Skip to contentSkip to conversation

EFI iounpam fix for acpi_os_unmap_memory

PATCH In review Edgar Hucek 4 participants 6 messages Feb 20, 2006 – Feb 22, 2006 ACPI

lore
Share & tools
lore
#0 lore raw

PATCH 1/1 EFI iounpam fix for acpi_os_unmap_memory

When EFI is enabled acpi_os_unmap_memory trys to unmap memory which was not mapped by acpi_os_map_memory.

This patch for it.

diff -uNr linux-2.6.16-rc4.orig/drivers/acpi/osl.c linux-2.6.16-rc4/drivers/acpi/osl.c
--- linux-2.6.16-rc4.orig/drivers/acpi/osl.c 2006-02-19
18:48:58.000000000 +0100
+++ linux-2.6.16-rc4/drivers/acpi/osl.c 2006-02-20 15:31:44.000000000 +0100 @@ -208,7 +208,13 @@

void acpi_os_unmap_memory(void __iomem * virt, acpi_size size) {
- iounmap(virt);
+ if(efi_enabled) {
+ if (!(EFI_MEMORY_WB &
efi_mem_attributes(virt_to_phys(virt)))) {
+ iounmap(virt);
+ }
+ } else {
+ iounmap(virt);
+ }
}
EXPORT_SYMBOL_GPL(acpi_os_unmap_memory);

#1 lore raw
Quoted 3 lines · Edgar Hucek

When EFI is enabled acpi_os_unmap_memory trys to unmap memory
which was not mapped by acpi_os_map_memory.

Your email client replaces tabs with spaces and wordwraps things.

The patch could be cleaned up a bit..

Matt, ACPi people: please ack or nack asap.

From: Edgar Hucek <[email protected]>

When EFI is enabled acpi_os_unmap_memory t] rys to unmap memory which was not mapped by acpi_os_map_memory.

---

 drivers/acpi/osl.c |    4 ++++
 1 files changed, 4 insertions(+)
drivers/acpi/osl.c renamed from drivers/acpi/osl.c~efi-iounpam-fix-for-acpi_os_unmap_memory +4 -0
diff -puN drivers/acpi/osl.c~efi-iounpam-fix-for-acpi_os_unmap_memory drivers/acpi/osl.c
@@ -208,6 +208,10 @@ EXPORT_SYMBOL_GPL(acpi_os_map_memory);
 
 void acpi_os_unmap_memory(void __iomem * virt, acpi_size size)
 {
+	/* Don't unmap memory which was not mapped by acpi_os_map_memory */
+	if (efi_enabled &&
+	    (efi_mem_attributes(virt_to_phys(virt)) & EFI_MEMORY_WB))
+		return;
 	iounmap(virt);
 }
 EXPORT_SYMBOL_GPL(acpi_os_unmap_memory);

_

#2 lore raw

drivers/acpi/osl.c EXPORT_SYMBOL_GPL(acpi_os_map_memory); · line 214

Quoted 7 lines · Andrew Morton

void acpi_os_unmap_memory(void __iomem * virt, acpi_size size) {

drivers/acpi/osl.c renamed from drivers/acpi/osl.c~efi-iounpam-fix-for-acpi_os_unmap_memory +4 -0
+	/* Don't unmap memory which was not mapped by acpi_os_map_memory */
+	if (efi_enabled &&
+	    (efi_mem_attributes(virt_to_phys(virt)) & EFI_MEMORY_WB))
+		return;

The patch is wrong because if the address came from ioremap virt_to_phys doesn't give the real physical address. Also looking at acpi_os_map_memory it doesn't quite match the logic there.

One working way to check for ioremap memory is
virt >= VMALLOC_START && virt < VMALLOC_END

-Andi

#3 lore raw
Quoted 16 lines · Andi Kleen

drivers/acpi/osl.c EXPORT_SYMBOL_GPL(acpi_os_map_memory); · line 214

Quoted 7 lines · Andrew Morton

void acpi_os_unmap_memory(void __iomem * virt, acpi_size size) {

drivers/acpi/osl.c renamed from drivers/acpi/osl.c~efi-iounpam-fix-for-acpi_os_unmap_memory +4 -0
+	/* Don't unmap memory which was not mapped by acpi_os_map_memory */
+	if (efi_enabled &&
+	    (efi_mem_attributes(virt_to_phys(virt)) & EFI_MEMORY_WB))
+		return;

The patch is wrong because if the address came from ioremap virt_to_phys doesn't give the real physical address. Also looking at acpi_os_map_memory it doesn't quite match the logic there.

