2008-01-28 09:10:28

by Yinghai Lu

[permalink] [raw]
Subject: [PATCH] x86_64: mark x86_cpu_to_node_map_init to __initdata like other xx_init

[PATCH] x86_64: mark x86_cpu_to_node_map_init to __initdata like other xx_init

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

diff --git a/arch/x86/mm/numa_64.c b/arch/x86/mm/numa_64.c
index f0e5cab..d7af3fd 100644
--- a/arch/x86/mm/numa_64.c
+++ b/arch/x86/mm/numa_64.c
@@ -31,7 +31,7 @@ bootmem_data_t plat_node_bdata[MAX_NUMNODES];

struct memnode memnode;

-int x86_cpu_to_node_map_init[NR_CPUS] = {
+int x86_cpu_to_node_map_init[NR_CPUS] __initdata = {
[0 ... NR_CPUS-1] = NUMA_NO_NODE
};
void *x86_cpu_to_node_map_early_ptr;
diff --git a/include/asm-x86/topology.h b/include/asm-x86/topology.h
index 8af05a9..d3340de 100644
--- a/include/asm-x86/topology.h
+++ b/include/asm-x86/topology.h
@@ -35,7 +35,7 @@ extern int cpu_to_node_map[];

#else
DECLARE_PER_CPU(int, x86_cpu_to_node_map);
-extern int x86_cpu_to_node_map_init[];
+extern int __initdata x86_cpu_to_node_map_init[];
extern void *x86_cpu_to_node_map_early_ptr;
/* Returns the number of the current Node. */
#define numa_node_id() (early_cpu_to_node(raw_smp_processor_id()))


2008-01-28 10:35:03

by Ingo Molnar

[permalink] [raw]
Subject: Re: [PATCH] x86_64: mark x86_cpu_to_node_map_init to __initdata like other xx_init


* Yinghai Lu <[email protected]> wrote:

> -int x86_cpu_to_node_map_init[NR_CPUS] = {
> +int x86_cpu_to_node_map_init[NR_CPUS] __initdata = {
> [0 ... NR_CPUS-1] = NUMA_NO_NODE
> };

i remember some linker warning here. While this array should indeed only
be used in early init, that decision is dynamic and our linker warnings
do not notice it. There's a special marker for such cases:
__initdata_refok. But ... i'm slightly nervous about turning off a vital
warning like that.

Sam, it would be nice to have a DEBUG_INITDATA mode of operation: in
this case free_initmem() would not truly free those pages but would
unmap them via:

kernel_map_pages(page, nrpages, 0);

could be made dependent on DEBUG_PAGEALLOC.

If this debugging is enabled then if any code references it, we get a
hard page fault. If we had a debug mode like that then bugs in this area
would not go unnoticed.

Or perhaps just make this part of normal DEBUG_PAGEALLOC. Like the patch
below on top of latest x86.git. Hm?

Ingo

------------------->
Subject: x86: init memory debugging
From: Ingo Molnar <[email protected]>

debug incorrect/late access to init memory, by permanently unmapping
the init memory ranges. Depends on CONFIG_DEBUG_PAGEALLOC=y.

Signed-off-by: Ingo Molnar <[email protected]>
---
arch/x86/mm/init_32.c | 12 ++++++++++++
arch/x86/mm/init_64.c | 11 +++++++++++
2 files changed, 23 insertions(+)

Index: linux-x86.q/arch/x86/mm/init_32.c
===================================================================
--- linux-x86.q.orig/arch/x86/mm/init_32.c
+++ linux-x86.q/arch/x86/mm/init_32.c
@@ -794,6 +794,18 @@ void free_init_pages(char *what, unsigne
unsigned long addr;

/*
+ * If debugging page accesses then do not free this memory but
+ * mark them not present - any buggy init-section access will
+ * create a kernel page fault:
+ */
+#ifdef CONFIG_DEBUG_PAGEALLOC
+ printk(KERN_INFO "debug: unmapping init memory %08lx..%08lx\n",
+ begin, PAGE_ALIGN(end));
+ set_memory_np(begin, (end - begin) >> PAGE_SHIFT);
+ return;
+#endif
+ set_memory_rw(begin, (end - begin) >> PAGE_SHIFT);
+ /*
* We just marked the kernel text read only above, now that
* we are going to free part of that, we need to make that
* writeable first.
Index: linux-x86.q/arch/x86/mm/init_64.c
===================================================================
--- linux-x86.q.orig/arch/x86/mm/init_64.c
+++ linux-x86.q/arch/x86/mm/init_64.c
@@ -580,6 +580,17 @@ void free_init_pages(char *what, unsigne
if (begin >= end)
return;

+ /*
+ * If debugging page accesses then do not free this memory but
+ * mark them not present - any buggy init-section access will
+ * create a kernel page fault:
+ */
+#ifdef CONFIG_DEBUG_PAGEALLOC
+ printk(KERN_INFO "debug: unmapping init memory %08lx..%08lx\n",
+ begin, PAGE_ALIGN(end));
+ set_memory_np(begin, (end - begin) >> PAGE_SHIFT);
+ return;
+#endif
printk(KERN_INFO "Freeing %s: %luk freed\n", what, (end - begin) >> 10);

for (addr = begin; addr < end; addr += PAGE_SIZE) {