[go: up one dir, main page]

|
|
Log in / Subscribe / Register

Device complexity

Device complexity

Posted Sep 22, 2024 7:01 UTC (Sun) by marcH (subscriber, #57642)
Parent article: Considering kernel pass-through interfaces

> But, he said, a device driver's job is to configure hardware properly; he wondered why this additional interface was needed. Gunthorpe answered that modern devices are hugely complex and must be configured to work within the environment in which they are used.

The term "device" has become somewhat misleading. A "device" sounds smaller and simpler than the CPU that "drives" it - because devices all used to be. But today a GPU has more TFLOPS drawing more power than the CPU. It may even run more code.

The Central "Processing" Unit is increasingly looking like a central.... "network hub"? You wouldn't want a hub requiring very fine-grained knowledge of the traffic going through it.


to post comments

Device complexity

Posted Sep 22, 2024 8:39 UTC (Sun) by Wol (subscriber, #4433) [Link] (1 responses)

> The Central "Processing" Unit is increasingly looking like a central.... "network hub"? You wouldn't want a hub requiring very fine-grained knowledge of the traffic going through it.

This sounds like we're moving to the mainframe model, where the CPU is one of the most under-powered parts of the system. Making it gate-keep everything else sounds like a great recipe for making linux obsolete.

Cheers,
Wol

CPU as gatekeeper

Posted Sep 23, 2024 0:04 UTC (Mon) by ringerc (subscriber, #3071) [Link]

Have been for decades.

SCSI, MMU, IOMMU, DMA, nVME, packetized PCIe, it's all been a trend toward the CPU handling access control, configuration and coordination.

With a few unfortunate missteps like some FireWire implementations' "plug in a device and do whatever you want with host host memory" it has worked well.


Copyright © 2026, Eklektix, Inc.
Comments and public postings are copyrighted by their creators.
Linux is a registered trademark of Linus Torvalds