above. In kernels built with CONFIG_RCU_NOCB_CPU=y, set the specified list of CPUs to be no-callback CPUs.
kernel/rcu/tree_plugin.h +4 -9
@@ -1463,20 +1463,15 @@ static void rcu_cleanup_after_idle(void)/* * Parse the boot-time rcu_nocb_mask CPU list from the kernel parameters. -* The string after the "rcu_nocbs=" is either "all" for all CPUs, or a -* comma-separated list of CPUs and/or CPU ranges. If an invalid list is -* given, a warning is emitted and all CPUs are offloaded. +* If the list is invalid, a warning is emitted and all CPUs are offloaded. */ static int __init rcu_nocb_setup (char *str) {alloc_bootmem_cpumask_var (&rcu_nocb_mask); -if (!strcasecmp (str,"all" )) +if (cpulist_parse (str, rcu_nocb_mask)) { +pr_warn ("rcu_nocbs= bad CPU range, all CPUs set\n" );cpumask_setall (rcu_nocb_mask); -else -if (cpulist_parse (str, rcu_nocb_mask)) { -pr_warn ("rcu_nocbs= bad CPU range, all CPUs set\n" ); -cpumask_setall (rcu_nocb_mask); - } + }return 1 ; }__setup ("rcu_nocbs=" , rcu_nocb_setup);
lib/cpumask.c +6 -0
@@ -2,6 +2,7 @@#include <linux/slab.h> #include <linux/kernel.h> #include <linux/bitops.h> +#include <linux/string.h> #include <linux/cpumask.h> #include <linux/export.h> #include <linux/memblock.h>
@@ -104,6 +105,11 @@ EXPORT_SYMBOL(cpumask_next_wrap);*/ int cpulist_parse (const char *buf,struct cpumask *dstp) { +if (!strcmp (buf,"all" )) { +cpumask_setall (dstp); +return 0 ; + } +return bitmap_parselist (buf,cpumask_bits (dstp), nr_cpumask_bits); }EXPORT_SYMBOL (cpulist_parse);
Signature · 1 line
2.25.1
PATCH v2 3/4 cpumask: add a "none" alias to complement "all"
In reply to Paul GortmakerWith global support for a CPU list alias of "all", it seems to just make sense to also trivially extend support for an opposite "none" specifier.
--- Documentation/admin-guide/kernel-parameters.rst | 6 ++++++ lib/cpumask.c | 5 +++++ 2 files changed, 11 insertions(+)
Documentation/admin-guide/kernel-parameters.rst +6 -0
@@ -76,6 +76,12 @@ is equivalent to "foo_cpus=0-N" -- where "N" is the numerically last CPU on the system, thus avoiding looking up the value in "/sys/devices/system/cpu" in advance on each deployed system. + foo_cpus=none + +will provide an empty/cleared cpu mask for the associated boot argument. + +Note that "all" and "none" are not necessarily valid/sensible input values +for each available parameter expecting a CPU list. This document may not be entirely up to date and comprehensive. The command "modinfo -p ${modulename}" shows a current list of all parameters of a loadable
lib/cpumask.c +5 -0
@@ -110,6 +110,11 @@ int cpulist_parse(const char *buf, struct cpumask *dstp)return 0 ; } +if (!strcmp (buf,"none" )) { +cpumask_clear (dstp); +return 0 ; + } +return bitmap_parselist (buf,cpumask_bits (dstp), nr_cpumask_bits); }EXPORT_SYMBOL (cpulist_parse);
Signature · 1 line
2.25.1
PATCH v2 4/4 cpumask: add "last" alias for cpu list specifications
In reply to Paul GortmakerIt seems that a common configuration is to use the 1st couple cores for housekeeping tasks, and or driving a busy peripheral that generates a lot of interrupts, or something similar.
This tends to leave the remaining ones to form a pool of similarly configured cores to take on the real workload of interest to the user.
So on machine A - with 32 cores, it could be 0-3 for "system" and then 4-31 being used in boot args like nohz_full=, or rcu_nocbs= as part of setting up the worker pool of CPUs.
But then newer machine B is added, and it has 48 cores, and so while the 0-3 part remains unchanged, the pool setup cpu list becomes 4-47.
Deployment would be easier if we could just simply replace 31 and 47 with "last" and let the system substitute in the actual number at boot; a number that it knows better than we do.
No need to have custom boot args per node, no need to do a trial boot in order to snoop /proc/cpuinfo and/or /sys/devices/system/cpu - no more fencepost errors of using 32 and 48 instead of 31 and 47.
A generic token replacement is used to substitute "last" with the number of CPUs present before handing off to bitmap processing. But it could just as easily be used to replace any placeholder token with any other token or value only known at/after boot.
