- 28 Jan, 2020 7 commits
-
-
Giovanni Gherdovich authored
Implement arch_scale_freq_capacity() for 'modern' x86. This function is used by the scheduler to correctly account usage in the face of DVFS. The present patch addresses Intel processors specifically and has positive performance and performance-per-watt implications for the schedutil cpufreq governor, bringing it closer to, if not on-par with, the powersave governor from the intel_pstate driver/framework. Large performance gains are obtained when the machine is lightly loaded and no regression are observed at saturation. The benchmarks with the largest gains are kernel compilation, tbench (the networking version of dbench) and shell-intensive workloads. 1. FREQUENCY INVARIANCE: MOTIVATION * Without it, a task looks larger if the CPU runs slower 2. PECULIARITIES OF X86 * freq invariance accounting requires knowing the ratio freq_curr/freq_max 2.1 CURRENT FREQUENCY * Use delta_APERF / delta_MPERF * freq_base (a.k.a "BusyMHz") 2.2 MAX FREQUENCY * It varies with time (turbo). As an approximation, we set it to a constant, i.e. 4-cores turbo frequency. 3. EFFECTS ON THE SCHEDUTIL FREQUENCY GOVERNOR * The invariant schedutil's formula has no feedback loop and reacts faster to utilization changes 4. KNOWN LIMITATIONS * In some cases tasks can't reach max util despite how hard they try 5. PERFORMANCE TESTING 5.1 MACHINES * Skylake, Broadwell, Haswell 5.2 SETUP * baseline Linux v5.2 w/ non-invariant schedutil. Tested freq_max = 1-2-3-4-8-12 active cores turbo w/ invariant schedutil, and intel_pstate/powersave 5.3 BENCHMARK RESULTS 5.3.1 NEUTRAL BENCHMARKS * NAS Parallel Benchmark (HPC), hackbench 5.3.2 NON-NEUTRAL BENCHMARKS * tbench (10-30% better), kernbench (10-15% better), shell-intensive-scripts (30-50% better) * no regressions 5.3.3 SELECTION OF DETAILED RESULTS 5.3.4 POWER CONSUMPTION, PERFORMANCE-PER-WATT * dbench (5% worse on one machine), kernbench (3% worse), tbench (5-10% better), shell-intensive-scripts (10-40% better) 6. MICROARCH'ES ADDRESSED HERE * Xeon Core before Scalable Performance processors line (Xeon Gold/Platinum etc have different MSRs semantic for querying turbo levels) 7. REFERENCES * MMTests performance testing framework, github.com/gormanm/mmtests +-------------------------------------------------------------------------+ | 1. FREQUENCY INVARIANCE: MOTIVATION +-------------------------------------------------------------------------+ For example; suppose a CPU has two frequencies: 500 and 1000 Mhz. When running a task that would consume 1/3rd of a CPU at 1000 MHz, it would appear to consume 2/3rd (or 66.6%) when running at 500 MHz, giving the false impression this CPU is almost at capacity, even though it can go faster [*]. In a nutshell, without frequency scale-invariance tasks look larger just because the CPU is running slower. [*] (footnote: this assumes a linear frequency/performance relation; which everybody knows to be false, but given realities its the best approximation we can make.) +-------------------------------------------------------------------------+ | 2. PECULIARITIES OF X86 +-------------------------------------------------------------------------+ Accounting for frequency changes in PELT signals requires the computation of the ratio freq_curr / freq_max. On x86 neither of those terms is readily available. 2.1 CURRENT FREQUENCY ==================== Since modern x86 has hardware control over the actual frequency we run at (because amongst other things, Turbo-Mode), we cannot simply use the frequency as requested through cpufreq. Instead we use the APERF/MPERF MSRs to compute the effective frequency over the recent past. Also, because reading MSRs is expensive, don't do so every time we need the value, but amortize the cost by doing it every tick. 2.2 MAX FREQUENCY ================= Obtaining freq_max is also non-trivial because at any time the hardware can provide a frequency boost to a selected subset of cores if the package has enough power to spare (eg: Turbo Boost). This means that the maximum frequency available to a given core changes with time. The approach taken in this change is to arbitrarily set freq_max to a constant value at boot. The value chosen is the "4-cores (4C) turbo frequency" on most microarchitectures, after evaluating the following candidates: * 1-core (1C) turbo frequency (the fastest turbo state available) * around base frequency (a.k.a. max P-state) * something in between, such as 4C turbo To interpret these options, consider that this is the denominator in freq_curr/freq_max, and that ratio will be used to scale PELT signals such as util_avg and load_avg. A large denominator will undershoot (util_avg looks a bit smaller than it really is), viceversa with a smaller denominator PELT signals will tend to overshoot. Given that PELT drives frequency selection in the schedutil governor, we will have: freq_max set to | effect on DVFS --------------------+------------------ 1C turbo | power efficiency (lower freq choices) base freq | performance (higher util_avg, higher freq requests) 4C turbo | a bit of both 4C turbo proves to be a good compromise in a number of benchmarks (see below). +-------------------------------------------------------------------------+ | 3. EFFECTS ON THE SCHEDUTIL FREQUENCY GOVERNOR +-------------------------------------------------------------------------+ Once an architecture implements a frequency scale-invariant utilization (the PELT signal util_avg), schedutil switches its frequency selection formula from freq_next = 1.25 * freq_curr * util [non-invariant util signal] to freq_next = 1.25 * freq_max * util [invariant util signal] where, in the second formula, freq_max is set to the 1C turbo frequency (max turbo). The advantage of the second formula, whose usage we unlock with this patch, is that freq_next doesn't depend on the current frequency in an iterative fashion, but can jump to any frequency in a single update. This absence of feedback in the formula makes it quicker to react to utilization changes and more robust against pathological instabilities. Compare it to the update formula of intel_pstate/powersave: freq_next = 1.25 * freq_max * Busy% where again freq_max is 1C turbo and Busy% is the percentage of time not spent idling (calculated with delta_MPERF / delta_TSC); essentially the same as invariant schedutil, and largely responsible for intel_pstate/powersave good reputation. The non-invariant schedutil formula is derived from the invariant one by approximating util_inv with util_raw * freq_curr / freq_max, but this has limitations. Testing shows improved performances due to better frequency selections when the machine is lightly loaded, and essentially no change in behaviour at saturation / overutilization. +-------------------------------------------------------------------------+ | 4. KNOWN LIMITATIONS +-------------------------------------------------------------------------+ It's been shown that it is possible to create pathological scenarios where a CPU-bound task cannot reach max utilization, if the normalizing factor freq_max is fixed to a constant value (see [Lelli-2018]). If freq_max is set to 4C turbo as we do here, one needs to peg at least 5 cores in a package doing some busywork, and observe that none of those task will ever reach max util (1024) because they're all running at less than the 4C turbo frequency. While this concern still applies, we believe the performance benefit of frequency scale-invariant PELT signals outweights the cost of this limitation. [Lelli-2018] https://lore.kernel.org/lkml/20180517150418.GF22493@localhost.localdomain/ +-------------------------------------------------------------------------+ | 5. PERFORMANCE TESTING +-------------------------------------------------------------------------+ 5.1 MACHINES ============ We tested the patch on three machines, with Skylake, Broadwell and Haswell CPUs. The details are below, together with the available turbo ratios as reported by the appropriate MSRs. * 8x-SKYLAKE-UMA: Single socket E3-1240 v5, Skylake 4 cores/8 threads Max EFFiciency, BASE frequency and