[dpdk-dev] CALL to eth PMD maintainers: complete closing of port

[Thomas Monjalon]

03/08/2020 20:50, Thomas Monjalon:
> 18/04/2019 12:59, Thomas Monjalon:
> > Hi all,
> > 
> > Since DPDK 18.11, the behaviour of the close operation is changed
> > if RTE_ETH_DEV_CLOSE_REMOVE is enabled in the driver:
> > port is released (i.e. totally freed and data erased) on close.
> > This new behaviour is enabled per driver for a migration period.
> > 
> > Looking at the code, you can see these comments:
> > /* old behaviour: only free queue arrays */
> > RTE_ETHDEV_LOG(DEBUG, "Port closing is using an old behaviour.\n"
> > 	"The driver %s should migrate to the new behaviour.\n",
> > /* new behaviour: send event + reset state + free all data */
> > 
> > You can find an advice in the commit:
> > 	http://git.dpdk.org/dpdk/commit/?id=23ea57a2a
> > "
> > When enabling RTE_ETH_DEV_CLOSE_REMOVE,
> > the PMD must free all its private resources for the port,
> > in its dev_close function.
> > It is advised to call the dev_close function in the remove function
> > in order to support removing a device without closing its ports.
> > "
> > 
> > It would be great to complete this migration for the next LTS
> > version, which will be 19.11.
> 
> For the record, it did not happen in 19.11.
> 
> > Following drivers should be migrated:
> > ( find drivers/net -mindepth 1 -maxdepth 1 -type d | cut -d/ -f3 ; git grep -l RTE_ETH_DEV_CLOSE_REMOVE drivers | cut -d/ -f3 ) | sort | uniq -u
> [...]
> 
> The progress in April 2019 was 4 of 46 (9%).
> 
> > Please let's progress smoothly on this topic, thanks.
> 
> More than one year later, the progress is 26 of 53 (49%).
> 
> > The concerned maintainers (Cc) can be found with the following command:
> > devtools/get-maintainer.sh $(( find drivers/net -mindepth 1 -maxdepth 1 -type d | cut -d/ -f-3 ; git grep -l RTE_ETH_DEV_CLOSE_REMOVE drivers ) | sort | uniq -u)
> 
> We cannot wait forever. Temporary cannot be longer than 2 years.
> I am going to send a deprecation notice to remove the "temporary"
> flag RTE_ETH_DEV_CLOSE_REMOVE.

The deprecation notice was merged in 20.08:
	http://mails.dpdk.org/archives/dev/2020-August/177314.html

> It will break drivers which are not migrated.
> It will probably help to find motivation in new priorities.
> 
> More details on what to do can be found in this mail thread:
> 	http://inbox.dpdk.org/dev/1748144.UFpUr2FPnr@xps/

Summary:

* The freeing of private port resources must be moved in the PMD
from the ".remove(device)" function to the ".dev_close(port)" function.

* If a generic resource (.mac_addrs or .hash_mac_addrs) cannot be freed,
it must be set to NULL in ".dev_close" PMD function to protect from
subsequent rte_eth_dev_release_port() freeing.

* Note 1:
The generic resources are freed in rte_eth_dev_release_port(),
after ".dev_close" is called in rte_eth_dev_close(), but not when
calling ".dev_close" directly from the ".remove" PMD function.
That's why rte_eth_dev_release_port() must still be called explicitly
from ".remove(device)" after calling the ".dev_close" PMD function.

* Note 2:
If a device can have multiple ports, the common resources must be freed
only in the ".remove(device)" function.

* Note 3:
The port is supposed to be in a stopped state when it is closed.
If it is not the case, it is free to the PMD implementation
how to react when trying to close a non-stopped port:
either try to stop it automatically or just return an error.