- 03 Oct, 2020 1 commit
-
-
Peter Oskolkov authored
Currently, pick_next_entity(...) has the following structure (simplified): [...] if (last_buddy_ok()) result = last_buddy; if (next_buddy_ok()) result = next_buddy; [...] The intended behavior is to prefer next buddy over last buddy; the current code somewhat obfuscates this, and also wastes cycles checking the last buddy when eventually the next buddy is picked up. So this patch refactors two 'ifs' above into [...] if (next_buddy_ok()) result = next_buddy; else if (last_buddy_ok()) result = last_buddy; [...] Signed-off-by: Peter Oskolkov <posk@google.com> Signed-off-by: Peter Zijlstra (Intel) <peterz@infradead.org> Reviewed-by: Vincent Guittot <vincent.guitttot@linaro.org> Link: https://lkml.kernel.org/r/20200930173532.1069092-1-posk@google.com
-
- 25 Sep, 2020 14 commits
-
-
Peter Oskolkov authored
Based on Google-internal RSEQ work done by Paul Turner and Andrew Hunter. This patch adds a selftest for MEMBARRIER_CMD_PRIVATE_EXPEDITED_RSEQ. The test quite often fails without the previous patch in this patchset, but consistently passes with it. Signed-off-by: Peter Oskolkov <posk@google.com> Signed-off-by: Peter Zijlstra (Intel) <peterz@infradead.org> Acked-by: Mathieu Desnoyers <mathieu.desnoyers@efficios.com> Link: https://lkml.kernel.org/r/20200923233618.2572849-3-posk@google.com
-
Peter Oskolkov authored
This patch adds rseq_offset_deref_addv() function to tools/testing/selftests/rseq/rseq-x86.h, to be used in a selftest in the next patch in the patchset. Once an architecture adds support for this function they should define "RSEQ_ARCH_HAS_OFFSET_DEREF_ADDV". Signed-off-by: Peter Oskolkov <posk@google.com> Signed-off-by: Peter Zijlstra (Intel) <peterz@infradead.org> Acked-by: Mathieu Desnoyers <mathieu.desnoyers@efficios.com> Link: https://lkml.kernel.org/r/20200923233618.2572849-2-posk@google.com
-
Peter Oskolkov authored
This patchset is based on Google-internal RSEQ work done by Paul Turner and Andrew Hunter. When working with per-CPU RSEQ-based memory allocations, it is sometimes important to make sure that a global memory location is no longer accessed from RSEQ critical sections. For example, there can be two per-CPU lists, one is "active" and accessed per-CPU, while another one is inactive and worked on asynchronously "off CPU" (e.g. garbage collection is performed). Then at some point the two lists are swapped, and a fast RCU-like mechanism is required to make sure that the previously active list is no longer accessed. This patch introduces such a mechanism: in short, membarrier() syscall issues an IPI to a CPU, restarting a potentially active RSEQ critical section on the CPU. Signed-off-by: Peter Oskolkov <posk@google.com> Signed-off-by: Peter Zijlstra (Intel) <peterz@infradead.org> Acked-by: Mathieu Desnoyers <mathieu.desnoyers@efficios.com> Link: https://lkml.kernel.org/r/20200923233618.2572849-1-posk@google.com
-
Barry Song authored
Barry Song noted the following Something is wrong. In find_busiest_group(), we are checking if src has higher load, however, in task_numa_find_cpu(), we are checking if dst will have higher load after balancing. It seems it is not sensible to check src. It maybe cause wrong imbalance value, for example, if dst_running = env->dst_stats.nr_running + 1 results in 3 or above, and src_running = env->src_stats.nr_running - 1 results in 1; The current code is thinking imbalance as 0 since src_running is smaller than 2. This is inconsistent with load balancer. Basically, in find_busiest_group(), the NUMA imbalance is ignored if moving a task "from an almost idle domain" to a "domain with spare capacity". This patch forbids movement "from a misplaced domain" to "an almost idle domain" as that is closer to what the CPU load balancer expects. This patch is not a universal win. The old behaviour was intended to allow a task from an almost idle NUMA node to migrate to its preferred node if the destination had capacity but there are corner cases. For example, a NAS compute load could be parallelised to use 1/3rd of available CPUs but not all those potential tasks are active at all times allowing this logic to trigger. An obvious example is specjbb 2005 running various numbers of warehouses on a 2 socket box with 80 cpus. specjbb 5.9.0-rc4 5.9.0-rc4 vanilla dstbalance-v1r1 Hmean tput-1 46425.00 ( 0.00%) 43394.00 * -6.53%* Hmean tput-2 98416.00 ( 0.00%) 96031.00 * -2.42%* Hmean tput-3 150184.00 ( 0.00%) 148783.00 * -0.93%* Hmean tput-4 200683.00 ( 0.00%) 197906.00 * -1.38%* Hmean tput-5 236305.00 ( 0.00%) 245549.00 * 3.91%* Hmean tput-6 281559.00 ( 0.00%) 285692.00 * 1.47%* Hmean tput-7 338558.00 ( 0.00%) 334467.00 * -1.21%* Hmean tput-8 340745.00 ( 0.00%) 372501.00 * 9.32%* Hmean tput-9 424343.00 ( 0.00%) 413006.00 * -2.67%* Hmean tput-10 421854.00 ( 0.00%) 434261.00 * 2.94%* Hmean tput-11 493256.00 ( 0.00%) 485330.00 * -1.61%* Hmean tput-12 549573.00 ( 0.00%) 529959.00 * -3.57%* Hmean tput-13 593183.00 ( 0.00%) 555010.00 * -6.44%* Hmean tput-14 588252.00 ( 0.00%) 599166.00 * 1.86%* Hmean tput-15 623065.00 ( 0.00%) 642713.00 * 3.15%* Hmean tput-16 703924.00 ( 0.00%) 660758.00 * -6.13%* Hmean tput-17 666023.00 ( 0.00%) 697675.00 * 4.75%* Hmean tput-18 761502.00 ( 0.00%) 758360.00 * -0.41%* Hmean tput-19 796088.00 ( 0.00%) 798368.00 * 0.29%* Hmean tput-20 733564.00 ( 0.00%) 823086.00 * 12.20%* Hmean tput-21 840980.00 ( 0.00%) 856711.00 * 1.87%* Hmean tput-22 804285.00 ( 0.00%) 872238.00 * 8.45%* Hmean tput-23 795208.00 ( 0.00%) 889374.00 * 11.84%* Hmean tput-24 848619.00 ( 0.00%) 966783.00 * 13.92%* Hmean tput-25 750848.00 ( 0.00%) 903790.00 * 20.37%* Hmean tput-26 780523.00 ( 0.00%) 962254.00 * 23.28%* Hmean tput-27 1042245.00 ( 0.00%) 991544.00 * -4.86%* Hmean tput-28 1090580.00 ( 0.00%) 1035926.00 * -5.01%* Hmean tput-29 999483.00 ( 0.00%) 1082948.00 * 8.35%* Hmean tput-30 1098663.00 ( 0.00%) 1113427.00 * 1.34%* Hmean tput-31 1125671.00 ( 0.00%) 1134175.00 * 0.76%* Hmean tput-32 968167.00 ( 0.00%) 1250286.00 * 29.14%* Hmean tput-33 1077676.00 ( 0.00%) 1060893.00 * -1.56%* Hmean tput-34 1090538.00 ( 0.00%) 1090933.00 * 0.04%* Hmean tput-35 967058.00 ( 0.00%) 1107421.00 * 14.51%* Hmean tput-36 1051745.00 ( 0.00%) 1210663.00 * 15.11%* Hmean tput-37 1019465.00 ( 0.00%) 1351446.00 * 32.56%* Hmean tput-38 1083102.00 ( 0.00%) 1064541.00 * -1.71%* Hmean tput-39 1232990.00 ( 0.00%) 1303623.00 * 5.73%* Hmean tput-40 1175542.00 ( 0.00%) 1340943.00 * 14.07%* Hmean tput-41 1127826.00 ( 0.00%) 1339492.00 * 18.77%* Hmean tput-42 1198313.00 ( 0.00%) 1411023.00 * 17.75%* Hmean tput-43 1163733.00 ( 0.00%) 1228253.00 * 5.54%* Hmean tput-44 1305562.00 ( 0.00%) 1357886.00 * 4.01%* Hmean tput-45 1326752.00 ( 0.00%) 1406061.00 * 5.98%* Hmean tput-46 1339424.00 ( 0.00%) 1418451.00 * 5.90%* Hmean tput-47 1415057.00 ( 0.00%) 1381570.00 * -2.37%* Hmean tput-48 1392003.00 ( 0.00%) 1421167.00 * 2.10%* Hmean tput-49 1408374.00 ( 0.00%) 1418659.00 * 0.73%* Hmean tput-50 1359822.00 ( 0.00%) 1391070.00 * 2.30%* Hmean tput-51 1414246.00 ( 0.00%) 1392679.00 * -1.52%* Hmean tput-52 1432352.00 ( 0.00%) 1354020.00 * -5.47%* Hmean tput-53 1387563.00 ( 0.00%) 1409563.00 * 1.59%* Hmean tput-54 1406420.00 ( 0.00%) 1388711.00 * -1.26%* Hmean tput-55 1438804.00 ( 0.00%) 1387472.00 * -3.57%* Hmean tput-56 1399465.00 ( 0.00%) 1400296.00 * 0.06%* Hmean tput-57 1428132.00 ( 0.00%) 1396399.00 * -2.22%* Hmean tput-58 1432385.00 ( 0.00%) 1386253.00 * -3.22%* Hmean tput-59 1421612.00 ( 0.00%) 1371416.00 * -3.53%* Hmean tput-60 1429423.00 ( 0.00%) 1389412.00 * -2.80%* Hmean tput-61 1396230.00 ( 0.00%) 1351122.00 * -3.23%* Hmean tput-62 1418396.00 ( 0.00%) 1383098.00 * -2.49%* Hmean tput-63 1409918.00 ( 0.00%) 1374662.00 * -2.50%* Hmean tput-64 1410236.00 ( 0.00%) 1376216.00 * -2.41%* Hmean tput-65 1396405.00 ( 0.00%) 1364418.00 * -2.29%* Hmean tput-66 1395975.00 ( 0.00%) 1357326.00 * -2.77%* Hmean tput-67 1392986.00 ( 0.00%) 1349642.00 * -3.11%* Hmean tput-68 1386541.00 ( 0.00%) 1343261.00 * -3.12%* Hmean tput-69 1374407.00 ( 0.00%) 1342588.00 * -2.32%* Hmean tput-70 1377513.00 ( 0.00%) 1334654.00 * -3.11%* Hmean tput-71 1369319.00 ( 0.00%) 1334952.00 * -2.51%* Hmean tput-72 1354635.00 ( 0.00%) 1329005.00 * -1.89%* Hmean tput-73 1350933.00 ( 0.00%) 1318942.00 * -2.37%* Hmean tput-74 1351714.00 ( 0.00%) 1316347.00 * -2.62%* Hmean tput-75 1352198.00 ( 0.00%) 1309974.00 * -3.12%* Hmean tput-76 1349490.00 ( 0.00%) 1286064.00 * -4.70%* Hmean tput-77 1336131.00 ( 0.00%) 1303684.00 * -2.43%* Hmean tput-78 1308896.00 ( 0.00%) 1271024.00 * -2.89%* Hmean tput-79 1326703.00 ( 0.00%) 1290862.00 * -2.70%* Hmean tput-80 1336199.00 ( 0.00%) 1291629.00 * -3.34%* The performance at the mid-point is better but not universally better. The patch is a mixed bag depending on the workload, machine and overall levels of utilisation. Sometimes it's better (sometimes much better), other times it is worse (sometimes much worse). Given that there isn't a universally good decision in this section and more people seem to prefer the patch then it may be best to keep the LB decisions consistent and revisit imbalance handling when the load