On 10/2/22, 8:30 AM, "Greg KH" gregkh@linuxfoundation.org wrote:
CAUTION: This email originated from outside of the organization. Do not click links or open attachments unless you can confirm the sender and know the content is safe.
On Thu, Sep 29, 2022 at 09:06:45PM +0000, Rishabh Bhatnagar wrote: > This patch series backports a bunch of patches related IRQ handling > with respect to freeing the irq line while IRQ is in flight at CPU > or at the hardware level. > Recently we saw this issue in serial 8250 driver where the IRQ was being > freed while the irq was in flight or not yet delivered to the CPU. As a > result the irqchip was going into a wedged state and IRQ was not getting > delivered to the cpu. These patches helped fixed the issue in 4.14 > kernel.
Why is the serial driver freeing an irq while the system is running? Ah, this could happen on a tty hangup, right? Yes, exactly during tty hangup we see this sequence happening. It doesn't happen on every hangup but can be reproduced within 10 tries. We didn't see the same behavior in 5.10 and hence found these commits.
> Let us know if more patches need backporting.
What hardware platform were these patches tested on to verify they work properly? And why can't they move to 4.19 or newer if they really need this fix? What's preventing that?
As Amazon doesn't seem to be testing 4.14.y -rc releases, I find it odd that you all did this backport. Is this a kernel that you all care about?
These were tested on Intel x86_64 (Xeon Platinum 8259). Amazon linux 2 still supports 4.14 kernel for our customers, so we would need to fix that.
thanks,
greg k-h