Quoted 3 lines

One working way to check for ioremap memory is
virt >= VMALLOC_START && virt < VMALLOC_END

OK, thanks. I don't think we actually know who is trying to unmap some memory which acpi didn't map.

Edgar, can you please describe the bug which you're trying to fix?

#4 lore raw
Quoted 17 lines · Andrew Morton
Quoted 16 lines · Andi Kleen

Review

Quoted 7 lines · Andrew Morton

void acpi_os_unmap_memory(void __iomem * virt, acpi_size size) {

(quoted diff) +4 -0
+	/* Don't unmap memory which was not mapped by acpi_os_map_memory */
+	if (efi_enabled &&
+	    (efi_mem_attributes(virt_to_phys(virt)) & EFI_MEMORY_WB))
+		return;

The patch is wrong because if the address came from ioremap virt_to_phys doesn't give the real physical address. Also looking at acpi_os_map_memory it doesn't quite match the logic there.

Quoted 8 lines
Quoted 3 lines

One working way to check for ioremap memory is
virt >= VMALLOC_START && virt < VMALLOC_END

OK, thanks. I don't think we actually know who is trying to unmap some memory which acpi didn't map.

Edgar, can you please describe the bug which you're trying to fix?

I think the bug is clear - the logic in acpi_os_unmap_memory needs to match what acpi_os_map_memory() does for EFI. In particular this means not calling iounmap.

He probably has a EFI system where this caused troubles.

But think even acpi_os_miap_memory is broken - I doubt the efi_mem_attributes thing guarantees that the memory is mapped. I guess this was added for IA64 because ioremap used to return uncached memory there and IA64 doesn't like this for memory accesses or something like that.

But Bjorn afaik recently fixed this by making ioremap handle it. So the right fix probably would be to just drop all the efi_enabled gunk in acpi_os_map_memory and just always use ioremap()

Also removed this ptr > ULONG_MAX check. Obvious it can never trigger.

<untested patch follows>

drivers/acpi/osl.c +4 -17
Index: linux/drivers/acpi/osl.c
===================================================================
@@ -182,23 +182,10 @@ acpi_status
 acpi_os_map_memory(acpi_physical_address phys, acpi_size size,
 		   void __iomem ** virt)
 {
-	if (efi_enabled) {
-		if (EFI_MEMORY_WB & efi_mem_attributes(phys)) {
-			*virt = (void __iomem *)phys_to_virt(phys);
-		} else {
-			*virt = ioremap(phys, size);
-		}
-	} else {
-		if (phys > ULONG_MAX) {
-			printk(KERN_ERR PREFIX "Cannot map memory that high\n");
-			return AE_BAD_PARAMETER;
-		}
-		/*
-		 * ioremap checks to ensure this is in reserved space
-		 */
-		*virt = ioremap((unsigned long)phys, size);
-	}
-
+	/*
+	 * ioremap checks to ensure this is in reserved space
+	 */
+	*virt = ioremap((unsigned long)phys, size);
 	if (!*virt)
 		return AE_NO_MEMORY;
 
Quoted 1 line
#5 lore raw
Quoted 5 lines · Andi Kleen
Quoted 3 lines · Andrew Morton

OK, thanks. I don't think we actually know who is trying to unmap some memory which acpi didn't map.

Quoted 7 lines
Quoted 1 line

Edgar, can you please describe the bug which you're trying to fix?

I think the bug is clear - the logic in acpi_os_unmap_memory needs to match what acpi_os_map_memory() does for EFI. In particular this means not calling iounmap.

He probably has a EFI system where this caused troubles.

This EFI system is the new Intel Core Duo based Apple iMac (Edgar described the process of booting Linux on this pretty box at
http://www.mactel-linux.org/)

In particular, the iounmap problem is visible in the logs at
http://www.mactel-linux.org/wiki/Dmesg

Stelian.

Signature · 6 lines
Stelian Pop <[email protected]>

-
To unsubscribe from this list: send the line "unsubscribe linux-acpi" in
the body of a message to [email protected]
More majordomo info at  http://vger.kernel.org/majordomo-info.html