Thunderbolt Networking Support Is Still Being Worked On For Linux

[wizard69]

I already have Ethernet.
Ethernet also have the advantage of being more secure as it does not have direct memory access (DMA).

Where are you getting this nonsense. DMA combined with Ethernet has bee a reality for decades now. Todays Ethernet controllers are pretty powerful chips in their own right.

[wizard69]

I really don't understand why Intel has let Thunderbolt continue for as long as they have, especially since it competes with their own technologies (like USB).

TB was never meant to compete with USB. It is an entirely different technology solving different technical issues.

There are hardly any compatible devices worth considering, it has been littered with problems in both Mac and Windows,

And USB hasn't? Frankly it took a long time for the first rounds of USB chips to evolve to the point of working correctly. This especially with USB 3 chips as many of the early ones where very buggy.

and the hardware evolves so quickly that it isn't really worth buying anything for it. Also, it seems to be less common than eSATA, which is saying a lot. It had some interesting potential,

It still has a lot of potential and frankly is being implemented in many applications where it makes sense. USB3 / TB will only take it further as the special external hardware is eliminated.

with things like external GPUs or daisy chaining a single cable for all external devices. But that stuff just didn't take off.

The idea of external GPU's is some idiots pipe dream. To put it bluntly it doesn't make any economic sense. There are rumors of Apple integrating a GPU into a video monitor but even that will likely blow out the cost of the monitor to the point it Never becomes popular.

Thunderbolt is effectively the successor to Firewire, and somehow Intel didn't realize that Firewire died off because nobody cared about it.

I don't see it that way. TB is effective a replacement for PCI-Express especially on devices that don't have room for traditional PCI Express slots. If you look at it that way you can see why nobody builds TB mice or TB based printers. The port simply was never intended to replace USB.

[wizard69]

I thought a tablet would be more likely to have a USB port than a Thunderbolt port.

Currently that is true.

Most tablets are ARM and I don't think there is any Thunderbolt-equipped ARM device.

Not at the moment but there is apparently noting stopping somebody from developing alternative hardware compatible with ARM processors.

Most smartphones and tablets seems to heading toward USB Type-C.

Yes but USB-C is a conglomeration of standards that includes TB. I suspect that this will end up confusing consumers more than anything. It is actually a port standard that also happens to support USB 3.x

[totoz]

My new Clevo laptop (P751DM) has a Thunderbolt 3/USB 3.1 combo port. Currently, I have neither TB3 nor USB 3.1 device to test with, but I wonder what it should look like in dmesg/lspci/lsusb. I don't know what to look for, and I have no idea about the name of the hardware implementing this combo. If someone has a hint, I'm interested.

At the moment (kernel 4.6), the only difference I could notice by disabling/enabling TB3 was an extra PCI bridge, but I don't know if it's enough to prove some TB3 support. And for USB 3.1, no sign of 3.1 anywhere.

[starshipeleven]

My new Clevo laptop (P751DM) has a Thunderbolt 3/USB 3.1 combo port. Currently, I have neither TB3 nor USB 3.1 device to test with, but I wonder what it should look like in dmesg/lspci/lsusb. I don't know what to look for, and I have no idea about the name of the hardware implementing this combo. If someone has a hint, I'm interested.

At the moment (kernel 4.6), the only difference I could notice by disabling/enabling TB3 was an extra PCI bridge, but I don't know if it's enough to prove some TB3 support. And for USB 3.1, no sign of 3.1 anywhere.

Any controller of USB and thunderbolt should show up with lshw run as root, as that should print all hardware info.

If they don't show up you probably lack drivers for themso they aren't even initialized/recognized.

If you want to post the output here I can have a look and see if something looks like them. Just redact the MAC addresses from the ethernet entries if you post here the full dump.

[W.Irrkopf]

The ethernet controller has DMA, direct memory access.
Whatever is connected to ethernet cable does not have DMA and can only talk to the ethernet controller.
Any kind of communication that does not broadly resemble ethernet can't get past the ethernet controller as it won't understand it.

Thunderbolt has pcie lines, and pcie lines allow ANYTHING you connect to them (an external port) to have DMA.

