DPDK patches and discussions
 help / color / mirror / Atom feed
From: David Marchand <david.marchand@redhat.com>
To: Feifei Wang <feifei.wang2@arm.com>
Cc: Ruifeng Wang <ruifeng.wang@arm.com>, dev <dev@dpdk.org>,
	nd <nd@arm.com>,  Jerin Jacob <jerinjacobk@gmail.com>,
	Stephen Hemminger <stephen@networkplumber.org>,
	 Thomas Monjalon <thomas@monjalon.net>,
	Mattias Rönnblom <mattias.ronnblom@ericsson.com>,
	 "Ananyev, Konstantin" <konstantin.ananyev@intel.com>
Subject: Re: [dpdk-dev] [PATCH v8 1/5] eal: add new definitions for wait scheme
Date: Sun, 31 Oct 2021 09:38:31 +0100
Message-ID: <CAJFAV8wJRZDLDKbwN-T-8bpUaQ3Z=VoftV6Dgm8sa0c5AYi_Zg@mail.gmail.com> (raw)
In-Reply-To: <20211029082021.945586-2-feifei.wang2@arm.com>

On Fri, Oct 29, 2021 at 10:20 AM Feifei Wang <feifei.wang2@arm.com> wrote:
>
> Introduce macros as generic interface for address monitoring.

The main point of this patch is to add a new generic helper.


>
> Add '__LOAD_EXC_128' for size of 128. For different size, encapsulate
> '__LOAD_EXC_16', '__LOAD_EXC_32', '__LOAD_EXC_64' and '__LOAD_EXC_128'
> into a new macro '__LOAD_EXC'.

ARM macros are just a result of introducing this new helper as a macro.
I would not mention them.


>
> Furthermore, to prevent compilation warning in arm:
> ----------------------------------------------
> 'warning: implicit declaration of function ...'
> ----------------------------------------------
> Delete 'undef' constructions for '__LOAD_EXC_xx', '__SEVL' and '__WFE'.
> And add ‘__RTE_ARM’ for these macros to fix the namespace.
> This is because original macros are undefine at the end of the file.
> If new macro 'rte_wait_event' calls them in other files, they will be
> seen as 'not defined'.


About this new helper, it's rather confusing:
- it is a macro, should be in capital letters,
- "rte_wait_event(addr, mask, cond, expected)" waits until "*addr &
mask cond expected" becomes false. I find this confusing. I would
invert the condition.
- so far, we had rte_wait_until_* helpers, rte_wait_event seems like a
step backward as it seems to talk about the ARM stuff (wfe),
- the masking part is artificial in some cases, at least let's avoid
using a too generic name, we can decide to add a non-masked helper
later.

For those reasons, I'd prefer we have something like:

/*
 * Wait until *addr & mask makes the condition true. With a relaxed memory
 * ordering model, the loads around this helper can be reordered.
 *
 * @param addr
 *  A pointer to the memory location.
 * @param mask
 *  A mask of *addr bits in interest.
 * @param cond
 *  A symbol representing the condition.
 * @param expected
 *  An expected value to be in the memory location.
 * @param memorder
 *  Two different memory orders that can be specified:
 *  __ATOMIC_ACQUIRE and __ATOMIC_RELAXED. These map to
 *  C++11 memory orders with the same names, see the C++11 standard or
 *  the GCC wiki on atomic synchronization for detailed definition.
 */
#define RTE_WAIT_UNTIL_MASKED(addr, mask, cond, expected, memorder)
           \
do {
           \
        RTE_BUILD_BUG_ON(memorder != __ATOMIC_ACQUIRE &&
           \
                memorder != __ATOMIC_RELAXED);
           \
        typeof(*(addr)) expected_value = expected;
           \
        while (!((__atomic_load_n(addr, memorder) & (mask)) cond
expected_value)) \
                rte_pause();
           \
} while (0)


Comments below.