--- .../admin-guide/kernel-parameters.rst | 7 ++ lib/cpumask.c | 112 +++++++++++++++++- 2 files changed, 117 insertions(+), 2 deletions(-)
Documentation/admin-guide/kernel-parameters.rst +7 -0
@@ -83,6 +83,13 @@ will provide an empty/cleared cpu mask for the associated boot argument. Note that "all" and "none" are not necessarily valid/sensible input values for each available parameter expecting a CPU list. + foo_cpus=1,3,5,16-last + +will at runtime, replace "last" with the number of the last (highest number) +present CPU on the system. Thus a common deployment can be used on multiple +systems with different total number of cores present, without needing to +evaluate the total core count in advance on each system. + This document may not be entirely up to date and comprehensive. The command "modinfo -p ${modulename}" shows a current list of all parameters of a loadable module. Loadable modules, after being loaded into the running kernel, also
lib/cpumask.c +110 -2
138 lines not on this page · raw message
Signature · 1 line
2.25.1
Paul E. McKenney
maintainerQuoted 11 lines · Paul Gortmaker
RFC/v1 ---> v2:
commit #1:
leave one line stub behind for !SMP solving build failures. Reported by Randy Dunlap and various build bots.commit #4
manage to remember '\0' char in strlen from one line to the next. Reported by Colin King.Original description from v1/RFC below remains unchanged...
Queued and this time kicking off testing that actually includes your patches! ;-)
Thanx, Paul
Quoted 92 lines
---
The basic objective here was to add support for "nohz_full=8-last" and/or "rcu_nocbs="4-last" -- essentially introduce "last" as a portable reference evaluated at boot/runtime for anything using a CPU list.
The thinking behind this, is that people carve off a few early CPUs to support housekeeping tasks, and perhaps dedicate one to a busy I/O peripheral, and then the remaining pool of CPUs out to the end are a part of a commonly configured pool used for the real work the user cares about.
Extend that logic out to a fleet of machines - some new, and some nearing EOL, and you've probably got a wide range of core counts to contend with - even though the early number of cores dedicated to the system overhead probably doesn't vary.
This change would enable sysadmins to have a common bootarg across all such systems, and would also avoid any off-by-one fencepost errors that happen for users who might briefly forget that core counts start at zero.
Looking around before starting, I noticed RCU already had a short-form abbreviation "all" -- but if we want to treat CPU lists in a uniform matter, then tokens shouldn't be implemented at a subsystem level and hence be subsystem specific; each with their own variations.
So I moved "all" to global use - for boot args, and for cgroups. Then I added the inverse "none" and finally, the one I wanted -- "last".
The use of "last" isn't a standalone word like "all" or "none". It will be a part of a complete range specification, possibly with CSV separate ranges, and possibly specified multiple times. So I had to be a bit more careful with string matching - and hence un-inlined the parse function as commit #1 in this series.
But it really is a generic support for "replace token ABC with known at boot value XYZ" - for example, it would be trivial to extend support to add "half" as a dynamic token to be replaced with 1/2 the core count, even though I wouldn't suggest that has a use case like "last" does.
I tested the string matching with a bunch of intentionally badly crafted strings in a user-space harness, and tested bootarg use with
nohz_fullandrcu_nocbs, and also the post-boot cgroup use case as per below:root@hackbox:/sys/fs/cgroup/cpuset# mkdir foo root@hackbox:/sys/fs/cgroup/cpuset# cd foo root@hackbox:/sys/fs/cgroup/cpuset/foo# cat cpuset.cpusroot@hackbox:/sys/fs/cgroup/cpuset/foo# /bin/echo 10-last > cpuset.cpus root@hackbox:/sys/fs/cgroup/cpuset/foo# cat cpuset.cpus 10-15 root@hackbox:/sys/fs/cgroup/cpuset/foo# /bin/echo all > cpuset.cpus root@hackbox:/sys/fs/cgroup/cpuset/foo# cat cpuset.cpus 0-15 root@hackbox:/sys/fs/cgroup/cpuset/foo# /bin/echo none > cpuset.cpus root@hackbox:/sys/fs/cgroup/cpuset/foo# cat cpuset.cpusroot@hackbox:/sys/fs/cgroup/cpuset/foo#
This was on a 16 core machine with
CONFIG_NR_CPUS=16in .config file.Note that the two use cases (boot and runtime) are why you see "early" parameter in the code - I entertained just sticking the string copy on the stack vs. the early alloc dance, but this felt more correct/robust. The cgroup and modular code using
cpulist_parse()are runtime cases.---
Cc Frederic Weisbecker <[email protected]>Cc "Paul E. McKenney" <[email protected]>Cc Josh Triplett <[email protected]>Cc Thomas Gleixner <[email protected]>Cc Ingo Molnar <[email protected]>Cc Li Zefan <[email protected]>Paul Gortmaker (4):
cpumask: un-inlinecpulist_parsefor SMP; prepare for ascii helpers
cpumask: make "all" alias global and not just RCU
cpumask: add a "none" alias to complement "all"
cpumask: add "last" alias for cpu list specifications.../admin-guide/kernel-parameters.rst | 20 +++ .../admin-guide/kernel-parameters.txt | 4 +- include/linux/cpumask.h | 8 ++ kernel/rcu/tree_plugin.h | 13 +- lib/cpumask.c | 132 ++++++++++++++++++ 5 files changed, 165 insertions(+), 12 deletions(-)Signature · 1 line
2.25.1
Quoted 10 lines · Paul E. McKenney
Quoted 9 lines · Paul Gortmaker
RFC/v1 ---> v2:
commit #1:
leave one line stub behind for !SMP solving build failures. Reported by Randy Dunlap and various build bots.commit #4
manage to remember '\0' char in strlen from one line to the next. Reported by Colin King.
