Re: CVS commit: src/sys/arch/m68k/m68k

[Nathanial Sloss <nathanialsloss%yahoo.com.au@localhost>]

On Mon, 31 Mar 2025 22:48:39 Tetsuya Isaki wrote:
> At Sun, 30 Mar 2025 04:44:26 +0000,
> 
> Nathanial Sloss wrote:
> > Module Name:	src
> > Committed By:	nat
> > Date:		Sun Mar 30 04:44:26 UTC 2025
> > 
> > Modified Files:
> > 	src/sys/arch/m68k/m68k: switch_subr.s
> > 
> > Log Message:
> > Sprinkle nop.
> > 
> > This is the kernel part of addressing the issue with fpu emulation on
> > lc040 cpus.
> 
> The originator of this PR port-mac68k/13078 said (in 2001):
> | Workaround: place a NOP before the F-line instruction.
> 
> And you responded (in Nov 2024):
> | The submitter is indeed right adding a nop before an fpu
> | instruction does fix the fpu emulation on the lc040...
> 
> But you add a NOP before RET.  Why?
> 
> Furthermore,
> 
> > --- switch_subr.s   31 Oct 2024 07:30:28 -0000  1.38
> > +++ switch_subr.s   30 Mar 2025 04:44:26 -0000  1.39
> > @@ -80,6 +80,7 @@
> > 
> >  ASENTRY_NOPROFILE(cpu_idle)
> >  
> >         stop    #PSL_LOWIPL
> >  
> >  GLOBAL(_Idle)                          /* For sun2/sun3's clock.c ... */
> > 
> > +       nop
> > 
> >         rts
> 
> 68LC040 runs this part but there are no FPU instructions.
> Is your 68LC040 affected if you remove this NOP?
> 
> > @@ -207,6 +208,7 @@
> > 
> >  .Lcpu_switch_nofprest:
> >         movl    %d1,%d0                 | return outgoing lwp
> >         movl    %d0,%a0                 | (in a0, too)
> > 
> > +       nop
> > 
> >         rts
> 
> This is in cpu_switchto.
> 68LC040 runs this part but never encounters FPU instructions.
> Is your 68LC040 affected if you remove this NOP?
> 
> > @@ -255,6 +257,7 @@
> > 
> >  #endif /* FPCOPROC */
> >  #endif /* !_M68K_CUSTOM_FPU_CTX */
> >  
> >         moveq   #0,%d0                  | return 0
> > 
> > +       nop
> > 
> >         rts
> 
> This is in savectx.
> 68LC040 runs this part but never encounters FPU instructions.
> Is your 68LC040 affected if you remove this NOP?
> 
> > @@ -286,6 +289,7 @@
> > 
> >         frestore (%sp)
> >         fnop
> >         addql   #4,%sp
> > 
> > +       nop
> > 
> >         rts
> >  
> >  #endif
> 
> This is in m68k_make_fpu_idle_frame().
> 68LC040 doesn't seem to call it.
> Is your 68LC040 affected if you remove this NOP?
> 
> > @@ -307,6 +311,7 @@
> > 
> >         fmovem  %fp0-%fp7,FPF_REGS(%a0) | save FP general registers
> >         fmovem  %fpcr/%fpsr/%fpi,FPF_FPCR(%a0) | save FP control
> >         registers
> >  
> >  .Lm68881sdone:
> > +       nop
> > 
> >         rts
> >  
> >  #endif
> >  #if defined(M68060)
> > 
> > @@ -318,6 +323,7 @@
> > 
> >         fmovem  %fpsr,FPF_FPSR(%a0)
> >         fmovem  %fpi,FPF_FPI(%a0)
> >  
> >  .Lm68060sdone:
> > +       nop
> > 
> >          rts
> >  
> >  #endif
> 
> These are in m68881_save().
> 68LC040 doesn't seem to call it.  In addition, the second chunk
> is not compiled because M68060 is not defined on mac68k.
> Is your 68LC040 affected if you remove these NOPs?
> 
> > @@ -335,6 +341,7 @@
> > 
> >         fmovem  FPF_REGS(%a0),%fp0-%fp7 | restore FP general registers
> >  
> >  .Lm68881rdone:
> >         frestore (%a0)                  | restore state
> > 
> > +       nop
> > 
> >         rts
> >  
> >  #endif
> >  #if defined(M68060)
> > 
> > @@ -347,6 +354,7 @@
> > 
> >         fmovem  FPF_REGS(%a0),%fp0-%fp7 | restore FP general registers
> >  
> >  .Lm68060fprdone:
> >         frestore (%a0)                  | restore state
> > 
> > +       nop
> > 
> >         rts
> >  
> >  #endif
> >  #endif
> 
> These are in m68881_restore().  Same as above.
> 
> You may be observing something different...
> ---
> Tetsuya Isaki <isaki%pastel-flower.jp@localhost / isaki%NetBSD.org@localhost>

Sorry for the late repy --- the message was sent to spam :(

The original submitter did not provide any patches to binutils or the kernel 
switching task, merely a supposition based on processcessor errata they were 
privy to.

Consider the following (including my gas patch for binutils):

movq.....
nop
fnop

Thats as it's assembled but at run time it could posslibly run like this

movq....
nop
(switch task)
fnop

So the nops are inserted before rts statements in the switcher just in case 
should the next instruction be an f-line instruction.

I'ts possible I've added too many (9 words in total) but I'm being over 
cautious--- its better to be safe than sorry!


With these changes (along with the modified assembler and recompiled userland) 
I've been running a PowerBook 520c (with an XC68LC040 buggy processor) for a 
few months now, flawlessly with the fpu emulation.

What I was after was a uniform approach to all m68k (buggy lc040's included).

It makes me wonder even why we build different packages for sun, amiga, x68k, 
and mac68k also....when they are all m68k :(

If you have an alternate proposal I'm very interested.


Best regards and happy hacking,

Nat