[PATCH] ppc64: Add iommu=on for enabling DART on small-mem machines

[PATCH] ppc64: Add iommu=on for enabling DART on small-mem machines

[olof]

Linus,

Below patch makes it possible for people like me with a small-mem G5 to
enable the DART. I see two reasons for wanting to do so:

1. To debug/test DART/iommu code itself (small audience, including
   myself).
2. To debug drivers on small-mem machines, since bad pci_map*() usage will
   be punished (possibly larger audience).


Thanks,

-Olof


===== arch/ppc64/kernel/prom.c 1.56 vs edited =====
--- 1.56/arch/ppc64/kernel/prom.c	Fri Feb 27 16:44:57 2004
+++ edited/arch/ppc64/kernel/prom.c	Fri Feb 27 15:48:10 2004
@@ -516,6 +516,9 @@
 	return mem;
 }

+#ifdef CONFIG_PMAC_DART
+static int dart_force_on;
+#endif

 static unsigned long __init
 prom_initialize_lmb(unsigned long mem)
@@ -539,10 +542,12 @@
 		prom_print(opt);
 		prom_print(RELOC("\n"));
 		opt += 6;
-		while(*opt && *opt == ' ')
+		while (*opt && *opt == ' ')
 			opt++;
 		if (!strncmp(opt, RELOC("off"), 3))
 			nodart = 1;
+		else if (!strncmp(opt, RELOC("on"), 2))
+			RELOC(dart_force_on) = 1;
 	}
 #else
 	nodart = 1;
@@ -763,8 +768,10 @@
 	extern unsigned long dart_tablebase;
 	extern unsigned long dart_tablesize;

-	/* Only reserve DART space if machine has more than 2Gb of RAM */
-	if (lmb_end_of_DRAM() <= 0x80000000ull)
+	/* Only reserve DART space if machine has more than 2GB of RAM
+	 * or if requested with iommu=on on cmdline.
+	 */
+	if (lmb_end_of_DRAM() <= 0x80000000ull && !RELOC(dart_force_on))
 		return;

 	/* 512 pages is max DART tablesize. */




Olof Johansson                                        Office: 4E002/905
Linux on Power Development                            IBM Systems Group
Email: olof@austin.ibm.com                          Phone: 512-838-9858
All opinions are my own and not those of IBM

Re: [PATCH] ppc64: Add iommu=on for enabling DART on small-mem machines

[Andi Kleen]

olof@austin.ibm.com writes:

> Linus,
> 
> Below patch makes it possible for people like me with a small-mem G5 to
> enable the DART. I see two reasons for wanting to do so:

Could you call it iommu=force ? 

It would be the same name as on x86-64 for the same thing then and
a consistent name may be easier to get to driver developers.

-Andi

Re: [PATCH] ppc64: Add iommu=on for enabling DART on small-mem machines

[Olof Johansson]

On 28 Feb 2004, Andi Kleen wrote:

> olof@austin.ibm.com writes:
>
> > Below patch makes it possible for people like me with a small-mem G5 to
> > enable the DART. I see two reasons for wanting to do so:
>
> Could you call it iommu=force ?
>
> It would be the same name as on x86-64 for the same thing then and
> a consistent name may be easier to get to driver developers.


Ack! I was wavering between the two options and finally went with =on. :)
Either way is fine with me.


-Olof

Olof Johansson                                        Office: 4E002/905
Linux on Power Development                            IBM Systems Group
Email: olof@austin.ibm.com                          Phone: 512-838-9858
All opinions are my own and not those of IBM

Re: [PATCH] ppc64: Add iommu=on for enabling DART on small-mem machines

[Mike Fedyk]

Olof Johansson wrote:
> On 28 Feb 2004, Andi Kleen wrote:
> 
> 
>>olof@austin.ibm.com writes:
>>
>>
>>>Below patch makes it possible for people like me with a small-mem G5 to
>>>enable the DART. I see two reasons for wanting to do so:
>>
>>Could you call it iommu=force ?
>>
>>It would be the same name as on x86-64 for the same thing then and
>>a consistent name may be easier to get to driver developers.
> 
> 
> 
> Ack! I was wavering between the two options and finally went with =on. :)
> Either way is fine with me.

And maybe even some generic code? :-D

Re: [PATCH] ppc64: Add iommu=on for enabling DART on small-mem machines

[Benjamin Herrenschmidt]

> And maybe even some generic code? :-D

No. It's not possible, at least not at this point. The decision of
allocating the DART space or not (and thus eventually stripping the
max available memory) is done very early. So early that the kernel
isn't yet running at it's link location ;) So the kind of code we
can actually run at this point is limited, with various tweaks to
deal with the relocation issue.

It's not very pretty, I hope we can improve that later, but that
is not something that will happen in the upcoming weeks .

Ben.