balancer code changes settle down. Jirka Hladky added the following observation. Our results are mostly in line with what you see. We observe big gains (20-50%) when the system is loaded to 1/3 of the maximum capacity and mixed results at the full load - some workloads benefit from the patch at the full load, others not, but performance changes at the full load are mostly within the noise of results (+/-5%). Overall, we think this patch is helpful. [mgorman@techsingularity.net: Rewrote changelog] Fixes: fb86f5b2 ("sched/numa: Use similar logic to the load balancer for moving between domains with spare capacity") Signed-off-by: Barry Song <song.bao.hua@hisilicon.com> Signed-off-by: Mel Gorman <mgorman@techsingularity.net> Signed-off-by: Peter Zijlstra (Intel) <peterz@infradead.org> Link: https://lkml.kernel.org/r/20200921221849.GI3179@techsingularity.net
-
Vincent Guittot authored
The busy_factor, which increases load balance interval when a cpu is busy, is set to 32 by default. This value generates some huge LB interval on large system like the THX2 made of 2 node x 28 cores x 4 threads. For such system, the interval increases from 112ms to 3584ms at MC level. And from 228ms to 7168ms at NUMA level. Even on smaller system, a lower busy factor has shown improvement on the fair distribution of the running time so let reduce it for all. Signed-off-by: Vincent Guittot <vincent.guittot@linaro.org> Signed-off-by: Peter Zijlstra (Intel) <peterz@infradead.org> Reviewed-by: Phil Auld <pauld@redhat.com> Link: https://lkml.kernel.org/r/20200921072424.14813-5-vincent.guittot@linaro.org
-
Vincent Guittot authored
sched domains tend to trigger simultaneously the load balance loop but the larger domains often need more time to collect statistics. This slowness makes the larger domain trying to detach tasks from a rq whereas tasks already migrated somewhere else at a sub-domain level. This is not a real problem for idle LB because the period of smaller domains will increase with its CPUs being busy and this will let time for higher ones to pulled tasks. But this becomes a problem when all CPUs are already busy because all domains stay synced when they trigger their LB. A simple way to minimize simultaneous LB of all domains is to decrement the the busy interval by 1 jiffies. Because of the busy_factor, the interval of larger domain will not be a multiple of smaller ones anymore. Signed-off-by: Vincent Guittot <vincent.guittot@linaro.org> Signed-off-by: Peter Zijlstra (Intel) <peterz@infradead.org> Reviewed-by: Phil Auld <pauld@redhat.com> Link: https://lkml.kernel.org/r/20200921072424.14813-4-vincent.guittot@linaro.org
-
Vincent Guittot authored
The 25% default imbalance threshold for DIE and NUMA domain is large enough to generate significant unfairness between threads. A typical example is the case of 11 threads running on 2x4 CPUs. The imbalance of 20% between the 2 groups of 4 cores is just low enough to not trigger the load balance between the 2 groups. We will have always the same 6 threads on one group of 4 CPUs and the other 5 threads on the other group of CPUS. With a fair time sharing in each group, we ends up with +20% running time for the group of 5 threads. Consider decreasing the imbalance threshold for overloaded case where we use the load to balance task and to ensure fair time sharing. Signed-off-by: Vincent Guittot <vincent.guittot@linaro.org> Signed-off-by: Peter Zijlstra (Intel) <peterz@infradead.org> Reviewed-by: Phil Auld <pauld@redhat.com> Acked-by: Hillf Danton <hdanton@sina.com> Link: https://lkml.kernel.org/r/20200921072424.14813-3-vincent.guittot@linaro.org
-
Vincent Guittot authored
Some UCs like 9 always running tasks on 8 CPUs can't be balanced and the load balancer currently migrates the waiting task between the CPUs in an almost random manner. The success of a rq pulling a task depends of the value of nr_balance_failed of its domains and its ability to be faster than others to detach it. This behavior results in an unfair distribution of the running time between tasks because some CPUs will run most of the time, if not always, the same task whereas others will share their time between several tasks. Instead of using nr_balance_failed as a boolean to relax the condition for detaching task, the LB will use nr_balanced_failed to relax the threshold between the tasks'load and the imbalance. This mecanism prevents the same rq or domain to always win the load balance fight. Signed-off-by: Vincent Guittot <vincent.guittot@linaro.org> Signed-off-by: Peter Zijlstra (Intel) <peterz@infradead.org> Reviewed-by: Phil Auld <pauld@redhat.com> Link: https://lkml.kernel.org/r/20200921072424.14813-2-vincent.guittot@linaro.org