available turbo levels (MHz): EFFIC 800 |******** BASE 3500 |*********************************** 4C 3700 |************************************* 3C 3800 |************************************** 2C 3900 |*************************************** 1C 3900 |*************************************** * 80x-BROADWELL-NUMA: Two sockets E5-2698 v4, 2x Broadwell 20 cores/40 threads Max EFFiciency, BASE frequency and available turbo levels (MHz): EFFIC 1200 |************ BASE 2200 |********************** 8C 2900 |***************************** 7C 3000 |****************************** 6C 3100 |******************************* 5C 3200 |******************************** 4C 3300 |********************************* 3C 3400 |********************************** 2C 3600 |************************************ 1C 3600 |************************************ * 48x-HASWELL-NUMA Two sockets E5-2670 v3, 2x Haswell 12 cores/24 threads Max EFFiciency, BASE frequency and available turbo levels (MHz): EFFIC 1200 |************ BASE 2300 |*********************** 12C 2600 |************************** 11C 2600 |************************** 10C 2600 |************************** 9C 2600 |************************** 8C 2600 |************************** 7C 2600 |************************** 6C 2600 |************************** 5C 2700 |*************************** 4C 2800 |**************************** 3C 2900 |***************************** 2C 3100 |******************************* 1C 3100 |******************************* 5.2 SETUP ========= * The baseline is Linux v5.2 with schedutil (non-invariant) and the intel_pstate driver in passive mode. * The rationale for choosing the various freq_max values to test have been to try all the 1-2-3-4C turbo levels (note that 1C and 2C turbo are identical on all machines), plus one more value closer to base_freq but still in the turbo range (8C turbo for both 80x-BROADWELL-NUMA and 48x-HASWELL-NUMA). * In addition we've run all tests with intel_pstate/powersave for comparison. * The filesystem is always XFS, the userspace is openSUSE Leap 15.1. * 8x-SKYLAKE-UMA is capable of HWP (Hardware-Managed P-States), so the runs with active intel_pstate on this machine use that. This gives, in terms of combinations tested on each machine: * 8x-SKYLAKE-UMA * Baseline: Linux v5.2, non-invariant schedutil, intel_pstate passive * intel_pstate active + powersave + HWP * invariant schedutil, freq_max = 1C turbo * invariant schedutil, freq_max = 3C turbo * invariant schedutil, freq_max = 4C turbo * both 80x-BROADWELL-NUMA and 48x-HASWELL-NUMA * [same as 8x-SKYLAKE-UMA, but no HWP capable] * invariant schedutil, freq_max = 8C turbo (which on 48x-HASWELL-NUMA is the same as 12C turbo, or "all cores turbo") 5.3 BENCHMARK RESULTS ===================== 5.3.1 NEUTRAL BENCHMARKS ------------------------ Tests that didn't show any measurable difference in performance on any of the test machines between non-invariant schedutil and our patch are: * NAS Parallel Benchmarks (NPB) using either MPI or openMP for IPC, any computational kernel * flexible I/O (FIO) * hackbench (using threads or processes, and using pipes or sockets) 5.3.2 NON-NEUTRAL BENCHMARKS ---------------------------- What follow are summary tables where each benchmark result is given a score. * A tilde (~) means a neutral result, i.e. no difference from baseline. * Scores are computed with the ratio result_new / result_baseline, so a tilde means a score of 1.00. * The results in the score ratio are the geometric means of results running the benchmark with different parameters (eg: for kernbench: using 1, 2, 4, ... number of processes; for pgbench: varying the number of clients, and so on). * The first three tables show higher-is-better kind of tests (i.e. measured in operations/second), the subsequent three show lower-is-better kind of tests (i.e. the workload is fixed and we measure elapsed time, think kernbench). * "gitsource" is a name we made up for the test consisting in running the entire unit tests suite of the Git SCM and measuring how long it takes. We take it as a typical example of shell-intensive serialized workload. * In the "I_PSTATE" column we have the results for intel_pstate/powersave. Other columns show invariant schedutil for different values of freq_max. 4C turbo is circled as it's the value we've chosen for the final implementation. 80x-BROADWELL-NUMA (comparison ratio; higher is better) +------+ I_PSTATE 1C 3C | 4C | 8C pgbench-ro 1.14 ~ ~ | 1.11 | 1.14 pgbench-rw ~ ~ ~ | ~ | ~ netperf-udp 1.06 ~ 1.06 | 1.05 | 1.07 netperf-tcp ~ 1.03 ~ | 1.01 | 1.02 tbench4 1.57 1.18 1.22 | 1.30 | 1.56 +------+ 8x-SKYLAKE-UMA (comparison ratio; higher is better) +------+ I_PSTATE/HWP 1C 3C | 4C | pgbench-ro ~ ~ ~ | ~ | pgbench-rw ~ ~ ~ | ~ | netperf-udp ~ ~ ~ | ~ | netperf-tcp ~ ~ ~ | ~ | tbench4 1.30 1.14 1.14 | 1.16 | +------+ 48x-HASWELL-NUMA (comparison ratio; higher is better) +------+ I_PSTATE 1C 3C | 4C | 12C pgbench-ro 1.15 ~ ~ | 1.06 | 1.16 pgbench-rw ~ ~ ~ | ~ | ~ netperf-udp 1.05 0.97 1.04 | 1.04 | 1.02 netperf-tcp 0.96 1.01 1.01 | 1.01 | 1.01 tbench4 1.50 1.05 1.13 | 1.13 | 1.25 +------+ In the table above we see that active intel_pstate is slightly better than our 4C-turbo patch (both in reference to the baseline non-invariant schedutil) on read-only pgbench and much better on tbench. Both cases are notable in which it shows that lowering our freq_max (to 8C-turbo and 12C-turbo on 80x-BROADWELL-NUMA and 48x-HASWELL-NUMA respectively) helps invariant schedutil to get closer. If we ignore active intel_pstate and focus on the comparison with baseline alone, there are several instances of double-digit performance improvement. 80x-BROADWELL-NUMA (comparison ratio; lower is better) +------+ I_PSTATE 1C 3C | 4C | 8C dbench4 1.23 0.95 0.95 | 0.95 | 0.95 kernbench 0.93 0.83 0.83 | 0.83 | 0.82 gitsource 0.98 0.49 0.49 | 0.49 | 0.48 +------+ 8x-SKYLAKE-UMA (comparison ratio; lower is better) +------+ I_PSTATE/HWP 1C 3C | 4C | dbench4 ~ ~ ~ | ~ | kernbench ~ ~ ~ | ~ | gitsource 0.92 0.55 0.55 | 0.55 | +------+ 48x-HASWELL-NUMA (comparison ratio; lower is better) +------+ I_PSTATE 1C 3C | 4C | 8C dbench4 ~ ~ ~ | ~ | ~ kernbench 0.94 0.90 0.89 | 0.90 | 0.90 gitsource 0.97 0.69 0.69 | 0.69 | 0.69 +------+ dbench is not very remarkable here, unless we notice how poorly active intel_pstate is performing on 80x-BROADWELL-NUMA: 23% regression versus non-invariant schedutil. We repeated that run getting consistent results. Out of scope for the patch at hand, but deserving future investigation. Other than that, we previously ran this campaign with Linux v5.0 and saw the patch doing better on dbench a the time. We haven't checked closely and can only speculate at this point. On the NUMA boxes kernbench gets 10-15% improvements on average; we'll see in the detailed tables that the gains concentrate on low process counts (lightly loaded machines). The test we call "gitsource" (running the git unit test suite, a long-running single-threaded shell script) appears rather spectacular in this table (gains of 30-50% depending on the machine). It is to be noted, however, that gitsource has no adjustable parameters (such as the number of jobs in kernbench, which we average over in order to get a single-number summary score) and is exactly the kind of low-parallelism workload that benefits the most from this patch. When looking at the detailed tables of kernbench or tbench4, at low process or client counts one can see similar numbers. 