This is a thorough rework of "Fix the NEC stop bug workaround" series.
The focus is on simple patches which fix one bug at a time, hopefully
for good, using localized changes touching only 1-2 related functions.
1/3 resolves the problem of infinite retries on NEC by limiting them
2/3 resolves broken cancellation on stream endpoints (purely a SW bug)
3/3 resolves almost all known cases of pointless retries, which is an
optimization at this point, as it falls back on 1/1 when it fails
At this point zero issues are seen on NEC, but other HCs with known HW
bugs are yet to be tested fully. However, ASM3142 has just passed the
"ifconfig up/down" test, so there is that. ASM1042 not tried at all.
If this works out, the ASMedia chips could be enabled. Or maybe all
chips could be enabled. Technically, retrying Stop Endpoint doesn't
seem to violate the spec. But we know the chips already have one bug,
so who knows if they don't have other bugs...
Maybe some internal operation scheduled by the doorbell ring could
interacs badly with commands. It's clear that some HC vendor haven't
thought very hard about what happens in those cases.
Regards,
Michal
