On some platforms such as the Dell XPS 13 laptop the firmware disables turbo
when the machine is disconnected from AC, and viceversa it enables it again
when it's reconnected. In these cases a _PPC ACPI notification is issued.
The scheduler needs to know freq_max for frequency-invariant calculations.
To account for turbo availability to come and go, record freq_max at boot as
if turbo was available and store it in a helper variable. Use a setter
function to swap between freq_base and freq_max every time turbo goes off or on.
Signed-off-by: Giovanni Gherdovich <ggherdovich@suse.cz>
Signed-off-by: Peter Zijlstra (Intel) <peterz@infradead.org>
Signed-off-by: Ingo Molnar <mingo@kernel.org>
Acked-by: Rafael J. Wysocki <rafael.j.wysocki@intel.com>
Link: https://lkml.kernel.org/r/20200122151617.531-7-ggherdovich@suse.cz
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.c
Suggested-by: Peter Zijlstra <peterz@infradead.org>
Signed-off-by: Giovanni Gherdovich <ggherdovich@suse.cz>
Signed-off-by: Peter Zijlstra (Intel) <peterz@infradead.org>
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
Pull misc x86 updates from Ingo Molnar:
"Misc changes:
- Enhance #GP fault printouts by distinguishing between canonical and
non-canonical address faults, and also add KASAN fault decoding.
- Fix/enhance the x86 NMI handler by putting the duration check into
a direct function call instead of an irq_work which we know to be
broken in some cases.
- Clean up do_general_protection() a bit"
* 'x86-core-for-linus' of git://git.kernel.org/pub/scm/linux/kernel/git/tip/tip:
x86/nmi: Remove irq_work from the long duration NMI handler
x86/traps: Cleanup do_general_protection()
x86/kasan: Print original address on #GP
x86/dumpstack: Introduce die_addr() for die() with #GP fault address
x86/traps: Print address on #GP
x86/insn-eval: Add support for 64-bit kernel mode
Pull x86 cleanups from Ingo Molnar:
"Misc cleanups all around the map"
* 'x86-cleanups-for-linus' of git://git.kernel.org/pub/scm/linux/kernel/git/tip/tip:
x86/CPU/AMD: Remove amd_get_topology_early()
x86/tsc: Remove redundant assignment
x86/crash: Use resource_size()
x86/cpu: Add a missing prototype for arch_smt_update()
x86/nospec: Remove unused RSB_FILL_LOOPS
x86/vdso: Provide missing include file
x86/Kconfig: Correct spelling and punctuation
Documentation/x86/boot: Fix typo
x86/boot: Fix a comment's incorrect file reference
x86/process: Remove set but not used variables prev and next
x86/Kconfig: Fix Kconfig indentation
Pull x86 asm updates from Ingo Molnar:
"Misc updates:
- Remove last remaining calls to exception_enter/exception_exit() and
simplify the entry code some more.
- Remove force_iret()
- Add support for "Fast Short Rep Mov", which is available starting
with Ice Lake Intel CPUs - and make the x86 assembly version of
memmove() use REP MOV for all sizes when FSRM is available.
- Micro-optimize/simplify the 32-bit boot code a bit.
- Use a more future-proof SYSRET instruction mnemonic"
* 'x86-asm-for-linus' of git://git.kernel.org/pub/scm/linux/kernel/git/tip/tip:
x86/boot: Simplify calculation of output address
x86/entry/64: Add instruction suffix to SYSRET
x86: Remove force_iret()
x86/cpufeatures: Add support for fast short REP; MOVSB
x86/context-tracking: Remove exception_enter/exit() from KVM_PV_REASON_PAGE_NOT_PRESENT async page fault
x86/context-tracking: Remove exception_enter/exit() from do_page_fault()
Pull perf updates from Ingo Molnar:
"Kernel side changes:
- Ftrace is one of the last W^X violators (after this only KLP is
left). These patches move it over to the generic text_poke()
interface and thereby get rid of this oddity. This requires a
surprising amount of surgery, by Peter Zijlstra.
- x86/AMD PMUs: add support for 'Large Increment per Cycle Events' to
count certain types of events that have a special, quirky hw ABI
(by Kim Phillips)
- kprobes fixes by Masami Hiramatsu
Lots of tooling updates as well, the following subcommands were
updated: annotate/report/top, c2c, clang, record, report/top TUI,
sched timehist, tests; plus updates were done to the gtk ui, libperf,
headers and the parser"
* 'perf-core-for-linus' of git://git.kernel.org/pub/scm/linux/kernel/git/tip/tip: (57 commits)
perf/x86/amd: Add support for Large Increment per Cycle Events
perf/x86/amd: Constrain Large Increment per Cycle events
perf/x86/intel/rapl: Add Comet Lake support
tracing: Initialize ret in syscall_enter_define_fields()
perf header: Use last modification time for timestamp
perf c2c: Fix return type for histogram sorting comparision functions
perf beauty sockaddr: Fix augmented syscall format warning
perf/ui/gtk: Fix gtk2 build
perf ui gtk: Add missing zalloc object
perf tools: Use %define api.pure full instead of %pure-parser
libperf: Setup initial evlist::all_cpus value
perf report: Fix no libunwind compiled warning break s390 issue
perf tools: Support --prefix/--prefix-strip
perf report: Clarify in help that --children is default
tools build: Fix test-clang.cpp with Clang 8+
perf clang: Fix build with Clang 9
kprobes: Fix optimize_kprobe()/unoptimize_kprobe() cancellation logic
tools lib: Fix builds when glibc contains strlcpy()
perf report/top: Make 'e' visible in the help and make it toggle showing callchains
perf report/top: Do not offer annotation for symbols without samples
...
