Allow to defer the flushing of the TLB when unmapping pages, which allows
to reduce the numbers of IPI and the number of sfence.vma.
The ubenchmarch used in commit 43b3dfdd04 ("arm64: support
batched/deferred tlb shootdown during page reclamation/migration") that
was multithreaded to force the usage of IPI shows good performance
improvement on all platforms:
* Unmatched: ~34%
* TH1520 : ~78%
* Qemu : ~81%
In addition, perf on qemu reports an important decrease in time spent
dealing with IPIs:
Before: 68.17% main [kernel.kallsyms] [k] __sbi_rfence_v02_call
After : 8.64% main [kernel.kallsyms] [k] __sbi_rfence_v02_call
* Benchmark:
int stick_this_thread_to_core(int core_id) {
int num_cores = sysconf(_SC_NPROCESSORS_ONLN);
if (core_id < 0 || core_id >= num_cores)
return EINVAL;
cpu_set_t cpuset;
CPU_ZERO(&cpuset);
CPU_SET(core_id, &cpuset);
pthread_t current_thread = pthread_self();
return pthread_setaffinity_np(current_thread,
sizeof(cpu_set_t), &cpuset);
}
static void *fn_thread (void *p_data)
{
int ret;
pthread_t thread;
stick_this_thread_to_core((int)p_data);
while (1) {
sleep(1);
}
return NULL;
}
int main()
{
volatile unsigned char *p = mmap(NULL, SIZE, PROT_READ | PROT_WRITE,
MAP_SHARED | MAP_ANONYMOUS, -1, 0);
pthread_t threads[4];
int ret;
for (int i = 0; i < 4; ++i) {
ret = pthread_create(&threads[i], NULL, fn_thread, (void *)i);
if (ret)
{
printf("%s", strerror (ret));
}
}
memset(p, 0x88, SIZE);
for (int k = 0; k < 10000; k++) {
/* swap in */
for (int i = 0; i < SIZE; i += 4096) {
(void)p[i];
}
/* swap out */
madvise(p, SIZE, MADV_PAGEOUT);
}
for (int i = 0; i < 4; i++)
{
pthread_cancel(threads[i]);
}
for (int i = 0; i < 4; i++)
{
pthread_join(threads[i], NULL);
}
return 0;
}
Signed-off-by: Alexandre Ghiti <alexghiti@rivosinc.com>
Reviewed-by: Jisheng Zhang <jszhang@kernel.org>
Tested-by: Jisheng Zhang <jszhang@kernel.org> # Tested on TH1520
Tested-by: Nam Cao <namcao@linutronix.de>
Link: https://lore.kernel.org/r/20240108193640.344929-1-alexghiti@rivosinc.com
Signed-off-by: Palmer Dabbelt <palmer@rivosinc.com>
On x86, batched and deferred tlb shootdown has lead to 90% performance
increase on tlb shootdown. on arm64, HW can do tlb shootdown without
software IPI. But sync tlbi is still quite expensive.
Even running a simplest program which requires swapout can
prove this is true,
#include <sys/types.h>
#include <unistd.h>
#include <sys/mman.h>
#include <string.h>
int main()
{
#define SIZE (1 * 1024 * 1024)
volatile unsigned char *p = mmap(NULL, SIZE, PROT_READ | PROT_WRITE,
MAP_SHARED | MAP_ANONYMOUS, -1, 0);
memset(p, 0x88, SIZE);
for (int k = 0; k < 10000; k++) {
/* swap in */
for (int i = 0; i < SIZE; i += 4096) {
(void)p[i];
}
/* swap out */
madvise(p, SIZE, MADV_PAGEOUT);
}
}
Perf result on snapdragon 888 with 8 cores by using zRAM
as the swap block device.
