Re: [PATCH - RFC] [1/2] 2.6 must-fix list - kernel error reporting

From: Jim Keniston (
Date: Tue Jul 08 2003 - 14:22:44 EST

Andrew Morton wrote:
> Jim Keniston <> wrote:
> >
> > The enclosed patches provide a mechanism for reporting error events
> > to user-mode applications via netlink.
> Seems sane to me.
> It needs to handle %z as well as %Z.

Yes, thanks. I missed that change to vsnprintf().

> The layout of `struct kern_log_entry' may be problematic. Think of the
> situation where a 64-bit kernel constructs one of these and sends it up to
> 32-bit userspace. Will the structure layout be the same under the 32-bit
> compiler as under the 64-bit one?

I think so. Nothing is bigger than 4 bytes except log_facility[] (16-byte
array of char, which doesn't induce padding at all on i386). But I will find a
64K/32U ppc machine and check that.

> How do you actually intend to use this?

I envision it being used by a configuration/status-monitoring system that monitors
hotplug events, sysfs, etc. for configuration changes, and listens to the
proposed interface for error events. Binary-only events (logged with evl_write())
would have to be interpreted based on knowledge existing entirely in user space (either
coded into the monitor program, or provided as supplementary information via a formatting
template or some such). PRINTF-format events can carry and/or be supplemented with
similar info, but have the error message built in.

> Will it be by adding new
> evt_printf() calls to particular drivers, or replacing existing printk's or
> both?

There have been a variety of suggestions for how error reporting could be improved.
Two common ones are:
1. Leave printks alone, and log additional info in whatever format you want via netlink.
(E.g., Dave Miller recommended something like this.) This proposal supports that.

2. Migrate from raw printks to "smarter" versions -- e.g., dev_printk and friends,
or the proposed netdev_printk. Such macros now call just printk (after adding
relevant info), but could be modified to call evl_printk as well with the same args.

There are a zillion variations on this, of course...

> Would it make sense for evt_printf() to fall back to printk() if nobody is
> listening to the log stream?

That certainly makes sense for evl_printf. For evl_write, just do a hex dump or something.
So evlog.c would query kerror.c to see if anybody's listening. Would kerror.c
consult nl_table[] directly, or is there an anybody_listening() function that
does this?

To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
the body of a message to
More majordomo info at
Please read the FAQ at

This archive was generated by hypermail 2b29 : Tue Jul 15 2003 - 22:00:28 EST