USB notes

From Helpful
Jump to navigation Jump to search


See Common plugs and connectors#USB

Hardware variants

PC controllers will be one of

  • OHCI (Open Host Controller Interface)
    • USB 1
    • Open standard
  • UHCI (Universal Host Controller Interface)
    • USB 1
    • Proprietary standard
    • Intel and VIA are usually UHCI
  • EHCI (Enhanced Host Controller Interface)
    • USB 2
    • public (open?) standard
  • XHCI (eXtensible Host Controller Interface)
    • USB 3
    • in theory replaces OHCI, UHCI, and EHCI

See also:


  • USB1 "Low speed"
    • ~178KByte/s (1.5 Mbit/s) of wire speed, max max ~120KByte/s of carried data due to 8b/10b coding (not yet accounting for protocol overhead)
    • Meant for devices that will never need much speed, such as keyboards
  • USB1 "Full speed"
    • 1.5MByte/s (12 Mbit/s) wire speed, ~1.2MByte/s of carried data
  • USB2 "High speed"
    • 60MByte/s (=480Mbit/s) wire speed, ~48MByte/s carried data
    • expect no more than 30MByte/s, and 25MB/s is more common (apparently because of offloading(verify))
  • USB3 "Superspeed+"
    • 625MByte/s (5Gbps) of wire speed, ~500MByte/s (4Gbps) carried data
    • Note: May do USB2 devices a little faster (as in, may reach 30-40MByte/s where many USB2 controller may stop at ~25MByte/s) (verify)
    • expect no more than ~400MB/s usually
  • USB3.1 "Superspeed+"
    • 1250MByte/s (10Gbps) of wire speed, before 128b/132b coding, ~1200MByte/s (~9Gbps) carried
    • expect no more than ~900MByte/s usually


  • Hubs share speed.
  • PC USB ports often come as two ports side by side on a hub, so share their speed
This means e.g. copying between two adjacent ports may happen at half the speed you expected.
(at least up to USB2 - USB3 became smarter about a few cases(verify))
  • USB3 modes are full-duplex, everything before was half-duplex
  • I'm ignoring USB4 for now

Related to use

Unplugging USB

This article/section is a stub — probably a pile of half-sorted notes and is probably a first version, is not well-checked, so may have incorrect bits. (Feel free to ignore, or tell me)

"Do you need to 'Safely Remove Hardware'?"

It's often not necessary.

...yet if you want to be sure, it's still a good idea to do. And somewhat simpler in that the explanation why is a bit longer than you want to remember.

At an electric level, and protocol level, USB was made to be hot-pluggable. Nothing will spark, and the driver and computer shouldn't crash.

Most drivers for USB devices are also forgiving about disconnect and reconnect - they mostly have to be.

Higher-level APIs (like sound and video) didn't use to be made with changing devices in mind. More recent variants are better.

Apps may still not be happy, but not dealing with this is to some degree considered a lazy bug.

And to some degree a question of "if it was still doing something, was that important to you?"

Say, a sound card, webcam, keyboard or mouse, even if it's still transferring, probably doesn't do or store anything you would care about losing.

Storage, on the other hand, often does - anything from a USB stick, memory card, external hard drive, connected cameras in storage mode.

Disconnecting a drive while running originally wasn't a thing, so filesystem APIs assume that a file stays open and available until the program that opened it says it is done.

Yanking while reading from a external storage will almost certainly make the reading program unhappy, yet the storage itself is typically fine.

Yanking while writing is usually bad.

Most filesystems write file data in-place (meaning they do not implement a guaranteed recovery)
A number of filesystems are a little more robust about filesystem metadata - but you shouldn't assume it

There are further notes.

Consider the following distinction

  • writeback cache describes the practice of an OSes of keeping a buffer "stuff to be written" the size of a few seconds work
it lets the system say "okay program I will do that for you (soon)" instead of making the program wait until it has actually done so (makes the program a little more responsive, and if the program really cares to say "now I need it on disk NOW" it can request that)
if the program writes a sequence of small chunks ("write 1K, write 1K, write 1K, write 1K"), the OS can merge that into a single "write 4K" - and related cleverness
  • writethrough descibes the OS treating a devices as "write any data as soon as possible"
This is often slower, but typically safer (because data doesn't hang around that system buffer where e.g. power loss would mean data loss)

When an OS knows a medium is removable, OSes usually opt for the safer-and-slower writethrough for external storage devices, but not always, and you can override this.

This is roughly why "if the USB stick it's no longer blinking it's probably safe enough to yank" is true enough

... in that it's probably in a consistent state. The filesystem may still record that you didn't unmount it, and the next system may read that out and want to do a consistency check - but it will probably find it perfectly fine.

