From: Jianbo Liu <jianbo.liu@linaro.org>
To: Shreyansh Jain <shreyansh.jain@nxp.com>
Cc: David Marchand <david.marchand@6wind.com>,
"dev@dpdk.org" <dev@dpdk.org>,
Thomas Monjalon <thomas.monjalon@6wind.com>
Subject: Re: [dpdk-dev] [PATCH 00/13] Introducing EAL Bus-Device-Driver Model
Date: Mon, 12 Dec 2016 22:35:43 +0800 [thread overview]
Message-ID: <CAP4Qi38CYGC612kQJpA49xWRF=jpqJ1COWg=Qe18HDzmiLVSuA@mail.gmail.com> (raw)
In-Reply-To: <1697fe66-962d-0848-5e68-615249b52dad@nxp.com>
Hi Shreyansh,
On 7 December 2016 at 21:10, Shreyansh Jain <shreyansh.jain@nxp.com> wrote:
> On Wednesday 07 December 2016 05:47 PM, David Marchand wrote:
>>
>> Hello Shreyansh,
>>
>> On Wed, Dec 7, 2016 at 10:55 AM, Shreyansh Jain <shreyansh.jain@nxp.com>
>> wrote:
>>>
>>> On Wednesday 07 December 2016 02:22 AM, David Marchand wrote:
>>>>>
>>>>> 0002~0003: Introducing the basic Bus model and associated test case
>>>>> 0005: Support insertion of device rather than addition to tail
>>>>
>>>>
>>>>
>>>> Patch 2 and 5 could be squashed.
>>>
>>>
>>>
>>> I deliberately kept them separate. I intent to extend the Patch 5 for
>>> hotplugging. But, if I don't end up adding support for that in this
>>> series,
>>> I will merge these two.
>>
>>
>> Fine.
>>
>>
>>>> The constructor priority stuff seems unneeded as long as we use
>>>> explicit reference to a global (or local, did not check) bus symbol
>>>> rather than a runtime lookup.
>>>
>>>
>>>
>>> I didn't understand your point here.
>>> IMO, constructor priority (or some other way to handle this) is
>>> important. I
>>> faced this issue while verifying it at my end when the drivers were
>>> getting
>>> registered before the bus.
>>>
>>> Can you elaborate more on '..use explicit reference to a global...'?
>>
>>
>> The drivers register themselves to a bus using this bus specific api.
>>
>> For pci, this is rte_eal_pci_register().
>> The pci_bus object must be moved to eal_common_pci.c (we can stil
>> internally expose for bsd / linux specific implementations).
>> Then, rte_eal_pci_register() can add the pci driver to the pci_bus
>> drivers list even if this pci_bus object is not registered yet to the
>> buses list.
>
>
> So, in eal_common_bus.c
>
> --->8---
>
> struct rte_bus *global_ptr_to_pci_bus = NULL;
>
> struct rte_bus pci_bus = { ... };
>
> rte_eal_pci_register() {
> if (global_ptr_to_pci_bus == NULL)
> rte_eal_bus_register(&pci_bus)
> else
> // continue as if PCI bus is registered
> }
>
> --->8---
>
> so, no RTE_REGISTER_BUS()?
>
> If yes, then RTE_REGISTER_BUS() should also check for an existing
> registration for duplication.
>
> I was banking on a model where bus handlers (or bus drivers) are independent
> entities, just like PMDs. So, we have a bus XYZ without any drivers
> necessarily based on it.
>
> By making registration dependent on driver registration, it becomes implicit
> that buses don't exist without drivers.
> I am not in favor of this - or maybe I lack enough reason for this (about
> how it will make framework/PMD life better).
>
>>
>> And no constructor order issue ?
>>
>>
>>>>
>>>>
>>>>> 0004: Add scan and match callbacks for the Bus and updated test
>>>>> case
>>>>
>>>>
>>>>
>>>> Why do you push back the bus object in the 'scan' method ?
>>>> This method is bus specific which means that the code "knows" the
>>>> object registered with the callback.
>>>
>>>
>>>
>>> This 'knows' is the grey area for me.
>>> The bus (for example, PCI) after scanning needs to call
>>> rte_eal_bus_add_device() to link the device in bus's device_list.
>>>
>>> Two options:
>>> 1. Have a global reference to "pci" bus (rte_bus) somewhere in eal_pci.c
>>> 2. Call rte_eal_get_bus() every time someone needs the reference.
>>> 3. C++ style, 'this->'.
>>>
>>> I have taken the 3rd path. It simplifies my code to not assume a handle
>>> as
>>> well as not allow for reference fetch calls every now and then.
