[dpdk-dev] [PATCH] app/testpmd: fix hotplug

[Ferruh Yigit]

On 1/23/2020 5:01 PM, Iremonger, Bernard wrote:
> Hi Ferruh,
> 
>> -----Original Message-----
>> From: Yigit, Ferruh <ferruh.yigit at intel.com>
>> Sent: Thursday, January 23, 2020 3:15 PM
>> To: Lu, Wenzhuo <wenzhuo.lu at intel.com>; Wu, Jingjing
>> <jingjing.wu at intel.com>; Iremonger, Bernard
>> <bernard.iremonger at intel.com>; Matan Azrad <matan at mellanox.com>
>> Cc: dev at dpdk.org; stable at dpdk.org
>> Subject: [PATCH] app/testpmd: fix hotplug
>>
> This is reverting the original patch, which breaks hotplug.
> I think the commit message should make this clear.
> This patch is not fixing hotplug, the original patch seems to have revealed a problem with hotplug.
> Reverting this patch will not fix the underlying hotplug issue.

What is the problem revealed by the original patch? Did you able to reproduce it?

> 
>> The 'port_id_is_invalid()' check in the 'detach_port_device()' is breaking the
>> hotplug support, since at that stage port will be closed and validity check
>> always fail [1] and removing the device is not really completed.
>>
>> But this cause the vfio request interrupt keep triggered continuously and
>> makes the application unusable, since port is closed but device is not
>> removed, the remove path keep generating error log:
>>
>> EAL: can not get port by device 0000:00:05.0!
>> EAL: can not get port by device 0000:00:05.0!
>> EAL: can not get port by device 0000:00:05.0!
>> EAL: can not get port by device 0000:00:05.0!
>> EAL: can not get port by device 0000:00:05.0!
>> EAL: can not get port by device 0000:00:05.0!
>>
>> Fixed by removing 'port_id_is_invalid()' check from 'detach_port_device()',
>> anyway it shouldn't be required. Without this check device remove works as
>> expected.
>>
>> Only "Invalid port_id=0" logs seen a few times, which is because the actual
>> removal not done synchronously but an alarm set for it, until the alarm fired
>> application may receive many interrupts, expect the first ones cause the
>> error.
>> So this patch also removes the logging from checking the invalid port in
>> 'rmv_port_callback()' to reduce the noise.
>>
>> [1]
>> rmv_port_callback()
>>   stop_port(port_id);
>>   close_port(port_id);
>>   detach_port_device(port_id);
>>
>> Fixes: 43d0e304980a ("app/testpmd: fix invalid port detaching")
>> Cc: stable at dpdk.org
>>
>> Signed-off-by: Ferruh Yigit <ferruh.yigit at intel.com>
>> ---
>>  app/test-pmd/testpmd.c | 6 ++----
>>  1 file changed, 2 insertions(+), 4 deletions(-)
>>
>> diff --git a/app/test-pmd/testpmd.c b/app/test-pmd/testpmd.c index
>> f9f4cd1d3..3323013bb 100644
>> --- a/app/test-pmd/testpmd.c
>> +++ b/app/test-pmd/testpmd.c
>> @@ -2641,9 +2641,6 @@ detach_port_device(portid_t port_id)
>>
>>  printf("Removing a device...\n");
>>
>> -if (port_id_is_invalid(port_id, ENABLED_WARN))
>> -return;
>> -
>>  dev = rte_eth_devices[port_id].device;
>>  if (dev == NULL) {
>>  printf("Device already removed\n");
>> @@ -2875,7 +2872,8 @@ rmv_port_callback(void *arg)
>>  int org_no_link_check = no_link_check;
>>  portid_t port_id = (intptr_t)arg;
>>
>> -RTE_ETH_VALID_PORTID_OR_RET(port_id);
>> +if (!rte_eth_dev_is_valid_port(port_id))
>> +return;
>>
>>  if (!test_done && port_is_forwarding(port_id)) {
>>  need_to_start = 1;
>> --
>> 2.24.1
> 
> Otherwise
> Acked-by: Bernard Iremonger <bernard.iremonger at intel.com>
>