>
> Signed-off-by: Feifei Wang <feifei.wang2@arm.com>
> Reviewed-by: Ruifeng Wang <ruifeng.wang@arm.com>
> Acked-by: Konstantin Ananyev <konstantin.ananyev@intel.com>
> ---
>  lib/eal/arm/include/rte_pause_64.h  | 202 +++++++++++++++++-----------
>  lib/eal/include/generic/rte_pause.h |  28 ++++
>  2 files changed, 154 insertions(+), 76 deletions(-)
>
> diff --git a/lib/eal/arm/include/rte_pause_64.h b/lib/eal/arm/include/rte_pause_64.h
> index e87d10b8cc..783c6aae87 100644
> --- a/lib/eal/arm/include/rte_pause_64.h
> +++ b/lib/eal/arm/include/rte_pause_64.h

[snip]

> +/*
> + * Atomic exclusive load from addr, it returns the 64-bit content of
> + * *addr while making it 'monitored', when it is written by someone
> + * else, the 'monitored' state is cleared and an event is generated
> + * implicitly to exit WFE.
> + */
> +#define __RTE_ARM_LOAD_EXC_64(src, dst, memorder) {      \
> +       if (memorder == __ATOMIC_RELAXED) {              \
> +               asm volatile("ldxr %x[tmp], [%x[addr]]"  \
> +                       : [tmp] "=&r" (dst)              \
> +                       : [addr] "r" (src)               \
> +                       : "memory");                     \
> +       } else {                                         \
> +               asm volatile("ldaxr %x[tmp], [%x[addr]]" \
> +                       : [tmp] "=&r" (dst)              \
> +                       : [addr] "r" (src)               \
> +                       : "memory");                     \
> +       } }
> +
> +/*
> + * Atomic exclusive load from addr, it returns the 128-bit content of
> + * *addr while making it 'monitored', when it is written by someone
> + * else, the 'monitored' state is cleared and an event is generated
> + * implicitly to exit WFE.
> + */
> +#define __RTE_ARM_LOAD_EXC_128(src, dst, memorder) {                    \
> +       volatile rte_int128_t *dst_128 = (volatile rte_int128_t *)&dst; \

dst needs some () protection => &(dst)
Is volatile necessary?


> +       if (memorder == __ATOMIC_RELAXED) {                             \
> +               asm volatile("ldxp %x[tmp0], %x[tmp1], [%x[addr]]"      \
> +                       : [tmp0] "=&r" (dst_128->val[0]),               \
> +                         [tmp1] "=&r" (dst_128->val[1])                \
> +                       : [addr] "r" (src)                              \
> +                       : "memory");                                    \
> +       } else {                                                        \
> +               asm volatile("ldaxp %x[tmp0], %x[tmp1], [%x[addr]]"     \
> +                       : [tmp0] "=&r" (dst_128->val[0]),               \
> +                         [tmp1] "=&r" (dst_128->val[1])                \
> +                       : [addr] "r" (src)                              \
> +                       : "memory");                                    \
> +       } }                                                             \
> +
> +#define __RTE_ARM_LOAD_EXC(src, dst, memorder, size) {          \
> +       RTE_BUILD_BUG_ON(size != 16 && size != 32 && size != 64 \
> +                               && size != 128);                \

Indent should be one tab (idem in other places of this patch).
Double tab is when we have line continuation in tests.


> +       if (size == 16)                                         \
> +               __RTE_ARM_LOAD_EXC_16(src, dst, memorder)       \
> +       else if (size == 32)                                    \
> +               __RTE_ARM_LOAD_EXC_32(src, dst, memorder)       \
> +       else if (size == 64)                                    \
> +               __RTE_ARM_LOAD_EXC_64(src, dst, memorder)       \
> +       else if (size == 128)                                   \
> +               __RTE_ARM_LOAD_EXC_128(src, dst, memorder)      \
> +}
> +

[snip]

> -#undef __LOAD_EXC_64
>
> -#undef __SEVL
> -#undef __WFE
> +#define rte_wait_event(addr, mask, cond, expected, memorder)              \
> +do {                                                                      \
> +       RTE_BUILD_BUG_ON(!__builtin_constant_p(memorder));                \

Is this check on memorder being constant necessary?
We have a build bug on, right after, would it not catch non constant cases?


