USB discussions getting a bit heated

From what I've read, it seems that the fundamental problem is that the BCM2835 SoC has a bare-bones USB host controller which requres a lot of the ARM processor.  Such a USB host controller works fine in BCM2835 intended applications, such as Roku, but if you try to use the ARM11 as a high-performance GNU/Linux engine talking to dozens of USB devices you will run into limitations.  It also sounds like documentation for the SoC's USB host controller is not available and that the low-level "open source" software suffers from "security through obscure code".  If this is indeed the case, it will be hard to improve things.

Please let me know if my impressions are incorrect.  I would be delighted to hear that open documentation for the USB host controller is available (link?) and that the low-level driver is perfectly understandable and well-documented (link?).

Apparently the documentation for the USB controller isn't even available

to Broadcom, and they are apparently trying to figure out how it works by

reading the code for the driver.

One possible problem is an interaction with something in the SD card code that uses a long timeout causing USB packets to be lost https://github.com/raspberrypi/linux/pull/72 but again, those fixes haven't been pulled yet, so even if that's the major fix very few people will be running it.

Either way, the synopsis OTG controller seems to have a lot of problems, but also a lot of users.  There were some rumblings a while back that the mainline kernel folks were looking at writing a better driver. That'll take time though and would likely be written against whatever the then 'current' kernel is, not the already out of date 3.1.x. There's a lot of churn in mainline arm right now with devicetree and such like, so forward or back porting stuff really isn't simple..

they are apparently trying to figure out how it works by

reading the code for the driver.

Some people, outside the rpf, looked at doing that to get a working driver into the 3.5 porting effort. They gave up after a while, the code seems to be a steaming pile of.......

John Beetem wrote:

 

From what I've read, it seems that the fundamental problem is that the BCM2835 SoC has a bare-bones USB host controller which requres a lot of the ARM processor.  Such a USB host controller works fine in BCM2835 intended applications, such as Roku, but if you try to use the ARM11 as a high-performance GNU/Linux engine talking to dozens of USB devices you will run into limitations.  It also sounds like documentation for the SoC's USB host controller is not available and that the low-level "open source" software suffers from "security through obscure code".  If this is indeed the case, it will be hard to improve things.

This is EXACTLY the same conclusion that I have reached.

The Pi is in reality not a general purpose computer.  It's half a computer, with the ARM being used to fill in the hardware gaps that let the SoC be designed at the lowest possible cost.  If you try to use the ARM at the same time for normal computing, you can kiss your SoC's correct functioning goodbye.

It's this that is wreaking havoc when you attach a correctly-working self-powered hub with added devices downstream.  Linux accepts it, because it doesn't realize what the purpose of the ARM is within the BCM2835.  In reality it should say "Sorry Dave, I don't have the CPU headroom for that" and refuse to attach.

Morgaine.

it should say "Sorry Dave, I don't have the CPU headroom for that" and refuse to attach.

LOL...  but does sound work well enough for it to sing the song ?

Those threads are terrible They downplay most of the problems people have encountered and add confusion.

But it's a timebomb waiting to go off...

I guess user Kernel sums it up pretty nicely http://www.raspberrypi.org/phpBB3/viewtopic.php?f=28&t=12097&start=197

 

Frankly even though the Foundation people are not making profit on Raspi, I think it doesn't make them any less responsible for the quality of the device they designed and have been (indirectly) selling.

 

Regardless of the cause (software, power, design) of Pi's problems, they are frustrating to deal with and are likely to cause many to lose interest at some point (even children, or perhaps especially them).

 

Thanks ! this is very informatic.

Those threads are terrible They downplay most of the problems people have encountered and add confusion.

Thing is, the issues are complex. A lot of the confusion is being caused by people who mostly don't have a grasp of either the hardware or software problems.  Lets face it, with a usb IP block that nobody seems to have much information about, we're all in the dark and what makes it tick may as well be black magic

Still, at this point the Pi really is a development board - there's lots of development happening and various people working on fixing issues. Unfortunately due to all the hype there's lots of them landing in the hands of people who are used to a device that's been in development for years with budgets in the billions and expect everything to 'just work' instantly, instead they have a device that was designed to a strict budget with all the compromises that entails. This was bound to cause some friction.

I suspect the RPF was very much unprepared for the amount of destructive stress testing a couple of hundred thousand users can do in such a short time with devices that were never thought of during the design.

selsinork wrote:

 

I suspect the RPF was very much unprepared for the amount of destructive stress testing a couple of hundred thousand users can do in such a short time with devices that were never thought of during the design.

I think everyone would agree that the Foundation was unprepared.  The Pi doesn't work properly even with very common devices and very unambitious setups that work on all other systems tested.

It should be pointed out that this isn't a case of the community being over-demanding, and the complaints aren't coming from people intent on destructive stress testing.  They come from ordinary users who are expecting (shock horror!) that when they attach a common USB device and it is recognized by an available driver, that it should then work.

Morgaine.