| lists.openwall.net | lists / announce owl-users owl-dev john-users john-dev passwdqc-users yescrypt popa3d-users / oss-security kernel-hardening musl sabotage tlsify passwords / crypt-dev xvendor / Bugtraq Full-Disclosure linux-kernel linux-netdev linux-ext4 linux-hardening linux-cve-announce PHC | |
|
Open Source and information security mailing list archives
| ||
|
Message-ID: <CAKwvOdnEfRFnO60mB9D7SAjCVijW5UBVYCGGXwhQ+EGw6c18BA@mail.gmail.com>
Date: Thu, 12 Nov 2020 11:39:07 -0800
From: Nick Desaulniers <ndesaulniers@...gle.com>
To: Magnus Karlsson <magnus.karlsson@...il.com>
Cc: kernel test robot <lkp@...el.com>,
"Karlsson, Magnus" <magnus.karlsson@...el.com>,
Björn Töpel <bjorn.topel@...el.com>,
Alexei Starovoitov <ast@...nel.org>,
Daniel Borkmann <daniel@...earbox.net>,
Network Development <netdev@...r.kernel.org>,
Jonathan Lemon <jonathan.lemon@...il.com>,
Jakub Kicinski <kuba@...nel.org>,
John Fastabend <john.fastabend@...il.com>,
kbuild-all@...ts.01.org,
clang-built-linux <clang-built-linux@...glegroups.com>,
bpf <bpf@...r.kernel.org>,
Jeff Kirsher <jeffrey.t.kirsher@...el.com>
Subject: Re: [PATCH bpf-next v2 5/5] i40e: use batched xsk Tx interfaces to
increase performance
On Wed, Nov 11, 2020 at 11:45 PM Magnus Karlsson
<magnus.karlsson@...il.com> wrote:
>
> On Wed, Nov 11, 2020 at 8:16 PM Nick Desaulniers
> <ndesaulniers@...gle.com> wrote:
> >
> > On Wed, Nov 11, 2020 at 3:57 AM Magnus Karlsson
> > <magnus.karlsson@...il.com> wrote:
> > >
> > > On Wed, Nov 11, 2020 at 2:38 AM kernel test robot <lkp@...el.com> wrote:
> > > >
> > > > Hi Magnus,
> > > >
> > > > I love your patch! Perhaps something to improve:
> > > >
> > > > [auto build test WARNING on bpf-next/master]
> > > >
> > > > url: https://github.com/0day-ci/linux/commits/Magnus-Karlsson/xsk-i40e-Tx-performance-improvements/20201110-190310
> > > > base: https://git.kernel.org/pub/scm/linux/kernel/git/bpf/bpf-next.git master
> > > > config: powerpc64-randconfig-r025-20201110 (attached as .config)
> > > > compiler: clang version 12.0.0 (https://github.com/llvm/llvm-project 4d81c8adb6ed9840257f6cb6b93f60856d422a15)
> >
> > ^ Note: clang
> >
> > > > reproduce (this is a W=1 build):
> > > > wget https://raw.githubusercontent.com/intel/lkp-tests/master/sbin/make.cross -O ~/bin/make.cross
> > > > chmod +x ~/bin/make.cross
> > > > # install powerpc64 cross compiling tool for clang build
> > > > # apt-get install binutils-powerpc64-linux-gnu
> > > > # https://github.com/0day-ci/linux/commit/b016bbeac6692a93e61b28efa430d64645032b5e
> > > > git remote add linux-review https://github.com/0day-ci/linux
> > > > git fetch --no-tags linux-review Magnus-Karlsson/xsk-i40e-Tx-performance-improvements/20201110-190310
> > > > git checkout b016bbeac6692a93e61b28efa430d64645032b5e
> > > > # save the attached .config to linux build tree
> > > > COMPILER_INSTALL_PATH=$HOME/0day COMPILER=clang make.cross ARCH=powerpc64
> > > >
> > > > If you fix the issue, kindly add following tag as appropriate
> > > > Reported-by: kernel test robot <lkp@...el.com>
> > > >
> > > > All warnings (new ones prefixed by >>):
> > > >
> > > > >> drivers/net/ethernet/intel/i40e/i40e_xsk.c:417:13: warning: unknown pragma ignored [-Wunknown-pragmas]
> > > > #pragma GCC unroll 4
> > > > ^
> > > > 1 warning generated.