> +       RTE_BUILD_BUG_ON(memorder != __ATOMIC_ACQUIRE &&                  \
> +                               memorder != __ATOMIC_RELAXED);            \
> +       const uint32_t size = sizeof(*(addr)) << 3;                       \
> +       typeof(*(addr)) expected_value = (expected);                      \

No need for () around expected.


> +       typeof(*(addr)) value;                                            \
> +       __RTE_ARM_LOAD_EXC((addr), value, memorder, size)                 \

No need for () around addr.


> +       if ((value & (mask)) cond expected_value) {                       \
> +               __RTE_ARM_SEVL()                                          \
> +               do {                                                      \
> +                       __RTE_ARM_WFE()                                   \
> +                       __RTE_ARM_LOAD_EXC((addr), value, memorder, size) \

Idem.


> +               } while ((value & (mask)) cond expected_value);           \
> +       }                                                                 \
> +} while (0)
>
>  #endif
>
> diff --git a/lib/eal/include/generic/rte_pause.h b/lib/eal/include/generic/rte_pause.h
> index 668ee4a184..d0c5b5a415 100644
> --- a/lib/eal/include/generic/rte_pause.h
> +++ b/lib/eal/include/generic/rte_pause.h
> @@ -111,6 +111,34 @@ rte_wait_until_equal_64(volatile uint64_t *addr, uint64_t expected,
>         while (__atomic_load_n(addr, memorder) != expected)
>                 rte_pause();
>  }

With this patch, ARM header goes though a conversion of assert() to
compilation checks (build bug on).
I don't see a reason not to do the same in generic header.

As a result of this conversion, #include <assert.h> then can be removed.
Though it triggers build failure on following files (afaics) who were
implictly relying on this inclusion:
drivers/net/ark/ark_ddm.c
drivers/net/ark/ark_udm.c
drivers/net/ice/ice_fdir_filter.c
drivers/net/ionic/ionic_rxtx.c
drivers/net/mlx4/mlx4_txq.c


-- 
David Marchand


  parent reply	other threads:[~2021-10-31  8:38 UTC|newest]