~ # perf record taskset -c 4 ./a.out
[ perf record: Woken up 10 times to write data ]
[ perf record: Captured and wrote 2.297 MB perf.data (60084 samples) ]
~ # perf report
# To display the perf.data header info, please use --header/--header-only options.
# To display the perf.data header info, please use --header/--header-only options.
#
#
# Total Lost Samples: 0
#
# Samples: 60K of event 'cycles'
# Event count (approx.): 35706225414
#
# Overhead Command Shared Object Symbol
# ........ ....... ................. ......
#
21.07% a.out [kernel.kallsyms] [k] _raw_spin_unlock_irq
8.23% a.out [kernel.kallsyms] [k] _raw_spin_unlock_irqrestore
6.67% a.out [kernel.kallsyms] [k] filemap_map_pages
6.16% a.out [kernel.kallsyms] [k] __zram_bvec_write
5.36% a.out [kernel.kallsyms] [k] ptep_clear_flush
3.71% a.out [kernel.kallsyms] [k] _raw_spin_lock
3.49% a.out [kernel.kallsyms] [k] memset64
1.63% a.out [kernel.kallsyms] [k] clear_page
1.42% a.out [kernel.kallsyms] [k] _raw_spin_unlock
1.26% a.out [kernel.kallsyms] [k] mod_zone_state.llvm.8525150236079521930
1.23% a.out [kernel.kallsyms] [k] xas_load
1.15% a.out [kernel.kallsyms] [k] zram_slot_lock
ptep_clear_flush() takes 5.36% CPU in the micro-benchmark swapping in/out
a page mapped by only one process. If the page is mapped by multiple
processes, typically, like more than 100 on a phone, the overhead would be
much higher as we have to run tlb flush 100 times for one single page.
Plus, tlb flush overhead will increase with the number of CPU cores due to
the bad scalability of tlb shootdown in HW, so those ARM64 servers should
expect much higher overhead.
Further perf annonate shows 95% cpu time of ptep_clear_flush is actually
used by the final dsb() to wait for the completion of tlb flush. This
provides us a very good chance to leverage the existing batched tlb in
kernel. The minimum modification is that we only send async tlbi in the
first stage and we send dsb while we have to sync in the second stage.
With the above simplest micro benchmark, collapsed time to finish the
program decreases around 5%.
Typical collapsed time w/o patch:
~ # time taskset -c 4 ./a.out
0.21user 14.34system 0:14.69elapsed
w/ patch:
~ # time taskset -c 4 ./a.out
0.22user 13.45system 0:13.80elapsed
Also tested with benchmark in the commit on Kunpeng920 arm64 server
and observed an improvement around 12.5% with command
`time ./swap_bench`.
w/o w/
real 0m13.460s 0m11.771s
user 0m0.248s 0m0.279s
sys 0m12.039s 0m11.458s
Originally it's noticed a 16.99% overhead of ptep_clear_flush()
which has been eliminated by this patch:
[root@localhost yang]# perf record -- ./swap_bench && perf report
[...]
16.99% swap_bench [kernel.kallsyms] [k] ptep_clear_flush
It is tested on 4,8,128 CPU platforms and shows to be beneficial on
large systems but may not have improvement on small systems like on
a 4 CPU platform.
Also this patch improve the performance of page migration. Using pmbench
and tries to migrate the pages of pmbench between node 0 and node 1 for
100 times for 1G memory, this patch decrease the time used around 20%
(prev 18.338318910 sec after 13.981866350 sec) and saved the time used
by ptep_clear_flush().
Link: https://lkml.kernel.org/r/20230717131004.12662-5-yangyicong@huawei.com
Tested-by: Yicong Yang <yangyicong@hisilicon.com>
Tested-by: Xin Hao <xhao@linux.alibaba.com>
Tested-by: Punit Agrawal <punit.agrawal@bytedance.com>
Signed-off-by: Barry Song <v-songbaohua@oppo.com>
Signed-off-by: Yicong Yang <yangyicong@hisilicon.com>
Reviewed-by: Kefeng Wang <wangkefeng.wang@huawei.com>
Reviewed-by: Xin Hao <xhao@linux.alibaba.com>
Reviewed-by: Anshuman Khandual <anshuman.khandual@arm.com>
Reviewed-by: Catalin Marinas <catalin.marinas@arm.com>
Cc: Anshuman Khandual <anshuman.khandual@arm.com>
Cc: Jonathan Corbet <corbet@lwn.net>
Cc: Nadav Amit <namit@vmware.com>
Cc: Mel Gorman <mgorman@suse.de>
Cc: Anshuman Khandual <khandual@linux.vnet.ibm.com>
Cc: Arnd Bergmann <arnd@arndb.de>
Cc: Barry Song <baohua@kernel.org>
Cc: Darren Hart <darren@os.amperecomputing.com>
Cc: Jonathan Cameron <Jonathan.Cameron@huawei.com>
Cc: lipeifeng <lipeifeng@oppo.com>
Cc: Mark Rutland <mark.rutland@arm.com>
Cc: Peter Zijlstra <peterz@infradead.org>
Cc: Ryan Roberts <ryan.roberts@arm.com>
Cc: Steven Miao <realmz6@gmail.com>
Cc: Will Deacon <will@kernel.org>
Cc: Zeng Tao <prime.zeng@hisilicon.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
The arch support status files don't match reality as of v5.19-rc1,
use the features-refresh.sh to refresh all the arch-support.txt files
in place. The main effect is to add entries for the new loong
architecture.