-
Xianting Tian authored
In the file fair.c, sometims update_tg_load_avg(cfs_rq, 0) is used, sometimes update_tg_load_avg(cfs_rq, false) is used. update_tg_load_avg() has the parameter force, but in current code, it never set 1 or true to it, so remove the force parameter. Signed-off-by: Xianting Tian <tian.xianting@h3c.com> Signed-off-by: Peter Zijlstra (Intel) <peterz@infradead.org> Link: https://lkml.kernel.org/r/20200924014755.36253-1-tian.xianting@h3c.com
-
Xunlei Pang authored
We've met problems that occasionally tasks with full cpumask (e.g. by putting it into a cpuset or setting to full affinity) were migrated to our isolated cpus in production environment. After some analysis, we found that it is due to the current select_idle_smt() not considering the sched_domain mask. Steps to reproduce on my 31-CPU hyperthreads machine: 1. with boot parameter: "isolcpus=domain,2-31" (thread lists: 0,16 and 1,17) 2. cgcreate -g cpu:test; cgexec -g cpu:test "test_threads" 3. some threads will be migrated to the isolated cpu16~17. Fix it by checking the valid domain mask in select_idle_smt(). Fixes: 10e2f1ac ("sched/core: Rewrite and improve select_idle_siblings()) Reported-by: Wetp Zhang <wetp.zy@linux.alibaba.com> Signed-off-by: Xunlei Pang <xlpang@linux.alibaba.com> Signed-off-by: Peter Zijlstra (Intel) <peterz@infradead.org> Reviewed-by: Jiang Biao <benbjiang@tencent.com> Reviewed-by: Vincent Guittot <vincent.guittot@linaro.org> Link: https://lkml.kernel.org/r/1600930127-76857-1-git-send-email-xlpang@linux.alibaba.com
-
YueHaibing authored
There is no caller in tree, so can remove it. Signed-off-by: YueHaibing <yuehaibing@huawei.com> Signed-off-by: Peter Zijlstra (Intel) <peterz@infradead.org> Reviewed-by: Dietmar Eggemann <dietmar.eggemann@arm.com> Link: https://lkml.kernel.org/r/20200922132410.48440-1-yuehaibing@huawei.com
-
Daniel Bristot de Oliveira authored
The RT_RUNTIME_SHARE sched feature enables the sharing of rt_runtime between CPUs, allowing a CPU to run a real-time task up to 100% of the time while leaving more space for non-real-time tasks to run on the CPU that lend rt_runtime. The problem is that a CPU can easily borrow enough rt_runtime to allow a spinning rt-task to run forever, starving per-cpu tasks like kworkers, which are non-real-time by design. This patch disables RT_RUNTIME_SHARE by default, avoiding this problem. The feature will still be present for users that want to enable it, though. Signed-off-by: Daniel Bristot de Oliveira <bristot@redhat.com> Signed-off-by: Peter Zijlstra (Intel) <peterz@infradead.org> Tested-by: Wei Wang <wvw@google.com> Link: https://lkml.kernel.org/r/b776ab46817e3db5d8ef79175fa0d71073c051c7.1600697903.git.bristot@redhat.com
-
Lucas Stach authored
When a boosted task gets throttled, what normally happens is that it's immediately enqueued again with ENQUEUE_REPLENISH, which replenishes the runtime and clears the dl_throttled flag. There is a special case however: if the throttling happened on sched-out and the task has been deboosted in the meantime, the replenish is skipped as the task will return to its normal scheduling class. This leaves the task with the dl_throttled flag set. Now if the task gets boosted up to the deadline scheduling class again while it is sleeping, it's still in the throttled state. The normal wakeup however will enqueue the task with ENQUEUE_REPLENISH not set, so we don't actually place it on the rq. Thus we end up with a task that is runnable, but not actually on the rq and neither a immediate replenishment happens, nor is the replenishment timer set up, so the task is stuck in forever-throttled limbo. Clear the dl_throttled flag before dropping back to the normal scheduling class to fix this issue. Signed-off-by: Lucas Stach <l.stach@pengutronix.de> Signed-off-by: Peter Zijlstra (Intel) <peterz@infradead.org> Acked-by: Juri Lelli <juri.lelli@redhat.com> Link: https://lkml.kernel.org/r/20200831110719.2126930-1-l.stach@pengutronix.de
-
Vincent Guittot authored
