Re: [patch 1/4] x86, pci, acpi: Redo sanity checks for root brigde probing

[Borislav Petkov]

On Thu, Dec 11, 2014 at 07:48:18PM -0000, Thomas Gleixner wrote:
> From: Yinghai Lu <yinghai@xxxxxxxxxx>
> 
> commit e22ce93870de ("x86, PCI, ACPI: Kill private function
> resource_to_addr() in arch/x86/pci/acpi.c") dropped the sanity checks
> for resources initialized via acpi_dev_resource_memory().
> 
> It further dropped a check for address_length = 0 for the resources
> which are initialized via acpi_dev_resource_address_space().
> 
> For translated addresses the range check operates on the non
> translated end address. Assign orig_end after adding the translation
> offset.
> 
> This causes that invalid resources are not dropped and the range check
> for translated addresses fails.
> 
> [ tglx: Rewrote changelog.
> 
>   Added the address_length check to the x86 code and not into
>   acpi_dev_resource_address_space(). While the acpi function should
>   check this, we want to have a proper check which sets the
>   IORESOURCE_DISABLED bit as it does for IO resources.
> 
>   Removed the pointless memset of addr. acpi_resource_to_address64()
>   either fails or fills it completely ]
> 
> Fixes: commit e22ce93870de ("x86, PCI, ACPI: Kill private function resource_to_addr() in arch/x86/pci/acpi.c").
> Signed-off-by: Yinghai Lu <yinghai@xxxxxxxxxx>
> Link: http://lkml.kernel.org/r/CAE9FiQWaL1Qu1eP-fQV784ucy+r-65AZ95VyiUrefn0dNtfLDw@xxxxxxxxxxxxxx
> Signed-off-by: Thomas Gleixner <tglx@xxxxxxxxxxxxx>
> 
> ---
>  arch/x86/pci/acpi.c |   63 ++++++++++++++++++++++++++--------------------------
>  1 file changed, 32 insertions(+), 31 deletions(-)
> 
> Index: tip/arch/x86/pci/acpi.c
> ===================================================================
> --- tip.orig/arch/x86/pci/acpi.c
> +++ tip/arch/x86/pci/acpi.c
> @@ -221,10 +221,9 @@ static void teardown_mcfg_map(struct pci
>  static acpi_status count_resource(struct acpi_resource *acpi_res, void *data)
>  {
>  	struct pci_root_info *info = data;
> -	struct resource r = {
> -		.flags = 0
> -	};
> +	struct resource r;
>  
> +	memset(&r, 0, sizeof(r));
>  	if (!acpi_dev_resource_memory(acpi_res, &r) &&
>  	    !acpi_dev_resource_address_space(acpi_res, &r))
>  		return AE_OK;
> @@ -239,14 +238,13 @@ static acpi_status setup_resource(struct
>  {
>  	struct pci_root_info *info = data;
>  	u64 translation_offset = 0;
> -	struct resource r = {
> -		.flags = 0
> -	};
> +	struct resource r;
> +	u64 orig_end;
>  
> +	memset(&r, 0, sizeof(r));
>  	if (acpi_dev_resource_memory(acpi_res, &r)) {
> -		r.flags &= IORESOURCE_MEM | IORESOURCE_IO;
> +		r.flags &= IORESOURCE_MEM;

As you pointed out on IRC, the &= is kinda ass-backwards logically but
I don't understand the big picture yet to know what happens with this
resource when it gets copied into the root a bit further down:

	...
        info->res[info->res_num] = r;

and what a *cleared* IORESOURCE_MEM flag for a memory resource means...

>  	} else if (acpi_dev_resource_address_space(acpi_res, &r)) {
> -		u64 orig_end;
>  		struct acpi_resource_address64 addr;
>  
>  		r.flags &= IORESOURCE_MEM | IORESOURCE_IO;
> @@ -256,40 +254,43 @@ static acpi_status setup_resource(struct
>  		if (ACPI_FAILURE(acpi_resource_to_address64(acpi_res, &addr)))
>  			return AE_OK;
>  
> +		if (!addr.address_length)
> +			return AE_OK;
> +

We have an
                if (r.flags == 0)

test above this - might change it to

		if (!r.flags)

for consistency.

