Re: [PATCH RESEND net-next] netlink: add tracepoint at NL_SET_ERR_MSG

On Tue, Feb 02, 2021 at 09:16:28AM -0700, David Ahern wrote:

On 2/2/21 5:30 AM, Marcelo Ricardo Leitner wrote:

quoted

Also, if the message is a common one, one may not be able to easily
distinguish them. Ideally this shouldn't happen, but when debugging
applications such as OVS, where lots of netlink requests are flying,
it saves us time. I can, for example, look at a perf capture and
search for cls_flower or so. Otherwise, it will all show up as
"af_netlink: <err_msg>"

Modules should be using the NL_SET_ERR_MSG_MOD variant, so the message
would be ""af_netlink: cls_flower: <err_msg>"

Ah, right. They don't always do, though (and that probably should be
fixed). Also, currently there is no _MOD variant for NL_SET_ERR_MSG_ATTR.

For example:
$ git grep NL_SET_ERR_MSG -- cls_flower.c
cls_flower.c:                   NL_SET_ERR_MSG_MOD(extack, "Failed to setup flow action");
cls_flower.c:           NL_SET_ERR_MSG_ATTR(extack,
cls_flower.c:           NL_SET_ERR_MSG_ATTR(extack,
cls_flower.c:           NL_SET_ERR_MSG(extack, "Missing MPLS option \"depth\"");
...

I get the value in knowing the call site, so not arguing against that.
Just hoping that your experience matches theory.

Okay.

Thanks,
Marcelo