[<prev] [next>] [<thread-prev] [thread-next>] [day] [month] [year] [list]
Message-ID: <20210203103241.7e86ca2b@kicinski-fedora-pc1c0hjn.dhcp.thefacebook.com>
Date: Wed, 3 Feb 2021 10:32:41 -0800
From: Jakub Kicinski <kuba@...nel.org>
To: "Kubalewski, Arkadiusz" <arkadiusz.kubalewski@...el.com>
Cc: "Nguyen, Anthony L" <anthony.l.nguyen@...el.com>,
"davem@...emloft.net" <davem@...emloft.net>,
"netdev@...r.kernel.org" <netdev@...r.kernel.org>,
"sassmann@...hat.com" <sassmann@...hat.com>,
"Topel, Bjorn" <bjorn.topel@...el.com>,
"Fijalkowski, Maciej" <maciej.fijalkowski@...el.com>,
"Karlsson, Magnus" <magnus.karlsson@...el.com>,
"Loktionov, Aleksandr" <aleksandr.loktionov@...el.com>,
"Kuruvinakunnel, George" <george.kuruvinakunnel@...el.com>
Subject: Re: [PATCH net-next 5/6] i40e: Add info trace at loading XDP
program
On Wed, 3 Feb 2021 10:00:07 +0000 Kubalewski, Arkadiusz wrote:
> >> diff --git a/drivers/net/ethernet/intel/i40e/i40e_main.c b/drivers/net/ethernet/intel/i40e/i40e_main.c
> >> index 521ea9df38d5..f35bd9164106 100644
> >> --- a/drivers/net/ethernet/intel/i40e/i40e_main.c
> >> +++ b/drivers/net/ethernet/intel/i40e/i40e_main.c
> >> @@ -12489,11 +12489,14 @@ static int i40e_xdp_setup(struct i40e_vsi *vsi,
> >> /* Kick start the NAPI context if there is an AF_XDP socket open
> >> * on that queue id. This so that receiving will start.
> >> */
> >> - if (need_reset && prog)
> >> + if (need_reset && prog) {
> >> + dev_info(&pf->pdev->dev,
> >> + "Loading XDP program, please note: XDP_REDIRECT action requires the same number of queues on both interfaces\n");
> >
> >We try to avoid spamming logs. This message will be helpful to users
> >only the first time, if at all.
>
> You are probably right, it would look like a spam to the one who is
> continuously loading and unloading the XDP programs.
> But still, want to remain as much user friendly as possible.
> Will use dev_info_once(...) instead.
Not exactly what I meant, I meant that it's only marginally useful the
first time the user sees it. Not first time since boot.
The two options that I think could be better are:
- work on improving the interfaces in terms of IRQ/queue config and
capabilities so the user is not confused in the first place;
- detect that the configuration is in fact problematic
(IOW #Qs < #CPUs) and setting extack. If you set the extact and
return 0 / success the extact will show as "Warning: " in iproute2
output.
Powered by blists - more mailing lists