[PATCH] ARM: conditional ioremap() fix for copy_oldmem_page()

Magnus Damm magnus.damm at gmail.com
Tue Jun 21 23:44:45 EDT 2011


From: Magnus Damm <damm at opensource.se>

Update the copy_oldmem_page() function to ioremap() only
when accessing memory that is outside the regular range
of system memory that is managed by the kernel.

Without this patch a warning is triggered in the ARM-specific
ioremap implementation, see WARN_ON(pfn_valid()) in ioremap.c

The copy_oldmem_page() function is used by the secondary crash
kernel to access memory using the /proc/vmcore code implemented
in fs/proc/vmcore.c. To pass information from the first kernel
to the secondary crash kernel a kernel command line option is
used to point out where the elf core hdr is located.

The crash kernel is loaded through kexec-tools which also contains
code that reserves memory for the elfcorehdr= option. This memory
block is reserved _inside_ the main system memory of the secondary
kernel. The /proc/vmcore code in the secondary kernel is however
using copy_oldmem_page() to access both this elfcorehdr area and
the rest of the memory used by the the first kernel.

So the copy_oldmem_page() function is used to access data that
may be located in system memory, or it may be outside. Always
using ioremap will not work, so this patch makes it conditional
based on pfn_valid().

For more details please look at the sh7372-based example here:
http://permalink.gmane.org/gmane.linux.ports.sh.devel/11502

Signed-off-by: Magnus Damm <damm at opensource.se>
---

 arch/arm/kernel/crash_dump.c |   14 ++++++++++----
 1 file changed, 10 insertions(+), 4 deletions(-)

--- 0001/arch/arm/kernel/crash_dump.c
+++ work/arch/arm/kernel/crash_dump.c	2011-06-18 20:59:49.000000000 +0900
@@ -39,9 +39,13 @@ ssize_t copy_oldmem_page(unsigned long p
 	if (!csize)
 		return 0;
 
-	vaddr = ioremap(pfn << PAGE_SHIFT, PAGE_SIZE);
-	if (!vaddr)
-		return -ENOMEM;
+	if (pfn_valid(pfn)) {
+		vaddr = phys_to_virt(pfn << PAGE_SHIFT);
+	} else {
+		vaddr = ioremap(pfn << PAGE_SHIFT, PAGE_SIZE);
+		if (!vaddr)
+			return -ENOMEM;
+	}
 
 	if (userbuf) {
 		if (copy_to_user(buf, vaddr + offset, csize)) {
@@ -52,6 +56,8 @@ ssize_t copy_oldmem_page(unsigned long p
 		memcpy(buf, vaddr + offset, csize);
 	}
 
-	iounmap(vaddr);
+	if (!pfn_valid(pfn))
+		iounmap(vaddr);
+
 	return csize;
 }



More information about the kexec mailing list