LWN: Comments on "The 3.17 kernel is out"

Broken libata blacklist in 3.17, fixed in 3.17.1

hmh — Wed, 15 Oct 2014 10:59:39 +0000

Fortunately, there was enough time for it to be late-added to 3.17.1, so kernel 3.17.1 fixes the issue.

Broken libata blacklist in 3.17 AND 3.17.1

hmh — Mon, 13 Oct 2014 11:05:00 +0000

The fix is unlikely to be in 3.17.1. Beware.

Broken libata blacklist in 3.17

NightMonkey — Wed, 08 Oct 2014 21:58:04 +0000

I just never use a .0 kernel release. If I'm impatient, I'll go with .1, and if not, wait for .2 and beyond. :)

Broken libata blacklist in 3.17

hmh — Tue, 07 Oct 2014 22:40:07 +0000

please apply:
http://www.spinics.net/lists/linux-ide/msg49669.html

Or wait for 3.17.1 before you try the shiny new kernel if any of your devices need the libata blacklist...

For example, owners of Micron/Crucial MX500 and MX550 SSDs must _not_ use kernel 3.17 without the above fix, on pain of data loss:
https://bugzilla.kernel.org/show_bug.cgi?id=71371

The 3.17 kernel is out

SEJeff — Mon, 06 Oct 2014 18:37:50 +0000

Perhaps you could reproduce it in a QEMU based VM (some of the issues at least)?

The 3.17 kernel is out

deater — Mon, 06 Oct 2014 17:38:04 +0000

The perf_event interface is fragile when it comes to fuzzing, especially with respect to race conditions across forks.

This has been a known issue for a while, and there were some last minute patches that made it into 3.17 final that seem to have improved things slightly.

However there are still various crashing bugs, and they don't seem to be deemed important enough to hold up a release, especially since they aren't technically regressions.

Part of the problem is that it's nearly impossible to make small reproducible testcases for these issues; it doesn't help that perf_event event support varies widely among processors so a simple fuzzer random seed isn't enough to reproduce an issue unless the other side has identical hardware.

The 3.17 kernel is out

rriggs — Mon, 06 Oct 2014 16:36:03 +0000

Does posting this information to KML before the official release announcement goes out help at all?

The 3.17 kernel is out

deater — Mon, 06 Oct 2014 15:32:12 +0000

Yet another kernel release that's trivial to lock up completely by running the perf_fuzzer for a little bit.

I wish it were easier to track down low-level bugs like this.