Re: [Qemu-devel] [PATCH] Floppy: Remove Sense Interrupt Hack

[Justin Chevrier]

--- On Sat, 12/20/08, Blue Swirl <address@hidden> wrote:

> From: Blue Swirl <address@hidden>
> Subject: Re: [Qemu-devel] [PATCH] Floppy: Remove Sense Interrupt Hack
> To: address@hidden
> Cc: address@hidden
> Date: Saturday, December 20, 2008, 5:54 AM
> On 12/19/08, Justin Chevrier <address@hidden>
> wrote:
> >
> >
> >
> >  --- On Fri, 12/19/08, Blue Swirl
> <address@hidden> wrote:
> >
> >  > From: Blue Swirl <address@hidden>
> >  > Subject: Re: [Qemu-devel] [PATCH] Floppy: Remove
> Sense Interrupt Hack
> >  > To: address@hidden, address@hidden
> >  > Date: Friday, December 19, 2008, 2:43 PM
> >
> > > On 12/18/08, Justin Chevrier
> <address@hidden>
> >  > wrote:
> >  > > Hey guys,
> >  > >
> >  > >  I'm tracking down the next issue(s)
> with the
> >  > Openserver installation. The current issue is
> that during
> >  > installation when the installer tries to link
> the BLTD (eg.
> >  > SCSI drivers) drivers into the kernel it is
> unable to access
> >  > the floppy drive. After reviewing the debug logs
> it turns
> >  > out that the installer dosen't like the
> value being
> >  > returned by Sense Interrupt. The cause of this
> is a hack in
> >  > the Sense Interrupt code that always returns an
> OR'd
> >  > value based on FD_SR0_SEEK instead of the actual
> Status0
> >  > regsiter. The hack identifies that the floppy
> code is broken
> >  > with regard to Status0 and Read/Writes. Outside
> of setting
> >  > particular values in case of CRC errors/etc I
> can't
> >  > identify from datasheets what we should be
> doing. I've
> >  > tracked down what I believe is the original
> mailing list
> >  > post regarding this hack and it seems to be
> related to
> >  > Windows 2000 and higher OS's recognizing the
> floppy
> >  > drive. I applied the patch below and verified
> that
> >  > >   detection/read/writes/formatting still
> work in
> >  > Windows XP SP2.
> >  > >
> >  > >  Original post:
> >  > >
> >  >
> http://lists.gnu.org/archive/html/qemu-devel/2004-10/msg00076.html
> >  > >
> >  > >  SVN change:
> >  > >
> >  >
> http://svn.savannah.gnu.org/viewvc?view=rev&root=qemu&revision=1118
> >  > >
> >  > >  Having said the above does anyone know
> what the
> >  > comment in the hack is referencing? Is it still
> needed? If
> >  > so what needs to be done to eliminate it?
> >  > >
> >  > >  If it's no longer needed I propose the
> patch
> >  > below.
> >  > >
> >  > >  Justin
> >  > >
> >  > >  Changelog:
> >  > >
> >  > >  Remove Sense Interrupt hack
> >  > >
> >  > >  Signed-off-by: Justin Chevrier
> >  > <address@hidden>
> >  >
> >  > It may be a hack, but if we remove it, Sparc64
> refuses to
> >  > boot and
> >  > prints endlessly:
> >  > SENSEI c0 00
> >  > SENSEI c0 00
> >  > SENSEI c0 00
> >  > SENSEI c0 00
> >
> >
> > Okay, thanks for the feedback!
> >
> >  I'll start looking into what SPARC64 expects
> here.
> 
> The message is printed here:
> http://tracker.coreboot.org/trac/openbios/browser/openbios-devel/drivers/floppy.c#L359

Thanks again!

Does the patch below work for you?

In this version we return 0xc0 (polling) for the first interrupt status after 
reset (Openserver appears to require this). After returning 0xc0 we set the 
Status0 register to 0x20 (seek complete). This should work as the spec says 
Sense Interrupt should be called up to 4 times to retrieve the queued status 
values. Tested successfully in both Windows 98 and XP SP2.

It also appears that the code referenced in the link loops infinitely due to a 
missing: "max_sensei--;" which is present in the Linux Kernel's equivalent code.

Justin

Index: fdc.c
===================================================================
--- fdc.c       (revision 6121)
+++ fdc.c       (working copy)
@@ -1601,20 +1601,13 @@
 {
     fdrive_t *cur_drv = get_cur_drv(fdctrl);

-#if 0
     fdctrl->fifo[0] =
         fdctrl->status0 | (cur_drv->head << 2) | GET_CUR_DRV(fdctrl);
-#else
-    /* XXX: status0 handling is broken for read/write
-       commands, so we do this hack. It should be suppressed
-       ASAP */
-    fdctrl->fifo[0] =
-        FD_SR0_SEEK | (cur_drv->head << 2) | GET_CUR_DRV(fdctrl);
-#endif
+
     fdctrl->fifo[1] = cur_drv->track;
     fdctrl_set_fifo(fdctrl, 2, 0);
     fdctrl_reset_irq(fdctrl);
-    fdctrl->status0 = FD_SR0_RDYCHG;
+    fdctrl->status0 = FD_SR0_SEEK;
 }

 static void fdctrl_handle_seek (fdctrl_t *fdctrl, int direction)