USB flash corrupts a little more easily while writing, because of how Flash works: writing small pieces of data to flash memory involves a read-erase-write cycle, leaving time in which the data is not on storage, and a non-obvious state to recover (SSDs, while based on Flash as well, have a more complex, smarter controller).

USB Device Not Recognized / Device has malfunctioned

Shown in a balloon near the task bar, with the addition that "One of the USB devices attached to this computer has malfunctioned, and windows does not recognize it."

(In some cases, if the balloon message is disabled, the device only shows up as an "Unknown Device" in the device manager.)(verify)

My summary so far is that there are many possible causes, it's hard to say which one it is, and most solutions are not very clearly good ones. Reinstalling usually won't help.

Possible problems/solutions

Things you can check easily enough:

  • broken device
    • test: try in other computers - and to be sure, also with varying OSes (some devices don't have drivers for older/specific OSes)
  • broken cable (cables get stepped, yanked, etc.)
    • if it's interchangeable, try another one.
    • If it's not interchangeable, try wiggling it to see if it accidentally makes a good contact. It's not a very good test in that it not working means nothing, but if it turns up temporarily, you know that a bit of DIYing may well fix the problem.
  • broken USB socket
    • Test: Try other things in the same port, and try this device in other ports
    • Note: each pair of ports is usually the same internal hub, so only trying the one right next to it is not the best test.
    • Basic mice are the most useful test devices, as they are low-power and do not require drivers to work
  • USB port in a weird state (e.g. the previous device drew too much power and triggered the polyfuse. But there is another reason or two)
    • (some are more sensitive/careful than others. Macbook polyfuses were apparently known for triggering quickly)
    • Test/fix: Turn computer off completely, and then leave the power cord/battery out for a few minutes (that much is rarely necessary, but sometimes it is and it takes less time to wait a bit longer once than to reboot again)
  • USB knows you are crossing the limit of power it can give out, and refuses the next device
    • more likely to happen on a non-powered hub
    • Test: try plugging the device into the computer itself (or into a powered hub that has no high-powered devices)
      • Front ports may sometimes be non-powered hubs, so a back (motherboard) port may be a better test.
    • Test: try a mouse in the same port - it's low-power and so very likely to work if the port's not broken.

Harder to check well:

  • stupid/missing driver for the particular device
    • Test: Try again after a restart, try with different OS, try with a different driver,
  • stupid power management bug (happens on some hardware)
    • There is/was a bug in XP(/Vista?) where power-suspended USB ports are not re-enabled properly, and there were mixed reports about Microsoft's patches (verify). Some people report that disable this port suspending feature helps.
  • windows USB driver set/setup got wonky.
    • Bit of a last resort thing to check
    • Test/fix: Uninstall as many related drivers as you can before plugging the devices back in.
  • some devices don't like sleep/hibernate, regardless of version of anything
and may stay confused until a reboot

It seems that certain devices are likelier to trigger this problem.

It seems that Windows's collection of drivers for USB devices (one or more) can get wonky. It may be that reinstalling all of these can help. Remove controllers drives last, so peripherals drivers first. Then rebooting and let it re-detect and re-install everything.

See also

Speed and latency

This article/section is a stub — probably a pile of half-sorted notes and is probably a first version, is not well-checked, so may have incorrect bits. (Feel free to ignore, or tell me)


"Do hubs, shared cables, etc. affect speed?"

On poll rate

Can we go lower?

Why do some devices only work on PC ports, not from hubs?

Notes on slow USB3

On USB power and charging

This article/section is a stub — probably a pile of half-sorted notes and is probably a first version, is not well-checked, so may have incorrect bits. (Feel free to ignore, or tell me)

USB only accidentally became a standard for charging things, which is why the history of transferring more power through it is interesting.

Shortish version


  • various things may choose to charge slower
  • without PD, the specific case may limit you to USB2 and USB3 limits (500mA; 900mA or 1,500mA for dual-lane), or not (3A) (verify) 5V, that means one of so 2.5W, 4.5W, 7.5W, 15W
and standard cable, it probably won't try for more than 3A at 20V (for 60W)
and a heavier cable, it can try up to 5A at 20V (for 100W)
..but there may be practical reasons any one case may not go that high - like that most devices don't need that

This means laptops can now use one of their USB-C plugs as a power plug, which is nice to minimize the amount of sockets, but note that not all chargers or cables will be equal here, and it's possibly there are preferred ports to do this on(verify).

before USB-C

  • On a PC:
a USB1 or USB2 root hub more or less guarantees 100mA, and may supply up to 500mA after negotiation (0.5A at 5V is 2.5W)
USB3 guarantees 150mA and may supply up to 900mA after negotiation
...'may', because the PC may deny it when it knows the total device draw is more than it has, which happens more easily when you add external hubs

  • generic wallplug chargers
when something doesn't present as a USB host at all, it's up to devices to behave reasonably
if a device cannot or does not check capability, they often draw at most 0.5A, but often less.