5.3.3 SELECTION OF DETAILED RESULTS ----------------------------------- Machine : 48x-HASWELL-NUMA Benchmark : tbench4 (i.e. dbench4 over the network, actually loopback) Varying parameter : number of clients Unit : MB/sec (higher is better) 5.2.0 vanilla (BASELINE) 5.2.0 intel_pstate 5.2.0 1C-turbo - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - Hmean 1 126.73 +- 0.31% ( ) 315.91 +- 0.66% ( 149.28%) 125.03 +- 0.76% ( -1.34%) Hmean 2 258.04 +- 0.62% ( ) 614.16 +- 0.51% ( 138.01%) 269.58 +- 1.45% ( 4.47%) Hmean 4 514.30 +- 0.67% ( ) 1146.58 +- 0.54% ( 122.94%) 533.84 +- 1.99% ( 3.80%) Hmean 8 1111.38 +- 2.52% ( ) 2159.78 +- 0.38% ( 94.33%) 1359.92 +- 1.56% ( 22.36%) Hmean 16 2286.47 +- 1.36% ( ) 3338.29 +- 0.21% ( 46.00%) 2720.20 +- 0.52% ( 18.97%) Hmean 32 4704.84 +- 0.35% ( ) 4759.03 +- 0.43% ( 1.15%) 4774.48 +- 0.30% ( 1.48%) Hmean 64 7578.04 +- 0.27% ( ) 7533.70 +- 0.43% ( -0.59%) 7462.17 +- 0.65% ( -1.53%) Hmean 128 6998.52 +- 0.16% ( ) 6987.59 +- 0.12% ( -0.16%) 6909.17 +- 0.14% ( -1.28%) Hmean 192 6901.35 +- 0.25% ( ) 6913.16 +- 0.10% ( 0.17%) 6855.47 +- 0.21% ( -0.66%) 5.2.0 3C-turbo 5.2.0 4C-turbo 5.2.0 12C-turbo - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - Hmean 1 128.43 +- 0.28% ( 1.34%) 130.64 +- 3.81% ( 3.09%) 153.71 +- 5.89% ( 21.30%) Hmean 2 311.70 +- 6.15% ( 20.79%) 281.66 +- 3.40% ( 9.15%) 305.08 +- 5.70% ( 18.23%) Hmean 4 641.98 +- 2.32% ( 24.83%) 623.88 +- 5.28% ( 21.31%) 906.84 +- 4.65% ( 76.32%) Hmean 8 1633.31 +- 1.56% ( 46.96%) 1714.16 +- 0.93% ( 54.24%) 2095.74 +- 0.47% ( 88.57%) Hmean 16 3047.24 +- 0.42% ( 33.27%) 3155.02 +- 0.30% ( 37.99%) 3634.58 +- 0.15% ( 58.96%) Hmean 32 4734.31 +- 0.60% ( 0.63%) 4804.38 +- 0.23% ( 2.12%) 4674.62 +- 0.27% ( -0.64%) Hmean 64 7699.74 +- 0.35% ( 1.61%) 7499.72 +- 0.34% ( -1.03%) 7659.03 +- 0.25% ( 1.07%) Hmean 128 6935.18 +- 0.15% ( -0.91%) 6942.54 +- 0.10% ( -0.80%) 7004.85 +- 0.12% ( 0.09%) Hmean 192 6901.62 +- 0.12% ( 0.00%) 6856.93 +- 0.10% ( -0.64%) 6978.74 +- 0.10% ( 1.12%) This is one of the cases where the patch still can't surpass active intel_pstate, not even when freq_max is as low as 12C-turbo. Otherwise, gains are visible up to 16 clients and the saturated scenario is the same as baseline. The scores in the summary table from the previous sections are ratios of geometric means of the results over different clients, as seen in this table. Machine : 80x-BROADWELL-NUMA Benchmark : kernbench (kernel compilation) Varying parameter : number of jobs Unit : seconds (lower is better) 5.2.0 vanilla (BASELINE) 5.2.0 intel_pstate 5.2.0 1C-turbo - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - Amean 2 379.68 +- 0.06% ( ) 330.20 +- 0.43% ( 13.03%) 285.93 +- 0.07% ( 24.69%) Amean 4 200.15 +- 0.24% ( ) 175.89 +- 0.22% ( 12.12%) 153.78 +- 0.25% ( 23.17%) Amean 8 106.20 +- 0.31% ( ) 95.54 +- 0.23% ( 10.03%) 86.74 +- 0.10% ( 18.32%) Amean 16 56.96 +- 1.31% ( ) 53.25 +- 1.22% ( 6.50%) 48.34 +- 1.73% ( 15.13%) Amean 32 34.80 +- 2.46% ( ) 33.81 +- 0.77% ( 2.83%) 30.28 +- 1.59% ( 12.99%) Amean 64 26.11 +- 1.63% ( ) 25.04 +- 1.07% ( 4.10%) 22.41 +- 2.37% ( 14.16%) Amean 128 24.80 +- 1.36% ( ) 23.57 +- 1.23% ( 4.93%) 21.44 +- 1.37% ( 13.55%) Amean 160 24.85 +- 0.56% ( ) 23.85 +- 1.17% ( 4.06%) 21.25 +- 1.12% ( 14.49%) 5.2.0 3C-turbo 5.2.0 4C-turbo 5.2.0 8C-turbo - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - Amean 2 284.08 +- 0.13% ( 25.18%) 283.96 +- 0.51% ( 25.21%) 285.05 +- 0.21% ( 24.92%) Amean 4 153.18 +- 0.22% ( 23.47%) 154.70 +- 1.64% ( 22.71%) 153.64 +- 0.30% ( 23.24%) Amean 8 87.06 +- 0.28% ( 18.02%) 86.77 +- 0.46% ( 18.29%) 86.78 +- 0.22% ( 18.28%) Amean 16 48.03 +- 0.93% ( 15.68%) 47.75 +- 1.99% ( 16.17%) 47.52 +- 1.61% ( 16.57%) Amean 32 30.23 +- 1.20% ( 13.14%) 30.08 +- 1.67% ( 13.57%) 30.07 +- 1.67% ( 13.60%) Amean 64 22.59 +- 2.02% ( 13.50%) 22.63 +- 0.81% ( 13.32%) 22.42 +- 0.76% ( 14.12%) Amean 128 21.37 +- 0.67% ( 13.82%) 21.31 +- 1.15% ( 14.07%) 21.17 +- 1.93% ( 14.63%) Amean 160 21.68 +- 0.57% ( 12.76%) 21.18 +- 1.74% ( 14.77%) 21.22 +- 1.00% ( 14.61%) The patch outperform active intel_pstate (and baseline) by a considerable margin; the summary table from the previous section says 4C turbo and active intel_pstate are 0.83 and 0.93 against baseline respectively, so 4C turbo is 0.83/0.93=0.89 against intel_pstate (~10% better on average). There is no noticeable difference with regard to the value of freq_max. Machine : 8x-SKYLAKE-UMA Benchmark : gitsource (time to run the git unit test suite) Varying parameter : none Unit : seconds (lower is better) 5.2.0 vanilla 5.2.0 intel_pstate/hwp 5.2.0 1C-turbo - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - Amean 858.85 +- 1.16% ( ) 791.94 +- 0.21% ( 7.79%) 474.95 ( 44.70%) 5.2.0 3C-turbo 5.2.0 4C-turbo - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - Amean 475.26 +- 0.20% ( 44.66%) 474.34 +- 0.13% ( 44.77%) In this test, which is of interest as representing shell-intensive (i.e. fork-intensive) serialized workloads, invariant schedutil outperforms intel_pstate/powersave by a whopping 40% margin. 5.3.4 POWER CONSUMPTION, PERFORMANCE-PER-WATT --------------------------------------------- The following table shows average power consumption in watt for each benchmark. Data comes from turbostat (package average), which in turn is read from the RAPL interface on CPUs. We know the patch affects CPU frequencies so it's reasonable to ignore other power consumers (such as memory or I/O). Also, we don't have a power meter available in the lab so RAPL is the best we have. turbostat sampled average power every 10 seconds for the entire duration of each benchmark. We took all those values and averaged them (i.e. with don't have detail on a per-parameter granularity, only on whole benchmarks). 80x-BROADWELL-NUMA (power consumption, watts) +--------+ BASELINE I_PSTATE 1C 3C | 4C | 8C pgbench-ro 130.01 142.77 131.11 132.45 | 134.65 | 136.84 pgbench-rw 68.30 60.83 71.45 71.70 | 71.65 | 72.54 dbench4 90.25 59.06 101.43 99.89 | 101.10 | 102.94 netperf-udp 65.70 69.81 66.02 68.03 | 68.27 | 68.95 netperf-tcp 88.08 87.96 88.97 88.89 | 88.85 | 88.20 tbench4 142.32 176.73 153.02 163.91 | 165.58 | 176.07 kernbench 92.94 101.95 114.91 115.47 | 115.52 | 115.10 gitsource 40.92 41.87 75.14 75.20 | 75.40 | 75.70 +--------+ 8x-SKYLAKE-UMA (power consumption, watts) +--------+ BASELINE I_PSTATE/HWP 1C 3C | 4C | pgbench-ro 46.49 46.68 46.56 46.59 | 46.52 | pgbench-rw 29.34 31.38 30.98 31.00 | 31.00 | dbench4 27.28 27.37 27.49 27.41 | 27.38 | netperf-udp 22.33 22.41 22.36 22.35 | 22.36 | netperf-tcp 27.29 27.29 27.30 27.31 | 27.33 | tbench4 41.13 45.61 43.10 43.33 | 43.56 | kernbench 42.56 42.63 43.01 43.01 | 43.01 | gitsource 13.32 13.69 17.33 17.30 | 17.35 | +--------+ 48x-HASWELL-NUMA (power consumption, watts) +--------+ BASELINE I_PSTATE 1C 3C | 4C | 12C pgbench-ro 128.84 136.04 129.87 132.43 | 132.30 | 134.86 pgbench-rw 37.68 37.92 37.17 37.74 | 37.73 | 37.31 dbench4 28.56 28.73 28.60 28.73 | 28.70 | 28.79 netperf-udp 56.70 60.44 56.79 57.42 | 57.54 | 57.52 netperf-tcp 75.49 75.27 75.87 76.02 | 76.01 | 75.95 tbench4 115.44 139.51 119.53 123.07 | 123.97 | 130.22 kernbench 83.23 91.55 95.58 95.69 | 95.72 | 96.04 gitsource 36.79 36.99 39.99 40.34 | 40.35 | 40.23 +--------+ A lower power consumption isn't necessarily better, it depends on what is done with that energy. Here are tables with the ratio of performance-per-watt on each machine and benchmark. Higher is always better; a tilde (~) means a neutral ratio (i.e. 1.00). 80x-BROADWELL-NUMA (performance-per-watt ratios; higher is better) +------+ I_PSTATE 1C 3C | 4C | 8C pgbench-ro 1.04 1.06 0.94 | 1.07 | 1.08 pgbench-rw 1.10 0.97 0.96 | 0.96 | 0.97 dbench4 1.24 0.94 0.95 | 0.94 | 0.92 netperf-udp ~ 1.02 1.02 | ~ | 1.02 netperf-tcp ~ 1.02 ~ | ~ | 1.02 tbench4 1.26 1.10 1.06 | 1.12 | 1.26 kernbench 0.98 0.97 0.97 | 0.97 | 0.98 gitsource ~ 1.11 1.11 | 1.11 | 1.13 +------+ 8x-SKYLAKE-UMA (performance-per-watt ratios; higher is better) +------+ I_PSTATE/HWP 1C 3C | 4C | pgbench-ro ~ ~ ~ | ~ | pgbench-rw 0.95 0.97 0.96 | 0.96 | dbench4 ~ ~ ~ | ~ | netperf-udp ~ ~ ~ | ~ | netperf-tcp ~ ~ ~ | ~ | tbench4 1.17 1.09 1.08 | 1.10 | kernbench ~ ~ ~ | ~ | gitsource 1.06 1.40 1.40 | 1.40 | +------+ 48x-HASWELL-NUMA (performance-per-watt ratios; higher is better) +------+ I_PSTATE 1C 3C | 4C | 12C pgbench-ro 1.09 ~ 1.09 | 1.03 | 1.11 pgbench-rw ~ 0.86 ~ | ~ | 0.86 dbench4 ~ 1.02 1.02 | 1.02 | ~ netperf-udp ~ 0.97 1.03 | 1.02 | ~ netperf-tcp 0.96 ~ ~ | ~ | ~ tbench4 1.24 ~ 1.06 | 1.05 | 1.11 kernbench 0.97 0.97 0.98 | 0.97 | 0.96 gitsource 1.03 1.33 1.32 | 1.32 | 1.33 +------+ These results are overall pleasing: in plenty of cases we observe performance-per-watt improvements. The few regressions (read/write pgbench and dbench on the Broadwell machine) are of small magnitude. kernbench loses a few