Pull EFI updates from Ingo Molnar:
"The main changes in this cycle were:
- Cleanup of the GOP [graphics output] handling code in the EFI stub
- Complete refactoring of the mixed mode handling in the x86 EFI stub
- Overhaul of the x86 EFI boot/runtime code
- Increase robustness for mixed mode code
- Add the ability to disable DMA at the root port level in the EFI
stub
- Get rid of RWX mappings in the EFI memory map and page tables,
where possible
- Move the support code for the old EFI memory mapping style into its
only user, the SGI UV1+ support code.
- plus misc fixes, updates, smaller cleanups.
... and due to interactions with the RWX changes, another round of PAT
cleanups make a guest appearance via the EFI tree - with no side
effects intended"
* 'efi-core-for-linus' of git://git.kernel.org/pub/scm/linux/kernel/git/tip/tip: (75 commits)
efi/x86: Disable instrumentation in the EFI runtime handling code
efi/libstub/x86: Fix EFI server boot failure
efi/x86: Disallow efi=old_map in mixed mode
x86/boot/compressed: Relax sed symbol type regex for LLVM ld.lld
efi/x86: avoid KASAN false positives when accessing the 1: 1 mapping
efi: Fix handling of multiple efi_fake_mem= entries
efi: Fix efi_memmap_alloc() leaks
efi: Add tracking for dynamically allocated memmaps
efi: Add a flags parameter to efi_memory_map
efi: Fix comment for efi_mem_type() wrt absent physical addresses
efi/arm: Defer probe of PCIe backed efifb on DT systems
efi/x86: Limit EFI old memory map to SGI UV machines
efi/x86: Avoid RWX mappings for all of DRAM
efi/x86: Don't map the entire kernel text RW for mixed mode
x86/mm: Fix NX bit clearing issue in kernel_map_pages_in_pgd
efi/libstub/x86: Fix unused-variable warning
efi/libstub/x86: Use mandatory 16-byte stack alignment in mixed mode
efi/libstub/x86: Use const attribute for efi_is_64bit()
efi: Allow disabling PCI busmastering on bridges during boot
efi/x86: Allow translating 64-bit arguments for mixed mode calls
...
Pull header cleanup from Ingo Molnar:
"This is a treewide cleanup, mostly (but not exclusively) with x86
impact, which breaks implicit dependencies on the asm/realtime.h
header and finally removes it from asm/acpi.h"
* 'core-headers-for-linus' of git://git.kernel.org/pub/scm/linux/kernel/git/tip/tip:
x86/ACPI/sleep: Move acpi_get_wakeup_address() into sleep.c, remove <asm/realmode.h> from <asm/acpi.h>
ACPI/sleep: Convert acpi_wakeup_address into a function
x86/ACPI/sleep: Remove an unnecessary include of asm/realmode.h
ASoC: Intel: Skylake: Explicitly include linux/io.h for virt_to_phys()
vmw_balloon: Explicitly include linux/io.h for virt_to_phys()
virt: vbox: Explicitly include linux/io.h to pick up various defs
efi/capsule-loader: Explicitly include linux/io.h for page_to_phys()
perf/x86/intel: Explicitly include asm/io.h to use virt_to_phys()
x86/kprobes: Explicitly include vmalloc.h for set_vm_flush_reset_perms()
x86/ftrace: Explicitly include vmalloc.h for set_vm_flush_reset_perms()
x86/boot: Explicitly include realmode.h to handle RM reservations
x86/efi: Explicitly include realmode.h to handle RM trampoline quirk
x86/platform/intel/quark: Explicitly include linux/io.h for virt_to_phys()
x86/setup: Enhance the comments
x86/setup: Clean up the header portion of setup.c
Pull timer updates from Thomas Gleixner:
"The timekeeping and timers departement provides:
- Time namespace support:
If a container migrates from one host to another then it expects
that clocks based on MONOTONIC and BOOTTIME are not subject to
disruption. Due to different boot time and non-suspended runtime
these clocks can differ significantly on two hosts, in the worst
case time goes backwards which is a violation of the POSIX
requirements.
The time namespace addresses this problem. It allows to set offsets
for clock MONOTONIC and BOOTTIME once after creation and before
tasks are associated with the namespace. These offsets are taken
into account by timers and timekeeping including the VDSO.
Offsets for wall clock based clocks (REALTIME/TAI) are not provided
by this mechanism. While in theory possible, the overhead and code
complexity would be immense and not justified by the esoteric
potential use cases which were discussed at Plumbers '18.
