kern/55671: USB drives fail to attach after appreciable uptime on -current

[david%gutteridge.ca@localhost]

>Number:         55671
>Category:       kern
>Synopsis:       USB drives fail to attach after appreciable uptime on -current
>Confidential:   no
>Severity:       serious
>Priority:       medium
>Responsible:    kern-bug-people
>State:          open
>Class:          sw-bug
>Submitter-Id:   net
>Arrival-Date:   Sat Sep 19 19:50:00 +0000 2020
>Originator:     David H. Gutteridge
>Release:        9.99.72
>Organization:
I've very little semblance of organization these days
>Environment:
NetBSD arcusxvi.nonus-porta.net 9.99.72 NetBSD 9.99.72 (DEBUG) #1: Wed Sep  9 10:18:47 EDT 2020  disciple%arcusxvi.nonus-porta.net@localhost:/home/disciple/netbsd-current/src/sys/arch/amd64/compile/obj/DEBUG amd64
>Description:
Relatively recent 9.99.x kernels will reproducibly fail to attach USB
pen drives after the laptop in question has been up for a while. On
boot, there is no problem; any pen drive I tested will attach without
issue or delay. After uptime of at least a day, the same drives will
consistently fail to attach, with this dmesg output:

[ 251998.035287] umass0 at uhub5 port 2 configuration 1 interface 0
[ 251998.035287] umass0: Verbatim (0x08ec) Store 'n' Go (0x0008), rev 2.00/1.00, addr 4
[ 251998.035287] umass0: using SCSI over Bulk-Only
[ 251998.055287] umass0: autoconfiguration error: failed to create xfers

I don't know if there is a "magic" uptime at which this occurs, or,
more likely, some extent of system activity (memory pressure?) is the
trigger for this.

(Where I say "relatively recent 9.99.x kernels", I didn't find this an
issue with 9.99.44, which I was running for a while on this laptop.
When I went to 9.99.69, and on, it started happening. Though, perhaps I
was just lucky before.)

I built a kernel with USB debugging defined on various devices, enabled
debugging output at "10", and have attached related "vmstat -u usbhist"
output at:

http://www.netbsd.org/~gutteridge/pen_drive_attach_failure.log

>How-To-Repeat:
1. Leave machine on for at least a day, maybe more.
During this time, "do stuff" on it, e.g. intensive compilation.
2. Insert USB pen drive.
>Fix: