Hi Wenzhuo,
Port hot-plug would require the DPDK app to
dynamically mange ports and the kernel to support PCI hot-plugging.
When multiple ports are detached/attached (SFP's plugged in/out),
would the DPDK port id's remain same? If not, app needs to handle port
id change for a give NIC.
While port hot-plug can be used to dynamically handle SFP
hot-plugging, I feel it's not the optimal solution.
Regards,
Srini
On Mon, Apr 24, 2017 at 12:45 PM, Lu, Wenzhuo <wenzhuo...@intel.com> wrote:
> Hi Srini,
>
>
>> -----Original Message-----
>> From: Srinivasan J [mailto:srinid...@gmail.com]
>> Sent: Monday, April 24, 2017 2:24 PM
>> To: Lu, Wenzhuo
>> Cc: Ananyev, Konstantin; dev@dpdk.org
>> Subject: Re: [PATCH] net/ixgbe: support detection of hot swapped SFP/SFP+
>>
>> Hi Wenzhuo,
>> I understand your concern. I had to do this change to
>> support hot
>> swap of SFP/SFP+ modules without restarting the DPDK app. If we have some
>> other mechanism in pipeline to support hot swap of SFP/SFP+ modules, then
>> my changes can be ignored. I just wanted to share the changes which worked
>> for me back to the community.
> I think restarting the APP is not necessary. You can uninit the port, then
> re-init the port.
> Please reference
> http://dpdk.org/doc/guides/prog_guide/port_hotplug_framework.html to see if
> it can help.
>
>>
>> Regards,
>> Srini
>>
>> On Fri, Apr 21, 2017 at 10:22 AM, Lu, Wenzhuo <wenzhuo...@intel.com>
>> wrote:
>> > Hi Srini,
>> >
>> >> -----Original Message-----
>> >> From: Srini J [mailto:srinid...@gmail.com]
>> >> Sent: Thursday, April 20, 2017 6:48 PM
>> >> To: Lu, Wenzhuo; Ananyev, Konstantin
>> >> Cc: dev@dpdk.org; Srinivasan Jayarajan
>> >> Subject: [PATCH] net/ixgbe: support detection of hot swapped SFP/SFP+
>> >>
>> >> From: Srinivasan Jayarajan <srinid...@gmail.com>
>> >>
>> >> Adds support to use a different SFP/SFP+ without restarting the
>> >> DPDK app. rte_eth_dev_stop()/rte_eth_dev_start() will need
>> >> to be called on the port to detect the SFP/SFP+ change.
>> >>
>> >> Signed-off-by: Srinivasan Jayarajan <srinid...@gmail.com>
>> >> ---
>> >> drivers/net/ixgbe/ixgbe_ethdev.c | 13 +++++++++++++
>> >> 1 file changed, 13 insertions(+)
>> >>
>> >> diff --git a/drivers/net/ixgbe/ixgbe_ethdev.c
>> >> b/drivers/net/ixgbe/ixgbe_ethdev.c
>> >> index c226e0a..85407a9 100644
>> >> --- a/drivers/net/ixgbe/ixgbe_ethdev.c
>> >> +++ b/drivers/net/ixgbe/ixgbe_ethdev.c
>> >> @@ -2520,6 +2520,19 @@ static int eth_ixgbevf_pci_remove(struct
>> >> rte_pci_device *pci_dev)
>> >> status = ixgbe_pf_reset_hw(hw);
>> >> if (status != 0)
>> >> return -1;
>> >> +
>> >> + /* Set phy type as unknown so that PHY scan is always done */
>> >> + hw->phy.type = ixgbe_phy_unknown;
>> >> +
>> >> + /* Identify PHY and related function pointers */
>> >> + status = hw->phy.ops.init(hw);
>> >> +
>> >> + if (status == IXGBE_ERR_SFP_NOT_SUPPORTED) {
>> >> + PMD_INIT_LOG(ERR, "Found unsupported SFP in "
>> >> + "ixgbe_dev_start(): %d", status);
>> >> + return -1;
>> >> + }
>> >> +
>> > I have the concern if it's a good idea to move the functions from dev_init
>> to dev_start. Especially this function named init.
>> > Anyway, let's listen to others opinion.
>> >
>> >
>> >> hw->mac.ops.start_hw(hw);
>> >> hw->mac.get_link_status = true;
>> >>
>> >> --
>> >> 1.8.1.4
>> >