The overhead for tasks in the root namespace (ie where host time
offsets = 0) is in the noise and great effort was made to ensure
that especially in the VDSO. If time namespace is disabled in the
kernel configuration the code is compiled out.
Kudos to Andrei Vagin and Dmitry Sofanov who implemented this
feature and kept on for more than a year addressing review
comments, finding better solutions. A pleasant experience.
- Overhaul of the alarmtimer device dependency handling to ensure
that the init/suspend/resume ordering is correct.
- A new clocksource/event driver for Microchip PIT64
- Suspend/resume support for the Hyper-V clocksource
- The usual pile of fixes, updates and improvements mostly in the
driver code"
* tag 'timers-core-2020-01-27' of git://git.kernel.org/pub/scm/linux/kernel/git/tip/tip: (71 commits)
alarmtimer: Make alarmtimer_get_rtcdev() a stub when CONFIG_RTC_CLASS=n
alarmtimer: Use wakeup source from alarmtimer platform device
alarmtimer: Make alarmtimer platform device child of RTC device
alarmtimer: Update alarmtimer_get_rtcdev() docs to reflect reality
hrtimer: Add missing sparse annotation for __run_timer()
lib/vdso: Only read hrtimer_res when needed in __cvdso_clock_getres()
MIPS: vdso: Define BUILD_VDSO32 when building a 32bit kernel
clocksource/drivers/hyper-v: Set TSC clocksource as default w/ InvariantTSC
clocksource/drivers/hyper-v: Untangle stimers and timesync from clocksources
clocksource/drivers/timer-microchip-pit64b: Fix sparse warning
clocksource/drivers/exynos_mct: Rename Exynos to lowercase
clocksource/drivers/timer-ti-dm: Fix uninitialized pointer access
clocksource/drivers/timer-ti-dm: Switch to platform_get_irq
clocksource/drivers/timer-ti-dm: Convert to devm_platform_ioremap_resource
clocksource/drivers/em_sti: Fix variable declaration in em_sti_probe
clocksource/drivers/em_sti: Convert to devm_platform_ioremap_resource
clocksource/drivers/bcm2835_timer: Fix memory leak of timer
clocksource/drivers/cadence-ttc: Use ttc driver as platform driver
clocksource/drivers/timer-microchip-pit64b: Add Microchip PIT64B support
clocksource/drivers/hyper-v: Reserve PAGE_SIZE space for tsc page
...
Pull power management updates from Rafael Wysocki:
"These add ACPI support to the intel_idle driver along with an admin
guide document for it, add support for CPR (Core Power Reduction) to
the AVS (Adaptive Voltage Scaling) subsystem, add new hardware support
in a few places, add some new sysfs attributes, debugfs files and
tracepoints, fix bugs and clean up a bunch of things all over.
Specifics:
- Update the ACPI processor driver in order to export
acpi_processor_evaluate_cst() to the code outside of it, add ACPI
support to the intel_idle driver based on that and clean up that
driver somewhat (Rafael Wysocki).
- Add an admin guide document for the intel_idle driver (Rafael
Wysocki).
- Clean up cpuidle core and drivers, enable compilation testing for
some of them (Benjamin Gaignard, Krzysztof Kozlowski, Rafael
Wysocki, Yangtao Li).
- Fix reference counting of OPP (operating performance points) table
structures (Viresh Kumar).
- Add support for CPR (Core Power Reduction) to the AVS (Adaptive
Voltage Scaling) subsystem (Niklas Cassel, Colin Ian King,
YueHaibing).
- Add support for TigerLake Mobile and JasperLake to the Intel RAPL
power capping driver (Zhang Rui).
- Update cpufreq drivers:
- Add i.MX8MP support to imx-cpufreq-dt (Anson Huang).
- Fix usage of a macro in loongson2_cpufreq (Alexandre Oliva).
- Fix cpufreq policy reference counting issues in s3c and
brcmstb-avs (chenqiwu).
- Fix ACPI table reference counting issue and HiSilicon quirk
handling in the CPPC driver (Hanjun Guo).
- Clean up spelling mistake in intel_pstate (Harry Pan).
- Convert the kirkwood and tegra186 drivers to using
devm_platform_ioremap_resource() (Yangtao Li).
- Update devfreq core:
- Add 'name' sysfs attribute for devfreq devices (Chanwoo Choi).
- Clean up the handing of transition statistics and allow them to
be reset by writing 0 to the 'trans_stat' devfreq device
attribute in sysfs (Kamil Konieczny).
- Add 'devfreq_summary' to debugfs (Chanwoo Choi).
- Clean up kerneldoc comments and Kconfig indentation (Krzysztof
Kozlowski, Randy Dunlap).
- Update devfreq drivers:
- Add dynamic scaling for the imx8m DDR controller and clean up
imx8m-ddrc (Leonard Crestez, YueHaibing).
- Fix DT node reference counting and nitialization error code path
in rk3399_dmc and add COMPILE_TEST and HAVE_ARM_SMCCC dependency
for it (Chanwoo Choi, Yangtao Li).