Thread overview: 113+ messages / expand[flat|nested]  mbox.gz  Atom feed  top
2021-09-02  5:32 [dpdk-dev] [RFC PATCH v1 0/5] add new API for wait until scheme Feifei Wang
2021-09-02  5:32 ` [dpdk-dev] [RFC PATCH v1 1/5] eal: " Feifei Wang
2021-09-02  5:32 ` [dpdk-dev] [RFC PATCH v1 2/5] eal: use wait until scheme for read pflock Feifei Wang
2021-09-02  5:32 ` [dpdk-dev] [RFC PATCH v1 3/5] eal: use wait until scheme for mcslock Feifei Wang
2021-09-02  5:32 ` [dpdk-dev] [RFC PATCH v1 4/5] lib/bpf: use wait until scheme for Rx/Tx iteration Feifei Wang
2021-09-02  5:32 ` [dpdk-dev] [RFC PATCH v1 5/5] lib/distributor: use wait until scheme Feifei Wang
2021-09-02 15:22 ` [dpdk-dev] [RFC PATCH v1 0/5] add new API for " Stephen Hemminger
2021-09-03  7:02   ` [dpdk-dev] 回复: " Feifei Wang
2021-09-23  9:58 ` [dpdk-dev] [RFC PATCH v2 0/5] add new definitions for wait scheme Feifei Wang
2021-09-23  9:58   ` [dpdk-dev] [RFC PATCH v2 1/5] eal: " Feifei Wang
2021-09-23  9:58   ` [dpdk-dev] [RFC PATCH v2 2/5] eal: use wait event for read pflock Feifei Wang
2021-09-23  9:59   ` [dpdk-dev] [RFC PATCH v2 3/5] eal: use wait event scheme for mcslock Feifei Wang
2021-09-23  9:59   ` [dpdk-dev] [RFC PATCH v2 4/5] lib/bpf: use wait event scheme for Rx/Tx iteration Feifei Wang
2021-09-24 18:07     ` Ananyev, Konstantin
2021-09-26  2:19       ` [dpdk-dev] 回复: " Feifei Wang
2021-09-23  9:59   ` [dpdk-dev] [RFC PATCH v2 5/5] lib/distributor: use wait event scheme Feifei Wang
2021-09-26  6:32 ` [dpdk-dev] [RFC PATCH v3 0/5] add new definitions for wait scheme Feifei Wang
2021-09-26  6:32   ` [dpdk-dev] [RFC PATCH v3 1/5] eal: " Feifei Wang
2021-10-07 16:18     ` Ananyev, Konstantin
2021-10-12  8:09       ` [dpdk-dev] 回复: " Feifei Wang
2021-10-13 15:03         ` [dpdk-dev] " Ananyev, Konstantin
2021-10-13 17:00           ` Stephen Hemminger
2021-10-14  3:14             ` [dpdk-dev] 回复: " Feifei Wang
2021-10-14  3:08           ` Feifei Wang
2021-09-26  6:32   ` [dpdk-dev] [RFC PATCH v3 2/5] eal: use wait event for read pflock Feifei Wang
2021-09-26  6:33   ` [dpdk-dev] [RFC PATCH v3 3/5] eal: use wait event scheme for mcslock Feifei Wang
2021-09-26  6:33   ` [dpdk-dev] [RFC PATCH v3 4/5] lib/bpf: use wait event scheme for Rx/Tx iteration Feifei Wang
2021-10-07 15:50     ` Ananyev, Konstantin
2021-10-07 17:40       ` Ananyev, Konstantin
2021-10-20  6:20         ` [dpdk-dev] 回复: " Feifei Wang
2021-09-26  6:33   ` [dpdk-dev] [RFC PATCH v3 5/5] lib/distributor: use wait event scheme Feifei Wang
2021-10-20  8:45   ` [dpdk-dev] [PATCH v4 0/5] add new definitions for wait scheme Feifei Wang
2021-10-20  8:45     ` [dpdk-dev] [PATCH v4 1/5] eal: " Feifei Wang
2021-10-21 16:24       ` Ananyev, Konstantin
2021-10-25  9:20         ` [dpdk-dev] 回复: " Feifei Wang
2021-10-25 14:28           ` [dpdk-dev] " Ananyev, Konstantin
2021-10-26  1:08             ` [dpdk-dev] 回复: " Feifei Wang
2021-10-22  0:10       ` [dpdk-dev] " Jerin Jacob
2021-10-25  9:30         ` [dpdk-dev] 回复: " Feifei Wang
2021-10-25  9:43           ` [dpdk-dev] " Jerin Jacob
2021-10-26  1:11             ` [dpdk-dev] 回复: " Feifei Wang
2021-10-20  8:45     ` [dpdk-dev] [PATCH v4 2/5] eal: use wait event for read pflock Feifei Wang
2021-10-20  8:45     ` [dpdk-dev] [PATCH v4 3/5] eal: use wait event scheme for mcslock Feifei Wang
2021-10-20  8:45     ` [dpdk-dev] [PATCH v4 4/5] lib/bpf: use wait event scheme for Rx/Tx iteration Feifei Wang
2021-10-20  8:45     ` [dpdk-dev] [PATCH v4 5/5] lib/distributor: use wait event scheme Feifei Wang
2021-10-26  8:01 ` [dpdk-dev] [PATCH v5 0/5] add new definitions for wait scheme Feifei Wang
2021-10-26  8:02   ` [dpdk-dev] [PATCH v5 1/5] eal: " Feifei Wang
2021-10-26  8:08     ` [dpdk-dev] 回复: " Feifei Wang
2021-10-26  9:46       ` [dpdk-dev] " Ananyev, Konstantin
2021-10-26  9:59         ` Ananyev, Konstantin
2021-10-27  6:56           ` [dpdk-dev] 回复: " Feifei Wang
2021-10-26  8:02   ` [dpdk-dev] [PATCH v5 2/5] eal: use wait event for read pflock Feifei Wang
2021-10-26  8:02   ` [dpdk-dev] [PATCH v5 3/5] eal: use wait event scheme for mcslock Feifei Wang
