On 8/4/21 10:47 AM, Sebastian Andrzej Siewior wrote: > On 2021-08-04 10:22:59 [-0600], Jens Axboe wrote: >> >> In that regard, I do still consider those patches out-of-tree, which >> they are. And while I'm more sympathetic to them compared to other >> out-of-tree code as there's a long term plan to get it all in, it's >> still out-of-tree. Best solution here is probably to just carry that >> particular change in the RT patchset for now. > > So today in the morning I learned that there is a memory allocation in > an IRQ-off section and now, a patch later, it is almost gone. So that > makes me actually happy :) 1 out of 2 is better than 0 ;-) > The spin_lock_irq() vs local_irq_disable() + spin_lock() is documented > in Documentation/locking/locktypes.rst. > That said I have no problem by carrying that patch in the RT-patchset > and revisit it later. Right, I suspect that was added as a pre RT patch dump at some point. It's a newer thing. Is it actually possible to set PREEMPT_RT in the mainline kernel? Looks like it depends on ARCH_SUPPORTS_RT and nobody sets that. So I agree that just carrying your solution in the RT patchset is fine for now, we can revisit later. -- Jens Axboe