shouldn't that be "Reported and fixed by Colin King"? ;-)
Quoted 100 lines
Quoted 2 lines
Original description from v1/RFC below remains unchanged...
Queued and this time kicking off testing that actually includes your patches! ;-)
Thanx, Paul
Quoted 92 lines
---
The basic objective here was to add support for "nohz_full=8-last" and/or "rcu_nocbs="4-last" -- essentially introduce "last" as a portable reference evaluated at boot/runtime for anything using a CPU list.
The thinking behind this, is that people carve off a few early CPUs to support housekeeping tasks, and perhaps dedicate one to a busy I/O peripheral, and then the remaining pool of CPUs out to the end are a part of a commonly configured pool used for the real work the user cares about.
Extend that logic out to a fleet of machines - some new, and some nearing EOL, and you've probably got a wide range of core counts to contend with - even though the early number of cores dedicated to the system overhead probably doesn't vary.
This change would enable sysadmins to have a common bootarg across all such systems, and would also avoid any off-by-one fencepost errors that happen for users who might briefly forget that core counts start at zero.
Looking around before starting, I noticed RCU already had a short-form abbreviation "all" -- but if we want to treat CPU lists in a uniform matter, then tokens shouldn't be implemented at a subsystem level and hence be subsystem specific; each with their own variations.
So I moved "all" to global use - for boot args, and for cgroups. Then I added the inverse "none" and finally, the one I wanted -- "last".
The use of "last" isn't a standalone word like "all" or "none". It will be a part of a complete range specification, possibly with CSV separate ranges, and possibly specified multiple times. So I had to be a bit more careful with string matching - and hence un-inlined the parse function as commit #1 in this series.
But it really is a generic support for "replace token ABC with known at boot value XYZ" - for example, it would be trivial to extend support to add "half" as a dynamic token to be replaced with 1/2 the core count, even though I wouldn't suggest that has a use case like "last" does.
I tested the string matching with a bunch of intentionally badly crafted strings in a user-space harness, and tested bootarg use with
nohz_fullandrcu_nocbs, and also the post-boot cgroup use case as per below:root@hackbox:/sys/fs/cgroup/cpuset# mkdir foo root@hackbox:/sys/fs/cgroup/cpuset# cd foo root@hackbox:/sys/fs/cgroup/cpuset/foo# cat cpuset.cpusroot@hackbox:/sys/fs/cgroup/cpuset/foo# /bin/echo 10-last > cpuset.cpus root@hackbox:/sys/fs/cgroup/cpuset/foo# cat cpuset.cpus 10-15 root@hackbox:/sys/fs/cgroup/cpuset/foo# /bin/echo all > cpuset.cpus root@hackbox:/sys/fs/cgroup/cpuset/foo# cat cpuset.cpus 0-15 root@hackbox:/sys/fs/cgroup/cpuset/foo# /bin/echo none > cpuset.cpus root@hackbox:/sys/fs/cgroup/cpuset/foo# cat cpuset.cpusroot@hackbox:/sys/fs/cgroup/cpuset/foo#
This was on a 16 core machine with
CONFIG_NR_CPUS=16in .config file.Note that the two use cases (boot and runtime) are why you see "early" parameter in the code - I entertained just sticking the string copy on the stack vs. the early alloc dance, but this felt more correct/robust. The cgroup and modular code using
cpulist_parse()are runtime cases.---
Cc Frederic Weisbecker <[email protected]>Cc "Paul E. McKenney" <[email protected]>Cc Josh Triplett <[email protected]>Cc Thomas Gleixner <[email protected]>Cc Ingo Molnar <[email protected]>Cc Li Zefan <[email protected]>Paul Gortmaker (4):
cpumask: un-inlinecpulist_parsefor SMP; prepare for ascii helpers
cpumask: make "all" alias global and not just RCU
cpumask: add a "none" alias to complement "all"
cpumask: add "last" alias for cpu list specifications.../admin-guide/kernel-parameters.rst | 20 +++ .../admin-guide/kernel-parameters.txt | 4 +- include/linux/cpumask.h | 8 ++ kernel/rcu/tree_plugin.h | 13 +- lib/cpumask.c | 132 ++++++++++++++++++ 5 files changed, 165 insertions(+), 12 deletions(-)Signature · 1 line
2.25.1