patches for DPDK stable branches
 help / color / mirror / Atom feed
From: Ferruh Yigit <ferruh.yigit@intel.com>
To: Andrew Rybchenko <andrew.rybchenko@oktetlabs.ru>,
	Thomas Monjalon <thomas@monjalon.net>
Cc: <dev@dpdk.org>, Olivier Matz <olivier.matz@6wind.com>,
	Ivan Ilchenko <ivan.ilchenko@oktetlabs.ru>, <stable@dpdk.org>,
	Andy Moreton <amoreton@xilinx.com>,
	Kuba Kozak <kubax.kozak@intel.com>
Subject: Re: [dpdk-stable] [PATCH v5 2/2] ethdev: fix docs of drivers callbacks getting xstats by IDs
Date: Thu, 30 Sep 2021 16:30:10 +0100	[thread overview]
Message-ID: <adb5d7c6-d74d-8337-f2d6-122cfecf2048@intel.com> (raw)
In-Reply-To: <5d9c9565-b3cf-8e29-309f-1aa3a2e91491@oktetlabs.ru>

On 9/30/2021 3:01 PM, Andrew Rybchenko wrote:
> On 9/30/21 3:08 PM, Ferruh Yigit wrote:
>> On 9/29/2021 12:54 PM, Andrew Rybchenko wrote:
>>> On 9/29/21 11:44 AM, Ferruh Yigit wrote:
>>>> On 9/28/2021 5:53 PM, Andrew Rybchenko wrote:
>>>>> On 9/28/21 7:50 PM, Ferruh Yigit wrote:
>>>>>> On 9/28/2021 1:05 PM, Andrew Rybchenko wrote:
>>>>>>> From: Ivan Ilchenko <ivan.ilchenko@oktetlabs.ru>
>>>>>>>
>>>>>>> Update xstats by IDs callbacks documentation in accordance with
>>>>>>> ethdev usage of these callbacks. Document valid combinations of
>>>>>>> input arguments to make driver implementation simpler.
>>>>>>>
>>>>>>> Fixes: 79c913a42f0 ("ethdev: retrieve xstats by ID")
>>>>>>> Cc: stable@dpdk.org
>>>>>>>
>>>>>>> Signed-off-by: Ivan Ilchenko <ivan.ilchenko@oktetlabs.ru>
>>>>>>> Signed-off-by: Andrew Rybchenko <andrew.rybchenko@oktetlabs.ru>
>>>>>>> Reviewed-by: Andy Moreton <amoreton@xilinx.com>
>>>>>>> ---
>>>>>>>  lib/ethdev/ethdev_driver.h | 42 ++++++++++++++++++++++++++++++++++++--
>>>>>>>  1 file changed, 40 insertions(+), 2 deletions(-)
>>>>>>>
>>>>>>> diff --git a/lib/ethdev/ethdev_driver.h b/lib/ethdev/ethdev_driver.h
>>>>>>> index 40e474aa7e..c89eefcc42 100644
>>>>>>> --- a/lib/ethdev/ethdev_driver.h
>>>>>>> +++ b/lib/ethdev/ethdev_driver.h
>>>>>>> @@ -187,11 +187,28 @@ typedef int (*eth_xstats_get_t)(struct rte_eth_dev *dev,
>>>>>>>  	struct rte_eth_xstat *stats, unsigned int n);
>>>>>>>  /**< @internal Get extended stats of an Ethernet device. */
>>>>>>>  
>>>>>>> +/**
>>>>>>> + * @internal
>>>>>>> + * Get extended stats of an Ethernet device.
>>>>>>> + *
>>>>>>> + * @param dev
>>>>>>> + *   ethdev handle of port.
>>>>>>> + * @param ids
>>>>>>> + *   IDs array to retrieve specific statistics. Must not be NULL.
>>>>>>> + * @param values
>>>>>>> + *   A pointer to a table to be filled with device statistics values.
>>>>>>> + *   Must not be NULL.
>>>>>>> + * @param n
>>>>>>> + *   Element count in @p ids and @p values.
