Port-mvme68k archive

[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index][Old Index]

Re: MVME177 panic: pmap_enter_ptpage (PR #45915 on 6.x)


> The kernel you built for me booted right up.  That patch seems good.  Thank 
> you.
> (dmesg below)

I've committed the fix and will send a pullup request for
netbsd-6 branches later.

> I tried the netboot too but it didn't work.  It was a bit odd.  Part
> of it was dumped
> to the console.  It looks like starting at offset 0x4F7A (20,346) it
> dumped to the
> console.
> Here's the first bit of console output:
> Network Boot File load in progress... To abort hit <BREAK>
> Bytes Received =&23784, Bytes Loaded =&23784
> Bytes/Second   =&23784, Elapsed Time =1 Second(s)
> [a bunch of unprintable...]NOp  @Nu>> %s MVME%x netboot (via %s) [%s]

After a few hours analysis, this seems caused by a bug of
gcc-4.5.x's __asm() statement.

Could you try the following netboot binary built from the same sources
but using gcc-4.1.3 from 5.x?

mvme68k bootloaders uses "MVMEPROM_OUTSTR" function for putchar():

The OUTSTR call takes two pointer args, start and end addresses
of strings.  6.x's gcc-4.5.x seems to pass an incorrect start address
so a bunch of garbages are put onto the screen.

gcc-4.1.3 in 5.x generates the following insns from
the mvmeprom_outstr() source:
 movel %sp@(4),%d1
 movel %sp@(8),%d0
 movel %d0,%sp@-
 movel %d1,%sp@-
 trap #15

but gcc-4.5.3 in 6.x emits:
 movel %sp@(8),%d0
 movel %d0,%sp@-
 movel %sp@(4),%d0
 movel %d0,%sp@-
 trap #15
so the second args are taken from a wrong (post-decremented) %sp address.
Disabling optimization (passing -O0 instead of -Os) doesn't help.

Does anyone have suggestion how this should be fixed?
Rewrite outstr (and the similar outln) functions in asm?

Izumi Tsutsui

Home | Main Index | Thread Index | Old Index