>  		if (addr.resource_type == ACPI_MEMORY_RANGE &&
>  		    addr.info.mem.caching == ACPI_PREFETCHABLE_MEMORY)
>  			r.flags |= IORESOURCE_PREFETCH;
>  
>  		translation_offset = addr.translation_offset;
> -		orig_end = r.end;
>  		r.start += translation_offset;
>  		r.end += translation_offset;
> -
> -		/* Exclude non-addressable range or non-addressable portion of range */
> -		r.end = min(r.end, iomem_resource.end);
> -		if (r.end <= r.start) {
> -			dev_info(&info->bridge->dev,
> -				"host bridge window [%#llx-%#llx] (ignored, not CPU addressable)\n",
> -				 (unsigned long long)r.start, orig_end);
> -			return AE_OK;
> -		} else if (orig_end != r.end) {
> -			dev_info(&info->bridge->dev,
> -				"host bridge window [%#llx-%#llx] ([%#llx-%#llx] ignored, not CPU addressable)\n",
> -				 (unsigned long long)r.start, orig_end,
> -				 (unsigned long long)r.end + 1, orig_end);
> -		}
> +	} else {
> +		return AE_OK;
>  	}
>  
> -	if (r.flags && resource_size(&r)) {
> -		r.name = info->name;
> -		info->res[info->res_num] = r;
> -		info->res_offset[info->res_num] = translation_offset;
> -		info->res_num++;
> -		if (!pci_use_crs)
> -			dev_printk(KERN_DEBUG, &info->bridge->dev,
> -				   "host bridge window %pR (ignored)\n", &r);
> +	/* Exclude non-addressable range or non-addressable portion of range */
> +	orig_end = r.end;
> +	r.end = min(r.end, iomem_resource.end);
> +	if (r.end <= r.start) {
> +		dev_info(&info->bridge->dev,
> +			"host bridge window [%#llx-%#llx] (ignored, not CPU addressable)\n",
> +			 (unsigned long long)r.start, orig_end);
> +		return AE_OK;
> +	} else if (orig_end != r.end) {
> +		dev_info(&info->bridge->dev,
> +			"host bridge window [%#llx-%#llx] ([%#llx-%#llx] ignored, not CPU addressable)\n",
> +			 (unsigned long long)r.start, orig_end,
> +			 (unsigned long long)r.end + 1, orig_end);

This hunk could be made a little more readable (diff ontop):

---
Index: b/arch/x86/pci/acpi.c
===================================================================
--- a/arch/x86/pci/acpi.c	2014-12-12 12:25:56.036682117 +0100
+++ b/arch/x86/pci/acpi.c	2014-12-12 12:25:11.892683404 +0100
@@ -270,26 +270,29 @@ static acpi_status setup_resource(struct
 
 	/* Exclude non-addressable range or non-addressable portion of range */
 	orig_end = r.end;
-	r.end = min(r.end, iomem_resource.end);
+	r.end	 = min(r.end, iomem_resource.end);
+
 	if (r.end <= r.start) {
 		dev_info(&info->bridge->dev,
 			"host bridge window [%#llx-%#llx] (ignored, not CPU addressable)\n",
 			 (unsigned long long)r.start, orig_end);
 		return AE_OK;
-	} else if (orig_end != r.end) {
+	}
+
+	if (orig_end != r.end) {
 		dev_info(&info->bridge->dev,
 			"host bridge window [%#llx-%#llx] ([%#llx-%#llx] ignored, not CPU addressable)\n",
 			 (unsigned long long)r.start, orig_end,
 			 (unsigned long long)r.end + 1, orig_end);
 	}
 
+	if (!pci_use_crs)
+		dev_printk(KERN_DEBUG, &info->bridge->dev, "host bridge window %pR (ignored)\n", &r);
+
 	r.name = info->name;
 	info->res[info->res_num] = r;
 	info->res_offset[info->res_num] = translation_offset;
 	info->res_num++;
-	if (!pci_use_crs)
-		dev_printk(KERN_DEBUG, &info->bridge->dev,
-			"host bridge window %pR (ignored)\n", &r);
 
 	return AE_OK;
 }

-- 
Regards/Gruss,
    Boris.

Sent from a fat crate under my desk. Formatting is fine.
--
--
To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
the body of a message to majordomo@xxxxxxxxxxxxxxx
More majordomo info at  http://vger.kernel.org/majordomo-info.html
Please read the FAQ at  http://www.tux.org/lkml/