Re: kern/56355: dtrace triggers double-fault in supervisor mode

[Taylor R Campbell <riastradh%NetBSD.org@localhost>]

The following reply was made to PR kern/56355; it has been noted by GNATS.

From: Taylor R Campbell <riastradh%NetBSD.org@localhost>
To: bch <brad.harder%gmail.com@localhost>
Cc: gnats-bugs%netbsd.org@localhost
Subject: Re: kern/56355: dtrace triggers double-fault in supervisor mode
Date: Sun, 15 Aug 2021 18:42:14 +0000

 > Date: Sun, 15 Aug 2021 09:05:52 -0700
 > From: bch <brad.harder%gmail.com@localhost>
 > 
 > On Sun, Aug 15, 2021 at 08:10 Taylor R Campbell <riastradh%netbsd.org@localhost>
 > wrote:
 > 
 > > > Date: Sun, 15 Aug 2021 15:08:11 +0000
 > > > From: Taylor R Campbell <riastradh%NetBSD.org@localhost>
 > > >
 > > > This patch changes the semantics of the code so that it excludes any
 > > > names that _start_ with `trap', `x86_curcpu', or `x86_curlwp', such as
 > > > `trapsignal'.  The use of strncmp instead of strcmp should be restored
 > > > (and the list should maybe be sorted lexicographically).
 > >
 > > Errr, I mean: this code should use strcmp, not strncmp.
 > 
 > If you read through to the MI exclusion code at fbt_exclude(), it's using
 > strncmp(). Should that be changed too, or otherwise, why pick strcmp() v
 > strncmp()?
 
 fbt_excluded is meant to exclude any function whose name starts with
 `dtrace_', `dtrace_safe_', `fbt_', &c., which is what, e.g.,
 strncmp(name, "fbt_", 4) does.  But we want to exclude the function
 `trap' specifically, not all functions that happen to start with
 `trap' like `trapsignal', hence strcmp(name, "trap").