Allocate sparsemem memmap above 4G on X86_64

On system with huge amount of physical memory.
VFS cache and memory memmap may eat all available system memory under 4G, then system may fail to allocated swiotlb bounce buffer.

There was a fix in arch/x86_64/mm/numa.c, but that fix does not cover sparsemem model.
This patch add fix to sparsemem model.

diff -Nraup a/include/asm-x86_64/mmzone.h b/include/asm-x86_64/mmzone.h

include/asm-x86_64/mmzone.h +5 -0

@@ -52,5 +52,10 @@ extern int pfn_valid(unsigned long pfn);
 #define FAKE_NODE_MIN_HASH_MASK	(~(FAKE_NODE_MIN_SIZE - 1uL))
 #endif
 
+#define ARCH_HAS_ALLOC_BOOTMEM_HIGH_NODE 1
+#define alloc_bootmem_high_node(pgdat,size) \
+({__alloc_bootmem_core(pgdat->bdata, size, SMP_CACHE_BYTES, (4UL*1024*1024*1024), 0);})
+
+
 #endif
 #endif

diff -Nraup a/include/linux/bootmem.h b/include/linux/bootmem.h

include/linux/bootmem.h +3 -0

@@ -131,5 +131,8 @@ extern void *alloc_large_system_hash(con
 #endif
 extern int hashdist;		/* Distribute hashes across NUMA nodes? */
 
+#ifndef ARCH_HAS_ALLOC_BOOTMEM_HIGH_NODE
+#define alloc_bootmem_high_node(pgdat, size) ({NULL;})
+#endif
 
 #endif /* _LINUX_BOOTMEM_H */

diff -Nraup a/mm/sparse.c b/mm/sparse.c

Quoted 1 line · On 17 May 2007 10:40:07 +0800 Zou Nan hai <[email protected]> wrote:

Please always prefer to use static inline functions rather than macros. They are more readable, they are more likely to have comments attached to them and they provide typechecking.

Please prefer to uninline functions by default. One reason for this is that adding inlines to headers increases include complexity. This code is all __init anyway, so the possible few bytes of text will get removed.

Try to avoid using the ARCH_HAS_FOO thing. We have two alternatives:

a) use __attribute__((weak))

b) do:

	extern void foo(void);
	#define foo foo

   then, elsewhere,

	#ifndef foo
	#define foo() bar()
	#endif

Both tricks avoid the introduction of two new symbols into the global namespace to solve a single problem.

Quoted 22 lines · Andrew Morton

Quoted 1 line · On 17 May 2007 10:40:07 +0800 Zou Nan hai <[email protected]> wrote:

Please always prefer to use static inline functions rather than macros. They are more readable, they are more likely to have comments attached to them and they provide typechecking.

Please prefer to uninline functions by default. One reason for this is that adding inlines to headers increases include complexity. This code is all __init anyway, so the possible few bytes of text will get removed.

Try to avoid using the ARCH_HAS_FOO thing. We have two alternatives:

a) use __attribute__((weak))

b) do:

	extern void foo(void);
	#define foo foo

Quoted 8 lines

   then, elsewhere,

	#ifndef foo
	#define foo() bar()
	#endif

Both tricks avoid the introduction of two new symbols into the global namespace to solve a single problem.

On systems with huge amount of physical memory, VFS cache and memory
memmap may eat all available system memory under 4G, then the system may fail to allocate swiotlb bounce buffer.
There was a fix for this issue in arch/x86_64/mm/numa.c, but that fix
dose not cover sparsemem model.
This patch add fix to sparsemem model by first try to allocate memmap
above 4G.

Signed-off-by Zou Nan hai <[email protected]>

Acked-by Suresh Siddha <[email protected]>

---
 arch/x86_64/mm/init.c |    6 ++++++
 mm/sparse.c           |   11 +++++++++++
 2 files changed, 17 insertions(+)

diff -Nraup a/arch/x86_64/mm/init.c b/arch/x86_64/mm/init.c

arch/x86_64/mm/init.c +6 -0

@@ -761,3 +761,9 @@ int in_gate_area_no_task(unsigned long a
 {
 	return (addr >= VSYSCALL_START) && (addr < VSYSCALL_END);
 }
+
+void *alloc_bootmem_high_node(pg_data_t *pgdat, unsigned long size)
+{
+	return __alloc_bootmem_core(pgdat->bdata, size,
+			SMP_CACHE_BYTES, (4UL*1024*1024*1024), 0);
+}

diff -Nraup a/mm/sparse.c b/mm/sparse.c

mm/sparse.c +6 -0

@@ -209,6 +209,12 @@ static int __meminit sparse_init_one_sec
 	return 1;
 }
 
+__attribute__((weak))
+void *alloc_bootmem_high_node(pg_data_t *pgdat, unsigned long size)
+{
+	return NULL;
+}
+
 static struct page __init *sparse_early_mem_map_alloc(unsigned long pnum)
 {
 	struct page *map;

@@ -219,6 +225,11 @@ static struct page __init *sparse_early_
 	if (map)