>>>>>>> + *
>>>>>>> + * @return
>>>>>>> + *   - A number of filled in stats.
>>>>>>> + *   - A negative value on error.
>>>>>>> + */
>>>>>>>  typedef int (*eth_xstats_get_by_id_t)(struct rte_eth_dev *dev,
>>>>>>>  				      const uint64_t *ids,
>>>>>>>  				      uint64_t *values,
>>>>>>>  				      unsigned int n);
>>>>>>> -/**< @internal Get extended stats of an Ethernet device. */
>>>>>>>  
>>>>>>>  /**
>>>>>>>   * @internal
>>>>>>> @@ -218,10 +235,31 @@ typedef int (*eth_xstats_get_names_t)(struct rte_eth_dev *dev,
>>>>>>>  	struct rte_eth_xstat_name *xstats_names, unsigned int size);
>>>>>>>  /**< @internal Get names of extended stats of an Ethernet device. */
>>>>>>>  
>>>>>>> +/**
>>>>>>> + * @internal
>>>>>>> + * Get names of extended stats of an Ethernet device.
>>>>>>> + * For name count, set @p xstats_names and @p ids to NULL.
>>>>>>
>>>>>> Why limiting this behavior to 'xstats_get_names_by_id'?
>>>>>>
>>>>>> Internally 'xstats_get_names_by_id' is used to get the count, but I think this
>>>>>> is not intentionally selected, just one of the xstats_*_by_id dev_ops used.
>>>>>>
>>>>>> I think it is confusing to have this support for one of the '_by_id' dev_ops but
>>>>>> not for other. Why not require both to support returning 'count'?
>>>>>
>>>>> Simply because it is dead code. There is no point to require
>>>>> from driver to have dead code.
>>>>>
>>>>
>>>> Let me step back a little, both ethdev APIs can be used to return xstats count
>>>> by providing 'values/names' & 'ids' pointers as NULL and 'size' as 0:
>>>> 'rte_eth_xstats_get_names_by_id()'
>>>> 'rte_eth_xstats_get_by_id()'
>>>>
>>>> But internally both APIs use 'xstats_get_names_by_id' dev_ops to get the count,
>>>> as said above I believe this selection is done unintentionally.
>>>>
>>>>
>>>> I am for below two options:
>>>>
>>>> a) Internally use 'xstats_get_names' || 'xstats_get' dev_ops to get the xstats
>>>> count, and doesn't support getting xstats count for both '_by_id' dev_ops, this
>>>> simplifies driver code. As far as I remember I suggested this before, still I
>>>> prefer this one.
>>>>
>>>> b) If we will support getting xstats count from '_by_id' dev_ops, I think both
>>>> should support it, to not make it more complex to figure out which one support
>>>> what. As sample both 'xstats_get_names' and 'xstats_get' supports getting xstats
>>>> count, not just one.
>>>>
>>>
>>> In (b) do you suggest to change ethdev to use xstats_get_by_id
>>> driver op if users asks for a number of xstats using
>>> rte_eth_xstats_get_by_id()? It will complicate ethdev and
>>> will complicate drivers. Just for the symmetry?
>>>
>>
>> Not just for symmetry, but simplify the logic. Both dev_ops are very similar and
> 
> I'm sorry, but could you point of which logic you'd
> like to simply. Less requirements on driver ops
> means less code required inside.
> 
>> they are having different behavior with same parameters is confusing.
> 
> Ah, logic from PMD maintainer point of view. Does it
> really worse to require extra code inside because of it?
> 
>> If you want to simplify the drivers why not go with option (a)? Option (a) also
>> doesn't change external API, and has only minor change in the ethdev layer.
> 
> Is two extra patches in v6 (discussed below) a step
> towards (a)?
> 

I am not sure of it, for (a) ethdev layer will handle getting all stats or
getting count using 'xstats_get_names' || 'xstats_get', and '*_by_id()' dev_ops
will become simpler since they won't accept NULL 'ids' & 'values'.

