On Fri, Jan 20, 2012 at 10:12 PM, Jack Wang [off-list ref] wrote:

quoted

Changes since v5: http://marc.info/?l=linux-scsi&m=132656475829637&w=2

1/ Reworked "libsas: let libata recover links that fail to transmit

initial

quoted

   sig-fis" to better handle problematic expanders and slow to attach ata
   devices.  libsas will now validate linkrate and take 'device-type' into
   account when determining if a new device is attached.

2/ Added support for asynchronously scanning ata devices.

3/ Reworked "libsas: fix sas_unregister_ports vs sas_drain_work" as

lockdep

quoted

   complained about the original.

4/ New patch "libsas: fix sas port naming" makes direct-attached device

naming

quoted

   more predictable and consistent with expander attached device naming.

5/ Made scan order more predictable by ensuring that regardless of device

type

quoted

   we always attach in discovery order.  Slow to link devices will still

be

quoted

   scanned out of order, link needs to be in at least the 'stp pending'

state

quoted

   before the device will be considered attached.

[Jack Wang]
Asynchronously scan really make the discovery far more quickly, with v4
reset expander will need more than 3 minites to get discovered, with this v6
patchset it takes less than 5 seconds.

Great!

Dear Dan,

Could you share how you test procedure to find out these bugs?

These fixes have primarily been driven by libsas ata error handling as
robust as libata, and teaching libsas that ata links need more care
and handling than it was giving before.

In the case of async probing, this is something that libata has had
since 2.6.29, but was never added to libsas.

--
Dan