RPI 4B on UEFI: xhci0 disconnects under high load

[Mark Millard]

On 2020-Sep-25, at 00:58, Robert Clausecker <fuz at fuz.su> wrote:

> Hi Mark,
> 
> Thanks for your quick response!  It appears that I had missed that this
> changeset was required.  Let me update to the most recent revision and try
> again.

My context is based on head -r365932 . In github terms, at:

https://github.com/freebsd/freebsd/commit/173c619

After that I do not know if anything new interferes.

> Yes, I have been using the github mirror.  Are there any other
> patches I should consider applying?

Most of my patches are for powerpc64 and powerpc (old PowerMacs).
The only aarch64-related patch I have is for:

/usr/src/sys/dev/acpica/acpi.c

from https://reviews.freebsd.org/D25219 . But you have
reported having this one in place. As I remember it is
required to have rpi4-uefi-devel work at all. But it
still requires that the uefi be configured to limit the
RAM to 3072 MiBytes if you want reliable behavior for
xhci use: FreeBSD does not correctly respect the DMA
limitations for xhci use for ACPI based booting.

(I have a type of test that fails without the 3072 MiByte
limitation imposed.)

You have reported having other patches in place that I
do not have. I do not know about the status of those.

> Yours,
> Robert Clausecker
> 
> On Thu, Sep 24, 2020 at 06:19:36PM -0700, Mark Millard wrote:
>> 
>> 
>> On 2020-Sep-24, at 15:47, Robert Clausecker <fuz at fuz.su> wrote:
>> 
>>> Good evening!
>>> 
>>> I have set up a FreeBSD system on a Raspberry Pi 4B as described
>>> in bug #249520 (https://bugs.freebsd.org/bugzilla/show_bug.cgi?id=249520).
>>> After setting up the USB drive on a USB 2.0 port, the system boots.
>>> However, when the system is under high I/O load (I tested this by
>>> compiling a Go toolchain), the USB controller eventually hangs and
>>> causes the system to effectively crash:
>>> 
>>> ---
>>> xhci_interrupt: host system error
>>> xhci0: Resetting controller
>> 
>> Looks like prior history to the above would be
>> appropriate. (The later messages likely are
>> consequences of the above.)
>> 
>> Also: ed6978a9a70 in github is from:
>> 
>> QUOTE
>> Author: ian
>> Date: Mon Sep 14 17:33:28 2020
>> New Revision: 365729
>> URL: 
>> https://svnweb.freebsd.org/changeset/base/365729
>> 
>> Log:
>>  Add product ID strings for a couple Microchip usb hubs.  Also, update the
>>  vendor ID string to say just "Microchip Technology" -- the buyout of
>>  Standard Microsystems happened in 2012 and the SMC/SMSC names are pretty
>>  much retired at this point.
>> END QUOTE
>> 
>> 
>> but there is a more recent check-in required to
>> avoid at least one way of getting "Resetting controller"
>> for -mcpu=cortex-a72 :
>> 
>> QUOTE
>> Author: hselasky
>> Date: Sat Sep 19 22:37:45 2020
>> New Revision: 365918
>> URL: 
>> https://svnweb.freebsd.org/changeset/base/365918
>> 
>> Log:
>>  Fix for use of the XHCI driver on Cortex-A72 by adding a missing cache
>>  flush operation before writing to the XHCI_ERSTBA_LO/HI register(s).
>> END QUOTE
>> 
>> [I do suggest that you report which git repository that you
>> are referencing since there are multiple ones right now that
>> have differing hashes. I guessed github from "(master)",
>> figuring that the cgit-beta.freebsd.org one would have
>> "(main)".]
>> 
>>> uhub1: at usbus0, port 1, addr 1 (disconnected)
>>> ugen0.2: <vendor 0x2109 USB2.0 Hub> at usbus0 (disconnected)
>>> uhub2: at uhub1, port 1, addr 1 (disconnected)
>>> ugen0.3: <ASIX Elec. Corp. AX88x72A> at usbus0 (disconnected)
>>> axe0: at uhub2, port 2, addr 2 (disconnected)
>>> ukphy0: detached
>>> miibus0: detached
>>> axe0: detached
>>> ugen0.4: <VLI Manufacture String VLI Product String> at usbus0 (disconnected)
>>> umass0: at uhub2, port 4, addr 3 (disconnected)
>>> (da0:umass-sim0:0:0:0): WRITE(10). CDB: 2a 00 01 85 d9 0d 00 00 80 00 
>>> (da0:umass-sim0:0:0:0): CAM status: CCB request completed with an error
>>> (da0:umass-sim0:0:0:0): Retrying command, 3 more tries remain
>>> da0 at umass-sim0 bus 0 scbus0 target 0 lun 0
>>> da0: <WDC WDS2 40G2G0B-00EP UJ43>  s/n ABCDEFA74566 detached
>>> Solaris: WARNING: Pool 'tau' has encountered an uncorrectable I/O failure and has been suspended.
>>> 
>>> Solaris: WARNING: Pool 'tau' has encountered an uncorrectable I/O failure and has been suspended.
>>> ---
>> 
>> The above messages I think are just consequences of earlier
>> problems.
>> 
>>> This is despite having applied D25219 and the D26493--D26496 series
>>> of patches which were supposed to address this sort of issue.  The same
>>> issue does not seem to appear with an older kernel to which the
>>> D26493--D26496 series of patches was not applied and which was not
>>> compiled with -mcpu=cortex-a72.  The older kernel identifies itself as
>>> 
>>>   FreeBSD 13.0-CURRENT #2 ed6978a9a70-c271559(master)-dirty
>>> 
>>> It's the one I described in my earlier mails to this list.  So it seems
>>> that in this case, pulling in patches meant to fix a bug seem to have
>>> introduced in this first place.  Any idea what could have happened?
>> 
>> I strongly suggest using a FreeBSD vintage that includes
>> the corrected XHCI driver.
> 

===
Mark Millard
marklmi at yahoo.com
( dsl-only.net went
away in early 2018-Mar)