[<prev] [next>] [<thread-prev] [thread-next>] [day] [month] [year] [list]
Message-ID: <87r0roxmvu.ffs@tglx>
Date: Wed, 10 May 2023 16:49:09 +0200
From: Thomas Gleixner <tglx@...utronix.de>
To: Shanker Donthineni <sdonthineni@...dia.com>
Cc: Yujie Liu <yujie.liu@...el.com>, oe-lkp@...ts.linux.dev,
lkp@...el.com, linux-kernel@...r.kernel.org,
Marc Zyngier <maz@...nel.org>,
Sebastian Andrzej Siewior <bigeasy@...utronix.de>,
Michael Walle <michael@...le.cc>,
Vikram Sethi <vsethi@...dia.com>
Subject: Re: [PATCH v3 3/3] genirq: Use the maple tree for IRQ descriptors
management
Shanker!
On Wed, May 10 2023 at 16:41, Thomas Gleixner wrote:
> On Wed, May 10 2023 at 15:24, Yujie Liu wrote:
> I decoded it by now and that maple_tree conversion is the culprit. It
> broke irq_get_next_irq() which is used during hotplug. It misses every
> other interrupt, so affinities are not fixed up.
I'm seriously grumpy. You throw that untested stuff over the fence,
pester me about merging it and then ignore the fallout.
This breaks cpuhotplug, debugfs, /proc/stat, x86/IOAPIC and some more.
It's not asked too much that if you change an iterator implementation to
validate that the outcome is still the same on the usage sites.
That change has never seen CPU hotplug testing. It reproduces
instantaneously in a VM even without running blktest.
I grant you that the documentation of mt_next() is incorrect, but that's
absolutely no excuse for not testing such a fundamental change at
all. It's neither an excuse for ignoring the fallout and wasting other
peoples time.
I'm dropping this from my to-merge list.
Yours grumpy
Thomas
Powered by blists - more mailing lists