Also when dev_ops merged, it forces PMDs to implement the getting by ids, but
now it is covered by ethdev layer for the PMDs that doesn't implement '_by_id()'.

>>> The patch does not change external API, does not change etcdev
>>> bahaviour. It just clarify requirements on driver ops to
>>> allow to check less in all drivers and support less cases
>>> in all drivers.
>>>
>>
>> It is not clarifying the requirements of dev_ops, but changing it. Previously
>> there was nothing saying only one of the '_by_id' dev_ops should support
>> returning element count.
> 
> I say "clarifying" since I adjust it a real source of
> truth for internals - implementation, usage of these
> callback by ethdev layer.
> 
> Yes, I agree that it changes documentation.
> 
>>
>>> If we make a one more step back, frankly speaking I think we
>>> have too many functions to retrieve statistics. I can
>>> understand from ethdev API point of view taking API stability
>>> into account etc. But why do we have so many complicated
>>> driver callbacks?
>>>> First of all I'd like to do one more cleanup:
>>> change eth_xstats_get_names_by_id_t prototype to
>>> have ids before xstats_names.
>>> I.e. eth_xstats_get_by_id_t(dev, ids, values, size)
>>> eth_xstats_get_names_by_id_t(dev, ids, names, size)
>>>
>>
>> +1
> 
> See patch 3/4 in v6
> 
>>
>>> Second, merge eth_xstats_get_names_t and eth_xstats_get_names_by_id_t
>>> callbacks to keep
>>> name of the first, but prototype from the second.
>>> The reason is equal functionality if ids==NULL,
>>> shorter name of the first and optional ids (i.e. no
>>> reason to mention optional parameter in name).
>>> Drivers which do not implement by_id_t,
>>> but implement eth_xstats_get_names_t, will simply
>>> return ENOTSUP if ids!=NULL.
>>>
>>
>> No objection, _by_id() version is already superset of the other.
> 
> See patch 4/4 in v6
> 
>>
>>> The case of values ops is more complicated,
>>> however since:
>>>
>>> 2834  * There is an assumption that 'xstat_names' and 'xstats' arrays
>>> are matched
>>> 2835  * by array index:
>>> 2836  *  xstats_names[i].name => xstats[i].value
>>> 2837  *
>>> 2838  * And the array index is same with id field of 'struct rte_eth_xstat':
>>> 2839  *  xstats[i].id == i
>>> 2840  *
>>> 2841  * This assumption makes key-value pair matching less flexible but
>>> simpler.
>>>
>>> we can switch to eth_xstats_get_by_id_t only callback as
>>> well and fill in stats->id equal to index on ethdev layer.
>>
>> When ids != NULL, the index from 'ids' can be used, isn't it.
> 
> Yes, of course, but above documentation is for API without IDs.
> 
>>
>>> However, it will require extra buffer for
>>> uint64_t *values and copying in the rte_eth_xstats_get()
>>> implementation. So, I doubt in this case.
>>>
>>
>> Overall merging xstats _by_id APIs doesn't look bad idea, but since it will
>> require change in applications, I am not really sure if benefit worth the
>> trouble it brings to users.
> 
> Yes, I agree. I'm not going to change external API.
> 
>>> In fact, it is sad that we still do not move forward
>>> in accordance with Thomas presentation made 2 years ago.
>>>
>>
>> In my experience things don't move forward without proper plan (who, what, when).
