CC few more LSI driver developers. > -----Original Message----- > From: Nicholas A. Bellinger [mailto:nab@xxxxxxxxxxxxxxx] > Sent: Monday, March 31, 2014 11:25 AM > To: Matthew Wilcox > Cc: linux-scsi@xxxxxxxxxxxxxxx; willy@xxxxxxxxxxxxxxx; Desai, Kashyap; > Saxena, Sumit; James Bottomley > Subject: Re: [PATCH 0/7] Performance improvements for LSI SCSI cards > > Hi Matthew, > > On Thu, 2014-03-27 at 16:40 -0400, Matthew Wilcox wrote: > > The host lock is a serious scalability problem on 2-socket and larger > > systems which are doing a lot of I/O. Before removing the temporary > > usgae of DEF_SCSI_QCMD, we need to remove all uses of serial_number. > > > > An unrelated performance issue is that reusing the most recent > > driver-specific data structure to track the I/O instead of the least > > recently used keeps the cache-hot lines in use, which is a nice > > performance improvement. It's already present in the mpt3sas driver, > > it just didn't make it into the fusion or mpt2sas drivers yet. > > > > Matthew Wilcox (7): > > mpt3sas: Remove uses of serial_number > > mpt3sas: Remove use of DEF_SCSI_QCMD > > mpt2sas: Remove uses of serial_number > > mpt2sas: Remove use of DEF_SCSI_QCMD > > mpt2sas: Add free smids to the head, not tail of list > > fusion: Add free msg frames to the head, not tail of list > > fusion: Remove use of DEF_SCSI_QCMD > > > > > > +1 to this long overdue series to enable host_lock-less mode with > mpt*sas + fusion. (CC'ing LSI folks + jejb) > > Reviewed-by: Nicholas Bellinger <nab@xxxxxxxxxxxxxxx> > ��.n��������+%������w��{.n�����{������ܨ}���Ơz�j:+v�����w����ޙ��&�)ߡ�a����z�ޗ���ݢj��w�f
