This code .... grew a bit since my previous pencil-sketch code.
The goal is really the final patch: using a llist without spinlocks to
handle dispatch of idle threads. To get there I found it necessary - or
at least helpful - to do a lot of refactoring.
This code passes some basic tests, but I haven't push it hard yet.
Even if other aren't convinced that llists are the best solution, I
think a lot of the refactoring is this valuable.
Comments welcome,
Thanks,
NeilBrown
---
NeilBr own (14):
lockd: remove SIGKILL handling.
nfsd: don't allow nfsd threads to be signalled.
SUNRPC: call svc_process() from svc_recv().
SUNRPC: change svc_recv() to return void.
SUNRPC: remove timeout arg from svc_recv()
SUNRPC: change various server-side #defines to enum
SUNRPC: refactor svc_recv()
SUNRPC: integrate back-channel processing with svc_recv() and svc_process()
SUNRPC: change how svc threads are asked to exit.
SUNRPC: change svc_pool_wake_idle_thread() to return nothing.
SUNRPC: add list of idle threads
SUNRPC: discard SP_CONGESTED
SUNRPC: change service idle list to be an llist
SUNRPC: only have one thread waking up at a time
fs/lockd/svc.c | 49 ++-----
fs/lockd/svclock.c | 9 +-
fs/nfs/callback.c | 59 +-------
fs/nfsd/nfs4proc.c | 10 +-
fs/nfsd/nfssvc.c | 22 +--
include/linux/llist.h | 2 +
include/linux/lockd/lockd.h | 4 +-
include/linux/sunrpc/cache.h | 11 +-
include/linux/sunrpc/svc.h | 87 +++++++++---
include/linux/sunrpc/svc_xprt.h | 39 +++---
include/linux/sunrpc/svcauth.h | 29 ++--
include/linux/sunrpc/svcsock.h | 2 +-
include/linux/sunrpc/xprtsock.h | 25 ++--
include/trace/events/sunrpc.h | 5 +-
lib/llist.c | 27 ++++
net/sunrpc/backchannel_rqst.c | 8 +-
net/sunrpc/svc.c | 71 ++++------
net/sunrpc/svc_xprt.c | 226 ++++++++++++++++--------------
net/sunrpc/xprtrdma/backchannel.c | 2 +-
19 files changed, 347 insertions(+), 340 deletions(-)
--
Signature