percentage points (it has a 10-15% performance improvement, but apparently the increase in power consumption is larger than that). tbench4 and gitsource, which benefit the most from the patch, keep a positive score in this table which is a welcome surprise; that suggests that in those particular workloads the non-invariant schedutil (and active intel_pstate, too) makes some rather suboptimal frequency selections. +-------------------------------------------------------------------------+ | 6. MICROARCH'ES ADDRESSED HERE +-------------------------------------------------------------------------+ The patch addresses Xeon Core processors that use MSR_PLATFORM_INFO and MSR_TURBO_RATIO_LIMIT to advertise their base frequency and turbo frequencies respectively. This excludes the recent Xeon Scalable Performance processors line (Xeon Gold, Platinum etc) whose MSRs have to be parsed differently. Subsequent patches will address: * Xeon Scalable Performance processors and Atom Goldmont/Goldmont Plus * Xeon Phi (Knights Landing, Knights Mill) * Atom Silvermont +-------------------------------------------------------------------------+ | 7. REFERENCES +-------------------------------------------------------------------------+ Tests have been run with the help of the MMTests performance testing framework, see github.com/gormanm/mmtests. The configuration file names for the benchmark used are: db-pgbench-timed-ro-small-xfs db-pgbench-timed-rw-small-xfs io-dbench4-async-xfs network-netperf-unbound network-tbench scheduler-unbound workload-kerndevel-xfs workload-shellscripts-xfs hpc-nas-c-class-mpi-full-xfs hpc-nas-c-class-omp-full All those benchmarks are generally available on the web: pgbench: https://www.postgresql.org/docs/10/pgbench.html netperf: https://hewlettpackard.github.io/netperf/ dbench/tbench: https://dbench.samba.org/ gitsource: git unit test suite, github.com/git/git NAS Parallel Benchmarks: https://www.nas.nasa.gov/publications/npb.html hackbench: https://people.redhat.com/mingo/cfs-scheduler/tools/hackbench.cSuggested-by:Peter Zijlstra <peterz@infradead.org> Signed-off-by:
Giovanni Gherdovich <ggherdovich@suse.cz> Signed-off-by:
Ingo Molnar <mingo@kernel.org> Acked-by:
Doug Smythies <dsmythies@telus.net> Acked-by:
Rafael J. Wysocki <rafael.j.wysocki@intel.com> Link: https://lkml.kernel.org/r/20200122151617.531-2-ggherdovich@suse.cz
-
Vincent Guittot authored
When a running task is moved on a throttled task group and there is no other task enqueued on the CPU, the task can keep running using 100% CPU whatever the allocated bandwidth for the group and although its cfs rq is throttled. Furthermore, the group entity of the cfs_rq and its parents are not enqueued but only set as curr on their respective cfs_rqs. We have the following sequence: sched_move_task -dequeue_task: dequeue task and group_entities. -put_prev_task: put task and group entities. -sched_change_group: move task to new group. -enqueue_task: enqueue only task but not group entities because cfs_rq is throttled. -set_next_task : set task and group_entities as current sched_entity of their cfs_rq. Another impact is that the root cfs_rq runnable_load_avg at root rq stays null because the group_entities are not enqueued. This situation will stay the same until an "external" event triggers a reschedule. Let trigger it immediately instead. Signed-off-by:Vincent Guittot <vincent.guittot@linaro.org> Signed-off-by:
Ben Segall <bsegall@google.com> Link: https://lkml.kernel.org/r/1579011236-31256-1-git-send-email-vincent.guittot@linaro.org
-
Wanpeng Li authored
On a machine, CPU 0 is used for housekeeping, the other 39 CPUs in the same socket are in nohz_full mode. We can observe huge time burn in the loop for seaching nearest busy housekeeper cpu by ftrace. 2) | get_nohz_timer_target() { 2) 0.240 us | housekeeping_test_cpu(); 2) 0.458 us | housekeeping_test_cpu(); ... 2) 0.292 us | housekeeping_test_cpu(); 2) 0.240 us | housekeeping_test_cpu(); 2) 0.227 us | housekeeping_any_cpu(); 2) + 43.460 us | } This patch optimizes the searching logic by finding a nearest housekeeper CPU in the housekeeping cpumask, it can minimize the worst searching time from ~44us to < 10us in my testing. In addition, the last iterated busy housekeeper can become a random candidate while current CPU is a better fallback if it is a housekeeper. Signed-off-by:Wanpeng Li <wanpengli@tencent.com> Signed-off-by:
Frederic Weisbecker <frederic@kernel.org> Link: https://lkml.kernel.org/r/1578876627-11938-1-git-send-email-wanpengli@tencent.com
-
Qais Yousef authored
The check to ensure that the new written value into cpu.uclamp.{min,max} is within range, [0:100], wasn't working because of the signed comparison 7301 if (req.percent > UCLAMP_PERCENT_SCALE) { 7302 req.ret = -ERANGE; 7303 return req; 7304 } # echo -1 > cpu.uclamp.min # cat cpu.uclamp.min 42949671.96 Cast req.percent into u64 to force the comparison to be unsigned and work as intended in capacity_from_percent(). # echo -1 > cpu.uclamp.min sh: write error: Numerical result out of range Fixes: 2480c093 ("sched/uclamp: Extend CPU's cgroup controller") Signed-off-by:Qais Yousef <qais.yousef@arm.com> Signed-off-by:
-
Mel Gorman authored
The CPU load balancer balances between different domains to spread load and strives to have equal balance everywhere. Communicating tasks can migrate so they are topologically close to each other but these decisions are independent. On a lightly loaded NUMA machine, two communicating tasks pulled together at wakeup time can be pushed apart by the load balancer. In isolation, the load balancer decision is fine but it ignores the tasks data locality and the wakeup/LB paths continually conflict. NUMA balancing is also a factor but it also simply conflicts with the load balancer. This patch allows a fixed degree of imbalance of two tasks to exist between NUMA domains regardless of utilisation levels. In many cases, this prevents communicating tasks being pulled apart. It was evaluated whether the imbalance should be scaled to the domain size. However, no additional benefit was measured across a range of workloads and machines and scaling adds the risk that lower domains have to be rebalanced. While this could change again in the future, such a change should specify the use case and benefit. The most obvious impact is on netperf TCP_STREAM -- two simple communicating tasks with some softirq offload depending on the transmission rate. 2-socket Haswell machine 48 core, HT enabled netperf-tcp -- mmtests config config-network-netperf-unbound baseline lbnuma-v3 Hmean 64 568.73 ( 0.00%) 577.56 * 1.55%* Hmean 128 1089.98 ( 0.00%) 1128.06 * 3.49%* Hmean 256 2061.72 ( 0.00%) 2104.39 * 2.07%* Hmean 1024 7254.27 ( 0.00%) 7557.52 * 4.18%* Hmean 2048 11729.20 ( 0.00%) 13350.67 * 13.82%* Hmean 3312 15309.08 ( 0.00%) 18058.95 * 17.96%* Hmean 4096 17338.75 ( 0.00%) 20483.66 * 18.14%* Hmean 8192 25047.12 ( 0.00%) 27806.84 * 11.02%* Hmean 16384 27359.55 ( 0.00%) 33071.88 * 20.88%* Stddev 64 2.16 ( 0.00%) 2.02 ( 6.53%) Stddev 128 2.31 ( 0.00%) 2.19 ( 5.05%) Stddev 256 11.88 ( 0.00%) 3.22 ( 72.88%) Stddev 1024 23.68 ( 0.00%) 7.24 ( 69.43%) Stddev 2048 79.46 ( 0.00%) 71.49 ( 10.03%) Stddev 3312 26.71 ( 0.00%) 57.80 (-116.41%) Stddev 4096 185.57 ( 0.00%) 96.15 ( 48.19%) Stddev 8192 245.80 ( 0.00%) 100.73 ( 59.02%) Stddev 16384 207.31 ( 0.00%) 141.65 ( 31.67%) In this case, there was a sizable improvement to performance and a general reduction in variance. However, this is not univeral. For most machines, the impact was roughly a 3% performance gain. Ops NUMA base-page range updates 19796.00 292.00 Ops NUMA PTE updates 19796.00 292.00 Ops NUMA PMD updates 0.00 0.00 Ops NUMA hint faults 16113.00 143.00 Ops NUMA hint local faults % 8407.00 142.00 Ops NUMA hint local percent 52.18 99.30 Ops NUMA pages migrated 4244.00 1.00 Without the patch, only 52.18% of sampled accesses are local. In an earlier changelog, 100% of sampled accesses are local and indeed on most machines, this was still the case. In this specific case, the local sampled rates was 99.3% but note the "base-page range updates" and "PTE updates". The activity