Re: [PATCH bpf-next v2 5/5] i40e: use batched xsk Tx interfaces to increase performance

[kernel test robot]

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)
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 <l...@intel.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.

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/kbuil...@lists.01.org

[Magnus Karlsson]

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.

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

[Nick Desaulniers]

On Wed, Nov 11, 2020 at 3:57 AM Magnus Karlsson
<magnus....@gmail.com> wrote:
>
> On Wed, Nov 11, 2020 at 2:38 AM kernel test robot <l...@intel.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

#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.

> --
> 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-li...@googlegroups.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

[Magnus Karlsson]

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

[Nick Desaulniers]

On Wed, Nov 11, 2020 at 11:45 PM Magnus Karlsson

Great job Magnus, I appreciate it!

--
Thanks,
~Nick Desaulniers