On 1/19/2018 4:19 PM, Ferruh Yigit wrote:
> On 1/18/2018 6:10 PM, Matan Azrad wrote:
>> Hi Ferruh
>>
>> From: Ferruh Yigit, Thursday, January 18, 2018 7:31 PM
>>> On 1/18/2018 11:27 AM, Matan Azrad wrote:
>>>> rte_eth_dev_is_removed API was added to detect a device removal
>>>> synchronously.
>>>>
>>>> When a device removal occurs during control command execution, many
>>>> different errors can be reported to the user.
>>>>
>>>> Adjust all ethdev APIs error reports to return -EIO in case of device
>>>> removal using rte_eth_dev_is_removed API.
>>>>
>>>> Signed-off-by: Matan Azrad <ma...@mellanox.com>
>>>> Acked-by: Thomas Monjalon <tho...@monjalon.net>
>>>> ---
>>>> lib/librte_ether/rte_ethdev.c | 192
>>>> +++++++++++++++++++++++++++---------------
>>>> lib/librte_ether/rte_ethdev.h | 51 ++++++++++-
>>>> 2 files changed, 170 insertions(+), 73 deletions(-)
>>>>
>>>> diff --git a/lib/librte_ether/rte_ethdev.c
>>>> b/lib/librte_ether/rte_ethdev.c index c93cec1..7044159 100644
>>>> --- a/lib/librte_ether/rte_ethdev.c
>>>> +++ b/lib/librte_ether/rte_ethdev.c
>>>> @@ -338,6 +338,16 @@ struct rte_eth_dev *
>>>> return -ENODEV;
>>>> }
>>>>
>>>> +static int
>>>> +eth_err(uint16_t port_id, int ret)
>>>> +{
>>>> + if (ret == 0)
>>>> + return 0;
>>>> + if (rte_eth_dev_is_removed(port_id))
>>>> + return -EIO;
>>>> + return ret;
>>>> +}
>>>> +
>>>> /* attach the new device, then store port_id of the device */ int
>>>> rte_eth_dev_attach(const char *devargs, uint16_t *port_id) @@ -492,7
>>>> +502,8 @@ struct rte_eth_dev *
>>>> return 0;
>>>> }
>>>>
>>>> - return dev->dev_ops->rx_queue_start(dev, rx_queue_id);
>>>> + return eth_err(port_id, dev->dev_ops->rx_queue_start(dev,
>>>> + rx_queue_id));
>>>>
>>>> }
>>>
>>> This patch updates *all* ethdev public APIs to add if device is removed
>>> check?
>>
>> Yes.
>>
>>> And each check goes to ethdev is_removed() dev_ops to ask if dev is
>>> removed.
>> Probably, if the REMOVED state setted in will not call device is_remove.
>>
>>> These must be better way of doing this, am I missing something.
>>
>> Suggest.
>
> With a silly analogy, this is like a blind person asking each time if he is
> dead
> before talking to other person.
>
> At first glance I can think of a kind of watchdog timer can be implemented in
> ethdev layer. It provides periodic checks and if device is dead it calls the
> registered user callback function.
>
> This method presented as synchronous method but not triggered from side where
> event happens, I mean not triggered from PMD but from application.
> So does application doing polling continuously if device is dead?
> Or if application is relying this patch to add a check in each API, what
> happens
> if device removed during data processing, will app rely on asynchronous
> method?
>
> I am including a few consumers of the ethdev to the mail thread, clearly I am
> not very supportive of this patch, but specially taking release is being close
> to the account, if there is no objection than me I will take as consensus to
> get
> the patch in.
It looks like there is no objection to the patch and it is already acked, I will
get latest version to next-net.
>
>>
>> This code will replace similar code in each PMD.
>>
>>> I definitely would like to see more comments for this patch.
>>>
>>> Another question is what happens if device removed while or before
>>> dev_ops called? There is no synchronizations in drivers for removal, right?
>>>
>>
>> Yes. You right, the device removal can be changed a moment after the call.
>> Actually the caller suspected in removal before call it(and want to validate
>> it) - so it makes sense.
>> From this reason the check in ethdev APIs is called generally in error flows.
>>
>>
>>> <...>
>