>>
> 
> +1
> 


  reply	other threads:[~2021-09-30 15:30 UTC|newest]

Thread overview: 64+ messages / expand[flat|nested]  mbox.gz  Atom feed  top
     [not found] <20210604144225.287678-1-andrew.rybchenko@oktetlabs.ru>
2021-06-04 14:42 ` [dpdk-stable] [PATCH 01/11] net/sfc: fix get xstats by ID callback to use MAC stats lock Andrew Rybchenko
2021-06-04 14:42 ` [dpdk-stable] [PATCH 02/11] net/sfc: fix reading adapter state without locking Andrew Rybchenko
2021-06-04 14:42 ` [dpdk-stable] [PATCH 03/11] ethdev: fix docs of functions getting xstats by IDs Andrew Rybchenko
2021-07-20 16:25   ` Ferruh Yigit
2021-07-22  9:12     ` Andrew Rybchenko
2021-07-23 14:19       ` Ferruh Yigit
2021-07-24 12:06         ` Andrew Rybchenko
2021-06-04 14:42 ` [dpdk-stable] [PATCH 04/11] ethdev: fix docs of drivers callbacks " Andrew Rybchenko
2021-07-20 16:51   ` Ferruh Yigit
2021-07-22  9:33     ` Andrew Rybchenko
2021-07-23 14:31       ` Ferruh Yigit
2021-07-23 18:47         ` Andrew Rybchenko
2021-06-04 14:42 ` [dpdk-stable] [PATCH 05/11] net/sfc: fix xstats by ID callbacks according to ethdev Andrew Rybchenko
2021-06-04 14:42 ` [dpdk-stable] [PATCH 06/11] net/sfc: fix accessing xstats by an unsorted list of IDs Andrew Rybchenko
2021-06-04 14:42 ` [dpdk-stable] [PATCH 07/11] net/sfc: fix MAC stats update to work for stopped device Andrew Rybchenko
     [not found] ` <20210722095433.1898589-1-andrew.rybchenko@oktetlabs.ru>
2021-07-22  9:54   ` [dpdk-stable] [PATCH v2 01/11] net/sfc: fix get xstats by ID callback to use MAC stats lock Andrew Rybchenko
2021-07-22  9:54   ` [dpdk-stable] [PATCH v2 02/11] net/sfc: fix reading adapter state without locking Andrew Rybchenko
2021-07-22  9:54   ` [dpdk-stable] [PATCH v2 03/11] ethdev: fix docs of functions getting xstats by IDs Andrew Rybchenko
2021-07-22  9:54   ` [dpdk-stable] [PATCH v2 04/11] ethdev: fix docs of drivers callbacks " Andrew Rybchenko
2021-07-22  9:54   ` [dpdk-stable] [PATCH v2 05/11] net/sfc: fix xstats by ID callbacks according to ethdev Andrew Rybchenko
2021-07-22  9:54   ` [dpdk-stable] [PATCH v2 06/11] net/sfc: fix accessing xstats by an unsorted list of IDs Andrew Rybchenko
2021-07-22  9:54   ` [dpdk-stable] [PATCH v2 07/11] net/sfc: fix MAC stats update to work for stopped device Andrew Rybchenko
     [not found] ` <20210723131515.2317168-1-andrew.rybchenko@oktetlabs.ru>
2021-07-23 13:15   ` [dpdk-stable] [PATCH v3 01/11] net/sfc: fix get xstats by ID callback to use MAC stats lock Andrew Rybchenko
2021-07-23 13:15   ` [dpdk-stable] [PATCH v3 02/11] net/sfc: fix reading adapter state without locking Andrew Rybchenko
2021-07-23 13:15   ` [dpdk-stable] [PATCH v3 03/11] ethdev: fix docs of functions getting xstats by IDs Andrew Rybchenko
2021-07-23 14:42     ` Ferruh Yigit
2021-07-24 12:07       ` Andrew Rybchenko
2021-07-23 13:15   ` [dpdk-stable] [PATCH v3 04/11] ethdev: fix docs of drivers callbacks " Andrew Rybchenko
2021-07-23 14:46     ` Ferruh Yigit
2021-07-23 13:15   ` [dpdk-stable] [PATCH v3 05/11] net/sfc: fix xstats by ID callbacks according to ethdev Andrew Rybchenko
2021-07-23 13:15   ` [dpdk-stable] [PATCH v3 06/11] net/sfc: fix accessing xstats by an unsorted list of IDs Andrew Rybchenko