>>>
>>> As a disadvantage: it means passing this as argument - and some cases
>>> maintaining it as __rte_unused.
>>>
>>> Taking (1) or (2) is not advantageous than this approach.
>>
>>
>> 1) is the simplest one.
>>
>> When you write a pci_scan method and embed it in you pci_bus object,
>> but this pci_scan method still wonders which bus object it is supposed
>> to work on, this is a bit like Schizophrenia ;-).
>
>
> :)
> This now is linked to the above issue of constructor priority and having a
> global bus reference. I don't personally prefer it.
> I will still give this a serious thought, though.
>
I'm also in favor of (3).
>>
>>
>>>> Is is that you want to have a single scan method used by multiple buses
>>>> ?
>>>
>>>
>>>
>>> Yes, but only as a use case. For example, platform devices are of various
>>> types - what if we have a south-bound bus over a platform bus. In which
>>> case, a hierarchical bus layout is possible.
>>> But, this is far-fetched idea for now.
>>
How to express the hierarchical bus layout as the bus in your design
is more like independent objects to hold drivers and their devices?
>>
>> Well, if you have no usecase at the moment, let's keep it simple, please.
>>
>
> Ok.
>
>>
>>>>
>>>>> 0006: Integrate bus scan/match with EAL, without any effective
>>>>> driver
>>>>
>>>>
>>>>
>>>> Hard to find a right balance in patch splittng, but patch 4 and 6 are
>>>> linked, I would squash them into one.
>>>
>>>
>>>
>>> Yes, it is hard and sometimes there is simply no strong rationale for
>>> splitting or merging. This is one of those cases.
>>> My idea was that one patch _only_ introduces Bus services (structures,
>>> functions etc) and another should enable the calls to it from EAL.
>>> In that sense, I still think 4 and 6 should remain separate, may be
>>> consecutive, though.
>>
>>
>> Ok, will see in next version of the patchset.
>
>
> Is there anything specific that you are looking for in patchset v2?
> I was thinking of:
> 0. fixing BSD compilation issue reported by CI
> 1. improving the test_pci.c
> 2. hotplugging
> 3. trying to move PCI to drives/bus/pci/linux/* and resolving how drivers
> link to it, and how EAL resources like devargs are consumed.
>
> Anything else?
>
>>
>>
>>>>
>>>>> 0007: rte_pci_driver->probe replaced with rte_driver->probe
>>>>
>>>>
>>>>
>>>> This patch is too big, please separate in two patches: eal changes
>>>> then ethdev/driver changes.
>>>
>>>
>>>
>>> I don't think that can be done. One change is incomplete without the
>>> other.
>>>
>>> Changes to all files are only for rte_pci_driver->probe to
>>> rte_driver->probe
>>> movement. EAL changes is to allow rte_eth_dev_pci_probe function after
>>> such
>>> a change as rte_driver->probe has different arguments as compared to
>>> rte_pci_driver->probe. The patches won't compile if I split.
>>>
>>> Am I missing something?
>>>>
>>>>
>>>> Why do you push back the driver object in the 'probe' method ? (idem
>>>> rte_bus->scan).
>>>
>>>
>>>
>>> I am assuming you are referring to rte_driver->probe().
>>> This is being done so that implementations (specific to drivers on a
>>> particular bus) can start extracting the rte_xxx_driver, if need be.
>>>
>>> For example, for e1000/em_ethdev.c, rte_driver->probe() have been set to
>>> rte_eth_dev_pci_probe() which requires rte_pci_driver to work with. In
>>> absence of the rte_driver object, this function cannot call
>>> rte_pci_driver->probe (for example) for driver specific operations.
>>
>>
>> Sorry, I am thinking a step ahead with eth_driver out of the picture.
>> But once eth_driver disappears, I can see no reason to keep this
>> driver in the probe method (Schizophrenia again).
>
>
> When eth_driver disappears, i was thinking of accomodating the eth_dev_init
> into the rte_pci_driver->probe/init.
> But, this is still a nascent thought.
> I am yet to start working on eth_driver.