Use runnable_avg to classify numa node state similarly to what is done for normal load balancer. This helps to ensure that numa and normal balancers use the same view of the state of the system. Large arm64system: 2 nodes / 224 CPUs: hackbench -l (256000/#grp) -g #grp grp tip/sched/core +patchset improvement 1 14,008(+/- 4,99 %) 13,800(+/- 3.88 %) 1,48 % 4 4,340(+/- 5.35 %) 4.283(+/- 4.85 %) 1,33 % 16 3,357(+/- 0.55 %) 3.359(+/- 0.54 %) -0,06 % 32 3,050(+/- 0.94 %) 3.039(+/- 1,06 %) 0,38 % 64 2.968(+/- 1,85 %) 3.006(+/- 2.92 %) -1.27 % 128 3,290(+/-12.61 %) 3,108(+/- 5.97 %) 5.51 % 256 3.235(+/- 3.95 %) 3,188(+/- 2.83 %) 1.45 % Signed-off-by: Vincent Guittot <vincent.guittot@linaro.org> Signed-off-by: Peter Zijlstra (Intel) <peterz@infradead.org> Reviewed-by: Mel Gorman <mgorman@suse.de> Link: https://lkml.kernel.org/r/20200921072959.16317-1-vincent.guittot@linaro.org
-
- 09 Sep, 2020 1 commit
-
-
Valentin Schneider authored
The last sd_flag_debug shuffle inadvertently moved its definition within an #ifdef CONFIG_SYSCTL region. While CONFIG_SYSCTL is indeed required to produce the sched domain ctl interface (which uses sd_flag_debug to output flag names), it isn't required to run any assertion on the sched_domain hierarchy itself. Move the definition of sd_flag_debug to a CONFIG_SCHED_DEBUG region of topology.c. Now at long last we have: - sd_flag_debug declared in include/linux/sched/topology.h iff CONFIG_SCHED_DEBUG=y - sd_flag_debug defined in kernel/sched/topology.c, conditioned by: - CONFIG_SCHED_DEBUG, with an explicit #ifdef block - CONFIG_SMP, as a requirement to compile topology.c With this change, all symbols pertaining to SD flag metadata (with the exception of __SD_FLAG_CNT) are now defined exclusively within topology.c Fixes: 8fca9494 ("sched/topology: Move sd_flag_debug out of linux/sched/topology.h") Reported-by: Randy Dunlap <rdunlap@infradead.org> Signed-off-by: Valentin Schneider <valentin.schneider@arm.com> Signed-off-by: Ingo Molnar <mingo@kernel.org> Link: https://lore.kernel.org/r/20200908184956.23369-1-valentin.schneider@arm.com
-
- 04 Sep, 2020 1 commit
-
-
Daniel Bristot de Oliveira authored
As discussed with Juri and Peter. Signed-off-by: Daniel Bristot de Oliveira <bristot@redhat.com> Signed-off-by: Ingo Molnar <mingo@kernel.org> Acked-by: Peter Zijlstra <a.p.zijlstra@chello.nl> Acked-by: Juri Lelli <juri.lelli@redhat.com> Link: https://lore.kernel.org/r/4476a6da70949913a59dab9aacfbd12162c1fbd7.1599146667.git.bristot@redhat.com
-
- 26 Aug, 2020 7 commits
-
-
Valentin Schneider authored
SD_DEGENERATE_GROUPS_MASK is only useful for sched/topology.c, but still gets defined for anyone who imports topology.h, leading to a flurry of unused variable warnings. Move it out of the header and place it next to the SD degeneration functions in sched/topology.c. Fixes: 4ee4ea44 ("sched/topology: Introduce SD metaflag for flags needing > 1 groups") Reported-by: Andy Shevchenko <andriy.shevchenko@linux.intel.com> Signed-off-by: Valentin Schneider <valentin.schneider@arm.com> Signed-off-by: Peter Zijlstra (Intel) <peterz@infradead.org> Link: https://lkml.kernel.org/r/20200825133216.9163-2-valentin.schneider@arm.com
-
Valentin Schneider authored
Defining an array in a header imported all over the place clearly is a daft idea, that still didn't stop me from doing it. Leave a declaration of sd_flag_debug in topology.h and move its definition to sched/debug.c. Fixes: b6e862f3 ("sched/topology: Define and assign sched_domain flag metadata") Reported-by: Andy Shevchenko <andriy.shevchenko@linux.intel.com> Signed-off-by: Valentin Schneider <valentin.schneider@arm.com> Signed-off-by: Peter Zijlstra (Intel) <peterz@infradead.org> Link: https://lkml.kernel.org/r/20200825133216.9163-1-valentin.schneider@arm.com
-
Sebastian Andrzej Siewior authored
sched_submit_work() is considered to be a hot path. The preempt_disable() instruction is a compiler barrier and forces the compiler to load task_struct::flags for the second comparison. By using a local variable, the compiler can load the value once and keep it in a register for the second comparison. Verified on x86-64 with gcc-10. Signed-off-by: Sebastian Andrzej Siewior <bigeasy@linutronix.de> Signed-off-by: Peter Zijlstra (Intel) <peterz@infradead.org> Link: https://lkml.kernel.org/r/20200819200025.lqvmyefqnbok5i4f@linutronix.de
-
Sebastian Andrzej Siewior authored
The bits PF_IO_WORKER and PF_WQ_WORKER are tested together in sched_submit_work() which is considered to be a hot path. If the two bits cross the 8 or 16 bit boundary then most architecture require multiple load instructions in order to create the constant value. Also, such a value can not be encoded within the compare opcode. By moving the bit definition within the same block, the compiler can create/use one immediate value. For some reason gcc-10 on ARM64 requires both bits to be next to each other in order to issue "tst reg, val; bne label". Otherwise the result is "mov reg1, val; tst reg, reg1; bne label". Move PF_VCPU out of the way so that PF_IO_WORKER can be next to PF_WQ_WORKER. Signed-off-by: Sebastian Andrzej Siewior <bigeasy@linutronix.de> Signed-off-by: Peter Zijlstra (Intel) <peterz@infradead.org> Link: https://lkml.kernel.org/r/20200819195505.y3fxk72sotnrkczi@linutronix.de