> > >
> > > And I was hoping that unknown pragmas would be ignored, but that will
> > > obviously not be the case with -Wunknown-pragmas added. The unrolling
> > > of this inner loop where the code spends most of its time gives me
> > > nearly 1 Mpps extra in performance which is substantial, so I would
> > > like to get this unrolled in some way, but without the warning. Need
> > > some advice please. Here are some options that comes in mind:
> > >
> > > #1: Suppress unknown pragma warnings in this file only by adding
> > > CFLAGS_i40e_xsk.o += -Wno-unknown-pragmas (or whatever that option
> > > might be) in the Makefile
> > >
> > > #2: Force the compiler to loop-unroll the loop with for example a
> > > switch statement with four cases that all fall through. This will make
> > > the code less readable.
> > >
> > > #3: Manually loop-unroll the loop. This will make the code even less
> > > readable than #2.
> >
> > #4 support both compilers. Note Clang's syntax is slightly different
> > here; it doesn't accept GCC specific pragmas, and uses a slightly
> > different form:
> > https://clang.llvm.org/docs/LanguageExtensions.html#loop-unrolling .
> > If you wrap that in a macro based on `#ifdef __clang__`, that should
> > do the trick.
>
> Yes, that did the trick. Tried it out with the compiler explorer at
> https://godbolt.org/ and it compiles nicely even for clang-powerpc64.
> Will spin a v3.
>
> Thank you: Magnus
Great job Magnus, I appreciate it!
>
> > >
> > > I prefer #1 as I like to keep the code readable, but you might have
> > > other better suggestions on how to tackle this.
> > >
> > > Thanks: Magnus
> > >
> > > > vim +417 drivers/net/ethernet/intel/i40e/i40e_xsk.c
> > > >
> > > > 408
> > > > 409 static void i40e_xmit_pkt_batch(struct i40e_ring *xdp_ring, struct xdp_desc *desc,
> > > > 410 unsigned int *total_bytes)
> > > > 411 {
> > > > 412 u16 ntu = xdp_ring->next_to_use;
> > > > 413 struct i40e_tx_desc *tx_desc;
> > > > 414 dma_addr_t dma;
> > > > 415 u32 i;
> > > > 416
> > > > > 417 #pragma GCC unroll 4
> > > > 418 for (i = 0; i < PKTS_PER_BATCH; i++) {
> > > > 419 dma = xsk_buff_raw_get_dma(xdp_ring->xsk_pool, desc[i].addr);
> > > > 420 xsk_buff_raw_dma_sync_for_device(xdp_ring->xsk_pool, dma, desc[i].len);
> > > > 421
> > > > 422 tx_desc = I40E_TX_DESC(xdp_ring, ntu++);
> > > > 423 tx_desc->buffer_addr = cpu_to_le64(dma);
> > > > 424 tx_desc->cmd_type_offset_bsz = build_ctob(I40E_TX_DESC_CMD_ICRC |
> > > > 425 I40E_TX_DESC_CMD_EOP,
> > > > 426 0, desc[i].len, 0);
> > > > 427
> > > > 428 *total_bytes += desc[i].len;
> > > > 429 }
> > > > 430
> > > > 431 xdp_ring->next_to_use = ntu;
> > > > 432 }
> > > > 433
> > > >
> > > > ---
> > > > 0-DAY CI Kernel Test Service, Intel Corporation
> > > > https://lists.01.org/hyperkitty/list/kbuild-all@lists.01.org
> > >
> > > --
> > > You received this message because you are subscribed to the Google Groups "Clang Built Linux" group.
> > > To unsubscribe from this group and stop receiving emails from it, send an email to clang-built-linux+unsubscribe@...glegroups.com.
> > > To view this discussion on the web visit https://groups.google.com/d/msgid/clang-built-linux/CAJ8uoz2aDjLPtcTgZ_pO-%3DS9TgXm3c57rN8TTPXdqT7HOOKrhA%40mail.gmail.com.
> >
> >
> >
> > --
> > Thanks,
> > ~Nick Desaulniers
--
Thanks,
~Nick Desaulniers
Powered by blists - more mailing lists