- Fix DT node reference counting in rockchip-dfi and make it use
devm_platform_ioremap_resource() (Yangtao Li).
- Fix excessive stack usage in exynos-ppmu (Arnd Bergmann).
- Fix initialization error code paths in exynos-bus (Yangtao Li).
- Clean up exynos-bus and exynos somewhat (Artur Świgoń, Krzysztof
Kozlowski).
- Add tracepoints for tracking usage_count updates unrelated to
status changes in PM-runtime (Michał Mirosław).
- Add sysfs attribute to control the "sync on suspend" behavior
during system-wide suspend (Jonas Meurer).
- Switch system-wide suspend tests over to 64-bit time (Alexandre
Belloni).
- Make wakeup sources statistics in debugfs cover deleted ones which
used to be the case some time ago (zhuguangqing).
- Clean up computations carried out during hibernation, update
messages related to hibernation and fix a spelling mistake in one
of them (Wen Yang, Luigi Semenzato, Colin Ian King).
- Add mailmap entry for maintainer e-mail address that has not been
functional for several years (Rafael Wysocki)"
* tag 'pm-5.6-rc1' of git://git.kernel.org/pub/scm/linux/kernel/git/rafael/linux-pm: (83 commits)
cpufreq: loongson2_cpufreq: adjust cpufreq uses of LOONGSON_CHIPCFG
intel_idle: Clean up irtl_2_usec()
intel_idle: Move 3 functions closer to their callers
intel_idle: Annotate initialization code and data structures
intel_idle: Move and clean up intel_idle_cpuidle_devices_uninit()
intel_idle: Rearrange intel_idle_cpuidle_driver_init()
intel_idle: Clean up NULL pointer check in intel_idle_init()
intel_idle: Fold intel_idle_probe() into intel_idle_init()
intel_idle: Eliminate __setup_broadcast_timer()
cpuidle: fix cpuidle_find_deepest_state() kerneldoc warnings
cpuidle: sysfs: fix warnings when compiling with W=1
cpuidle: coupled: fix warnings when compiling with W=1
cpufreq: brcmstb-avs: fix imbalance of cpufreq policy refcount
PM: suspend: Add sysfs attribute to control the "sync on suspend" behavior
PM / devfreq: Add debugfs support with devfreq_summary file
Documentation: admin-guide: PM: Add intel_idle document
cpuidle: arm: Enable compile testing for some of drivers
PM-runtime: add tracepoints for usage_count changes
cpufreq: intel_pstate: fix spelling mistake: "Whethet" -> "Whether"
PM: hibernate: fix spelling mistake "shapshot" -> "snapshot"
...
Add a helper, lookup_address_in_mm(), to traverse the page tables of a
given mm struct. KVM will use the helper to retrieve the host mapping
level, e.g. 4k vs. 2mb vs. 1gb, of a compound (or DAX-backed) page
without having to resort to implementation specific metadata. E.g. KVM
currently uses different logic for HugeTLB vs. THP, and would add a
third variant for DAX-backed files.
Cc: Dan Williams <dan.j.williams@intel.com>
Signed-off-by: Sean Christopherson <sean.j.christopherson@intel.com>
Reviewed-by: Thomas Gleixner <tglx@linutronix.de>
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
The helper x86_set_memory_region() is only used in vmx_set_tss_addr()
and kvm_arch_destroy_vm(). Push the lock upper in both cases. With
that, drop x86_set_memory_region().
This prepares to allow __x86_set_memory_region() to return a HVA
mapped, because the HVA will need to be protected by the lock too even
after __x86_set_memory_region() returns.
Signed-off-by: Peter Xu <peterx@redhat.com>
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
Current SVM implementation does not have support for handling PKU. Guests
running on a host with future AMD cpus that support the feature will read
garbage from the PKRU register and will hit segmentation faults on boot as
memory is getting marked as protected that should not be. Ensure that cpuid
from SVM does not advertise the feature.
Signed-off-by: John Allen <john.allen@amd.com>
Cc: stable@vger.kernel.org
Fixes: 0556cbdc2f ("x86/pkeys: Don't check if PKRU is zero before writing it")
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
Pull dmaengine updates from Vinod Koul:
"This time we have a bunch of core changes to support dynamic channels,
hotplug of controllers, new apis for metadata ops etc along with new
drivers for Intel data accelerators, TI K3 UDMA, PLX DMA engine and
hisilicon Kunpeng DMA engine. Also usual assorted updates to drivers.
