Re: [PATCH 0/1] tools/nolibc: tests: use volatile to force stack smashing

[Paul E. McKenney]

On Tue, Mar 28, 2023 at 08:51:52PM +0200, Willy Tarreau wrote:
> On Tue, Mar 28, 2023 at 11:29:53AM -0700, Paul E. McKenney wrote:
> > I have queued this for the v6.5 merge window, thank you!  If urgency
> > does develop in the next couple of days, please let me know, and I will
> > see what I can do about moving it to the v6.4 pile.
> 
> Thank you!
> 
> > I got this from "make run" (after merging with v6.3-rc3 as discussed
> > earlier):
> > 
> > 	make[1]: Leaving directory '/home/git/linux-build'
> > 	126 test(s) passed.
> > 
> > This differs from your results, so please see below for the run.out file.
> > (I see 126 instances of "[OK]".)
> 
> Oh you're right! I indeed found no FAIL so it was OK for me and I didn't
> pay attention but it's "just" a matter of message appearing on the console
> in the middle of the test:
> 
>   $ diff -u paul.out willy.out  |less
>   --- paul.out    2023-03-28 20:38:40.079920385 +0200
>   +++ willy.out   2023-03-28 20:39:04.534900530 +0200
>   @@ -130,11 +130,11 @@
>    Errors during this test: 0
>    
>    Running test 'protection'
>   -0 -fstackprotector                                               [OK]
>   +0 -fstackprotector [    2.696920] init (47) used greatest stack depth: 14536 bytes left
>   +                                              [OK]
>    Errors during this test: 0
> 
> We've had a few occurrences of garbaged outputs like this, so I think I
> should improve the test to count OK/FAIL/SKIPPED so that we can be more
> confident in the output when seeing 0 FAIL for example. I suspect that
> above it could be related to the long chain we've seen during the 6.3-rc1
> crash, that went down into the random code, because probably that this
> first-time initialization can enlarge the stack a little bit.
> 
> In my case, I just applied all the nolibc patches on top of 6.3-rc4 to
> run the test so our kernels are slightly different (since my branch
> based on rcu-03.20a did still originate from the 6.3-rc1 thus it was
> failing to boot like you faced previously).
> 
> Maybe I should also improve the grep to try to look for patterns looking
> exactly like this (test numer and name followed by a warning).

Good point!

> > But this from "make run-user":
> > 
> > 	  CC      nolibc-test
> > 	124 test(s) passed
> > 
> > The output of "grep -v "\[OK]" run.out" is as follows:
> > 
> > 	Running test 'syscall'
> > 	18 chroot_root                                                  [SKIPPED]
> > 	43 link_dir                                                     [SKIPPED]
> > 	Errors during this test: 0
> > 
> > 	Running test 'stdlib'
> > 	Errors during this test: 0
> > 
> > 	Running test 'protection'
> > 	Errors during this test: 0
> > 
> > 	Total number of errors: 0
> > 	Exiting with status 0
> > 
> > I am guessing is that this is because I am too cowardly to run this
> > test with root privileges, but thought I should run it by you.
> 
> Yes exactly, that's why I've added getuid() support lately, in order to
> skip these two tests when not root (because I cowardly refuse to run
> this test as root as well and don't want to get used to seeing "normal"
> failures).
> 
> Thus what you're seeing is OK overall. There's just this message that
> appears now on top of -rc4, I'll retry later on top of -rc3 (probably
> not before this week-end, I'm still having other stuff to do), but so
> far so good.

Thank you for looking this over!

							Thanx, Paul