stack synchronization
Boris Dinkevich
dinkevich at gmail.com
Sun Jan 9 23:07:51 PST 2005
Hello everyone,
I see no reference to giant locking in 5.3 (except in event/task threads)
Does this not affect performance when multiple devices are connected to the
host ?
Also, I have seen a modification to the bsd stack by Hans Petter Selasky,
where he uses per bus mtxes for locking,
is this not the better approach to the problem ?
Interrupt is synced via Giant ?
Best Regards
Boris Dinkevich
At 05:09 AM 10/1/2005, M. Warner Losh wrote:
>In message: <41E1C8C5.4070605 at elischer.org>
> Julian Elischer <julian at elischer.org> writes:
>: M. Warner Losh wrote:
>: > In message: <6.1.2.0.2.20050109174427.0802c640 at pop3>
>: > Boris Dinkevich <dinkevich at gmail.com> writes:
>: > : Hello everyone,
>: > :
>: > : When going over the usb-host stack, it appears that synchronization
>is done
>: > : via splusb/x.
>: > : But in the 5.3 release, these functions are implemented at stubs.
>: > :
>: > : How is the sync done then ?
>: > :
>: > : Also, with soft_interrupts, it appears there is no need for spls, am I
>: > : correct ?
>: >
>: > Giant locking.
>: >
>: > I have patches to move the interrupt outside of giant, but the rest of
>: > the stack is still under giant locking.
>:
>: it may be worth just adding a single "usb system" lock..
>
>well duh! However, there's a lot of things that can trivially be done
>with giant. Locking with a usb subsystem lock takes a lot more
>effort...
>
>Warner
More information about the freebsd-usb
mailing list