Core:
- Support for dynamic channels
- Removal of various slave wrappers
- Make few slave request APIs as private to dmaengine
- Symlinks between channels and slaves
- Support for hotplug of controllers
- Support for metadata_ops for dma_async_tx_descriptor
- Reporting DMA cached data amount
- Virtual dma channel locking updates
New drivers/device/feature support support:
- Driver for Intel data accelerators
- Driver for TI K3 UDMA
- Driver for PLX DMA engine
- Driver for hisilicon Kunpeng DMA engine
- Support for eDMA support for QorIQ LS1028A in fsl edma driver
- Support for cyclic dma in sun4i driver
- Support for X1830 in JZ4780 driver"
* tag 'dmaengine-5.6-rc1' of git://git.infradead.org/users/vkoul/slave-dma: (62 commits)
dmaengine: Create symlinks between DMA channels and slaves
dmaengine: hisilicon: Add Kunpeng DMA engine support
dmaengine: idxd: add char driver to expose submission portal to userland
dmaengine: idxd: connect idxd to dmaengine subsystem
dmaengine: idxd: add descriptor manipulation routines
dmaengine: idxd: add sysfs ABI for idxd driver
dmaengine: idxd: add configuration component of driver
dmaengine: idxd: Init and probe for Intel data accelerators
dmaengine: add support to dynamic register/unregister of channels
dmaengine: break out channel registration
x86/asm: add iosubmit_cmds512() based on MOVDIR64B CPU instruction
dmaengine: ti: k3-udma: fix spelling mistake "limted" -> "limited"
dmaengine: s3c24xx-dma: fix spelling mistake "to" -> "too"
dmaengine: Move dma_get_{,any_}slave_channel() to private dmaengine.h
dmaengine: Remove dma_request_slave_channel_compat() wrapper
dmaengine: Remove dma_device_satisfies_mask() wrapper
dt-bindings: fsl-imx-sdma: Add i.MX8MM/i.MX8MN/i.MX8MP compatible string
dmaengine: zynqmp_dma: fix burst length configuration
dmaengine: sun4i: Add support for cyclic requests with dedicated DMA
dmaengine: fsl-qdma: fix duplicated argument to &&
...
Pull x86 platform driver updates from Andy Shevchenko:
- Enable thermal policy for ASUS TUF FX705DY/FX505DY
- Support left round button on ASUS N56VB
- Support new Mellanox platforms of basic class VMOD0009 and VMOD0010
- Intel Comet Lake, Tiger Lake and Elkhart Lake support in the PMC
driver
- Big clean-up to Intel PMC core, PMC IPC and SCU IPC drivers
- Touchscreen support for the PiPO W11 tablet
* tag 'platform-drivers-x86-v5.6-1' of git://git.infradead.org/linux-platform-drivers-x86: (64 commits)
platform/x86: intel_pmc_ipc: Switch to use driver->dev_groups
platform/x86: intel_pmc_ipc: Propagate error from kstrtoul()
platform/x86: intel_pmc_ipc: Use octal permissions in sysfs attributes
platform/x86: intel_pmc_ipc: Get rid of unnecessary includes
platform/x86: intel_pmc_ipc: Drop ipc_data_readb()
platform/x86: intel_pmc_ipc: Drop intel_pmc_gcr_read() and intel_pmc_gcr_write()
platform/x86: intel_pmc_ipc: Make intel_pmc_ipc_raw_cmd() static
platform/x86: intel_pmc_ipc: Make intel_pmc_ipc_simple_command() static
platform/x86: intel_pmc_ipc: Make intel_pmc_gcr_update() static
platform/x86: intel_scu_ipc: Reformat kernel-doc comments of exported functions
platform/x86: intel_scu_ipc: Drop intel_scu_ipc_raw_command()
platform/x86: intel_scu_ipc: Drop intel_scu_ipc_io[read|write][8|16]()
platform/x86: intel_scu_ipc: Drop unused macros
platform/x86: intel_scu_ipc: Drop unused prototype intel_scu_ipc_fw_update()
platform/x86: intel_scu_ipc: Sleeping is fine when polling
platform/x86: intel_scu_ipc: Drop intel_scu_ipc_i2c_cntrl()
platform/x86: intel_scu_ipc: Remove Lincroft support
platform/x86: intel_scu_ipc: Add constants for register offsets
platform/x86: intel_scu_ipc: Fix interrupt support
platform/x86: intel_scu_ipcutil: Remove default y from Kconfig
...
Pull x86 microcode update from Borislav Petkov:
"Another boring branch this time around: mark a stub function inline,
by Valdis Kletnieks"
* 'x86-microcode-for-linus' of git://git.kernel.org/pub/scm/linux/kernel/git/tip/tip:
x86/microcode/AMD: Make stub function static inline
Move the kvm_cpu_{un}init() calls to common x86 code as an intermediate
step to removing kvm_cpu_{un}init() altogether.
Note, VMX'x alloc_apic_access_page() and init_rmode_identity_map() are
per-VM allocations and are intentionally kept if vCPU creation fails.
They are freed by kvm_arch_destroy_vm().
No functional change intended.
Signed-off-by: Sean Christopherson <sean.j.christopherson@intel.com>
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
Move allocation of VMX and SVM vcpus to common x86. Although the struct
being allocated is technically a VMX/SVM struct, it can be interpreted
directly as a 'struct kvm_vcpu' because of the pre-existing requirement
that 'struct kvm_vcpu' be located at offset zero of the arch/vendor vcpu
struct.
Remove the message from the build-time assertions regarding placement of
the struct, as compatibility with the arch usercopy region is no longer
the sole dependent on 'struct kvm_vcpu' being at offset zero.