-
Jiang Biao authored
The code in reweight_entity() can be simplified. For a sched entity on the rq, the entity accounting can be replaced by cfs_rq instantaneous load updates currently called from within the entity accounting. Even though an entity on the rq can't represent a task in reweight_entity() (a task is always dequeued before calling this function) and so the numa task accounting and the rq->cfs_tasks list management of the entity accounting are never called, the redundant cfs_rq->nr_running decrement/increment will be avoided. Signed-off-by: Jiang Biao <benbjiang@tencent.com> Signed-off-by: Peter Zijlstra (Intel) <peterz@infradead.org> Reviewed-by: Dietmar Eggemann <dietmar.eggemann@arm.com> Reviewed-by: Vincent Guittot <vincent.guittot@linaro.org> Link: https://lkml.kernel.org/r/20200811113209.34057-1-benbjiang@tencent.com
-
Lukasz Luba authored
In find_energy_efficient_cpu() 'cpu_cap' could be less that 'util'. It might be because of RT, DL (so higher sched class than CFS), irq or thermal pressure signal, which reduce the capacity value. In such situation the result of 'cpu_cap - util' might be negative but stored in the unsigned long. Then it might be compared with other unsigned long when uclamp_rq_util_with() reduced the 'util' such that is passes the fits_capacity() check. Prevent this situation and make the arithmetic more safe. Fixes: 1d42509e ("sched/fair: Make EAS wakeup placement consider uclamp restrictions") Signed-off-by: Lukasz Luba <lukasz.luba@arm.com> Signed-off-by: Peter Zijlstra (Intel) <peterz@infradead.org> Reviewed-by: Valentin Schneider <valentin.schneider@arm.com> Link: https://lkml.kernel.org/r/20200810083004.26420-1-lukasz.luba@arm.com
-
Josh Don authored
SMT siblings share caches, so cache hotness should be irrelevant for cross-sibling migration. Signed-off-by: Josh Don <joshdon@google.com> Proposed-by: Venkatesh Pallipadi <venki@google.com> Signed-off-by: Peter Zijlstra (Intel) <peterz@infradead.org> Link: https://lkml.kernel.org/r/20200804193413.510651-1-joshdon@google.com
-
- 19 Aug, 2020 16 commits
-
-
Valentin Schneider authored
There would be no point in preserving a sched_domain with a single group just because it has this flag set. Add it to SD_DEGENERATE_GROUPS_MASK. Signed-off-by: Valentin Schneider <valentin.schneider@arm.com> Signed-off-by: Ingo Molnar <mingo@kernel.org> Acked-by: Peter Zijlstra <a.p.zijlstra@chello.nl> Link: https://lore.kernel.org/r/20200817113003.20802-17-valentin.schneider@arm.com
-
Valentin Schneider authored
A sched_domain can only have overlapping sched_groups if it has more than one group. Signed-off-by: Valentin Schneider <valentin.schneider@arm.com> Signed-off-by: Ingo Molnar <mingo@kernel.org> Acked-by: Peter Zijlstra <a.p.zijlstra@chello.nl> Link: https://lore.kernel.org/r/20200817113003.20802-16-valentin.schneider@arm.com
-
Valentin Schneider authored
Being a load-balancing flag, it requires 2+ groups to have any effect. Signed-off-by: Valentin Schneider <valentin.schneider@arm.com> Signed-off-by: Ingo Molnar <mingo@kernel.org> Acked-by: Peter Zijlstra <a.p.zijlstra@chello.nl> Link: https://lore.kernel.org/r/20200817113003.20802-15-valentin.schneider@arm.com
-
Valentin Schneider authored
There would be no point in preserving a sched_domain with a single group just because it has this flag set. Add it to SD_DEGENERATE_GROUPS_MASK. Signed-off-by: Valentin Schneider <valentin.schneider@arm.com> Signed-off-by: Ingo Molnar <mingo@kernel.org> Acked-by: Peter Zijlstra <a.p.zijlstra@chello.nl> Link: https://lore.kernel.org/r/20200817113003.20802-14-valentin.schneider@arm.com
-
Valentin Schneider authored
Even if no mainline topology uses this flag, it is a load balancing flag just like SD_BALANCE_FORK and requires 2+ groups to have any effect. Signed-off-by: Valentin Schneider <valentin.schneider@arm.com> Signed-off-by: Ingo Molnar <mingo@kernel.org> Acked-by: Peter Zijlstra <a.p.zijlstra@chello.nl> Link: https://lore.kernel.org/r/20200817113003.20802-13-valentin.schneider@arm.com
-
Valentin Schneider authored
