[PATCH] pnpacpi mishandles port io ADDRESS resources

[PATCH] pnpacpi mishandles port io ADDRESS resources

[Matthew Wilcox]

ACPI ADDRESSn resources can describe both memory and port io, but the
current code assumes they're descibing memory, which isn't true for HP's
ia64 systems.

Signed-off-by: Matthew Wilcox <matthew@wil.cx>

--- a/drivers/pnp/pnpacpi/rsparser.c	4 Feb 2006 04:51:56 -0000	1.7
+++ b/drivers/pnp/pnpacpi/rsparser.c	10 Jun 2006 16:27:48 -0000
@@ -221,19 +221,34 @@ static acpi_status pnpacpi_allocated_res
 				res->data.fixed_memory32.address_length);
 		break;
 	case ACPI_RESOURCE_TYPE_ADDRESS16:
-		pnpacpi_parse_allocated_memresource(res_table,
-				res->data.address16.minimum,
-				res->data.address16.address_length);
+		if (res->data.address.resource_type == 0)
+			pnpacpi_parse_allocated_memresource(res_table,
+					res->data.address16.minimum,
+					res->data.address16.address_length);
+		else if (res->data.address.resource_type == 1)
+			pnpacpi_parse_allocated_ioresource(res_table,
+					res->data.address16.minimum,
+					res->data.address16.address_length);
 		break;
 	case ACPI_RESOURCE_TYPE_ADDRESS32:
-		pnpacpi_parse_allocated_memresource(res_table,
-				res->data.address32.minimum,
-				res->data.address32.address_length);
+		if (res->data.address.resource_type == 0)
+			pnpacpi_parse_allocated_memresource(res_table,
+					res->data.address32.minimum,
+					res->data.address32.address_length);
+		else if (res->data.address.resource_type == 1)
+			pnpacpi_parse_allocated_ioresource(res_table,
+					res->data.address32.minimum,
+					res->data.address32.address_length);
 		break;
 	case ACPI_RESOURCE_TYPE_ADDRESS64:
-		pnpacpi_parse_allocated_memresource(res_table,
-		res->data.address64.minimum,
-		res->data.address64.address_length);
+		if (res->data.address.resource_type == 0)
+			pnpacpi_parse_allocated_memresource(res_table,
+					res->data.address64.minimum,
+					res->data.address64.address_length);
+		else if (res->data.address.resource_type == 1)
+			pnpacpi_parse_allocated_ioresource(res_table,
+					res->data.address64.minimum,
+					res->data.address64.address_length);
 		break;
 
 	case ACPI_RESOURCE_TYPE_EXTENDED_ADDRESS64:

Re: [PATCH] pnpacpi mishandles port io ADDRESS resources

[Matthieu CASTET]

Le Sat, 10 Jun 2006 10:41:50 -0600, Matthew Wilcox a écrit :

> 
> ACPI ADDRESSn resources can describe both memory and port io, but the
> current code assumes they're descibing memory, which isn't true for HP's
> ia64 systems.
> 
There already a patch for that in -mm kernel

Matthieu

ACPI patch process problems again

[Matthew Wilcox]

On Sat, Jun 10, 2006 at 10:51:13PM +0200, Matthieu CASTET wrote:
> Le Sat, 10 Jun 2006 10:41:50 -0600, Matthew Wilcox a ?crit?:
> > ACPI ADDRESSn resources can describe both memory and port io, but the
> > current code assumes they're descibing memory, which isn't true for HP's
> > ia64 systems.
> > 
> There already a patch for that in -mm kernel

Yes, the original author just contacted me to tell me.  It seems to me
that we have a process problem when a bug gets fixed on March 28th,
but not propagated upstream by June 10th.  That's almost 11 weeks,
and I'm sure it'll be more by the time it's finally merged.  I thought
the days of ACPI process problems were behind us ;-(

Re: ACPI patch process problems again

[matthieu castet]

Hi,

Matthew Wilcox wrote:
> On Sat, Jun 10, 2006 at 10:51:13PM +0200, Matthieu CASTET wrote:
> 
>>Le Sat, 10 Jun 2006 10:41:50 -0600, Matthew Wilcox a ?crit?:
>>
>>>ACPI ADDRESSn resources can describe both memory and port io, but the
>>>current code assumes they're descibing memory, which isn't true for HP's
>>>ia64 systems.
>>>
>>
>>There already a patch for that in -mm kernel
> 
> 
> Yes, the original author just contacted me to tell me.  It seems to me
> that we have a process problem when a bug gets fixed on March 28th,
> but not propagated upstream by June 10th.  That's almost 11 weeks,
> and I'm sure it'll be more by the time it's finally merged.  I thought
> the days of ACPI process problems were behind us ;-(

The problem is that this patch (and your patch) introduce a regression 
for some boxes.

See http://bugzilla.kernel.org/show_bug.cgi?id=6292 for more info.

I submited today a fix for this regression.

Matthieu