[Qemu-devel] Re: [PATCH] lsi53c895a: Add support for OS/2 Warp SYM8XX.ADD driver

[Nicholas A. Bellinger]

On Thu, 2010-09-30 at 15:23 +0200, Hannes Reinecke wrote:
> Nicholas A. Bellinger wrote:
> > From: Nicholas Bellinger <address@hidden>
> > 
> > Greetings Paul, Jan, Kevin and co,
> > 
> > This series is against my v0.12.5 qemu-kvm.git that contains QEMU SCSI layer
> > SGL passthrough from Gerd Hoffman, 8708EM2 MegaSas emulation from Dr. Hannes
> > Reinecke, and well as my own hw/scsi-bsg.c support.  This tree is located 
> > here:
> > 
> > http://git.kernel.org/?p=virt/kvm/nab/qemu-kvm.git;a=summary
> > 
> > This first patch adds a missing qdev->reset() NOP caller in 
> > hw/scsi-generic.c that
> > is now expected by lsi53c895a.c in >= v0.12.5 code.  You will want to apply 
> > this to
> > all >= v0.12.5 QEMU trees so scsi-generic does not segfault with lsi53c895a 
> > expecting
> > a valid qdev->reset().
> > 
> > The second item is a bit more exotic.. 8-)  So I have been thinking about 
> > how to get
> > qemu-kvm.git scsi-generic <-> TCM_Loop to function with OS/2 Warp v4 (SP15) 
> > in guest
> > for a while now, and I am happy to report that after sending some time in 
> > the last weeks
> > getting OS/2 setup (hey, it has been +13 years) and finding a functioning 
> > sym53c895a
> > driver, and finally finding a working SYM8XX.ADD and being able to fill in 
> > missing
> > informational registers and adding a workaround to fix a bogus Destination 
> > ID register
> > WRITE from the now +10 year old SYM8XX.ADD driver code.
> > 
> Hey, and while you're at it:
> The lsi53c895a emulation is missing support for 'abort' and 'device
> reset' TMFs; newer Linux kernel have the habit of sending them
> accordingly. Should be fairly straightforward, only I gave up on it
> after reading the first 10 odd pages of the manual ...
> 

Yeah, I was looking at adding that earlier but also ended up getting
lost in the specific details..  After the last few patches to fix
breakage in lsi53c895a.c with v0.12.5 <-> SCSI SGL passthrough and this
series, how this might TMFs might function is slightly more clear to me
now for sym53c8xx_2 guest driver code.

I might take a stab at adding TMFs at some point in the future, but
perhaps opening up the discussion about merging Gerd's QEMU SCSI SGL
passthrough (and how it makes lsi53c895a.c and megasas.c code cleaner)
might be in order now before we start adding new features into existing
lsi53c895a.c code.

I say this because the lsi53c895a.c driver is what is currently
diverging the quickest between the mainline qemu-kvm.git tree and the
SCSI-SGL-Passthrough/MegaSAS/BSG qemu-kvm.git tree.

Best,

--nab