#
bd60438e |
|
31-Jul-2023 |
Willem de Bruijn <willemb@google.com> |
selftests/net: report rcv_mss in tcp_mmap tcp_mmap tests TCP_ZEROCOPY_RECEIVE. If 0% of data is received using mmap, this may be due to mss. Report rcv_mss to identify this cause. Output of a run failed due to too small mss: received 32768 MB (0 % mmap'ed) in 8.40458 s, 32.7057 Gbit cpu usage user:0.027922 sys:8.21126, 251.44 usec per MB, 3252 c-switches, rcv_mss 1428 Output on a successful run: received 32768 MB (99.9507 % mmap'ed) in 4.69023 s, 58.6064 Gbit cpu usage user:0.029172 sys:2.56105, 79.0473 usec per MB, 57591 c-switches, rcv_mss 4096 Suggested-by: Eric Dumazet <edumazet@google.com> Signed-off-by: Willem de Bruijn <willemb@google.com> Signed-off-by: David S. Miller <davem@davemloft.net>
|
#
905a9eb5 |
|
05-Apr-2023 |
Eric Dumazet <edumazet@google.com> |
selftests/net: fix typo in tcp_mmap kernel test robot reported the following warning: All warnings (new ones prefixed by >>): tcp_mmap.c: In function 'child_thread': >> tcp_mmap.c:211:61: warning: 'lu' may be used uninitialized in this function [-Wmaybe-uninitialized] 211 | zc.length = min(chunk_size, FILE_SZ - lu); We want to read FILE_SZ bytes, so the correct expression should be (FILE_SZ - total) Fixes: 5c5945dc695c ("selftests/net: Add SHA256 computation over data sent in tcp_mmap") Reported-by: kernel test robot <lkp@intel.com> Link: https://lore.kernel.org/oe-kbuild-all/202304042104.UFIuevBp-lkp@intel.com/ Signed-off-by: Eric Dumazet <edumazet@google.com> Cc: Xiaoyan Li <lixiaoyan@google.com> Cc: Kuniyuki Iwashima <kuniyu@amazon.com> Link: https://lore.kernel.org/r/20230405071556.1019623-1-edumazet@google.com Signed-off-by: Paolo Abeni <pabeni@redhat.com>
|
#
5c5945dc |
|
21-Mar-2023 |
Xiaoyan Li <lixiaoyan@google.com> |
selftests/net: Add SHA256 computation over data sent in tcp_mmap Add option to compute and send SHA256 over data sent (-i). This is to ensure the correctness of data received. Data is randomly populated from /dev/urandom. Tested: ./tcp_mmap -s -z -i ./tcp_mmap -z -H $ADDR -i SHA256 is correct ./tcp_mmap -s -i ./tcp_mmap -H $ADDR -i SHA256 is correct Signed-off-by: Coco Li <lixiaoyan@google.com> Reviewed-by: Eric Dumazet <edumazet@google.com> Link: https://lore.kernel.org/r/20230321081202.2370275-2-lixiaoyan@google.com Signed-off-by: Paolo Abeni <pabeni@redhat.com>
|
#
057fb031 |
|
20-Jan-2023 |
Eric Dumazet <edumazet@google.com> |
selftests: net: tcp_mmap: populate pages in send path In commit 72653ae5303c ("selftests: net: tcp_mmap: Use huge pages in send path") I made a change to use hugepages for the buffer used by the client (tx path) Today, I understood that the cause for poor zerocopy performance was that after a mmap() for a 512KB memory zone, kernel uses a single zeropage, mapped 128 times. This was really the reason for poor tx path performance in zero copy mode, because this zero page refcount is under high pressure, especially when TCP ACK packets are processed on another cpu. We need either to force a COW on all the memory range, or use MAP_POPULATE so that a zero page is not abused. Signed-off-by: Eric Dumazet <edumazet@google.com> Link: https://lore.kernel.org/r/20230120181136.3764521-1-edumazet@google.com Signed-off-by: Jakub Kicinski <kuba@kernel.org>
|
#
59c0d319 |
|
20-Aug-2020 |
Eric Dumazet <edumazet@google.com> |
selftests: net: tcp_mmap: Use huge pages in receive path One down side of using TCP rx zerocopy is one extra TLB miss per page after the mapping operation. While if the application is using hugepages, the non zerocopy recvmsg() will not have to pay these TLB costs. This patch allows server side to use huge pages for the non zero copy case, to allow fair comparisons when both solutions use optimal conditions. Signed-off-by: Eric Dumazet <edumazet@google.com> Cc: Arjun Roy <arjunroy@google.com> Cc: Soheil Hassas Yeganeh <soheil@google.com> Signed-off-by: David S. Miller <davem@davemloft.net>