with the patch is negligible as were the number of faults. The small number of pages migrated were related to shared libraries. A 2-socket Broadwell showed better results on average but are not presented for brevity as the performance was similar except it showed 100% of the sampled NUMA hints were local. The patch holds up for a 4-socket Haswell, an AMD EPYC and AMD Epyc 2 machine. For dbench, the impact depends on the filesystem used and the number of clients. On XFS, there is little difference as the clients typically communicate with workqueues which have a separate class of scheduler problem at the moment. For ext4, performance is generally better, particularly for small numbers of clients as NUMA balancing activity is negligible with the patch applied. A more interesting example is the Facebook schbench which uses a number of messaging threads to communicate with worker threads. In this configuration, one messaging thread is used per NUMA node and the number of worker threads is varied. The 50, 75, 90, 95, 99, 99.5 and 99.9 percentiles for response latency is then reported. Lat 50.00th-qrtle-1 44.00 ( 0.00%) 37.00 ( 15.91%) Lat 75.00th-qrtle-1 53.00 ( 0.00%) 41.00 ( 22.64%) Lat 90.00th-qrtle-1 57.00 ( 0.00%) 42.00 ( 26.32%) Lat 95.00th-qrtle-1 63.00 ( 0.00%) 43.00 ( 31.75%) Lat 99.00th-qrtle-1 76.00 ( 0.00%) 51.00 ( 32.89%) Lat 99.50th-qrtle-1 89.00 ( 0.00%) 52.00 ( 41.57%) Lat 99.90th-qrtle-1 98.00 ( 0.00%) 55.00 ( 43.88%) Lat 50.00th-qrtle-2 42.00 ( 0.00%) 42.00 ( 0.00%) Lat 75.00th-qrtle-2 48.00 ( 0.00%) 47.00 ( 2.08%) Lat 90.00th-qrtle-2 53.00 ( 0.00%) 52.00 ( 1.89%) Lat 95.00th-qrtle-2 55.00 ( 0.00%) 53.00 ( 3.64%) Lat 99.00th-qrtle-2 62.00 ( 0.00%) 60.00 ( 3.23%) Lat 99.50th-qrtle-2 63.00 ( 0.00%) 63.00 ( 0.00%) Lat 99.90th-qrtle-2 68.00 ( 0.00%) 66.00 ( 2.94% For higher worker threads, the differences become negligible but it's interesting to note the difference in wakeup latency at low utilisation and mpstat confirms that activity was almost all on one node until the number of worker threads increase. Hackbench generally showed neutral results across a range of machines. This is different to earlier versions of the patch which allowed imbalances for higher degrees of utilisation. perf bench pipe showed negligible differences in overall performance as the differences are very close to the noise. An earlier prototype of the patch showed major regressions for NAS C-class when running with only half of the available CPUs -- 20-30% performance hits were measured at the time. With this version of the patch, the impact is negligible with small gains/losses within the noise measured. This is because the number of threads far exceeds the small imbalance the aptch cares about. Similarly, there were report of regressions for the autonuma benchmark against earlier versions but again, normal load balancing now applies for that workload. In general, the patch simply seeks to avoid unnecessary cross-node migrations in the basic case where imbalances are very small. For low utilisation communicating workloads, this patch generally behaves better with less NUMA balancing activity. For high utilisation, there is no change in behaviour. Signed-off-by:
Mel Gorman <mgorman@techsingularity.net> Signed-off-by:
Valentin Schneider <valentin.schneider@arm.com> Reviewed-by:
Srikar Dronamraju <srikar@linux.vnet.ibm.com> Acked-by:
Phil Auld <pauld@redhat.com> Tested-by:
-
Peter Zijlstra (Intel) authored
The way loadavg is tracked during nohz only pays attention to the load upon entering nohz. This can be particularly noticeable if full nohz is entered while non-idle, and then the cpu goes idle and stays that way for a long time. Use the remote tick to ensure that full nohz cpus report their deltas within a reasonable time. [ swood: Added changelog and removed recheck of stopped tick. ] Signed-off-by:
Scott Wood <swood@redhat.com> Signed-off-by:
-
Scott Wood authored
This will be used in the next patch to get a loadavg update from nohz cpus. The delta check is skipped because idle_sched_class doesn't update se.exec_start. Signed-off-by:
-
- 20 Jan, 2020 1 commit
-
-
Viresh Kumar authored
sched_idle_cpu() isn't used for non SMP configuration and with a recent change, we have started getting following warning: kernel/sched/fair.c:5221:12: warning: ‘sched_idle_cpu’ defined but not used [-Wunused-function] Fix that by defining sched_idle_cpu() only for SMP configurations. Fixes: 323af6de ("sched/fair: Load balance aggressively for SCHED_IDLE CPUs") Reported-by:
Stephen Rothwell <sfr@canb.auug.org.au> Signed-off-by:
Viresh Kumar <viresh.kumar@linaro.org> Signed-off-by:
-
- 17 Jan, 2020 14 commits
-
-
Valentin Schneider authored
topology.c::get_group() relies on the assumption that non-NUMA domains do not partially overlap. Zeng Tao pointed out in [1] that such topology descriptions, while completely bogus, can end up being exposed to the scheduler. In his example (8 CPUs, 2-node system), we end up with: MC span for CPU3 == 3-7 MC span for CPU4 == 4-7 The first pass through get_group(3, sdd@MC) will result in the following sched_group list: 3 -> 4 -> 5 -> 6 -> 7 ^ / `----------------' And a later pass through get_group(4, sdd@MC) will "corrupt" that to: 3 -> 4 -> 5 -> 6 -> 7 ^ / `-----------' which will completely break things like 'while (sg != sd->groups)' when using CPU3's base sched_domain. There already are some architecture-specific checks in place such as x86/kernel/smpboot.c::topology.sane(), but this is something we can detect in the core scheduler, so it seems worthwhile to do so. Warn and abort the construction of the sched domains if such a broken topology description is detected. Note that this is somewhat expensive (O(t.c²), 't' non-NUMA topology levels and 'c' CPUs) and could be gated under SCHED_DEBUG if deemed necessary. Testing ======= Dietmar managed to reproduce this using the following qemu incantation: $ qemu-system-aarch64 -kernel ./Image -hda ./qemu-image-aarch64.img \ -append 'root=/dev/vda console=ttyAMA0 loglevel=8 sched_debug' -smp \ cores=8 --nographic -m 512 -cpu cortex-a53 -machine virt -numa \ node,cpus=0-2,nodeid=0 -numa node,cpus=3-7,nodeid=1 alongside the following drivers/base/arch_topology.c hack (AIUI wouldn't be needed if '-smp cores=X, sockets=Y' would work with qemu): 8<--- @@ -465,6 +465,9 @@ void update_siblings_masks(unsigned int cpuid) if (cpuid_topo->package_id != cpu_topo->package_id) continue; + if ((cpu < 4 && cpuid > 3) || (cpu > 3 && cpuid < 4)) + continue; + cpumask_set_cpu(cpuid, &cpu_topo->core_sibling); cpumask_set_cpu(cpu, &cpuid_topo->core_sibling); 8<--- [1]: https://lkml.kernel.org/r/1577088979-8545-1-git-send-email-prime.zeng@hisilicon.comReported-by:Zeng Tao <prime.zeng@hisilicon.com> Signed-off-by:
-
Hewenliang authored
There is a spelling misake in comments of cpuidle_idle_call. Fix it. Signed-off-by:
Hewenliang <hewenliang4@huawei.com> Signed-off-by:
Steven Rostedt (VMware) <rostedt@goodmis.org> Link: https://lkml.kernel.org/r/20200110025604.34373-1-hewenliang4@huawei.com
-
Vincent Guittot authored
With commit bef69dd8 ("sched/cpufreq: Move the cfs_rq_util_change() call to cpufreq_update_util()") update_load_avg() has become the central point for calling cpufreq (not including the update of blocked load). This change helps to simplify further the number of calls to cpufreq_update_util() and to remove last redundant ones. With update_load_avg(), we are now sure that cpufreq_update_util() will be called after every task attachment to a cfs_rq and especially after propagating this event down to the util_avg of the root cfs_rq, which is the level that is used by cpufreq governors like schedutil to set the frequency of a CPU. The SCHED_CPUFREQ_MIGRATION flag forces an early call to cpufreq when the migration happens in a cgroup whereas util_avg of root cfs_rq is not yet updated and this call is duplicated with the one that happens immediately after when the migration event reaches the root cfs_rq. The dedicated flag SCHED_CPUFREQ_MIGRATION is now useless and can be removed. The interface of attach_entity_load_avg() can also be simplified accordingly. Signed-off-by:
-
Wang Long authored
when CONFIG_PSI_DEFAULT_DISABLED set to N or the command line set psi=0, I think we should not create /proc/pressure and /proc/pressure/{io|memory|cpu}. In the future, user maybe determine whether the psi feature is enabled by checking the existence of the /proc/pressure dir or /proc/pressure/{io|memory|cpu} files. Signed-off-by:Wang Long <w@laoqinren.net> Signed-off-by:
Johannes Weiner <hannes@cmpxchg.org> Link: https://lkml.kernel.org/r/1576672698-32504-1-git-send-email-w@laoqinren.net
-
Peng Liu authored
commit bf475ce0 ("sched/fair: Add per-CPU min capacity to sched_group_capacity") introduced per-cpu min_capacity. commit e3d6d0cb ("sched/fair: Add sched_group per-CPU max capacity") introduced per-cpu max_capacity. In the SD_OVERLAP case, the local variable 'capacity' represents the sum of CPU capacity of all CPUs in the first sched group (sg) of the sched domain (sd). It is erroneously used to calculate sg's min and max CPU capacity. To fix this use capacity_of(cpu) instead of 'capacity'. The code which achieves this via cpu_rq(cpu)->sd->groups->sgc->capacity (for rq->sd != NULL) can be removed since it delivers the same value as capacity_of(cpu) which is currently only used for the (!rq->sd) case (see update_cpu_capacity()). An sg of the lowest sd (rq->sd or sd->child == NULL) represents a single CPU (and hence sg->sgc->capacity == capacity_of(cpu)). Signed-off-by:
Peng Liu <iwtbavbm@gmail.com> Signed-off-by:
-
Peng Wang authored
Move the code of calculation for delta_sum/delta_avg to where it is really needed to be done. Signed-off-by:
Peng Wang <rocking@linux.alibaba.com> Signed-off-by:
-
Alex Shi authored
Every time we call irqtime_account_process_tick() is in a interrupt, Every caller will get and assign a parameter rq = this_rq(), This is unnecessary and increase the code size a little bit. Move the rq getting action to irqtime_account_process_tick internally is better. base with this patch cputime.o 578792 bytes 577888 bytes Signed-off-by:Alex Shi <alex.shi@linux.alibaba.com> Signed-off-by:
-
Yangtao Li authored
The function stop_cpus() is only used internally by the stop_machine for stop multiple cpus. Make it static. Signed-off-by:
Yangtao Li <tiny.windzz@gmail.com> Signed-off-by:
-
Wei Li authored
Lengthy output of sysrq-t may take a lot of time on slow serial console with lots of processes and CPUs. So we need to reset NMI-watchdog to avoid spurious lockup messages, and we also reset softlockup watchdogs on all other CPUs since another CPU might be blocked waiting for us to process an IPI or stop_machine. Add to sysrq_sched_debug_show() as what we did in show_state_filter(). Signed-off-by:
Wei Li <liwei391@huawei.com> Signed-off-by:
-
Li Guanglei authored
rq::uclamp is an array of struct uclamp_rq, make sure we clear the whole thing. Fixes: 69842cba ("sched/uclamp: Add CPU's clamp buckets refcountinga") Signed-off-by:
Li Guanglei <guanglei.li@unisoc.com> Signed-off-by:
-
Qais Yousef authored
When a new cgroup is created, the effective uclamp value wasn't updated with a call to cpu_util_update_eff() that looks at the hierarchy and update to the most restrictive values. Fix it by ensuring to call cpu_util_update_eff() when a new cgroup becomes online. Without this change, the newly created cgroup uses the default root_task_group uclamp values, which is 1024 for both uclamp_{min, max}, which will cause the rq to to be clamped to max, hence cause the system to run at max frequency. The problem was observed on Ubuntu server and was reproduced on Debian and Buildroot rootfs. By default, Ubuntu and Debian create a cpu controller cgroup hierarchy and add all tasks to it - which creates enough noise to keep the rq uclamp value at max most of the time. Imitating this behavior makes the problem visible in Buildroot too which otherwise looks fine since it's a minimal userspace. Fixes: 0b60ba2d ("sched/uclamp: Propagate parent clamps") Reported-by: -
Viresh Kumar authored
The fair scheduler performs periodic load balance on every CPU to check if it can pull some tasks from other busy CPUs. The duration of this periodic load balance is set to sd->balance_interval for the idle CPUs and is calculated by multiplying the sd->balance_interval with the sd->busy_factor (set to 32 by default) for the busy CPUs. The multiplication is done for busy CPUs to avoid doing load balance too often and rather spend more time executing actual task. While that is the right thing to do for the CPUs busy with SCHED_OTHER or SCHED_BATCH tasks, it may not be the optimal thing for CPUs running only SCHED_IDLE tasks. With the recent enhancements in the fair scheduler around SCHED_IDLE CPUs, we now prefer to enqueue a newly-woken task to a SCHED_IDLE CPU instead of other busy or idle CPUs. The same reasoning should be applied to the load balancer as well to make it migrate tasks more aggressively to a SCHED_IDLE CPU, as that will reduce the scheduling latency of the migrated (SCHED_OTHER) tasks. This patch makes minimal changes to the fair scheduler to do the next load balance soon after the last non SCHED_IDLE task is dequeued from a runqueue, i.e. making the CPU SCHED_IDLE. Also the sd->busy_factor is ignored while calculating the balance_interval for such CPUs. This is done to avoid delaying the periodic load balance by few hundred milliseconds for SCHED_IDLE CPUs. This is tested on ARM64 Hikey620 platform (octa-core) with the help of rt-app and it is verified, using kernel traces, that the newly SCHED_IDLE CPU does load balancing shortly after it becomes SCHED_IDLE and pulls tasks from other busy CPUs. Signed-off-by:
-
Vincent Guittot authored
Similarly to calculate_imbalance() and find_busiest_group(), using the number of idle CPUs when there is only 1 CPU in the group is not efficient because we can't make a difference between a CPU running 1 task and a CPU running dozens of small tasks competing for the same CPU but not enough to overload it. More generally speaking, we should use the number of running tasks when there is the same number of idle CPUs in a group instead of blindly select the 1st one. When the groups have spare capacity and the same number of idle CPUs, we compare the number of running tasks to select the busiest group. Signed-off-by:
-
Jisheng Zhang authored
After commit 9cf57731 ("watchdog/softlockup: Replace "watchdog/%u" threads with cpu_stop_work"), the percpu soft_lockup_hrtimer_cnt is not used any more, so remove it and related code. Signed-off-by:
Jisheng Zhang <Jisheng.Zhang@synaptics.com> Signed-off-by:
-
- 25 Dec, 2019 9 commits
-
-
Qais Yousef authored
Capacity Awareness refers to the fact that on heterogeneous systems (like Arm big.LITTLE), the capacity of the CPUs is not uniform, hence when placing tasks we need to be aware of this difference of CPU capacities. In such scenarios we want to ensure that the selected CPU has enough capacity to meet the requirement of the running task. Enough capacity means here that capacity_orig_of(cpu) >= task.requirement. The definition of task.requirement is dependent on the scheduling class. For CFS, utilization is used to select a CPU that has >= capacity value than the cfs_task.util. capacity_orig_of(cpu) >= cfs_task.util DL isn't capacity aware at the moment but can make use of the bandwidth reservation to implement that in a similar manner CFS uses utilization. The following patchset implements that: https://lore.kernel.org/lkml/20190506044836.2914-1-luca.abeni@santannapisa.it/ capacity_orig_of(cpu)/SCHED_CAPACITY >= dl_deadline/dl_runtime For RT we don't have a per task utilization signal and we lack any information in general about what performance requirement the RT task needs. But with the introduction of uclamp, RT tasks can now control that by setting uclamp_min to guarantee a minimum performance point. ATM the uclamp value are only used for frequency selection; but on heterogeneous systems this is not enough and we need to ensure that the capacity of the CPU is >= uclamp_min. Which is what implemented here. capacity_orig_of(cpu) >= rt_task.uclamp_min Note that by default uclamp.min is 1024, which means that RT tasks will always be biased towards the big CPUs, which make for a better more predictable behavior for the default case. Must stress that the bias acts as a hint rather than a definite placement strategy. For example, if all big cores are busy executing other RT tasks we can't guarantee that a new RT task will be placed there. On non-heterogeneous systems the original behavior of RT should be retained. Similarly if uclamp is not selected in the config. [ mingo: Minor edits to comments. ] Signed-off-by:
Dietmar Eggemann <dietmar.eggemann@arm.com> Reviewed-by:
-
Valentin Schneider authored
task_fits_capacity() has just been made uclamp-aware, and find_energy_efficient_cpu() needs to go through the same treatment. Things are somewhat different here however - using the task max clamp isn't sufficient. Consider the following setup: The target runqueue, rq: rq.cpu_capacity_orig = 512 rq.cfs.avg.util_avg = 200 rq.uclamp.max = 768 // the max p.uclamp.max of all enqueued p's is 768 The waking task, p (not yet enqueued on rq): p.util_est = 600 p.uclamp.max = 100 Now, consider the following code which doesn't use the rq clamps: util = uclamp_task_util(p); // Does the task fit in the spare CPU capacity? cpu = cpu_of(rq); fits_capacity(util, cpu_capacity(cpu) - cpu_util(cpu)) This would lead to: util = 100; fits_capacity(100, 512 - 200) fits_capacity() would return true. However, enqueuing p on that CPU *will* cause it to become overutilized since rq clamp values are max-aggregated, so we'd remain with rq.uclamp.max = 768 which comes from the other tasks already enqueued on rq. Thus, we could select a high enough frequency to reach beyond 0.8 * 512 utilization (== overutilized) after enqueuing p on rq. What find_energy_efficient_cpu() needs here is uclamp_rq_util_with() which lets us peek at the future utilization landscape, including rq-wide uclamp values. Make find_energy_efficient_cpu() use uclamp_rq_util_with() for its fits_capacity() check. This is in line with what compute_energy() ends up using for estimating utilization. Tested-By:Quentin Perret <qperret@google.com> Signed-off-by:
-
Valentin Schneider authored
task_fits_capacity() drives CPU selection at wakeup time, and is also used to detect misfit tasks. Right now it does so by comparing task_util_est() with a CPU's capacity, but doesn't take into account uclamp restrictions. There's a few interesting uses that can come out of doing this. For instance, a low uclamp.max value could prevent certain tasks from being flagged as misfit tasks, so they could merrily remain on low-capacity CPUs. Similarly, a high uclamp.min value would steer tasks towards high capacity CPUs at wakeup (and, should that fail, later steered via misfit balancing), so such "boosted" tasks would favor CPUs of higher capacity. Introduce uclamp_task_util() and make task_fits_capacity() use it. Tested-By:
-
Valentin Schneider authored
The current helper returns (CPU) rq utilization with uclamp restrictions taken into account. A uclamp task utilization helper would be quite helpful, but this requires some renaming. Prepare the code for the introduction of a uclamp_task_util() by renaming the existing uclamp_util_with() to uclamp_rq_util_with(). Tested-By:
-
Valentin Schneider authored
Vincent pointed out recently that the canonical type for utilization values is 'unsigned long'. Internally uclamp uses 'unsigned int' values for cache optimization, but this doesn't have to be exported to its users. Make the uclamp helpers that deal with utilization use and return unsigned long values. Tested-By:
-
Valentin Schneider authored
The sole user of uclamp_util(), schedutil_cpu_util(), was made to use uclamp_util_with() instead in commit: af24bde8 ("sched/uclamp: Add uclamp support to energy_compute()") From then on, uclamp_util() has remained unused. Being a simple wrapper around uclamp_util_with(), we can get rid of it and win back a few lines. Tested-By:
-
Viresh Kumar authored
There are instances where we keep searching for an idle CPU despite already having a sched-idle CPU (in find_idlest_group_cpu(), select_idle_smt() and select_idle_cpu() and then there are places where we don't necessarily do that and return a sched-idle CPU as soon as we find one (in select_idle_sibling()). This looks a bit inconsistent and it may be worth having the same policy everywhere. On the other hand, choosing a sched-idle CPU over a idle one shall be beneficial from performance and power point of view as well, as we don't need to get the CPU online from a deep idle state which wastes quite a lot of time and energy and delays the scheduling of the newly woken up task. This patch tries to simplify code around sched-idle CPU selection and make it consistent throughout. Testing is done with the help of rt-app on hikey board (ARM64 octa-core, 2 clusters, 0-3 and 4-7). The cpufreq governor was set to performance to avoid any side affects from CPU frequency. Following are the tests performed: Test 1: 1-cfs-task: A single SCHED_NORMAL task is pinned to CPU5 which runs for 2333 us out of 7777 us (so gives time for the cluster to go in deep idle state). Test 2: 1-cfs-1-idle-task: A single SCHED_NORMAL task is pinned on CPU5 and single SCHED_IDLE task is pinned on CPU6 (to make sure cluster 1 doesn't go in deep idle state). Test 3: 1-cfs-8-idle-task: A single SCHED_NORMAL task is pinned on CPU5 and eight SCHED_IDLE tasks are created which run forever (not pinned anywhere, so they run on all CPUs). Checked with kernelshark that as soon as NORMAL task sleeps, the SCHED_IDLE task starts running on CPU5. And here are the results on mean latency (in us), using the "st" tool. $ st 1-cfs-task/rt-app-cfs_thread-0.log N min max sum mean stddev 642 90 592 197180 307.134 109.906 $ st 1-cfs-1-idle-task/rt-app-cfs_thread-0.log N min max sum mean stddev 642 67 311 113850 177.336 41.4251 $ st 1-cfs-8-idle-task/rt-app-cfs_thread-0.log N min max sum mean stddev 643 29 173 41364 64.3297 13.2344 The mean latency when we need to: - wakeup from deep idle state is 307 us. - wakeup from shallow idle state is 177 us. - preempt a SCHED_IDLE task is 64 us. Signed-off-by:
-
Qian Cai authored
This commit left behind an unused variable: 5443a0be ("sched: Use fair:prio_changed() instead of ad-hoc implementation") left behind an unused variable. kernel/sched/core.c: In function 'set_user_nice': kernel/sched/core.c:4507:16: warning: variable 'delta' set but not used int old_prio, delta; ^~~~~ Signed-off-by:
Qian Cai <cai@lca.pw> Signed-off-by:
-
Ingo Molnar authored
Signed-off-by:
-
- 23 Dec, 2019 2 commits
-
-
Linus Torvalds authored
-
git://git.kernel.org/pub/scm/linux/kernel/git/viro/vfsLinus Torvalds authored
Pull vfs fixes from Al Viro: "Eric's s_inodes softlockup fixes + Jan's fix for recent regression from pipe rework" * 'fixes' of git://git.kernel.org/pub/scm/linux/kernel/git/viro/vfs: fs: call fsnotify_sb_delete after evict_inodes fs: avoid softlockups in s_inodes iterators pipe: Fix bogus dereference in iov_iter_alignment()
-
- 22 Dec, 2019 7 commits
-
-
git://git.kernel.org/pub/scm/fs/xfs/xfs-linuxLinus Torvalds authored
Pull xfs fixes from Darrick Wong: "Fix a few bugs that could lead to corrupt files, fsck complaints, and filesystem crashes: - Minor documentation fixes - Fix a file corruption due to read racing with an insert range operation. - Fix log reservation overflows when allocating large rt extents - Fix a buffer log item flags check - Don't allow administrators to mount with sunit= options that will cause later xfs_repair complaints about the root directory being suspicious because the fs geometry appeared inconsistent - Fix a non-static helper that should have been static" * tag 'xfs-5.5-fixes-2' of git://git.kernel.org/pub/scm/fs/xfs/xfs-linux: xfs: Make the symbol 'xfs_rtalloc_log_count' static xfs: don't commit sunit/swidth updates to disk if that would cause repair failures xfs: split the sunit parameter update into two parts xfs: refactor agfl length computation function libxfs: resync with the userspace libxfs xfs: use bitops interface for buf log item AIL flag check xfs: fix log reservation overflows when allocating large rt extents xfs: stabilize insert range start boundary to avoid COW writeback race xfs: fix Sphinx documentation warning -