2021-10-26  8:02   ` [dpdk-dev] [PATCH v5 4/5] lib/bpf: use wait event scheme for Rx/Tx iteration Feifei Wang
2021-10-26  8:18     ` [dpdk-dev] 回复: " Feifei Wang
2021-10-26  9:43       ` [dpdk-dev] " Ananyev, Konstantin
2021-10-26 12:56         ` Ananyev, Konstantin
2021-10-27  7:04           ` [dpdk-dev] 回复: " Feifei Wang
2021-10-27  7:31             ` Feifei Wang
2021-10-27 14:47             ` [dpdk-dev] " Ananyev, Konstantin
2021-10-28  6:24               ` [dpdk-dev] 回复: " Feifei Wang
2021-10-26  8:02   ` [dpdk-dev] [PATCH v5 5/5] lib/distributor: use wait event scheme Feifei Wang
2021-10-27  8:10 ` [dpdk-dev] [PATCH v6 0/4] add new definitions for wait scheme Feifei Wang
2021-10-27  8:10   ` [dpdk-dev] [PATCH v6 1/4] eal: " Feifei Wang
2021-10-27  8:10   ` [dpdk-dev] [PATCH v6 2/4] eal: use wait event for read pflock Feifei Wang
2021-10-27  8:10   ` [dpdk-dev] [PATCH v6 3/4] eal: use wait event scheme for mcslock Feifei Wang
2021-10-27 11:16     ` Mattias Rönnblom
2021-10-28  6:32       ` [dpdk-dev] 回复: " Feifei Wang
2021-10-27  8:10   ` [dpdk-dev] [PATCH v6 4/4] lib/distributor: use wait event scheme Feifei Wang
2021-10-27 10:57   ` [dpdk-dev] [PATCH v6 0/4] add new definitions for wait scheme Jerin Jacob
2021-10-28  6:33     ` [dpdk-dev] 回复: " Feifei Wang
2021-10-28  6:56 ` [dpdk-dev] [PATCH v7 0/5] " Feifei Wang
2021-10-28  6:56   ` [dpdk-dev] [PATCH v7 1/5] eal: " Feifei Wang
2021-10-28  7:15     ` Jerin Jacob
2021-10-28  7:40       ` [dpdk-dev] 回复: " Feifei Wang
2021-10-28  7:51         ` [dpdk-dev] " Jerin Jacob
2021-10-28  9:27           ` [dpdk-dev] 回复: " Feifei Wang
2021-10-28 13:14     ` [dpdk-dev] " Ananyev, Konstantin
2021-10-28  6:56   ` [dpdk-dev] [PATCH v7 2/5] eal: use wait event for read pflock Feifei Wang
2021-10-28  6:56   ` [dpdk-dev] [PATCH v7 3/5] eal: use wait event scheme for mcslock Feifei Wang
2021-10-28  7:02     ` Jerin Jacob
2021-10-28  7:14       ` [dpdk-dev] 回复: " Feifei Wang
2021-10-28  6:56   ` [dpdk-dev] [PATCH v7 4/5] lib/bpf: use wait event scheme for Rx/Tx iteration Feifei Wang
2021-10-28 13:15     ` Ananyev, Konstantin
2021-10-28  6:56   ` [dpdk-dev] [PATCH v7 5/5] lib/distributor: use wait event scheme Feifei Wang
2021-10-29  8:20 ` [dpdk-dev] [PATCH v8 0/5] add new definitions for wait scheme Feifei Wang
2021-10-29  8:20   ` [dpdk-dev] [PATCH v8 1/5] eal: " Feifei Wang
2021-10-29 13:54     ` Jerin Jacob
2021-10-31  8:38     ` David Marchand [this message]
2021-11-01  2:29       ` [dpdk-dev] 回复: " Feifei Wang
2021-10-29  8:20   ` [dpdk-dev] [PATCH v8 2/5] eal: use wait event for read pflock Feifei Wang
2021-10-29 13:55     ` Jerin Jacob
2021-10-31  8:37     ` David Marchand
2021-10-29  8:20   ` [dpdk-dev] [PATCH v8 3/5] eal: use wait event scheme for mcslock Feifei Wang
2021-10-29 13:55     ` Jerin Jacob
2021-10-31  8:37     ` David Marchand
2021-10-29  8:20   ` [dpdk-dev] [PATCH v8 4/5] lib/bpf: use wait event scheme for Rx/Tx iteration Feifei Wang
2021-10-31  8:37     ` David Marchand
2021-10-29  8:20   ` [dpdk-dev] [PATCH v8 5/5] lib/distributor: use wait event scheme Feifei Wang
2021-10-29 13:58     ` Jerin Jacob
2021-10-31  8:38       ` David Marchand
2021-11-01 12:44       ` David Hunt
2021-11-01  6:00 ` [dpdk-dev] [PATCH v9 0/5] add new helper for wait scheme Feifei Wang
2021-11-01  6:00   ` [dpdk-dev] [PATCH v9 1/5] eal: add a new generic " Feifei Wang
2021-11-01  6:00   ` [dpdk-dev] [PATCH v9 2/5] pflock: use wait until scheme for read pflock Feifei Wang
2021-11-03 14:46     ` David Marchand
2021-11-04  1:24       ` [dpdk-dev] 回复: " Feifei Wang
2021-11-01  6:00   ` [dpdk-dev] [PATCH v9 3/5] mcslock: use wait until scheme for mcslock Feifei Wang
2021-11-01  6:00   ` [dpdk-dev] [PATCH v9 4/5] bpf: use wait until scheme for Rx/Tx iteration Feifei Wang
2021-11-01  6:00   ` [dpdk-dev] [PATCH v9 5/5] distributor: use wait until scheme Feifei Wang
2021-11-01 16:05     ` Pattan, Reshma
2021-11-02  2:00       ` [dpdk-dev] 回复: " Feifei Wang
2021-11-03 14:55   ` [dpdk-dev] [PATCH v9 0/5] add new helper for wait scheme David Marchand