2021-07-23 13:15   ` [dpdk-stable] [PATCH v3 07/11] net/sfc: fix MAC stats update to work for stopped device Andrew Rybchenko
2021-07-24 12:33 ` [dpdk-stable] [PATCH v4 1/2] ethdev: fix docs of functions getting xstats by IDs Andrew Rybchenko
2021-07-24 12:33   ` [dpdk-stable] [PATCH v4 2/2] ethdev: fix docs of drivers callbacks " Andrew Rybchenko
2021-07-26 10:13     ` [dpdk-stable] [dpdk-dev] " Olivier Matz
2021-09-28 12:04       ` Andrew Rybchenko
2021-07-26 10:13   ` [dpdk-stable] [dpdk-dev] [PATCH v4 1/2] ethdev: fix docs of functions " Olivier Matz
2021-09-28 12:01     ` Andrew Rybchenko
2021-09-28 12:05 ` [dpdk-stable] [PATCH v5 " Andrew Rybchenko
2021-09-28 12:05   ` [dpdk-stable] [PATCH v5 2/2] ethdev: fix docs of drivers callbacks " Andrew Rybchenko
2021-09-28 16:50     ` Ferruh Yigit
2021-09-28 16:53       ` Andrew Rybchenko
2021-09-29  8:44         ` Ferruh Yigit
2021-09-29 11:54           ` Andrew Rybchenko
2021-09-30 12:08             ` Ferruh Yigit
2021-09-30 14:01               ` Andrew Rybchenko
2021-09-30 15:30                 ` Ferruh Yigit [this message]
2021-09-30 16:01                   ` Andrew Rybchenko
2021-09-28 16:46   ` [dpdk-stable] [PATCH v5 1/2] ethdev: fix docs of functions " Ferruh Yigit
2021-09-30 14:04 ` [dpdk-stable] [PATCH v6 1/4] " Andrew Rybchenko
2021-09-30 14:04   ` [dpdk-stable] [PATCH v6 2/4] ethdev: fix docs of drivers callbacks " Andrew Rybchenko
     [not found] ` <20210930160156.961041-1-andrew.rybchenko@oktetlabs.ru>
2021-09-30 16:01   ` [dpdk-stable] [PATCH v7 1/5] ethdev: fix docs of functions " Andrew Rybchenko
2021-09-30 16:01   ` [dpdk-stable] [PATCH v7 2/4] ethdev: fix docs of drivers callbacks " Andrew Rybchenko
2021-09-30 16:01   ` [dpdk-stable] [PATCH v7 3/5] " Andrew Rybchenko
2021-09-30 16:05 ` [dpdk-stable] [PATCH v8 1/5] ethdev: fix docs of functions " Andrew Rybchenko
2021-09-30 16:05   ` [dpdk-stable] [PATCH v8 3/5] ethdev: fix docs of drivers callbacks " Andrew Rybchenko
2021-09-30 16:33     ` Ferruh Yigit
2021-10-01  9:07       ` Andrew Rybchenko
2021-10-01  9:07 ` [dpdk-stable] [PATCH v9 1/5] ethdev: fix docs of functions " Andrew Rybchenko
2021-10-01  9:07   ` [dpdk-stable] [PATCH v9 3/5] ethdev: fix docs of drivers callbacks " Andrew Rybchenko
2021-10-01  9:42     ` Ferruh Yigit
2021-10-01 10:22       ` Andrew Rybchenko
2021-10-06 10:37         ` Ferruh Yigit
2021-10-06 11:08   ` [dpdk-stable] [PATCH v9 1/5] ethdev: fix docs of functions " Ferruh Yigit

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=adb5d7c6-d74d-8337-f2d6-122cfecf2048@intel.com \
    --to=ferruh.yigit@intel.com \
    --cc=amoreton@xilinx.com \
    --cc=andrew.rybchenko@oktetlabs.ru \
    --cc=dev@dpdk.org \
    --cc=ivan.ilchenko@oktetlabs.ru \
    --cc=kubax.kozak@intel.com \
    --cc=olivier.matz@6wind.com \
    --cc=stable@dpdk.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
Be sure your reply has a Subject: header at the top and a blank line before the message body.
This is a public inbox, see mirroring instructions
for how to clone and mirror all data and code used for this inbox;
as well as URLs for NNTP newsgroup(s).