Linux 6.6 To Deal With Unresponsive Intel QAT Devices

[phoronix]

Phoronix: Linux 6.6 To Deal With Unresponsive Intel QAT Devices

Linux has supported Quick Assist Technology (QAT) devices from the start whether it be QAT PCIe adapters or QAT support found within select Atom and Xeon CPUs as well as the latest-generation Sapphire Rapids CPUs. Only now though with the upcoming Linux 6.6 kernel is it adding a heartbeat feature for determining if a QAT device becomes unresponsive so that it can be acted upon...

https://www.phoronix.com/news/Intel-QAT-Heartbeat-Linux

[tildearrow]

Typypo:

a heartbeat feature is being addeded to deal with potentially unresponsive QAT devices.

[tildearrow]

Having recovery is great, but how about you ship stable, bug-free, well-tested firmware?

[anybody]

Having recovery is great, but how about you ship stable, bug-free, well-tested firmware?

From the team that brought us i225 with 3 revisions, all of which were problematic and the first two actually defective… actually working hardware can no longer be expected. Now the CPU has to constantly check if Intel network devices or network accelerators are still alive. This is what >5GHz are meant for.

[arQon]

Yeah, this is just embarrassing. If your "accelerator" is so defective that it locks up at random and needs a watchdog to reset it, not only are you not getting a chunk of the "acceleration" you bought the things for in the first place, but I'm not sure I'd trust it to reliably be producing correct results either - which is kind of a big deal for something that's supposed to be running crypto.

[CommunityMember]

Having recovery is great, but how about you ship stable, bug-free, well-tested firmware?

I would agree. Has your companies software never, ever, needed to ship fixes?

[cynic]

From the team that brought us i225 with 3 revisions, all of which were problematic and the first two actually defective… actually working hardware can no longer be expected. Now the CPU has to constantly check if Intel network devices or network accelerators are still alive. This is what >5GHz are meant for.

I would agree. Has your companies software never, ever, needed to ship fixes?

The need to ship fixes is one thing.
Shipping low quality software to start with is another thing.

[pegasus]

Anyone remembers tcp checksum offload on network cards cca 15y ago? And how often it needed to be disabled as it was causing difficult to trace down mess?