>
>>
>>
>>
>
next prev parent reply other threads:[~2016-12-12 14:35 UTC|newest]
Thread overview: 132+ messages / expand[flat|nested] mbox.gz Atom feed top
2016-12-04 10:11 Shreyansh Jain
2016-12-04 10:11 ` [dpdk-dev] [PATCH 01/13] eal: define container_of macro Shreyansh Jain
2016-12-04 10:11 ` [dpdk-dev] [PATCH 02/13] eal/bus: introduce bus abstraction Shreyansh Jain
2016-12-04 10:11 ` [dpdk-dev] [PATCH 03/13] test: add basic bus infrastructure tests Shreyansh Jain
2016-12-04 10:11 ` [dpdk-dev] [PATCH 04/13] eal/bus: add scan and match support Shreyansh Jain
2016-12-04 10:11 ` [dpdk-dev] [PATCH 05/13] eal/bus: add support for inserting a device on a bus Shreyansh Jain
2016-12-04 10:11 ` [dpdk-dev] [PATCH 06/13] eal: integrate bus scan and probe with EAL Shreyansh Jain
2016-12-04 10:11 ` [dpdk-dev] [PATCH 07/13] pci: replace probe and remove handlers with rte_driver Shreyansh Jain
2016-12-08 17:50 ` Ferruh Yigit
2016-12-09 4:59 ` Shreyansh Jain
2016-12-04 10:11 ` [dpdk-dev] [PATCH 08/13] eal: enable probe and remove from bus infrastructure Shreyansh Jain
2016-12-06 10:45 ` Shreyansh Jain
2016-12-04 10:11 ` [dpdk-dev] [PATCH 09/13] pci: split match and probe function Shreyansh Jain
2016-12-04 10:11 ` [dpdk-dev] [PATCH 10/13] eal/pci: generalize args of PCI scan/match towards RTE device/driver Shreyansh Jain
2016-12-04 10:11 ` [dpdk-dev] [PATCH 11/13] pci: Pass rte_pci_addr to functions instead of separate args Shreyansh Jain
2016-12-04 10:11 ` [dpdk-dev] [PATCH 12/13] eal: enable PCI bus Shreyansh Jain
2016-12-04 10:11 ` [dpdk-dev] [PATCH 13/13] eal/pci: remove PCI probe and init calls Shreyansh Jain
2016-12-06 20:52 ` [dpdk-dev] [PATCH 00/13] Introducing EAL Bus-Device-Driver Model David Marchand
2016-12-07 9:55 ` Shreyansh Jain
2016-12-07 12:17 ` David Marchand
2016-12-07 13:10 ` Shreyansh Jain
2016-12-07 13:24 ` Thomas Monjalon
2016-12-08 5:04 ` Shreyansh Jain
2016-12-08 7:21 ` Thomas Monjalon
2016-12-08 7:53 ` Shreyansh Jain
2016-12-12 14:35 ` Jianbo Liu [this message]
2016-12-13 6:56 ` Shreyansh Jain
2016-12-13 13:37 ` [dpdk-dev] [PATCH v2 00/12] " Shreyansh Jain
2016-12-13 13:37 ` [dpdk-dev] [PATCH v2 01/12] eal: define container_of macro Shreyansh Jain
2016-12-13 22:24 ` Jan Blunck
2016-12-14 5:12 ` Shreyansh Jain
2016-12-16 8:14 ` Jan Blunck
2016-12-16 9:23 ` Adrien Mazarguil
2016-12-16 10:47 ` Jan Blunck
2016-12-16 11:21 ` Adrien Mazarguil
2016-12-16 11:54 ` Shreyansh Jain
2016-12-13 13:37 ` [dpdk-dev] [PATCH v2 02/12] eal/bus: introduce bus abstraction Shreyansh Jain
2016-12-13 13:37 ` [dpdk-dev] [PATCH v2 03/12] test: add basic bus infrastructure tests Shreyansh Jain
2016-12-13 13:37 ` [dpdk-dev] [PATCH v2 04/12] eal/bus: add scan, match and insert support Shreyansh Jain
2016-12-13 13:37 ` [dpdk-dev] [PATCH v2 05/12] eal: integrate bus scan and probe with EAL Shreyansh Jain
2016-12-13 13:37 ` [dpdk-dev] [PATCH v2 06/12] eal: add probe and remove support for rte_driver Shreyansh Jain
2016-12-13 13:37 ` [dpdk-dev] [PATCH v2 07/12] eal: enable probe from bus infrastructure Shreyansh Jain
2016-12-13 13:37 ` [dpdk-dev] [PATCH v2 08/12] pci: split match and probe function Shreyansh Jain
2016-12-13 13:37 ` [dpdk-dev] [PATCH v2 09/12] eal/pci: generalize args of PCI scan/match towards RTE device/driver Shreyansh Jain