Signed-off-by: Sean Christopherson <sean.j.christopherson@intel.com>
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
With the introduction of MOVDIR64B instruction, there is now an instruction
that can write 64 bytes of data atomically.
Quoting from Intel SDM:
"There is no atomicity guarantee provided for the 64-byte load operation
from source address, and processor implementations may use multiple
load operations to read the 64-bytes. The 64-byte direct-store issued
by MOVDIR64B guarantees 64-byte write-completion atomicity. This means
that the data arrives at the destination in a single undivided 64-byte
write transaction."
We have identified at least 3 different use cases for this instruction in
the format of func(dst, src, count):
1) Clear poison / Initialize MKTME memory
@dst is normal memory.
@src in normal memory. Does not increment. (Copy same line to all
targets)
@count (to clear/init multiple lines)
2) Submit command(s) to new devices
@dst is a special MMIO region for a device. Does not increment.
@src is normal memory. Increments.
@count usually is 1, but can be multiple.
3) Copy to iomem in big chunks
@dst is iomem and increments
@src in normal memory and increments
@count is number of chunks to copy
Add support for case #2 to support device that will accept commands via
this instruction. We provide a @count in order to submit a batch of
preprogrammed descriptors in virtually contiguous memory. This
allows the caller to submit multiple descriptors to a device with a single
submission. The special device requires the entire 64bytes descriptor to
be written atomically and will accept MOVDIR64B instruction.
Signed-off-by: Dave Jiang <dave.jiang@intel.com>
Acked-by: Borislav Petkov <bp@suse.de>
Link: https://lore.kernel.org/r/157965022175.73301.10174614665472962675.stgit@djiang5-desk3.ch.intel.com
Signed-off-by: Vinod Koul <vkoul@kernel.org>
From: Dave Hansen <dave.hansen@linux.intel.com>
MPX is being removed from the kernel due to a lack of support
in the toolchain going forward (gcc).
This removes all the remaining (dead at this point) MPX handling
code remaining in the tree. The only remaining code is the XSAVE
support for MPX state which is currently needd for KVM to handle
VMs which might use MPX.
Cc: Peter Zijlstra (Intel) <peterz@infradead.org>
Cc: Andy Lutomirski <luto@kernel.org>
Cc: x86@kernel.org
Cc: Linus Torvalds <torvalds@linux-foundation.org>
Cc: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Dave Hansen <dave.hansen@linux.intel.com>
There are no existing users for this functionality so drop it from the
driver completely. This also means we don't need to keep the struct
intel_scu_ipc_pdata_t around anymore so remove that as well.
Signed-off-by: Mika Westerberg <mika.westerberg@linux.intel.com>
Reviewed-by: Andy Shevchenko <andriy.shevchenko@linux.intel.com>
Signed-off-by: Andy Shevchenko <andriy.shevchenko@linux.intel.com>
Add feature-specific helpers for querying guest CPUID support from the
emulator instead of having the emulator do a full CPUID and perform its
own bit tests. The primary motivation is to eliminate the emulator's
usage of bit() so that future patches can add more extensive build-time
assertions on the usage of bit() without having to expose yet more code
to the emulator.
Note, providing a generic guest_cpuid_has() to the emulator doesn't work
due to the existing built-time assertions in guest_cpuid_has(), which
require the feature being checked to be a compile-time constant.
No functional change intended.
Signed-off-by: Sean Christopherson <sean.j.christopherson@intel.com>
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
ICR and TSCDEADLINE MSRs write cause the main MSRs write vmexits in our
product observation, multicast IPIs are not as common as unicast IPI like
RESCHEDULE_VECTOR and CALL_FUNCTION_SINGLE_VECTOR etc.
This patch introduce a mechanism to handle certain performance-critical
WRMSRs in a very early stage of KVM VMExit handler.
This mechanism is specifically used for accelerating writes to x2APIC ICR
that attempt to send a virtual IPI with physical destination-mode, fixed
delivery-mode and single target. Which was found as one of the main causes
of VMExits for Linux workloads.
The reason this mechanism significantly reduce the latency of such virtual
IPIs is by sending the physical IPI to the target vCPU in a very early stage
of KVM VMExit handler, before host interrupts are enabled and before expensive
operations such as reacquiring KVM’s SRCU lock.
Latency is reduced even more when KVM is able to use APICv posted-interrupt
mechanism (which allows to deliver the virtual IPI directly to target vCPU
without the need to kick it to host).
Testing on Xeon Skylake server:
The virtual IPI latency from sender send to receiver receive reduces
more than 200+ cpu cycles.
Reviewed-by: Liran Alon <liran.alon@oracle.com>
Cc: Paolo Bonzini <pbonzini@redhat.com>
Cc: Radim Krčmář <rkrcmar@redhat.com>
Cc: Sean Christopherson <sean.j.christopherson@intel.com>
Cc: Vitaly Kuznetsov <vkuznets@redhat.com>
Cc: Liran Alon <liran.alon@oracle.com>
Signed-off-by: Wanpeng Li <wanpengli@tencent.com>
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
We carry a quirk in the x86 EFI code to switch back to an older
method of mapping the EFI runtime services memory regions, because
it was deemed risky at the time to implement a new method without
providing a fallback to the old method in case problems arose.
