x86: check boundary in count/setup_resource called by get_current_resources

PATCH In review Yinghai Lu 3 participants 5 messages Nov 1, 2007

Subsystems (3)

x86 ACPI PCI

lore

Share & tools

lore

Yinghai Lu

author

2007-11-01 08:19 UTC #0 lore raw

PATCH 1/1 x86: check boundary in count/setup_resource called by get_current_resources

[PATCH] x86: check boundary in count/setup_resource called by get_current_resources

need to check info->res_num less than PCI_BUS_NUM_RESOURCES, so info->bus->resource[info->res_num] = res will not beyond of bus resource array when acpi resutrn too many resource entries.

Signed-off-by Yinghai Lu <[email protected]>

arch/x86/pci/acpi.c +7 -0

Index: linux-2.6/arch/x86/pci/acpi.c
===================================================================
@@ -77,9 +77,13 @@ count_resource(struct acpi_resource *acp
 	struct acpi_resource_address64 addr;
 	acpi_status status;
 
+	if (info->res_num >= PCI_BUS_NUM_RESOURCES)
+		return AE_OK;
+
 	status = resource_to_addr(acpi_res, &addr);
 	if (ACPI_SUCCESS(status))
 		info->res_num++;
+
 	return AE_OK;
 }

@@ -93,6 +97,9 @@ setup_resource(struct acpi_resource *acp
 	unsigned long flags;
 	struct resource *root;
 
+	if (info->res_num >= PCI_BUS_NUM_RESOURCES)
+		return AE_OK;
+
 	status = resource_to_addr(acpi_res, &addr);
 	if (!ACPI_SUCCESS(status))
 		return AE_OK;

Andrew Morton

2007-11-01 08:33 UTC #1 lore raw

Quoted 6 lines · On Thu, 01 Nov 2007 01:20:29 -0700 Yinghai Lu <[email protected]> wrote:

[PATCH] x86: check boundary in count/setup_resource called by get_current_resources
need to check info->res_num less than PCI_BUS_NUM_RESOURCES, so info->bus->resource[info->res_num] = res will not beyond of bus resource array when acpi resutrn too many resource entries.

Isn't this a bit of a problem? It sounds like PCI_BUS_NUM_RESOURCES is to small for that system? If so, some sort of dynamic allocation might be needed.

Quoted 8 lines

Index: linux-2.6/arch/x86/pci/acpi.c
===================================================================
arch/x86/pci/acpi.c +0 -0
@@ -77,9 +77,13 @@ count_resource(struct acpi_resource *acp
 	struct acpi_resource_address64 addr;
 	acpi_status status;

arch/x86/pci/acpi.c count_resource(struct acpi_resource *acp · line 88

Quoted 10 lines

arch/x86/pci/acpi.c +4 -0

 
+	if (info->res_num >= PCI_BUS_NUM_RESOURCES)
+		return AE_OK;
+
 	status = resource_to_addr(acpi_res, &addr);
 	if (ACPI_SUCCESS(status))
 		info->res_num++;
+
 	return AE_OK;
 }

grump. I don't know why people like a blank line before `return': it's just a waste of screen space. And the surrounding code in arch/x86/pci/acpi.c doesn't do this either.

arch/x86/pci/acpi.c setup_resource(struct acpi_resource *acp · line 101

Quoted 6 lines

arch/x86/pci/acpi.c +2 -0

@@ -93,6 +97,9 @@ setup_resource(struct acpi_resource *acp
 	unsigned long flags;
 	struct resource *root;
 
+	if (info->res_num >= PCI_BUS_NUM_RESOURCES)
+		return AE_OK;

And should we really be silently ignoring this problem? Should we at least report it?

