diff options
author | Stefan Richter <stefanr@s5r6.in-berlin.de> | 2005-12-12 23:03:19 -0500 |
---|---|---|
committer | Jody McIntyre <scjody@modernduck.com> | 2005-12-12 23:03:19 -0500 |
commit | 51c1d80e929bace26d2d795bd77fcc14b02ba3bb (patch) | |
tree | 0da6fdaf7b83263d3bcebc380d99d7001b3ff21e /drivers/ieee1394 | |
parent | e38dc0ae24635a2a8a68d87cd0f4a13e74a52d98 (diff) | |
download | op-kernel-dev-51c1d80e929bace26d2d795bd77fcc14b02ba3bb.zip op-kernel-dev-51c1d80e929bace26d2d795bd77fcc14b02ba3bb.tar.gz |
ieee1394: run high-level updates before high-level probes
After a bus reset, let nodemgr call high-level update hooks first for nodes
which do not need to be probed. The main benefit is for a bus with more
than one SBP-2 device: SBP-2 reconnects will be performed before SBP-2
logins, thus have a much higher chance to succeed, and their SCSI devices
will not be blocked much longer than necessary. This was demonstrated for
Linux 2.4 by Dave Cinege a while ago.
A better approach would be to perform time-consuming probes in parallel by a
subthread. I actually plan to implement this for sbp2 but it may take a
while to get that done and tested. Until then, this tweak is a huge
improvement for users with multiple SBP-2 devices.
Signed-off-by: Stefan Richter <stefanr@s5r6.in-berlin.de>
Signed-off-by: Jody McIntyre <scjody@modernduck.com>
Diffstat (limited to 'drivers/ieee1394')
-rw-r--r-- | drivers/ieee1394/nodemgr.c | 20 |
1 files changed, 17 insertions, 3 deletions
diff --git a/drivers/ieee1394/nodemgr.c b/drivers/ieee1394/nodemgr.c index 01ab2bf..0ec2987 100644 --- a/drivers/ieee1394/nodemgr.c +++ b/drivers/ieee1394/nodemgr.c @@ -1407,14 +1407,28 @@ static void nodemgr_node_probe(struct host_info *hi, int generation) struct hpsb_host *host = hi->host; struct class *class = &nodemgr_ne_class; struct class_device *cdev; + struct node_entry *ne; /* Do some processing of the nodes we've probed. This pulls them * into the sysfs layer if needed, and can result in processing of * unit-directories, or just updating the node and it's - * unit-directories. */ + * unit-directories. + * + * Run updates before probes. Usually, updates are time-critical + * while probes are time-consuming. (Well, those probes need some + * improvement...) */ + down_read(&class->subsys.rwsem); - list_for_each_entry(cdev, &class->children, node) - nodemgr_probe_ne(hi, container_of(cdev, struct node_entry, class_dev), generation); + list_for_each_entry(cdev, &class->children, node) { + ne = container_of(cdev, struct node_entry, class_dev); + if (!ne->needs_probe) + nodemgr_probe_ne(hi, ne, generation); + } + list_for_each_entry(cdev, &class->children, node) { + ne = container_of(cdev, struct node_entry, class_dev); + if (ne->needs_probe) + nodemgr_probe_ne(hi, ne, generation); + } up_read(&class->subsys.rwsem); |