  • specific chargers
proprietary, often specific to brands
amounts to specific devices checking for specific chargers.
and sometimes negotiating for a higher voltage as well.
they may draw up to 1A or 2A, at 5V or sometimes higher.
The range here is 5W to at most 20W -- largely because wires used in USB2 cables cannot support more.
there were various non-standardized ways of doing this (many involved putting some voltage dividers on the data lines) - Battery Charging came later.
Names seem to include
QuickCharge (from Quallcomm)
TurboPower (from Motorola)
Adaptive Fast Charging (from Samsung)
SuperCharge (from Huawei)
Pump Express (from Mediatek)
PowerIQ (from Anker)
VOOC (from OPPO), a.k.a. Dash Charge
AiPower (from Aukey)
VoltIQ (from TronSmart)
iSmart (from RavPower)
...and more variations, as well as more rebrandings. These still tend to be designed for at most 10W, 15W or 20W (because of the mentioned wire thing) but are more likely to actually do more than standard 2.5W.
introduced later (well after many not-so-standards, mentioned above)
up to 1.5A (which is 7.5W), but devices may play it safe and don't try more than 0.5A (2.5W).
no specific logo, either labeled or you have to know(verify)
I should read it, but I'm guessing it says you should only draw current until the voltage starts drooping - basically not going beyond current capability

Using USB to charge batteries

In general, don't count on classical USB supplying more than 2.5W (standard)

...or 5W or 10W if slightly special (not so standard).


phones may have batteries on the order of 10Wh, that means charge will rarely take less than 4 hours - and easily double that when some part of the system is being careful
laptops have batteries on the order of ~100Wh, that means a charge would take dozens of hours -- which is why we don't use USB to charge them (before USB-C, anyway)

PC port limits

It's often mentioned that USB ports have a polyfuse - a physical component specced so that drawing more than e.g. 500mA will lead to that power being disconnected.

(There are other solutions. A few implementations actually monitor current, and use transistors to disconnect(verify), and in some rare cases the protection is missing)

Implementations play it a little wider than that, to avoid disconnecting devices due to transients and slight deviation -- because polyfuses may, by nature, not reset until power is removed, which is why you sometimes need to completely power off a device to get that port working again.

So USB2 polyfuses seem to have a trip current around 1.5A, and also a higher-than-500mA hold current (I've seen 750mA mentioned)

Roughly twice that for regular USB3 ports (ports can supply 0.9A as per specs)

More so for charging ports (how much?(verify)).

DIYers will know from trial and error that you can get away with just ignoring the intended negotiation and drawing 0.5A from a USB2 PC port without requesting it (0.9A from USB3 as per specs)

...because most hosts/hubs just assume USB devices act according to specs, they assume you would always ask not just draw it. And if you ever made your project into a product, it would need to actually request it to pass USB certification.

Note also that since you are defeating the thing that lets USB guarantee power is actually available, so by cheating this way may be able to cause all USB devices that sharing that power to brown out, something normally prevented by USB.

Hub limits

USB-C won't charge

USB-C to USB-C doesn't charge

USB-A to USB-C doesn't charge


Why do some cables charge faster than others?


Related to coding

on libusb and platforms

A primer, some terminology


Keyboard limitations

Can I write and use my own USB drivers for a device?

Can I give my DIY device a custom name?

This article/section is a stub — probably a pile of half-sorted notes and is probably a first version, is not well-checked, so may have incorrect bits. (Feel free to ignore, or tell me)

tl;dr: not really

This would mostly just matter to what is shown in device manager. This might be nice for the convenience of admins.

Or, in theory, to avoid getting confused which of the many attached Arduinos you need to upload to, except that the name for a device is essentially settled by its VID and PID, so that won't work.

In Windows there is a registry trick to effectively override the name reported by a driver - but that would be for all devices that that driver applies to.

To get it to look distinct, you would really need a different PID. Which which reduces the answer to the one as "can you supply your own USB drivers", because that's roughly what it would amount to.

That said, changing only the name apparently means you don't need to alter the driver itself, just the INF around it[1] which is within reach.