Slashdot Mirror
USB 3.2 Work Is On The Way For The Linux 4.18 Kernel: Report (phoronix.com)
An anonymous reader shares a report: USB 3.2 was announced last summer as an incremental update to the USB standard to double the bandwidth for existing USB Type-C cables. We haven't seen much in the way of USB 3.2 mentions in the Linux kernel yet but then again we haven't really seen USB 3.2 devices yet. USB 3.2 brings a multi-lane operation mode for hosts and devices using existing Type-C cables as well as a minor update to the USB hub specification. USB 3.2 allows for new 10 Gbit/s and 20 Gbit/s rates using two lanes, USB 3.2 Gen 1x2 and USB 3.2 Gen 2x2, respectively. It looks like kernel developers are now working on getting their USB 3.2 Linux support in order. We were tipped off that as of last week there are some USB 3.2 patches queued in the usb-next tree maintained by Greg Kroah-Hartman's.
Not really.
A parallel bus uses a parallel data lane: an 8-pin parallel data lane sends one byte by sending 8 bits all at once, such that the single clock across all buses synchronizes all bits.
A multi-lane serial bus is sending data in packets, such that the data coming down any one lane is self-consistent. For example: bonded ethernet adapters send entire frames down each link, rather than spreading a frame out across multiple links in an alternation of bits.
That means any one lane is sending a complete signal, and any interference causing errors down another lane don't affect the unaffected lane (in parallel buses, an error in one line would affect the entire signal: if you send a packet and one line has noise, you get an erroneous packet--all of the data sent down all lines is erroneous, even though most lines are noise-free).
Weird, huh?
Support my political activism on Patreon.
The printer ports or the ISA bus on your old computer that is gathering dust in a corner of your basement are true parallel connection:
Each bit of a 8bit byte is traveling in parallel in a neighboring conductor.
(But this would be a technical nightmare on modern speed used in modern system, as you'd need to make sure that all the bits arrive at the exact same time in perfect unison at the destination. lenght of conductor and speed of light/speed of electricity in a medium start to matter a lot.
See the complicated squiggly traces between DIMMs and memory controllers to make sure that the path takes the exact same amount of time.
That's why it's being abandoned in modern buses)
DVI, SATA-Express, PCI-Express, USB 3, etc. have all multiple conductor, but they are all in essence still serial.
Each bit of a word is travelling one after the other in the same conductor.
And the above standard just happen to have multiple serial-links that can be used concurrently : the system can send multiple data packet more or less at the same time, each sent serially along a different line.
(And in the cases of standards such as PCIe, the various serial links might not be even talking to the same device).
Unlike the parallel situation, you do not need to make sure that the bits travel in unison down different conductors : each serial link is sending different data packets, they can arrive with slightly different timing.
(Basically, is like plugging 2 network cables in 2 ports of the same server using link aggregation. You didn't suddenly turn ethernet into a parallel bus, you just have more interfaces to spread the load of sending your packet over).
So no. Latest iteration USB are still serial. They can just have more independent serial-links used concurrently, but the bits still travel one after the other on the same link, the other link is used to send a different data packet concurrently)
At best you could invent a new term like "multi-serial" or "concurrent serial".
"Sufficiently advanced satire is indistinguishable from reality." - [Tips: 1DrYakQDKCQ6y52z6QbnkxHXAocMZJE61o ]
We've gone from 480Mb/sec up to 6Gb, 10Gb, now 20Gb, and we still can't use a native USB cable to network two linux boxes.
Thunderbolt on Mac has this, and there was a very limited solution for linux that was never production ready.
We're still stuck at 1Gb/sec between machines for any networking that isn't cost-prohibitive, impractical, or both.
The cables exist, I look forward to the day when I can plug two machines together and run a network between them over this cable.
Will this ever happen? Or is this a case of hardware vendors blocking open source so that they don't compete with their own jurassic and overpriced 10Gb products?
Aside from people just fucking it up(downright ubiquitious in USB power delivery scenarios, sometimes of the 'likely hardware damage' flavor) the trouble is that the alt-modes move what a USB port might be capable of outside the realm of the USB implementation itself.
Alt mode support is a USB thing; so compliant USB chipsets Must correctly handle the relevant signalling and handing off; but the behavior of whatever is handed off to is outside the USB spec and often in the realm of parts that are expensive enough that there is no way the USB-IF could get any serious traction even if they tried to twist arms.
Even on expensive gear with substantial punch there typically aren't enough resources for the alt-modes to be supported without various compromises and non-obvious limitations (it isn't a 100% perfect match; because the Mac Pro predates USB-C connector thunderbolt; but it's a good example of the mismatch between various types of resources at the system's disposal: it's sorta plug and play; but "attach displays to different Thunderbolt busses when possible. Don't attach more than two displays to any bus"). It's quite typical for a computer to have 10 or more USB ports; Even if the Thunderbolt silicon were free, which it isn't, it's not so typical for a computer to have an extra 20 or 40 PCIe lanes; nor is it common to have more than two or three displayport outputs(unless you are loading up on GPUs or using classy workstation cards). Even if the extra expense and board space is provided so that any USB-C port can be used for any alt-mode for which system resources are available, that still means that "you can only plug monitors into two of them; unless you have two displayport monitors and one HDMI, in which case you can use three; but if you do that only Thunderbolt devices that use PCIe exclusively will work on the remaining ports, no Thunderbolt displays" is the order of the day.
Once you get down to realistically cheap devices; where things like not having a PCIe bus; or only having HDMI because you are a cheap SoC designed for set top boxes; or not wanted to put a fancy matrix switch in so that multi-Gb/s busses are available from any USB port are to be expected the limits of what any given USB port can do are just going to get tighter.
If(and it's a big if) some reasonably sane scheme is designed to make it clear what a device can or can't do it's still arguably a win over fixed function ports(if a laptop, say, can only actually charge via one of its ports and output video over one or two others; that's still a laptop that has USB ports where it would otherwise have a vendor barrel plug and an HDMI port, which is nice to have); but it's pretty much impossible to expect reasonably comprehensive support out of devices without making them mostly expensive and overqualified.