|
#
72653ae5 |
|
20-Aug-2020 |
Eric Dumazet <edumazet@google.com> |
selftests: net: tcp_mmap: Use huge pages in send path There are significant gains using huge pages when available, as shown in [1]. This patch adds mmap_large_buffer() and uses it in client side (tx path of this reference tool) Following patch will use the feature for server side. [1] https://patchwork.ozlabs.org/project/netdev/patch/20200820154359.1806305-1-edumazet@google.com/ Signed-off-by: Eric Dumazet <edumazet@google.com> Cc: Arjun Roy <arjunroy@google.com> Cc: Soheil Hassas Yeganeh <soheil@google.com> Signed-off-by: David S. Miller <davem@davemloft.net>
|
#
0d894193 |
|
20-Aug-2020 |
Eric Dumazet <edumazet@google.com> |
selftests: net: tcp_mmap: use madvise(MADV_DONTNEED) When TCP_ZEROCOPY_RECEIVE operation has been added, I made the mistake of automatically un-mapping prior content before mapping new pages. This has the unfortunate effect of adding potentially long MMU operations (like TLB flushes) while socket lock is held. Using madvise(MADV_DONTNEED) right after pages has been used has two benefits : 1) This releases pages sooner, allowing pages to be recycled if they were part of a page pool in a NIC driver. 2) No more long unmap operations while preventing immediate processing of incoming packets. The cost of the added system call is small enough. Arjun will submit a kernel patch allowing to opt out from the unmap attempt in tcp_zerocopy_receive() Signed-off-by: Eric Dumazet <edumazet@google.com> Cc: Arjun Roy <arjunroy@google.com> Cc: Soheil Hassas Yeganeh <soheil@google.com> Signed-off-by: David S. Miller <davem@davemloft.net>
|
#
94b6c13b |
|
26-Jul-2020 |
Tanner Love <tannerlove@google.com> |
selftests/net: tcp_mmap: fix clang warning for target arch PowerPC When size_t maps to unsigned int (e.g. on 32-bit powerpc), then the comparison with 1<<35 is always true. Clang 9 threw: warning: result of comparison of constant 34359738368 with \ expression of type 'size_t' (aka 'unsigned int') is always true \ [-Wtautological-constant-out-of-range-compare] while (total < FILE_SZ) { Tested: make -C tools/testing/selftests TARGETS="net" run_tests Fixes: 192dc405f308 ("selftests: net: add tcp_mmap program") Signed-off-by: Tanner Love <tannerlove@google.com> Acked-by: Willem de Bruijn <willemb@google.com> Reviewed-by: Eric Dumazet <edumazet@google.com> Signed-off-by: David S. Miller <davem@davemloft.net>
|
#
a8472417 |
|
06-May-2020 |
Eric Dumazet <edumazet@google.com> |
selftests: net: tcp_mmap: fix SO_RCVLOWAT setting Since chunk_size is no longer an integer, we can not use it directly as an argument of setsockopt(). This patch should fix tcp_mmap for Big Endian kernels. Fixes: 597b01edafac ("selftests: net: avoid ptl lock contention in tcp_mmap") Signed-off-by: Eric Dumazet <edumazet@google.com> Cc: Soheil Hassas Yeganeh <soheil@google.com> Cc: Arjun Roy <arjunroy@google.com> Acked-by: Soheil Hassas Yeganeh <soheil@google.com> Signed-off-by: David S. Miller <davem@davemloft.net>
|
#
bf5525f3 |
|
05-May-2020 |
Eric Dumazet <edumazet@google.com> |
selftests: net: tcp_mmap: clear whole tcp_zerocopy_receive struct We added fields in tcp_zerocopy_receive structure, so make sure to clear all fields to not pass garbage to the kernel. We were lucky because recent additions added 'out' parameters, still we need to clean our reference implementation, before folks copy/paste it. Fixes: c8856c051454 ("tcp-zerocopy: Return inq along with tcp receive zerocopy.") Fixes: 33946518d493 ("tcp-zerocopy: Return sk_err (if set) along with tcp receive zerocopy.") Signed-off-by: Eric Dumazet <edumazet@google.com> Cc: Arjun Roy <arjunroy@google.com> Cc: Soheil Hassas Yeganeh <soheil@google.com> Acked-by: Soheil Hassas Yeganeh <soheil@google.com> Signed-off-by: David S. Miller <davem@davemloft.net>