SD_PREFER_SIBLING is currently considered in sd_parent_degenerate() but not in sd_degenerate(). It too hinges on load balancing, and thus won't have any effect when set on a domain with a single group. Add it to SD_DEGENERATE_GROUPS_MASK. Signed-off-by: Valentin Schneider <valentin.schneider@arm.com> Signed-off-by: Ingo Molnar <mingo@kernel.org> Acked-by: Peter Zijlstra <a.p.zijlstra@chello.nl> Link: https://lore.kernel.org/r/20200817113003.20802-12-valentin.schneider@arm.com
-
Valentin Schneider authored
We currently set this flag *only* on domains whose topology level exactly match the level where we detect asymmetry (as returned by asym_cpu_capacity_level()). This is rather problematic. Say there are two clusters in the system, one with a lone big CPU and the other with a mix of big and LITTLE CPUs (as is allowed by DynamIQ): DIE [ ] MC [ ][ ] 0 1 2 3 4 L L B B B asym_cpu_capacity_level() will figure out that the MC level is the one where all CPUs can see a CPU of max capacity, and we will thus set SD_ASYM_CPUCAPACITY at MC level for all CPUs. That lone big CPU will degenerate its MC domain, since it would be alone in there, and will end up with just a DIE domain. Since the flag was only set at MC, this CPU ends up not seeing any SD with the flag set, which is broken. Rather than clearing dflags at every topology level, clear it before entering the topology level loop. This will properly propagate upwards flags that are set starting from a certain level. Signed-off-by: Valentin Schneider <valentin.schneider@arm.com> Signed-off-by: Ingo Molnar <mingo@kernel.org> Reviewed-by: Quentin Perret <qperret@google.com> Reviewed-by: Dietmar Eggemann <dietmar.eggemann@arm.com> Acked-by: Peter Zijlstra <a.p.zijlstra@chello.nl> Link: https://lore.kernel.org/r/20200817113003.20802-11-valentin.schneider@arm.com
-
Valentin Schneider authored
If there is only a single NUMA node in the system, the only NUMA topology level that will be generated will be NODE (identity distance), which doesn't have SD_SERIALIZE. This means we don't need this special case in sd_parent_degenerate(), as having the NODE level "naturally" covers it. Thus, remove it. Suggested-by: Peter Zijlstra <peterz@infradead.org> Signed-off-by: Valentin Schneider <valentin.schneider@arm.com> Signed-off-by: Ingo Molnar <mingo@kernel.org> Reviewed-by: Dietmar Eggemann <dietmar.eggemann@arm.com> Acked-by: Peter Zijlstra <a.p.zijlstra@chello.nl> Link: https://lore.kernel.org/r/20200817113003.20802-10-valentin.schneider@arm.com
-
Valentin Schneider authored
Leverage SD_DEGENERATE_GROUPS_MASK in sd_degenerate() and sd_parent_degenerate(). Suggested-by: Peter Zijlstra <peterz@infradead.org> Signed-off-by: Valentin Schneider <valentin.schneider@arm.com> Signed-off-by: Ingo Molnar <mingo@kernel.org> Acked-by: Peter Zijlstra <a.p.zijlstra@chello.nl> Link: https://lore.kernel.org/r/20200817113003.20802-9-valentin.schneider@arm.com
-
Valentin Schneider authored
In preparation of cleaning up the sd_degenerate*() functions, mark flags used in sd_degenerate() with the new SDF_NEEDS_GROUPS flag. With this, build a compile-time mask of those SD flags. Note that sd_parent_degenerate() uses an extra flag in its mask, SD_PREFER_SIBLING, which remains singled out for now. Suggested-by: Peter Zijlstra <peterz@infradead.org> Signed-off-by: Valentin Schneider <valentin.schneider@arm.com> Signed-off-by: Ingo Molnar <mingo@kernel.org> Reviewed-by: Dietmar Eggemann <dietmar.eggemann@arm.com> Acked-by: Peter Zijlstra <a.p.zijlstra@chello.nl> Link: https://lore.kernel.org/r/20200817113003.20802-8-valentin.schneider@arm.com
-
Valentin Schneider authored
Decoding the output of /proc/sys/kernel/sched_domain/cpu*/domain*/flags has always been somewhat annoying, as one needs to go fetch the bit -> name mapping from the source code itself. This encoding can be saved in a script somewhere, but that isn't safe from flags being added, removed or even shuffled around. What matters for debugging purposes is to get *which* flags are set in a given domain, their associated value is pretty much meaningless. Make the sd flags debug file output flag names. Signed-off-by: Valentin Schneider <valentin.schneider@arm.com> Signed-off-by: Ingo Molnar <mingo@kernel.org> Acked-by: Peter Zijlstra <a.p.zijlstra@chello.nl> Link: https://lore.kernel.org/r/20200817113003.20802-7-valentin.schneider@arm.com
-
Valentin Schneider authored
Now that we have some description of what we expect the flags layout to be, we can use that to assert at runtime that the actual layout is sane. Signed-off-by: Valentin Schneider <valentin.schneider@arm.com> Signed-off-by: Ingo Molnar <mingo@kernel.org> Reviewed-by: Dietmar Eggemann <dietmar.eggemann@arm.com> Acked-by: Peter Zijlstra <a.p.zijlstra@chello.nl> Link: https://lore.kernel.org/r/20200817113003.20802-6-valentin.schneider@arm.com
