linux.git/arch/powerpc/include/asm/cache.h, branch v4.11

powerpc: Add new cache geometry aux vectors

2017-02-06T08:46:04+00:00

This adds AUX vectors for the L1I,D, L2 and L3 cache levels
providing for each cache level the size of the cache in bytes
and the geometry (line size and number of ways).

We chose to not use the existing alpha/sh definition which
packs all the information in a single entry per cache level as
it is too restricted to represent some of the geometries used
on POWER.

Signed-off-by: Benjamin Herrenschmidt 
Signed-off-by: Michael Ellerman

powerpc/64: Add L2 and L3 cache shape info

2017-02-06T08:46:04+00:00

Retrieved from device-tree when available

Signed-off-by: Benjamin Herrenschmidt 
Signed-off-by: Michael Ellerman

powerpc/64: Clean up ppc64_caches using a struct per cache

2017-02-06T08:46:04+00:00

We have two set of identical struct members for the I and D sides
and mostly identical bunches of code to parse the device-tree to
populate them. Instead make a ppc_cache_info structure with one
copy for I and one for D

Signed-off-by: Benjamin Herrenschmidt 
Signed-off-by: Michael Ellerman

powerpc/64: Retrieve number of L1 cache sets from device-tree

2017-02-06T08:46:04+00:00

It will be used to calculate the associativity

Signed-off-by: Benjamin Herrenschmidt 
Signed-off-by: Michael Ellerman

powerpc/64: Fix naming of cache block vs. cache line

2017-02-06T08:46:04+00:00

In a number of places we called "cache line size" what is actually
the cache block size, which in the powerpc architecture, means the
effective size to use with cache management instructions (it can
be different from the actual cache line size).

We fix the naming across the board and properly retrieve both
pieces of information when available in the device-tree.

Signed-off-by: Benjamin Herrenschmidt 
Signed-off-by: Michael Ellerman

powerpc/64s: Reduce exception alignment

2016-11-14T00:11:51+00:00

Exception handlers are aligned to 128 bytes (L1 cache) on 64s, which is
overkill. It can reduce the icache footprint of any individual exception
path. However taken as a whole, the expansion in icache footprint seems
likely to be counter-productive and cause more total misses.

Create IFETCH_ALIGN_SHIFT/BYTES, which should give optimal ifetch
alignment with much more reasonable alignment. This saves 1792 bytes
from head_64.o text with an allmodconfig build.

Other subarchitectures should define appropriate IFETCH_ALIGN_SHIFT
values if this becomes more widely used.

Signed-off-by: Nicholas Piggin 
Signed-off-by: Michael Ellerman

powerpc: add inline functions for cache related instructions

2016-03-11T23:20:11+00:00

This patch adds inline functions to use dcbz, dcbi, dcbf, dcbst
from C functions

Signed-off-by: Christophe Leroy 
Signed-off-by: Scott Wood

powerpc: Revert "Use the POWER8 Micro Partition Prefetch Engine in KVM HV on POWER8"

2015-10-21T09:50:30+00:00

This reverts commit 9678cdaae939 ("Use the POWER8 Micro Partition
Prefetch Engine in KVM HV on POWER8") because the original commit had
multiple, partly self-cancelling bugs, that could cause occasional
memory corruption.

In fact the logmpp instruction was incorrectly using register r0 as the
source of the buffer address and operation code, and depending on what
was in r0, it would either do nothing or corrupt the 64k page pointed to
by r0.

The logmpp instruction encoding and the operation code definitions could
be corrected, but then there is the problem that there is no clearly
defined way to know when the hardware has finished writing to the
buffer.

The original commit attempted to work around this by aborting the
write-out before starting the prefetch, but this is ineffective in the
case where the virtual core is now executing on a different physical
core from the one where the write-out was initiated.

These problems plus advice from the hardware designers not to use the
function (since the measured performance improvement from using the
feature was actually mostly negative), mean that reverting the code is
the best option.