2016-12-13 13:37 ` [dpdk-dev] [PATCH v2 10/12] pci: Pass rte_pci_addr to functions instead of separate args Shreyansh Jain
2016-12-13 13:37 ` [dpdk-dev] [PATCH v2 11/12] eal: enable PCI bus Shreyansh Jain
2016-12-13 13:37 ` [dpdk-dev] [PATCH v2 12/12] drivers: update PMDs to use rte_driver probe and remove Shreyansh Jain
2016-12-13 13:52 ` Andrew Rybchenko
2016-12-13 15:07 ` Ferruh Yigit
2016-12-14 5:14 ` Shreyansh Jain
2016-12-14 5:11 ` Shreyansh Jain
2016-12-14 9:49 ` Shreyansh Jain
2016-12-15 21:36 ` Jan Blunck
2016-12-26 9:14 ` Shreyansh Jain
2016-12-16 13:10 ` [dpdk-dev] [PATCH v3 00/12] Introducing EAL Bus-Device-Driver Model Shreyansh Jain
2016-12-16 13:10 ` [dpdk-dev] [PATCH v3 01/12] eal: define container_of macro Shreyansh Jain
2016-12-16 13:10 ` [dpdk-dev] [PATCH v3 02/12] eal/bus: introduce bus abstraction Shreyansh Jain
2016-12-20 12:37 ` Hemant Agrawal
2016-12-20 13:17 ` Jan Blunck
2016-12-20 13:51 ` Shreyansh Jain
2016-12-20 17:11 ` Stephen Hemminger
2016-12-21 7:11 ` Shreyansh Jain
2016-12-21 15:38 ` Jan Blunck
2016-12-21 23:33 ` Stephen Hemminger
2016-12-22 5:12 ` Shreyansh Jain
2016-12-22 5:52 ` Shreyansh Jain
2016-12-25 17:39 ` Shreyansh Jain
2016-12-16 13:10 ` [dpdk-dev] [PATCH v3 03/12] test: add basic bus infrastructure tests Shreyansh Jain
2016-12-16 13:10 ` [dpdk-dev] [PATCH v3 04/12] eal/bus: add scan, match and insert support Shreyansh Jain
2016-12-16 13:25 ` Shreyansh Jain
2016-12-16 13:10 ` [dpdk-dev] [PATCH v3 05/12] eal: integrate bus scan and probe with EAL Shreyansh Jain
2016-12-16 13:10 ` [dpdk-dev] [PATCH v3 06/12] eal: add probe and remove support for rte_driver Shreyansh Jain
2016-12-16 13:10 ` [dpdk-dev] [PATCH v3 07/12] eal: enable probe from bus infrastructure Shreyansh Jain
2016-12-16 13:10 ` [dpdk-dev] [PATCH v3 08/12] pci: split match and probe function Shreyansh Jain
2016-12-16 13:10 ` [dpdk-dev] [PATCH v3 09/12] eal/pci: generalize args of PCI scan/match towards RTE device/driver Shreyansh Jain
2016-12-16 13:10 ` [dpdk-dev] [PATCH v3 10/12] pci: Pass rte_pci_addr to functions instead of separate args Shreyansh Jain
2016-12-16 13:10 ` [dpdk-dev] [PATCH v3 11/12] eal: enable PCI bus and PCI test framework Shreyansh Jain
2016-12-16 13:20 ` Shreyansh Jain
2016-12-16 13:10 ` [dpdk-dev] [PATCH v3 12/12] drivers: update PMDs to use rte_driver probe and remove Shreyansh Jain
2016-12-26 12:50 ` [dpdk-dev] [PATCH v4 00/12] Introducing EAL Bus-Device-Driver Model Shreyansh Jain
2016-12-26 12:50 ` [dpdk-dev] [PATCH v4 01/12] eal/bus: introduce bus abstraction Shreyansh Jain
2016-12-26 12:50 ` [dpdk-dev] [PATCH v4 02/12] test: add basic bus infrastructure tests Shreyansh Jain
2016-12-26 12:50 ` [dpdk-dev] [PATCH v4 03/12] eal/bus: add scan, match and insert support Shreyansh Jain
2016-12-26 13:27 ` Shreyansh Jain
2016-12-26 12:50 ` [dpdk-dev] [PATCH v4 04/12] eal: integrate bus scan and probe with EAL Shreyansh Jain
2016-12-26 12:50 ` [dpdk-dev] [PATCH v4 05/12] eal: add probe and remove support for rte_driver Shreyansh Jain
2016-12-26 12:50 ` [dpdk-dev] [PATCH v4 06/12] eal: enable probe from bus infrastructure Shreyansh Jain
2016-12-26 12:50 ` [dpdk-dev] [PATCH v4 07/12] pci: split match and probe function Shreyansh Jain
