Re: WWW Linux Sparc-32 SMP

[Robert Dinse <nanook@eskimo.com>]

     I am very sad to report, it did not change the behavior at all.  After
booting and mounting the swap, the system skipped all of the single user init
stuff and went immediately into run level 3, which failed miserably without
file systems being mounted, network interfaces being turned up, and all the
other things that normally happen.  Then the system went into spin_lock
deadlock just like before.  In short, no change in behavior whatsoever. 

On Thu, 22 Jun 2000, Anton Blanchard wrote:
>  
> >      To put it mildly, it did not work on my SS-10 with Hypersparc.  It is
> > VERY freaky.  The first time I tried to boot it would hang when it tried to 
> > add swap.  I hadn't stripped vmlinux yet, so I went and tried that, then it 
> > got past adding swap but freaked when it should have run the init scripts.
> 
> Yes this was my fault. Hypersparc doesn't have context tags on its icache
> so we must flush it each time we change contexts.
> 
> Cheers,
> Anton
> 
> --- linux/arch/sparc/mm/srmmu.c	Thu Jun 22 08:45:28 2000
> +++ linux_sparc32/arch/sparc/mm/srmmu.c	Thu Jun 22 10:43:17 2000
> @@ -88,6 +88,8 @@
>  int viking_mxcc_present = 0;
>  spinlock_t srmmu_context_spinlock = SPIN_LOCK_UNLOCKED;
>  
> +int is_hypersparc;
> +
>  /*
>   * In general all page table modifications should use the V8 atomic
>   * swap instruction.  This insures the mmu and the cpu are in sync
> @@ -542,7 +544,10 @@
>  		spin_unlock(&srmmu_context_spinlock);
>  		srmmu_ctxd_set(&srmmu_context_table[mm->context], mm->pgd);
>  	}
> -	/* XXX should we hyper_flush_whole_icache() here - Anton */
> +
> +	if (is_hypersparc)
> +		hyper_flush_whole_icache();
> +
>  	srmmu_set_context(mm->context);
>  }
>  
> @@ -1448,6 +1453,8 @@
>  	srmmu_name = "ROSS HyperSparc";
>  
>  	init_vac_layout();
> +
> +	is_hypersparc = 1;
>  
>  	BTFIXUPSET_CALL(pte_clear, srmmu_pte_clear, BTFIXUPCALL_NORM);
>  	BTFIXUPSET_CALL(pmd_clear, srmmu_pmd_clear, BTFIXUPCALL_NORM);
>