git://git.kernel.org/pub/scm/linux/kernel/git/tytso/ext4Linus Torvalds authored
Pull ext4 bug fixes from Ted Ts'o: "Ext4 bug fixes, including a regression fix" * tag 'ext4_for_linus_stable' of git://git.kernel.org/pub/scm/linux/kernel/git/tytso/ext4: ext4: clarify impact of 'commit' mount option ext4: fix unused-but-set-variable warning in ext4_add_entry() jbd2: fix kernel-doc notation warning ext4: use RCU API in debug_print_tree ext4: validate the debug_want_extra_isize mount option at parse time ext4: reserve revoke credits in __ext4_new_inode ext4: unlock on error in ext4_expand_extra_isize() ext4: optimize __ext4_check_dir_entry() ext4: check for directory entries too close to block end ext4: fix ext4_empty_dir() for directories with holes
-
git://git.kernel.dk/linux-blockLinus Torvalds authored
Pull block fixes from Jens Axboe: "Let's try this one again, this time without the compat_ioctl changes. We've got those fixed up, but that can go out next week. This contains: - block queue flush lockdep annotation (Bart) - Type fix for bsg_queue_rq() (Bart) - Three dasd fixes (Stefan, Jan) - nbd deadlock fix (Mike) - Error handling bio user map fix (Yang) - iocost fix (Tejun) - sbitmap waitqueue addition fix that affects the kyber IO scheduler (David)" * tag 'block-5.5-20191221' of git://git.kernel.dk/linux-block: sbitmap: only queue kyber's wait callback if not already active block: fix memleak when __blk_rq_map_user_iov() is failed s390/dasd: fix typo in copyright statement s390/dasd: fix memleak in path handling error case s390/dasd/cio: Interpret ccw_device_get_mdc return value correctly block: Fix a lockdep complaint triggered by request queue flushing block: Fix the type of 'sts' in bsg_queue_rq() block: end bio with BLK_STS_AGAIN in case of non-mq devs and REQ_NOWAIT nbd: fix shutdown and recv work deadlock v2 iocost: over-budget forced IOs should schedule async delay -
git://git.kernel.org/pub/scm/virt/kvm/kvmLinus Torvalds authored
Pull KVM fixes from Paolo Bonzini: "PPC: - Fix a bug where we try to do an ultracall on a system without an ultravisor KVM: - Fix uninitialised sysreg accessor - Fix handling of demand-paged device mappings - Stop spamming the console on IMPDEF sysregs - Relax mappings of writable memslots - Assorted cleanups MIPS: - Now orphan, James Hogan is stepping down x86: - MAINTAINERS change, so long Radim and thanks for all the fish - supported CPUID fixes for AMD machines without SPEC_CTRL" * tag 'for-linus' of git://git.kernel.org/pub/scm/virt/kvm/kvm: MAINTAINERS: remove Radim from KVM maintainers MAINTAINERS: Orphan KVM for MIPS kvm: x86: Host feature SSBD doesn't imply guest feature AMD_SSBD kvm: x86: Host feature SSBD doesn't imply guest feature SPEC_CTRL_SSBD KVM: PPC: Book3S HV: Don't do ultravisor calls on systems without ultravisor KVM: arm/arm64: Properly handle faulting of device mappings KVM: arm64: Ensure 'params' is initialised when looking up sys register KVM: arm/arm64: Remove excessive permission check in kvm_arch_prepare_memory_region KVM: arm64: Don't log IMP DEF sysreg traps KVM: arm64: Sanely ratelimit sysreg messages KVM: arm/arm64: vgic: Use wrapper function to lock/unlock all vcpus in kvm_vgic_create() KVM: arm/arm64: vgic: Fix potential double free dist->spis in __kvm_vgic_destroy() KVM: arm/arm64: Get rid of unused arg in cpu_init_hyp_mode() -
git://git.kernel.org/pub/scm/linux/kernel/git/riscv/linuxLinus Torvalds authored
Pull RISC-V fixes from Paul Walmsley: "Several fixes, and one cleanup, for RISC-V. Fixes: - Fix an error in a Kconfig file that resulted in an undefined Kconfig option "CONFIG_CONFIG_MMU" - Fix undefined Kconfig option "CONFIG_CONFIG_MMU" - Fix scratch register clearing in M-mode (affects nommu users) - Fix a mismerge on my part that broke the build for CONFIG_SPARSEMEM_VMEMMAP users Cleanup: - Move SiFive L2 cache-related code to drivers/soc, per request" * tag 'riscv/for-v5.5-rc3' of git://git.kernel.org/pub/scm/linux/kernel/git/riscv/linux: riscv: move sifive_l2_cache.c to drivers/soc riscv: define vmemmap before pfn_to_page calls riscv: fix scratch register clearing in M-mode. riscv: Fix use of undefined config option CONFIG_CONFIG_MMU -
git://git.kernel.org/pub/scm/linux/kernel/git/netdev/netLinus Torvalds authored
Pull networking fixes from David Miller: 1) Several nf_flow_table_offload fixes from Pablo Neira Ayuso, including adding a missing ipv6 match description. 2) Several heap overflow fixes in mwifiex from qize wang and Ganapathi Bhat. 3) Fix uninit value in bond_neigh_init(), from Eric Dumazet. 4) Fix non-ACPI probing of nxp-nci, from Stephan Gerhold. 5) Fix use after free in tipc_disc_rcv(), from Tuong Lien. 6) Enforce limit of 33 tail calls in mips and riscv JIT, from Paul Chaignon. 7) Multicast MAC limit test is off by one in qede, from Manish Chopra. 8) Fix established socket lookup race when socket goes from TCP_ESTABLISHED to TCP_LISTEN, because there lacks an intervening RCU grace period. From Eric Dumazet. 9) Don't send empty SKBs from tcp_write_xmit(), also from Eric Dumazet. 10) Fix active backup transition after link failure in bonding, from Mahesh Bandewar. 11) Avoid zero sized hash table in gtp driver, from Taehee Yoo. 12) Fix wrong interface passed to ->mac_link_up(), from Russell King. 13) Fix DSA egress flooding settings in b53, from Florian Fainelli. 14) Memory leak in gmac_setup_txqs(), from Navid Emamdoost. 15) Fix double free in dpaa2-ptp code, from Ioana Ciornei. 16) Reject invalid MTU values in stmmac, from Jose Abreu. 17) Fix refcount leak in error path of u32 classifier, from Davide Caratti. 18) Fix regression causing iwlwifi firmware crashes on boot, from Anders Kaseorg. 19) Fix inverted return value logic in llc2 code, from Chan Shu Tak. 20) Disable hardware GRO when XDP is attached to qede, frm Manish Chopra. 21) Since we encode state in the low pointer bits, dst metrics must be at least 4 byte aligned, which is not necessarily true on m68k. Add annotations to fix this, from Geert Uytterhoeven. * git://git.kernel.org/pub/scm/linux/kernel/git/netdev/net: (160 commits) sfc: Include XDP packet headroom in buffer step size. sfc: fix channel allocation with brute force net: dst: Force 4-byte alignment of dst_metrics selftests: pmtu: fix init mtu value in description hv_netvsc: Fix unwanted rx_table reset net: phy: ensure that phy IDs are correctly typed mod_devicetable: fix PHY module format qede: Disable hardware gro when xdp prog is installed net: ena: fix issues in setting interrupt moderation params in ethtool net: ena: fix default tx interrupt moderation interval net/smc: unregister ib devices in reboot_event net: stmmac: platform: Fix MDIO init for platforms without PHY llc2: Fix return statement of llc_stat_ev_rx_null_dsap_xid_c (and _test_c) net: hisilicon: Fix a BUG trigered by wrong bytes_compl net: dsa: ksz: use common define for tag len s390/qeth: don't return -ENOTSUPP to userspace s390/qeth: fix promiscuous mode after reset s390/qeth: handle error due to unsupported transport mode cxgb4: fix refcount init for TC-MQPRIO offload tc-testing: initial tdc selftests for cls_u32 ... -
Jan Stancek authored
LTP pipeio_1 test is hanging with v5.5-rc2-385-gb8e382a1, with read side observing empty pipe and sleeping and write side running out of space and then sleeping as well. In this scenario there are 5 writers and 1 reader. Problem is that after pipe_write() reacquires pipe lock, it re-checks for empty pipe with potentially stale 'head' and doesn't wake up read side anymore. pipe->tail can advance beyond 'head', because there are multiple writers. Use pipe->head for empty pipe check after reacquiring lock to observe current state. Testing: With patch, LTP pipeio_1 ran successfully in loop for 1 hour. Without patch it hanged within a minute. Fixes: 1b6b26ae ("pipe: fix and clarify pipe write wakeup logic") Reported-by:
Rachel Sibley <rasibley@redhat.com> Signed-off-by:
Jan Stancek <jstancek@redhat.com> Signed-off-by:
Linus Torvalds <torvalds@linux-foundation.org>
-