2016-12-26 12:50 ` [dpdk-dev] [PATCH v4 08/12] eal/pci: generalize args of PCI scan/match towards RTE device/driver Shreyansh Jain
2016-12-26 12:50 ` [dpdk-dev] [PATCH v4 09/12] pci: Pass rte_pci_addr to functions instead of separate args Shreyansh Jain
2016-12-26 12:50 ` [dpdk-dev] [PATCH v4 10/12] eal: enable PCI bus and PCI test framework Shreyansh Jain
2016-12-26 12:50 ` [dpdk-dev] [PATCH v4 11/12] drivers: update PMDs to use rte_driver probe and remove Shreyansh Jain
2016-12-26 12:50 ` [dpdk-dev] [PATCH v4 12/12] eal/bus: add bus iteration macros Shreyansh Jain
2016-12-26 13:23 ` [dpdk-dev] [PATCH v5 00/12] Introducing EAL Bus-Device-Driver Model Shreyansh Jain
2016-12-26 13:23 ` [dpdk-dev] [PATCH v5 01/12] eal/bus: introduce bus abstraction Shreyansh Jain
2017-01-03 21:52 ` Thomas Monjalon
2017-01-06 10:31 ` Shreyansh Jain
2017-01-06 14:55 ` Thomas Monjalon
2017-01-09 6:24 ` Shreyansh Jain
2017-01-09 15:22 ` Ferruh Yigit
2017-01-10 4:07 ` Shreyansh Jain
2016-12-26 13:23 ` [dpdk-dev] [PATCH v5 02/12] test: add basic bus infrastructure tests Shreyansh Jain
2016-12-26 13:23 ` [dpdk-dev] [PATCH v5 03/12] eal/bus: add scan, match and insert support Shreyansh Jain
2016-12-26 13:23 ` [dpdk-dev] [PATCH v5 04/12] eal: integrate bus scan and probe with EAL Shreyansh Jain
2017-01-03 21:46 ` Thomas Monjalon
2017-01-06 10:38 ` Shreyansh Jain
2017-01-06 12:00 ` Shreyansh Jain
2017-01-06 13:46 ` Thomas Monjalon
2017-01-09 6:35 ` Shreyansh Jain
2017-01-08 12:21 ` Rosen, Rami
2017-01-09 6:34 ` Shreyansh Jain
2016-12-26 13:23 ` [dpdk-dev] [PATCH v5 05/12] eal: add probe and remove support for rte_driver Shreyansh Jain
2017-01-03 22:05 ` Thomas Monjalon
2017-01-06 11:44 ` Shreyansh Jain
2017-01-06 15:26 ` Thomas Monjalon
2017-01-09 6:28 ` Shreyansh Jain
2016-12-26 13:23 ` [dpdk-dev] [PATCH v5 06/12] eal: enable probe from bus infrastructure Shreyansh Jain
2016-12-26 13:24 ` [dpdk-dev] [PATCH v5 07/12] pci: split match and probe function Shreyansh Jain
2017-01-03 22:08 ` Thomas Monjalon
2016-12-26 13:24 ` [dpdk-dev] [PATCH v5 08/12] eal/pci: generalize args of PCI scan/match towards RTE device/driver Shreyansh Jain
2017-01-03 22:13 ` Thomas Monjalon
2017-01-06 12:03 ` Shreyansh Jain
2016-12-26 13:24 ` [dpdk-dev] [PATCH v5 09/12] pci: Pass rte_pci_addr to functions instead of separate args Shreyansh Jain
2016-12-26 13:24 ` [dpdk-dev] [PATCH v5 10/12] eal: enable PCI bus and PCI test framework Shreyansh Jain
2016-12-26 13:24 ` [dpdk-dev] [PATCH v5 11/12] drivers: update PMDs to use rte_driver probe and remove Shreyansh Jain
2017-01-09 15:19 ` Ferruh Yigit
2017-01-09 16:18 ` Ferruh Yigit
2017-01-10 4:09 ` Shreyansh Jain
2016-12-26 13:24 ` [dpdk-dev] [PATCH v5 12/12] eal/bus: add bus iteration macros Shreyansh Jain
2017-01-03 22:15 ` Thomas Monjalon
2017-01-03 22:22 ` [dpdk-dev] [PATCH v5 00/12] Introducing EAL Bus-Device-Driver Model Thomas Monjalon
2017-01-06 6:27 ` Shreyansh Jain
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='CAP4Qi38CYGC612kQJpA49xWRF=jpqJ1COWg=Qe18HDzmiLVSuA@mail.gmail.com' \
--to=jianbo.liu@linaro.org \
--cc=david.marchand@6wind.com \
--cc=dev@dpdk.org \
--cc=shreyansh.jain@nxp.com \
--cc=thomas.monjalon@6wind.com \
/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).