Signed-off-by: Zheng Zengkai <zhengzengkai@huawei.com>
Link: https://lore.kernel.org/r/20220609025656.143460-1-zhengzengkai@huawei.com
Signed-off-by: Jonathan Corbet <corbet@lwn.net>
* 'remove-h8300' of git://git.infradead.org/users/hch/misc:
remove the h8300 architecture
This is clearly the least actively maintained architecture we have at
the moment, and probably the least useful. It is now the only one that
does not support MMUs at all, and most of the boards only support 4MB
of RAM, out of which the defconfig kernel needs more than half just
for .text/.data.
Guenter Roeck did the original patch to remove the architecture in 2013
after it had already been obsolete for a while, and Yoshinori Sato brought
it back in a much more modern form in 2015. Looking at the git history
since the reinstantiation, it's clear that almost all commits in the tree
are build fixes or cross-architecture cleanups:
$ git log --no-merges --format=%an v4.5.. arch/h8300/ | sort | uniq
-c | sort -rn | head -n 12
25 Masahiro Yamada
18 Christoph Hellwig
14 Mike Rapoport
9 Arnd Bergmann
8 Mark Rutland
7 Peter Zijlstra
6 Kees Cook
6 Ingo Molnar
6 Al Viro
5 Randy Dunlap
4 Yury Norov
Signed-off-by: Arnd Bergmann <arnd@arndb.de>
The nds32 architecture, also known as AndeStar V3, is a custom 32-bit
RISC target designed by Andes Technologies. Support was added to the
kernel in 2016 as the replacement RISC-V based V5 processors were
already announced, and maintained by (current or former) Andes
employees.
As explained by Alan Kao, new customers are now all using RISC-V,
and all known nds32 users are already on longterm stable kernels
provided by Andes, with no development work going into mainline
support any more.
While the port is still in a reasonably good shape, it only gets
worse over time without active maintainers, so it seems best
to remove it before it becomes unusable. As always, if it turns
out that there are mainline users after all, and they volunteer
to maintain the port in the future, the removal can be reverted.
Link: https://lore.kernel.org/linux-mm/YhdWNLUhk+x9RAzU@yamatobi.andestech.com/
Link: https://lore.kernel.org/lkml/20220302065213.82702-1-alankao@andestech.com/
Link: https://www.andestech.com/en/products-solutions/andestar-architecture/
Signed-off-by: Alan Kao <alankao@andestech.com>
[arnd: rewrite changelog to provide more background]
Signed-off-by: Arnd Bergmann <arnd@arndb.de>
BATCHED_UNMAP_TLB_FLUSH is used on x86 to do batched tlb shootdown by
sending one IPI to TLB flush all entries after unmapping pages rather
than sending an IPI to flush each individual entry.
On arm64, tlb shootdown is done by hardware. Flush instructions are
innershareable. The local flushes are limited to the boot (1 per CPU)
and when a task is getting a new ASID.
So marking this feature as "TODO" is not proper. ".." isn't good as
well. So this patch adds a "N/A" for this kind of features which are
not needed on some architectures.
Signed-off-by: Barry Song <song.bao.hua@hisilicon.com>
Acked-by: Will Deacon <will@kernel.org>
Cc: Mel Gorman <mgorman@suse.de>
Cc: Andy Lutomirski <luto@kernel.org>
Cc: Catalin Marinas <catalin.marinas@arm.com>
Cc: Will Deacon <will@kernel.org>
Link: https://lore.kernel.org/r/20210223003230.11976-1-song.bao.hua@hisilicon.com
Signed-off-by: Jonathan Corbet <corbet@lwn.net>
The unicore32 port do not seem maintained for a long time now, there is no
upstream toolchain that can create unicore32 binaries and all the links to
prebuilt toolchains for unicore32 are dead. Even compilers that were
available are not supported by the kernel anymore.
Guenter Roeck says:
I have stopped building unicore32 images since v4.19 since there is no
available compiler that is still supported by the kernel. I am surprised
that support for it has not been removed from the kernel.
Remove unicore32 port.
Signed-off-by: Mike Rapoport <rppt@linux.ibm.com>
Acked-by: Arnd Bergmann <arnd@arndb.de>
Acked-by: Guenter Roeck <linux@roeck-us.net>
Now that the script 'features-refresh.sh' is available, uses this script
to refresh all the arch-support.txt files in place.
Signed-off-by: Andrea Parri <andrea.parri@amarulasolutions.com>
Cc: Ingo Molnar <mingo@kernel.org>
Cc: Jonathan Corbet <corbet@lwn.net>
Cc: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Jonathan Corbet <corbet@lwn.net>
A number of architecture ports are obsolete and getting dropped,
so we no longer want to track the respective features.
We already removed the lines for metag and mn10300, this does
the same edits for all the others.
For the remaining 21 architectures, this shows how many are known
to implement each given feature:
19 time/modern-timekeeping/arch-support.txt
19 time/clockevents/arch-support.txt
15 core/tracehook/arch-support.txt
14 core/generic-idle-thread/arch-support.txt
13 locking/lockdep/arch-support.txt
12 io/dma-api-debug/arch-support.txt
11 debug/kgdb/arch-support.txt
10 time/virt-cpuacct/arch-support.txt
9 debug/kretprobes/arch-support.txt
9 debug/kprobes/arch-support.txt
8 vm/THP/arch-support.txt
8 vm/pte_special/arch-support.txt
8 vm/numa-memblock/arch-support.txt
8 io/sg-chain/arch-support.txt
7 perf/kprobes-event/arch-support.txt
7 locking/rwsem-optimized/arch-support.txt
7 debug/gcov-profile-all/arch-support.txt
7 core/jump-labels/arch-support.txt
7 core/BPF-JIT/arch-support.txt
6 vm/ELF-ASLR/arch-support.txt
6 time/context-tracking/arch-support.txt
6 seccomp/seccomp-filter/arch-support.txt
6 debug/stackprotector/arch-support.txt
5 time/irq-time-acct/arch-support.txt
5 io/dma-contiguous/arch-support.txt
5 debug/uprobes/arch-support.txt
4 vm/ioremap_prot/arch-support.txt
4 time/arch-tick-broadcast/arch-support.txt
4 perf/perf-stackdump/arch-support.txt
4 perf/perf-regs/arch-support.txt
3 debug/KASAN/arch-support.txt
2 vm/PG_uncached/arch-support.txt
2 vm/huge-vmap/arch-support.txt
2 sched/numa-balancing/arch-support.txt
2 sched/membarrier-sync-core/arch-support.txt
2 locking/cmpxchg-local/arch-support.txt
2 debug/optprobes/arch-support.txt
2 debug/kprobes-on-ftrace/arch-support.txt
1 vm/TLB/arch-support.txt
1 locking/queued-spinlocks/arch-support.txt
1 locking/queued-rwlocks/arch-support.txt
1 debug/user-ret-profiler/arch-support.txt
0 lib/strncasecmp/arch-support.txt
Note that the list does not include riscv or nds32 yet, these still
need to be added.
Signed-off-by: Arnd Bergmann <arnd@arndb.de>
Remove any remaining references to the Meta architecture in
Documentation/, primarily from Documentation/features/.
Signed-off-by: James Hogan <jhogan@kernel.org>
Cc: Jonathan Corbet <corbet@lwn.net>
Cc: linux-metag@vger.kernel.org
Cc: linux-doc@vger.kernel.org
The AVR32 architecture support has been removed from the Linux kernel,
hence remove all references to it from Documentation.
Signed-off-by: Hans-Christian Noren Egtvedt <egtvedt@samfundet.no>
Signed-off-by: Håvard Skinnemoen <hskinnemoen@gmail.com>
Signed-off-by: Nicolas Ferre <nicolas.ferre@microchip.com>
Acked-by: Andy Shevchenko <andy.shevchenko@gmail.com>
Acked-by: Boris Brezillon <boris.brezillon@free-electrons.com>