|
#
597b01ed |
|
15-Nov-2019 |
Eric Dumazet <edumazet@google.com> |
selftests: net: avoid ptl lock contention in tcp_mmap tcp_mmap is used as a reference program for TCP rx zerocopy, so it is important to point out some potential issues. If multiple threads are concurrently using getsockopt(... TCP_ZEROCOPY_RECEIVE), there is a chance the low-level mm functions compete on shared ptl lock, if vma are arbitrary placed. Instead of letting the mm layer place the chunks back to back, this patch enforces an alignment so that each thread uses a different ptl lock. Performance measured on a 100 Gbit NIC, with 8 tcp_mmap clients launched at the same time : $ for f in {1..8}; do ./tcp_mmap -H 2002:a05:6608:290:: & done In the following run, we reproduce the old behavior by requesting no alignment : $ tcp_mmap -sz -C $((128*1024)) -a 4096 received 32768 MB (100 % mmap'ed) in 9.69532 s, 28.3516 Gbit cpu usage user:0.08634 sys:3.86258, 120.511 usec per MB, 171839 c-switches received 32768 MB (100 % mmap'ed) in 25.4719 s, 10.7914 Gbit cpu usage user:0.055268 sys:21.5633, 659.745 usec per MB, 9065 c-switches received 32768 MB (100 % mmap'ed) in 28.5419 s, 9.63069 Gbit cpu usage user:0.057401 sys:23.8761, 730.392 usec per MB, 14987 c-switches received 32768 MB (100 % mmap'ed) in 28.655 s, 9.59268 Gbit cpu usage user:0.059689 sys:23.8087, 728.406 usec per MB, 18509 c-switches received 32768 MB (100 % mmap'ed) in 28.7808 s, 9.55074 Gbit cpu usage user:0.066042 sys:23.4632, 718.056 usec per MB, 24702 c-switches received 32768 MB (100 % mmap'ed) in 28.8259 s, 9.5358 Gbit cpu usage user:0.056547 sys:23.6628, 723.858 usec per MB, 23518 c-switches received 32768 MB (100 % mmap'ed) in 28.8808 s, 9.51767 Gbit cpu usage user:0.059357 sys:23.8515, 729.703 usec per MB, 14691 c-switches received 32768 MB (100 % mmap'ed) in 28.8879 s, 9.51534 Gbit cpu usage user:0.047115 sys:23.7349, 725.769 usec per MB, 21773 c-switches New behavior (automatic alignment based on Hugepagesize), we can see the system overhead being dramatically reduced. $ tcp_mmap -sz -C $((128*1024)) received 32768 MB (100 % mmap'ed) in 13.5339 s, 20.3103 Gbit cpu usage user:0.122644 sys:3.4125, 107.884 usec per MB, 168567 c-switches received 32768 MB (100 % mmap'ed) in 16.0335 s, 17.1439 Gbit cpu usage user:0.132428 sys:3.55752, 112.608 usec per MB, 188557 c-switches received 32768 MB (100 % mmap'ed) in 17.5506 s, 15.6621 Gbit cpu usage user:0.155405 sys:3.24889, 103.891 usec per MB, 226652 c-switches received 32768 MB (100 % mmap'ed) in 19.1924 s, 14.3222 Gbit cpu usage user:0.135352 sys:3.35583, 106.542 usec per MB, 207404 c-switches received 32768 MB (100 % mmap'ed) in 22.3649 s, 12.2906 Gbit cpu usage user:0.142429 sys:3.53187, 112.131 usec per MB, 250225 c-switches received 32768 MB (100 % mmap'ed) in 22.5336 s, 12.1986 Gbit cpu usage user:0.140654 sys:3.61971, 114.757 usec per MB, 253754 c-switches received 32768 MB (100 % mmap'ed) in 22.5483 s, 12.1906 Gbit cpu usage user:0.134035 sys:3.55952, 112.718 usec per MB, 252997 c-switches received 32768 MB (100 % mmap'ed) in 22.6442 s, 12.139 Gbit cpu usage user:0.126173 sys:3.71251, 117.147 usec per MB, 253728 c-switches Signed-off-by: Eric Dumazet <edumazet@google.com> Cc: Soheil Hassas Yeganeh <soheil@google.com> Cc: Arjun Roy <arjunroy@google.com> Acked-by: Soheil Hassas Yeganeh <soheil@google.com> Signed-off-by: David S. Miller <davem@davemloft.net>
|
#
20021578 |
|
14-Nov-2019 |
Eric Dumazet <edumazet@google.com> |
selftests: net: tcp_mmap should create detached threads Since we do not plan using pthread_join() in the server do_accept() loop, we better create detached threads, or risk increasing memory footprint over time. Fixes: 192dc405f308 ("selftests: net: add tcp_mmap program") Signed-off-by: Eric Dumazet <edumazet@google.com> Signed-off-by: David S. Miller <davem@davemloft.net>