So yeah, it's not the same thing at all.

Not according to this entry on Wikipedia:

However, where Thunderbolt is used as a system interconnection (IPoTB supported on OS X Mavericks), then the IP implementation runs on the underlying Thunderbolt low-latency packet-switching fabric, and the PCI Express protocol is not present on the cable. That means that if IPoTB networking is used between a group of computers, there is no threat of such DMA attack between them.

[starshipeleven]

Not according to this entry on Wikipedia:

My point was that thunderbolt allows pcie lines and thus DMA to be accessed from an external port.

Of course while the cable is in use to do something else that does not involve using pcie lanes (like also the displayport function or the USB 3.0 one), the pcie lines are not used thus there is no risk of DMA.

Point is, how do you tell that a device using a Thunderbolt port isn't malicious (i.e. using pcie lanes to do bad things too)? You break out your hardware logic analyzers and sniff the communication?

[schmidtbag]

TB was never meant to compete with USB. It is an entirely different technology solving different technical issues.

I'm aware. I said before that it's the successor to Firewire. But to the consumer, it effectively fulfills most of the same purposes as USB.

And USB hasn't? Frankly it took a long time for the first rounds of USB chips to evolve to the point of working correctly. This especially with USB 3 chips as many of the early ones where very buggy.

I don't really remember how the first generations of USB performed, but I don't recall USB 1.0, 1.1, or 2.0 having issues that resulted in kernel panics or significant data loss. As for USB 3.0, the only issues I've ever heard of were disappointing data rates or high CPU usage.

It still has a lot of potential and frankly is being implemented in many applications where it makes sense. USB3 / TB will only take it further as the special external hardware is eliminated.

Agreed - TB does have a lot of potential, and it always did. But nobody is taking advantage of it in ways that USB 3.x can't do either. Sure, maybe some tasks TB can do better but it's not usually worth the price difference. As much as you like to remind me that TB and USB aren't the same, the fact of the matter is most things that TB can do, another interface (such as USB) can also do equally well for a lower cost and greater availability.

The idea of external GPU's is some idiots pipe dream. To put it bluntly it doesn't make any economic sense. There are rumors of Apple integrating a GPU into a video monitor but even that will likely blow out the cost of the monitor to the point it Never becomes popular.

You apparently don't know what the home PC market is like. The vast majority of home users prefer laptops. Laptops tend to have real crappy GPUs, and even when they're good, they're throttled to meet power and thermal conditions. There is a huge market for external GPUs, to the point that there have been multiple attempts. The problem is there has never been an easy or cost-effective way of dynamically adding a GPU to a computer. Seeing as TB is, in one perspective, just an external PCIe slot, it is one of the best interfaces to do an external GPU. With the exception of ExpressPort, there has never been another interface that was even remotely capable of handling even the crappiest of GPUs, usually because of latency, CPU overhead, bandwidth, cost, or any combination of those. TB has everything needed to do an external GPU setup, with the exception of widespread availability.

[totoz]

Any controller of USB and thunderbolt should show up with lshw run as root, as that should print all hardware info.

If they don't show up you probably lack drivers for themso they aren't even initialized/recognized.

If you want to post the output here I can have a look and see if something looks like them. Just redact the MAC addresses from the ethernet entries if you post here the full dump.

Here it is: http://pastebin.com/1tKNdx2L

[starshipeleven]

Here it is: http://pastebin.com/1tKNdx2L

I'm not seeing anything that looks like it. I think it isn't recognized at all (how unexpected).

Looking around, it seems that another guy with a skylake thunderbolt system can see the thunderbolt in linux ONLY if at boot there is something connected to the thunderbolt port https://bugzilla.kernel.org/show_bug.cgi?id=115121

If you look at his "lspci-connected-at-boot.log"

you see the last entry (not present in the other log)

"06:00.0 PCI bridge: Intel Corporation Device 1576 (prog-if 00 [Normal decode])"

the device "1576" happens to be a thunderbolt controller, "DSL6340 Thunderbolt 3 Bridge [Alpine Ridge 2C 2015]" https://pci-ids.ucw.cz/read/PC/8086