Reply instructions:

You may reply publicly to this message via plain-text email
using any one of the following methods:

* Save the following mbox file, import it into your mail client,
  and reply-to-all from there: mbox

  Avoid top-posting and favor interleaved quoting:
  https://en.wikipedia.org/wiki/Posting_style#Interleaved_style

* Reply using the --to, --cc, and --in-reply-to
  switches of git-send-email(1):

  git send-email \
    --in-reply-to='CAJFAV8wJRZDLDKbwN-T-8bpUaQ3Z=VoftV6Dgm8sa0c5AYi_Zg@mail.gmail.com' \
    --to=david.marchand@redhat.com \
    --cc=dev@dpdk.org \
    --cc=feifei.wang2@arm.com \
    --cc=jerinjacobk@gmail.com \
    --cc=konstantin.ananyev@intel.com \
    --cc=mattias.ronnblom@ericsson.com \
    --cc=nd@arm.com \
    --cc=ruifeng.wang@arm.com \
    --cc=stephen@networkplumber.org \
    --cc=thomas@monjalon.net \
    /path/to/YOUR_REPLY

  https://kernel.org/pub/software/scm/git/docs/git-send-email.html

* If your mail client supports setting the In-Reply-To header
  via mailto: links, try the mailto: link

DPDK patches and discussions

This inbox may be cloned and mirrored by anyone:

	git clone --mirror http://inbox.dpdk.org/dev/0 dev/git/0.git

	# If you have public-inbox 1.1+ installed, you may
	# initialize and index your mirror using the following commands:
	public-inbox-init -V2 dev dev/ http://inbox.dpdk.org/dev \
		dev@dpdk.org
	public-inbox-index dev

Example config snippet for mirrors.
Newsgroup available over NNTP:
	nntp://inbox.dpdk.org/inbox.dpdk.dev


AGPL code for this site: git clone https://public-inbox.org/public-inbox.git