Quoted 4 lines

arch/x86/pci/acpi.c +0 -0

 	status = resource_to_addr(acpi_res, &addr);
 	if (!ACPI_SUCCESS(status))
 		return AE_OK;

Yinghai Lu

author

2007-11-01 18:03 UTC #2 lore raw

Quoted 4 lines · Andrew Morton

Quoted 2 lines · On Thu, 01 Nov 2007 01:20:29 -0700 Yinghai Lu <[email protected]> wrote:
[PATCH] x86: check boundary in count/setup_resource called by get_current_resources

Quoted 8 lines

Quoted 4 lines
need to check info->res_num less than PCI_BUS_NUM_RESOURCES, so info->bus->resource[info->res_num] = res will not beyond of bus resource array when acpi resutrn too many resource entries.
Isn't this a bit of a problem? It sounds like PCI_BUS_NUM_RESOURCES is to small for that system? If so, some sort of dynamic allocation might be needed.

sound reasonable...
i have one local patch for amd64 that will get resources from pci conf. and it will use all 8 slots for bus 0. and transparent bus under it only can copy 5 of them.

Gary Hade

2007-11-01 20:10 UTC #3 lore raw

Quoted 12 lines · Yinghai Lu

Quoted 11 lines · Andrew Morton
Quoted 8 lines · On Thu, 01 Nov 2007 01:20:29 -0700 Yinghai Lu <[email protected]> wrote:
[PATCH] x86: check boundary in count/setup_resource called by get_current_resources
need to check info->res_num less than PCI_BUS_NUM_RESOURCES, so info->bus->resource[info->res_num] = res will not beyond of bus resource array
when acpi resutrn too many resource entries.

Quoted 8 lines

Quoted 3 lines
Isn't this a bit of a problem? It sounds like PCI_BUS_NUM_RESOURCES is to small for that system? If so, some sort of dynamic allocation might be needed.
sound reasonable...
i have one local patch for amd64 that will get resources from pci conf. and it will use all 8 slots for bus 0.
and transparent bus under it only can copy 5 of them.

Yea, with the current fixed size pci_bus resource array I believe you would need to increase PCI_BUS_NUM_RESOURCES from 8 to 11 for the transparent bridge child bus to get all 8 _CRS returned resources.

Gary

Signature · 6 lines

Gary Hade
System x Enablement
IBM Linux Technology Center
503-578-4503  IBM T/L: 775-4503
[email protected]
http://www.ibm.com/linux/ltc

Gary Hade

2007-11-01 19:13 UTC #4 lore raw

In reply to Andrew Morton

Quoted 4 lines · Andrew Morton

Quoted 2 lines · On Thu, 01 Nov 2007 01:20:29 -0700 Yinghai Lu <[email protected]> wrote:
[PATCH] x86: check boundary in count/setup_resource called by get_current_resources

Quoted 8 lines

Quoted 4 lines
need to check info->res_num less than PCI_BUS_NUM_RESOURCES, so info->bus->resource[info->res_num] = res will not beyond of bus resource array when acpi resutrn too many resource entries.
Isn't this a bit of a problem? It sounds like PCI_BUS_NUM_RESOURCES is to small for that system? If so, some sort of dynamic allocation might be needed.

I should have considered the possible resource array overrun when I created these functions. I had assumed (apparently incorrectly) that the old PCI_BUS_NUM_RESOURCES value of 4 was based on a spec defined limit on the maximum number of resources that _CRS can return.

I recently noticed the potential overrun myself while backporting the code to kernel source where PCI_BUS_NUM_RESOURCES was initially defined as 4. This happened on a system where the _CRS associated with one of the root bridges returned 5 resources with the 5th causing a write beyond the end of the array. Increasing PCI_BUS_NUM_RESOURCES to the current value of 8 eliminated the overrun that I experienced but after discovering that there is apparently no limit on the number of resources that _CRS can return I had intended to post a change similar to what Yinghai Lu is proposing.

With the current PCI_BUS_NUM_RESOURCES value, _CRS can return up to 8 resources before the pci_bus resource array is totally saturated but it should be noted that if a transparent bridge is present below the root bridge it's child bus will only see the first 5 resources.

The current fixed pci_bus resource array size of 8 is adequate (for storing _CRS returned resource and visibility across transparent bridges) on those systems that I work on but without a bound on the number of resources returned by _CRS some sort of dynamic allocation certainly makes sense.

Note that exposure to this issue is currently limited to those that use the 'pci=use_crs' kernel option.

Gary

Signature · 6 lines

Gary Hade
System x Enablement
IBM Linux Technology Center
503-578-4503  IBM T/L: 775-4503
[email protected]
http://www.ibm.com/linux/ltc