Such problems did arise, but they appear to be limited to SGI UV1
machines, and so these are the only ones for which the fallback gets
enabled automatically (via a DMI quirk). The fallback can be enabled
manually as well, by passing efi=old_map, but there is very little
evidence that suggests that this is something that is being relied
upon in the field.
Given that UV1 support is not enabled by default by the distros
(Ubuntu, Fedora), there is no point in carrying this fallback code
all the time if there are no other users. So let's move it into the
UV support code, and document that efi=old_map now requires this
support code to be enabled.
Note that efi=old_map has been used in the past on other SGI UV
machines to work around kernel regressions in production, so we
keep the option to enable it by hand, but only if the kernel was
built with UV support.
Signed-off-by: Ard Biesheuvel <ardb@kernel.org>
Signed-off-by: Ingo Molnar <mingo@kernel.org>
Link: https://lore.kernel.org/r/20200113172245.27925-8-ardb@kernel.org
Reshuffle the x86 stub code a bit so that we can tag the efi_is_64bit()
function with the 'const' attribute, which permits the compiler to
optimize away any redundant calls. Since we have two different entry
points for 32 and 64 bit firmware in the startup code, this also
simplifies the C code since we'll enter it with the efi_is64 variable
already set.
Signed-off-by: Ard Biesheuvel <ardb@kernel.org>
Signed-off-by: Ingo Molnar <mingo@kernel.org>
Link: https://lore.kernel.org/r/20200113172245.27925-2-ardb@kernel.org
Add support for a new version of the Load Store unit bank type as
indicated by its McaType value, which will be present in future SMCA
systems.
Add the new (HWID, MCATYPE) tuple. Reuse the same name, since this is
logically the same to the user.
Also, add the new error descriptions to edac_mce_amd.
Signed-off-by: Yazen Ghannam <yazen.ghannam@amd.com>
Signed-off-by: Borislav Petkov <bp@suse.de>
Link: https://lkml.kernel.org/r/20200110015651.14887-2-Yazen.Ghannam@amd.com
To support time namespaces in the VDSO with a minimal impact on regular non
time namespace affected tasks, the namespace handling needs to be hidden in
a slow path.
The most obvious place is vdso_seq_begin(). If a task belongs to a time
namespace then the VVAR page which contains the system wide VDSO data is
replaced with a namespace specific page which has the same layout as the
VVAR page. That page has vdso_data->seq set to 1 to enforce the slow path
and vdso_data->clock_mode set to VCLOCK_TIMENS to enforce the time
namespace handling path.
The extra check in the case that vdso_data->seq is odd, e.g. a concurrent
update of the VDSO data is in progress, is not really affecting regular
tasks which are not part of a time namespace as the task is spin waiting
for the update to finish and vdso_data->seq to become even again.
If a time namespace task hits that code path, it invokes the corresponding
time getter function which retrieves the real VVAR page, reads host time
and then adds the offset for the requested clock which is stored in the
special VVAR page.
Allocate the time namespace page among VVAR pages and place vdso_data on
it. Provide __arch_get_timens_vdso_data() helper for VDSO code to get the
code-relative position of VVARs on that special page.
Co-developed-by: Andrei Vagin <avagin@openvz.org>
Signed-off-by: Andrei Vagin <avagin@openvz.org>
Signed-off-by: Dmitry Safonov <dima@arista.com>
Signed-off-by: Thomas Gleixner <tglx@linutronix.de>
Link: https://lore.kernel.org/r/20191112012724.250792-23-dima@arista.com
VDSO support for time namespaces needs to set up a page with the same
layout as VVAR. That timens page will be placed on position of VVAR page
inside namespace. That page has vdso_data->seq set to 1 to enforce
the slow path and vdso_data->clock_mode set to VCLOCK_TIMENS to enforce
the time namespace handling path.
To prepare the time namespace page the kernel needs to know the vdso_data
offset. Provide arch_get_vdso_data() helper for locating vdso_data on VVAR
page.
Co-developed-by: Andrei Vagin <avagin@openvz.org>
Signed-off-by: Andrei Vagin <avagin@openvz.org>
Signed-off-by: Dmitry Safonov <dima@arista.com>
Signed-off-by: Thomas Gleixner <tglx@linutronix.de>
Link: https://lore.kernel.org/r/20191112012724.250792-22-dima@arista.com
Provide stubs for perf_guest_get_msrs() and intel_pt_handle_vmx() when
building without support for Intel CPUs, i.e. CPU_SUP_INTEL=n. Lack of
stubs is not currently a problem as the only user, KVM_INTEL, takes a
dependency on CPU_SUP_INTEL=y. Provide the stubs for all CPUs so that
KVM_INTEL can be built for any CPU with compatible hardware support,
e.g. Centuar and Zhaoxin CPUs.