Fixes: 9678cdaae939 ("Use the POWER8 Micro Partition Prefetch Engine in KVM HV on POWER8")
Signed-off-by: Paul Mackerras 
Signed-off-by: Michael Ellerman

powerpc: Remove duplicate cacheable_memcpy/memzero functions

2015-03-17T00:25:50+00:00

These functions are only used from one place each.  If the cacheable_*
versions really are more efficient, then those changes should be
migrated into the common code instead.

NOTE: The old routines are just flat buggy on kernels that support
      hardware with different cacheline sizes.

Signed-off-by: Kyle Moffett 
Signed-off-by: Benjamin Herrenschmidt

Use the POWER8 Micro Partition Prefetch Engine in KVM HV on POWER8

2014-07-28T13:23:17+00:00

The POWER8 processor has a Micro Partition Prefetch Engine, which is
a fancy way of saying "has way to store and load contents of L2 or
L2+MRU way of L3 cache". We initiate the storing of the log (list of
addresses) using the logmpp instruction and start restore by writing
to a SPR.

The logmpp instruction takes parameters in a single 64bit register:
- starting address of the table to store log of L2/L2+L3 cache contents
  - 32kb for L2
  - 128kb for L2+L3
  - Aligned relative to maximum size of the table (32kb or 128kb)
- Log control (no-op, L2 only, L2 and L3, abort logout)

We should abort any ongoing logging before initiating one.

To initiate restore, we write to the MPPR SPR. The format of what to write
to the SPR is similar to the logmpp instruction parameter:
- starting address of the table to read from (same alignment requirements)
- table size (no data, until end of table)
- prefetch rate (from fastest possible to slower. about every 8, 16, 24 or
  32 cycles)

The idea behind loading and storing the contents of L2/L3 cache is to
reduce memory latency in a system that is frequently swapping vcores on
a physical CPU.

The best case scenario for doing this is when some vcores are doing very
cache heavy workloads. The worst case is when they have about 0 cache hits,
so we just generate needless memory operations.

This implementation just does L2 store/load. In my benchmarks this proves
to be useful.

Benchmark 1:
 - 16 core POWER8
 - 3x Ubuntu 14.04LTS guests (LE) with 8 VCPUs each
 - No split core/SMT
 - two guests running sysbench memory test.
   sysbench --test=memory --num-threads=8 run
 - one guest running apache bench (of default HTML page)
   ab -n 490000 -c 400 http://localhost/

This benchmark aims to measure performance of real world application (apache)
where other guests are cache hot with their own workloads. The sysbench memory
benchmark does pointer sized writes to a (small) memory buffer in a loop.

In this benchmark with this patch I can see an improvement both in requests
per second (~5%) and in mean and median response times (again, about 5%).
The spread of minimum and maximum response times were largely unchanged.

benchmark 2:
 - Same VM config as benchmark 1
 - all three guests running sysbench memory benchmark

This benchmark aims to see if there is a positive or negative affect to this
cache heavy benchmark. Although due to the nature of the benchmark (stores) we
may not see a difference in performance, but rather hopefully an improvement
in consistency of performance (when vcore switched in, don't have to wait
many times for cachelines to be pulled in)

The results of this benchmark are improvements in consistency of performance
rather than performance itself. With this patch, the few outliers in duration
go away and we get more consistent performance in each guest.

benchmark 3:
 - same 3 guests and CPU configuration as benchmark 1 and 2.
 - two idle guests
 - 1 guest running STREAM benchmark

This scenario also saw performance improvement with this patch. On Copy and
Scale workloads from STREAM, I got 5-6% improvement with this patch. For
Add and triad, it was around 10% (or more).

benchmark 4:
 - same 3 guests as previous benchmarks
 - two guests running sysbench --memory, distinctly different cache heavy
   workload
 - one guest running STREAM benchmark.

Similar improvements to benchmark 3.

benchmark 5:
 - 1 guest, 8 VCPUs, Ubuntu 14.04
 - Host configured with split core (SMT8, subcores-per-core=4)
 - STREAM benchmark

In this benchmark, we see a 10-20% performance improvement across the board
of STREAM benchmark results with this patch.

Based on preliminary investigation and microbenchmarks
by Prerna Saxena 

Signed-off-by: Stewart Smith 
Acked-by: Paul Mackerras 
Signed-off-by: Alexander Graf