[PATCH 0/63] convert usb_control_msg() and usb_bulk_msg() to use msecs
nacc at us.ibm.com
Wed Feb 9 01:02:12 CET 2005
The following patchset (affecting USB, Bluetooth, IrDA, Sound, DVB and W1 --
hence the long To: an Cc: lists...) converts the usb_control_msg() and
usb_bulk_msg() functions to use milliseconds in the timeout parameter instead
of jiffies. This is not a problem for almost all cases, as the requested delays are usually quite long wrt. these two functions.
I see several reasons for committing these changes:
1) Expressing time in milliseconds makes the code cleaner and easier to
understand. It also makes the delay independent (from a code perspective) of
HZ's value, which would be important for dynamic-HZ or tickless systems.
2) Also for dynamic-HZ/tickless systems, it becomes unclear what exactly HZ
represents. Milliseconds have meaning in the human world beyond the kernel. This
ties in pretty strongly with 1)'s point about understanding.
3) As the time/timer subsystem evolves and changes, the overall message I am
seeing is that jiffies are not a proper expression of time. Thus, I am hoping
these patches minimize this particular interface's usage of jiffies in this
way; while, in the end, the call to usb_start_wait_for_urb ends up with the
same jiffies-based time, if the timer subsystem were to, for instance, convert
to nanoseconds, we could instead pass on the parameter multiplied by a million.
It simply makes adjusting to any new time system that much easier.
Exceptions do exist, of course, and the following files do not use a
HZ-relative timeout variable; thus I was unsure how exactly to proceed.
Given that HZ==1000 in i386, the constants used technically correspond
directly to milliseconds. I am unsure if that is the intent of the author
in all cases, though. I left the hard-coded non-HZ related values alone in all
As always, any comments or suggestions are greatly appreciated.
More information about the lm-sensors