|
#
e698a237 |
|
22-Oct-2019 |
Masami Hiramatsu <mhiramat@kernel.org> |
selftests: net: Use size_t and ssize_t for counting file size Use size_t and ssize_t correctly for counting send file size instead of unsigned long and long, because long is 32bit on 32bit arch, which is not enough for counting long file size (>4GB). Signed-off-by: Masami Hiramatsu <mhiramat@kernel.org> Cc: Eric Dumazet <edumazet@google.com> Cc: David S. Miller <davem@davemloft.net> Signed-off-by: Shuah Khan <skhan@linuxfoundation.org>
|
#
fc44ef5a |
|
28-May-2019 |
Thomas Gleixner <tglx@linutronix.de> |
treewide: Replace GPLv2 boilerplate/reference with SPDX - rule 222 Based on 1 normalized pattern(s): license gplv2 this program is free software you can redistribute it and or modify it under the terms and conditions of the gnu general public license version 2 as published by the free software foundation this program is distributed in the hope it will be useful but without any warranty without even the implied warranty of merchantability or fitness for a particular purpose see the gnu general public license for more details you should have received a copy of the gnu general public license along with this program if not write to the free software foundation inc 51 franklin st fifth floor boston ma 02110 1301 usa extracted by the scancode license scanner the SPDX license identifier GPL-2.0-only has been chosen to replace the boilerplate/reference in 4 file(s). Signed-off-by: Thomas Gleixner <tglx@linutronix.de> Reviewed-by: Alexios Zavras <alexios.zavras@intel.com> Reviewed-by: Steve Winslow <swinslow@gmail.com> Reviewed-by: Allison Randal <allison@lohutok.net> Cc: linux-spdx@vger.kernel.org Link: https://lkml.kernel.org/r/20190528171440.038486796@linutronix.de Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
|
#
258fe208 |
|
02-Aug-2018 |
Maninder Singh <maninder1.s@samsung.com> |
selftest/net: fix protocol family to work for IPv4. use actual protocol family passed by user rather than hardcoded AF_INTE6 to cerate sockets. current code is not working for IPv4. Signed-off-by: Maninder Singh <maninder1.s@samsung.com> Signed-off-by: Vaneet Narang <v.narang@samsung.com> Reviewed-by: Eric Dumazet <edumazet@google.com> Signed-off-by: David S. Miller <davem@davemloft.net>
|
#
aacb0c2e |
|
27-Apr-2018 |
Eric Dumazet <edumazet@google.com> |
selftests: net: tcp_mmap must use TCP_ZEROCOPY_RECEIVE After prior kernel change, mmap() on TCP socket only reserves VMA. We have to use getsockopt(fd, IPPROTO_TCP, TCP_ZEROCOPY_RECEIVE, ...) to perform the transfert of pages from skbs in TCP receive queue into such VMA. struct tcp_zerocopy_receive { __u64 address; /* in: address of mapping */ __u32 length; /* in/out: number of bytes to map/mapped */ __u32 recv_skip_hint; /* out: amount of bytes to skip */ }; After a successful getsockopt(...TCP_ZEROCOPY_RECEIVE...), @length contains number of bytes that were mapped, and @recv_skip_hint contains number of bytes that should be read using conventional read()/recv()/recvmsg() system calls, to skip a sequence of bytes that can not be mapped, because not properly page aligned. Signed-off-by: Eric Dumazet <edumazet@google.com> Cc: Andy Lutomirski <luto@kernel.org> Acked-by: Soheil Hassas Yeganeh <soheil@google.com> Signed-off-by: David S. Miller <davem@davemloft.net>
|
#
192dc405 |
|
16-Apr-2018 |
Eric Dumazet <edumazet@google.com> |
selftests: net: add tcp_mmap program This is a reference program showing how mmap() can be used on TCP flows to implement receive zero copy. Signed-off-by: Eric Dumazet <edumazet@google.com> Signed-off-by: David S. Miller <davem@davemloft.net>
|