Skip to contentSkip to conversation

kexec-tools and multiboot breakage

PATCH In review Peter Chubb 2 participants 4 messages Jan 15, 2014 – Jan 16, 2014

Subsystems (2)
lore
Share & tools
lore
#0 lore raw

Hi Simon,
When you get a moment, please check out
http://bugs.debian.org/cgi-bin/bugreport.cgi?bug=735360

Quick summary:
If your kernel is configured to protect the first few pages of physical memory, then low memory doesn't start at location 0 --- so when scanning the memory ranges from /proc/iomem, the code in kexec/arch/i386/kexec-multiboot-x86.c never fills in mem_lower.

I think it's safe to assume that if a memory segment starts at 64k or lower, it can be extended to location 0 because of the kernel protection on a PC99-style architecture.

Peter C
--
Dr Peter Chubb peter.chubb AT nicta.com.au
http://www.ssrg.nicta.com.au Software Systems Research Group/NICTA

_______________________________________________
kexec mailing list
[email protected]
http://lists.infradead.org/mailman/listinfo/kexec

#1 lore raw
Quoted 10 lines · Peter Chubb

Hi Simon,
When you get a moment, please check out
http://bugs.debian.org/cgi-bin/bugreport.cgi?bug=735360

Quick summary:
If your kernel is configured to protect the first few pages of physical memory, then low memory doesn't start at location 0 --- so when scanning the memory ranges from /proc/iomem, the code in kexec/arch/i386/kexec-multiboot-x86.c never fills in mem_lower.

Quoted 3 lines

I think it's safe to assume that if a memory segment starts at 64k or lower, it can be extended to location 0 because of the kernel protection on a PC99-style architecture.

Hi Peter,

that approach seems reasonable to me. Could you formally post the patch to the kexec list with me CCed? I would like a Signed-off-by line. Thanks.

_______________________________________________
kexec mailing list
[email protected]
http://lists.infradead.org/mailman/listinfo/kexec

#2 lore raw

PATCH 1/1 Fix value of mbi->mem_lower for multiboot-x86

In the multiboot header, there is a field, `mem_lower' that is meant to contain the size of memory starting at zero and ending below 640k. If your kernel is compiled with CONFIG_X86_RESERVE_LOW non zero (the usual case), then a hole is inserted into kernel's physical memory map at zero, so the test to find the size of this region in kexec/arch/i386/kexec-multiboot-x86.c never succeeds, so the value is always zero.

On a PC99 architecture, there is always memory at physycal address zero; assume that a region that starts below 64k actually starts at zero, and use it for the mem_lower variable.

---
 kexec/arch/i386/kexec-multiboot-x86.c |   14 +++++++++++---
 1 file changed, 11 insertions(+), 3 deletions(-)
kexec/arch/i386/kexec-multiboot-x86.c +11 -3
Index: kexec-tools-2.0.4/kexec/arch/i386/kexec-multiboot-x86.c
===================================================================
@@ -261,10 +261,18 @@ int multiboot_x86_load(int argc, char **
 		mmap[i].length_high    = length >> 32;
 		if (range[i].type == RANGE_RAM) {
 			mmap[i].Type = 1; /* RAM */
-			/* Is this the "low" memory? */
-			if ((range[i].start == 0)
-			    && (range[i].end > mem_lower))
+			/*
+                         * Is this the "low" memory?  Can't just test
+                         * against zero, because Linux protects (and
+                         * hides) the first few pages of physical
+                         * memory.
+                         */
+
+			if ((range[i].start <= 64*1024)
+			    && (range[i].end > mem_lower)) {
+                                range[i].start = 0;
 				mem_lower = range[i].end;
+                        }
 			/* Is this the "high" memory? */
 			if ((range[i].start <= 0x100000)
 			    && (range[i].end > mem_upper + 0x100000))

--
Dr Peter Chubb peter.chubb AT nicta.com.au
http://www.ssrg.nicta.com.au Software Systems Research Group/NICTA

_______________________________________________
kexec mailing list
[email protected]
http://lists.infradead.org/mailman/listinfo/kexec

#3 lore raw

Re: [PATCH] Fix value of mbi->mem_lower for multiboot-x86

Quoted 9 lines · Peter Chubb

In the multiboot header, there is a field, `mem_lower' that is meant to contain the size of memory starting at zero and ending below 640k. If your kernel is compiled with CONFIG_X86_RESERVE_LOW non zero (the usual case), then a hole is inserted into kernel's physical memory map at zero, so the test to find the size of this region in kexec/arch/i386/kexec-multiboot-x86.c never succeeds, so the value is always zero.

Quoted 5 lines

On a PC99 architecture, there is always memory at physycal address zero; assume that a region that starts below 64k actually starts at zero, and use it for the mem_lower variable.

Thanks Peter, I have applied this.

Quoted 33 lines
---
 kexec/arch/i386/kexec-multiboot-x86.c |   14 +++++++++++---
 1 file changed, 11 insertions(+), 3 deletions(-)

Index: kexec-tools-2.0.4/kexec/arch/i386/kexec-multiboot-x86.c
===================================================================

kexec/arch/i386/kexec-multiboot-x86.c +11 -4
@@ -261,10 +261,18 @@ int multiboot_x86_load(int argc, char **
 		mmap[i].length_high    = length >> 32;
 		if (range[i].type == RANGE_RAM) {
 			mmap[i].Type = 1; /* RAM */
-			/* Is this the "low" memory? */
-			if ((range[i].start == 0)
-			    && (range[i].end > mem_lower))
+			/*
+                         * Is this the "low" memory?  Can't just test
+                         * against zero, because Linux protects (and
+                         * hides) the first few pages of physical
+                         * memory.
+                         */
+
+			if ((range[i].start <= 64*1024)
+			    && (range[i].end > mem_lower)) {
+                                range[i].start = 0;
 				mem_lower = range[i].end;
+                        }
 			/* Is this the "high" memory? */
 			if ((range[i].start <= 0x100000)
 			    && (range[i].end > mem_upper + 0x100000))
 
--
Quoted 3 lines

Dr Peter Chubb peter.chubb AT nicta.com.au
http://www.ssrg.nicta.com.au Software Systems Research Group/NICTA

_______________________________________________
kexec mailing list
[email protected]
http://lists.infradead.org/mailman/listinfo/kexec