-test10/PPC still broken on PowerMac 8500

-test10/PPC still broken on PowerMac 8500

[John Mock]

Most of these are SCSI issues and the last one makes -test10 hard to debug.

* MESH gets SLAB errors during startup, CDROM eject
* "mac53c94: module license 'unspecified' taints kernel."
* "53C94 did not call scsi_unregister" [sorry, should have filed bug report]
* 'swim3.c' doesn't compile properly
* Switching from X to text console ('controlfb' frame buffer) loses video 
  sync.

The 53C94 problems probably aren't hard to fix.  For the floppy code (that 
is, 'swim3.c'), 'benh' has a version of 'swim3' which may only need further
testing.  The MESH issue looks like a buffer alignment problem, and worked 
without complaint in the 2.4 kernels.  

The video mode problem is a real nuisance and is the biggest reason i'm not
doing more than intermittent testing of 2.6.0/PPC.

			         -- JM


P.S.  I came across a large pile of floppies during a massive cleanup (why
i've been so busy) and i can run some more tests of the 'swim3' code after 
the Thanksgiving break.

Re: -test10/PPC still broken on PowerMac 8500

[Benjamin Herrenschmidt]

On Wed, 2003-11-26 at 11:55, John Mock wrote:
> Most of these are SCSI issues and the last one makes -test10 hard to debug.
> 
> * MESH gets SLAB errors during startup, CDROM eject

Those are still the same good old issues, you have to disable SLAB
debugging unfortunately, at least until there is a better fix, but
so far there isn't.

> * "mac53c94: module license 'unspecified' taints kernel."

Ok.

> * "53C94 did not call scsi_unregister" [sorry, should have filed bug report]

Ok.

> * 'swim3.c' doesn't compile properly

Yup, known. My 8500 is sitll on the boat from france and I had my
time lately taken 200% by the G5 port.

> * Switching from X to text console ('controlfb' frame buffer) loses video 
>   sync.

Known.

> The 53C94 problems probably aren't hard to fix.  For the floppy code (that 
> is, 'swim3.c'), 'benh' has a version of 'swim3' which may only need further
> testing.  The MESH issue looks like a buffer alignment problem, and worked 
> without complaint in the 2.4 kernels.  

It's a problem. The SCSI stack is passing us unaligned buffers when slab
debugging is enabled, thus triggering a HW issue with those unaligned
buffers (the chip writes before the beginning of the buffer)

> The video mode problem is a real nuisance and is the biggest reason i'm not
> doing more than intermittent testing of 2.6.0/PPC.

I'd appreciate if you could track it down as I won't have access to the
8500 for a few weeks still.

> 			         -- JM
> 
> 
> P.S.  I came across a large pile of floppies during a massive cleanup (why
> i've been so busy) and i can run some more tests of the 'swim3' code after 
> the Thanksgiving break

Ok.

Ben.

swim3 testing [was: -test10/PPC still broken on PowerMac 8500]

[John Mock]

OK, compare your current 'swim3.c' against 2.4.21/PPC and -test11/i386.
It generally reads the same floppies that the VAIO reads, and if it gets
errors, the Sony gets the same errors.  If anything, maybe because of a
better Mac floppy drive, it reads a few floppies the Sony cannot.  When
it does get errors, it reports them in the same manner as the current
i386 floppy driver (albeit on the one floppy that i had intentionally
created a bad track on for test purposes, they disagreed by one sector 
as exactly where the error occurred, which i'm not concerned about).

Comparing the your current driver against the 2.4.21 swim3 driver, the
error reporting is different and it takes much longer to report errors
under 2.4.21 (3 min. total vs. 40 sec total for my intentionally bad 
floppy), and 2.4.21 reports errors for each sector after the bad one, 
while the i386 and the current 'swim3' driver flag the first error and 
seem to discard the rest.

I can't say what the correct behavior is.  However, your current 'swim3'
driver seems to be more useful to me than the older one, and its behavior
is much closer to the current i386 driver (i.e. having more reproducible
results across platforms makes life easier for many people).

So aside from the losing video sync switching back from X, the PowerMac
8500 under -test10 seems to be alot better now, assuming your current
driver is acceptable for 2.6.0-test11.  (I wish i could say the same for
i386 software suspend, alas, but that'll just have to wait...)  Thanks 
again for the clues!
				 -- JM