-
Valentin Schneider authored
There are some expectations regarding how sched domain flags should be laid out, but none of them are checked or asserted in sched_domain_debug_one(). After staring at said flags for a while, I've come to realize there's two repeating patterns: - Shared with children: those flags are set from the base CPU domain upwards. Any domain that has it set will have it set in its children. It hints at "some property holds true / some behaviour is enabled until this level". - Shared with parents: those flags are set from the topmost domain downwards. Any domain that has it set will have it set in its parents. It hints at "some property isn't visible / some behaviour is disabled until this level". There are two outliers that (currently) do not map to either of these: o SD_PREFER_SIBLING, which is cleared below levels with SD_ASYM_CPUCAPACITY. The change was introduced by commit: 9c63e84d ("sched/core: Disable SD_PREFER_SIBLING on asymmetric CPU capacity domains") as it could break misfit migration on some systems. In light of this, we might want to change it back to make it fit one of the two categories and fix the issue another way. o SD_ASYM_CPUCAPACITY, which gets set on a single level and isn't propagated up nor down. From a topology description point of view, it really wants to be SDF_SHARED_PARENT; this will be rectified in a later patch. Tweak the sched_domain flag declaration to assign each flag an expected layout, and include the rationale for each flag "meta type" assignment as a comment. Consolidate the flag metadata into an array; the index of a flag's metadata can easily be found with log2(flag), IOW __ffs(flag). Signed-off-by: Valentin Schneider <valentin.schneider@arm.com> Signed-off-by: Ingo Molnar <mingo@kernel.org> Reviewed-by: Dietmar Eggemann <dietmar.eggemann@arm.com> Acked-by: Peter Zijlstra <a.p.zijlstra@chello.nl> Link: https://lore.kernel.org/r/20200817113003.20802-5-valentin.schneider@arm.com
-
Valentin Schneider authored
To associate the SD flags with some metadata, we need some more structure in the way they are declared. Rather than shove that in a free-standing macro list, move the declaration in a separate file that can be re-imported with different SD_FLAG definitions. This is inspired by what is done with the syscall table (see uapi/asm/unistd.h and sys_call_table). The value assigned to a given SD flag now depends on the order it appears in sd_flags.h. No change in functionality. Signed-off-by: Valentin Schneider <valentin.schneider@arm.com> Signed-off-by: Ingo Molnar <mingo@kernel.org> Reviewed-by: Dietmar Eggemann <dietmar.eggemann@arm.com> Acked-by: Peter Zijlstra <a.p.zijlstra@chello.nl> Link: https://lore.kernel.org/r/20200817113003.20802-4-valentin.schneider@arm.com
-
Valentin Schneider authored
The ARM-specific GMC level is meant to be built using the thread sibling mask, but no devicetree in arch/arm/boot/dts uses the 'thread' cpu-map binding. With SD_SHARE_POWERDOMAIN gone, this topology level can be removed, at which point ARM no longer benefits from having a custom defined topology table. Delete the GMC topology level by making ARM use the default scheduler topology table. This essentially reverts commit: fb2aa855 ("sched, ARM: Create a dedicated scheduler topology table") No change in functionality is expected. Suggested-by: Dietmar Eggemann <dietmar.eggemann@arm.com> Signed-off-by: Valentin Schneider <valentin.schneider@arm.com> Signed-off-by: Ingo Molnar <mingo@kernel.org> Reviewed-by: Dietmar Eggemann <dietmar.eggemann@arm.com> Acked-by: Peter Zijlstra <a.p.zijlstra@chello.nl> Link: https://lore.kernel.org/r/20200817113003.20802-3-valentin.schneider@arm.com
-
Valentin Schneider authored
This flag was introduced in 2014 by commit: d77b3ed5 ("sched: Add a new SD_SHARE_POWERDOMAIN for sched_domain") but AFAIA it was never leveraged by the scheduler. The closest thing I can think of is EAS caring about frequency domains, and it does that by leveraging performance domains. Remove the flag. No change in functionality is expected. Suggested-by: Morten Rasmussen <morten.rasmussen@arm.com> Signed-off-by: Valentin Schneider <valentin.schneider@arm.com> Signed-off-by: Ingo Molnar <mingo@kernel.org> Reviewed-by: Dietmar Eggemann <dietmar.eggemann@arm.com> Acked-by: Peter Zijlstra <a.p.zijlstra@chello.nl> Link: https://lore.kernel.org/r/20200817113003.20802-2-valentin.schneider@arm.com
-