Note, the existing stub for perf_guest_get_msrs() is essentially dead
code as KVM selects CONFIG_PERF_EVENTS, i.e. the only user guarantees
the full implementation is built.
Signed-off-by: Sean Christopherson <sean.j.christopherson@intel.com>
Signed-off-by: Borislav Petkov <bp@suse.de>
Link: https://lkml.kernel.org/r/20191221044513.21680-19-sean.j.christopherson@intel.com
Add a new feature flag, X86_FEATURE_MSR_IA32_FEAT_CTL, to track whether
IA32_FEAT_CTL has been initialized. This will allow KVM, and any future
subsystems that depend on IA32_FEAT_CTL, to rely purely on cpufeatures
to query platform support, e.g. allows a future patch to remove KVM's
manual IA32_FEAT_CTL MSR checks.
Various features (on platforms that support IA32_FEAT_CTL) are dependent
on IA32_FEAT_CTL being configured and locked, e.g. VMX and LMCE. The
MSR is always configured during boot, but only if the CPU vendor is
recognized by the kernel. Because CPUID doesn't incorporate the current
IA32_FEAT_CTL value in its reporting of relevant features, it's possible
for a feature to be reported as supported in cpufeatures but not truly
enabled, e.g. if the CPU supports VMX but the kernel doesn't recognize
the CPU.
As a result, without the flag, KVM would see VMX as supported even if
IA32_FEAT_CTL hasn't been initialized, and so would need to manually
read the MSR and check the various enabling bits to avoid taking an
unexpected #GP on VMXON.
Signed-off-by: Sean Christopherson <sean.j.christopherson@intel.com>
Signed-off-by: Borislav Petkov <bp@suse.de>
Link: https://lkml.kernel.org/r/20191221044513.21680-14-sean.j.christopherson@intel.com
Add an entry in struct cpuinfo_x86 to track VMX capabilities and fill
the capabilities during IA32_FEAT_CTL MSR initialization.
Make the VMX capabilities dependent on IA32_FEAT_CTL and
X86_FEATURE_NAMES so as to avoid unnecessary overhead on CPUs that can't
possibly support VMX, or when /proc/cpuinfo is not available.
Signed-off-by: Sean Christopherson <sean.j.christopherson@intel.com>
Signed-off-by: Borislav Petkov <bp@suse.de>
Link: https://lkml.kernel.org/r/20191221044513.21680-11-sean.j.christopherson@intel.com
Add a VMX-specific variant of X86_FEATURE_* flags, which will eventually
supplant the synthetic VMX flags defined in cpufeatures word 8. Use the
Intel-defined layouts for the major VMX execution controls so that their
word entries can be directly populated from their respective MSRs, and
so that the VMX_FEATURE_* flags can be used to define the existing bit
definitions in asm/vmx.h, i.e. force developers to define a VMX_FEATURE
flag when adding support for a new hardware feature.
The majority of Intel's (and compatible CPU's) VMX capabilities are
enumerated via MSRs and not CPUID, i.e. querying /proc/cpuinfo doesn't
naturally provide any insight into the virtualization capabilities of
VMX enabled CPUs. Commit
e38e05a858 ("x86: extended "flags" to show virtualization HW feature
in /proc/cpuinfo")
attempted to address the issue by synthesizing select VMX features into
a Linux-defined word in cpufeatures.
Lack of reporting of VMX capabilities via /proc/cpuinfo is problematic
because there is no sane way for a user to query the capabilities of
their platform, e.g. when trying to find a platform to test a feature or
debug an issue that has a hardware dependency. Lack of reporting is
especially problematic when the user isn't familiar with VMX, e.g. the
format of the MSRs is non-standard, existence of some MSRs is reported
by bits in other MSRs, several "features" from KVM's point of view are
enumerated as 3+ distinct features by hardware, etc...
The synthetic cpufeatures approach has several flaws:
- The set of synthesized VMX flags has become extremely stale with
respect to the full set of VMX features, e.g. only one new flag
(EPT A/D) has been added in the the decade since the introduction of
the synthetic VMX features. Failure to keep the VMX flags up to
date is likely due to the lack of a mechanism that forces developers
to consider whether or not a new feature is worth reporting.
- The synthetic flags may incorrectly be misinterpreted as affecting
kernel behavior, i.e. KVM, the kernel's sole consumer of VMX,
completely ignores the synthetic flags.
- New CPU vendors that support VMX have duplicated the hideous code
that propagates VMX features from MSRs to cpufeatures. Bringing the
synthetic VMX flags up to date would exacerbate the copy+paste
trainwreck.
Define separate VMX_FEATURE flags to set the stage for enumerating VMX
capabilities outside of the cpu_has() framework, and for adding
functional usage of VMX_FEATURE_* to help ensure the features reported
via /proc/cpuinfo is up to date with respect to kernel recognition of
VMX capabilities.
Signed-off-by: Sean Christopherson <sean.j.christopherson@intel.com>
Signed-off-by: Borislav Petkov <bp@suse.de>
Link: https://lkml.kernel.org/r/20191221044513.21680-10-sean.j.christopherson@intel.com
