[PATCH V4 06/10] MIPS: Reserve extra memory for crash dump

[Huacai Chen <chenhc@xxxxxxxxxx>]

Traditionally, MIPS's contiguous low memory can be as less as 256M, so
crashkernel=X@Y may be unable to large enough in some cases. Moreover,
for the "multi numa node + sparse memory model" case, it is attempt to
allocate section_mem_maps on every node. Thus, if the total memory of a
node is more than 1GB, we reserve the top 128MB for the crash kernel.

Signed-off-by: Huacai Chen <chenhc@xxxxxxxxxx>
---
 arch/mips/kernel/setup.c | 55 ++++++++++++++++++++++++++++++++++++++++++++++++
 1 file changed, 55 insertions(+)

diff --git a/arch/mips/kernel/setup.c b/arch/mips/kernel/setup.c
index c71d1eb..9f2d5d9 100644
--- a/arch/mips/kernel/setup.c
+++ b/arch/mips/kernel/setup.c
@@ -28,6 +28,7 @@
 #include <linux/dma-contiguous.h>
 #include <linux/decompress/generic.h>
 #include <linux/of_fdt.h>
+#include <linux/crash_dump.h>
 
 #include <asm/addrspace.h>
 #include <asm/bootinfo.h>
@@ -830,6 +831,52 @@ static void __init request_crashkernel(struct resource *res)
 #define BUILTIN_EXTEND_WITH_PROM	\
 	IS_ENABLED(CONFIG_MIPS_CMDLINE_BUILTIN_EXTEND)
 
+/* Traditionally, MIPS's contiguous low memory is 256M, so crashkernel=X@Y is
+ * unable to be large enough in some cases. Thus, if the total memory of a node
+ * is more than 1GB, we reserve the top 128MB for the crash kernel */
+static void reserve_crashm_region(int node, unsigned long s0, unsigned long e0)
+{
+#ifdef CONFIG_KEXEC
+	if (crashk_res.start == crashk_res.end)
+		return;
+
+	if ((e0 - s0) <= (SZ_1G >> PAGE_SHIFT))
+		return;
+
+	s0 = e0 - (SZ_128M >> PAGE_SHIFT);
+
+	reserve_bootmem_node(NODE_DATA(node), PFN_PHYS(s0),
+			(e0 - s0) << PAGE_SHIFT, BOOTMEM_DEFAULT);
+#endif
+}
+
+static void reserve_oldmem_region(int node, unsigned long s0, unsigned long e0)
+{
+#ifdef CONFIG_CRASH_DUMP
+	unsigned long s1, e1;
+
+	if (!is_kdump_kernel())
+		return;
+
+	if ((e0 - s0) > (SZ_1G >> PAGE_SHIFT))
+		e0 = e0 - (SZ_128M >> PAGE_SHIFT);
+
+	/* boot_mem_map.map[0] is crashk_res reserved by primary kernel */
+	s1 = PFN_UP(boot_mem_map.map[0].addr);
+	e1 = PFN_DOWN(boot_mem_map.map[0].addr + boot_mem_map.map[0].size);
+
+	if (node == 0) {
+		reserve_bootmem_node(NODE_DATA(node), PFN_PHYS(s0),
+				(s1 - s0) << PAGE_SHIFT, BOOTMEM_DEFAULT);
+		reserve_bootmem_node(NODE_DATA(node), PFN_PHYS(e1),
+				(e0 - e1) << PAGE_SHIFT, BOOTMEM_DEFAULT);
+	} else {
+		reserve_bootmem_node(NODE_DATA(node), PFN_PHYS(s0),
+				(e0 - s0) << PAGE_SHIFT, BOOTMEM_DEFAULT);
+	}
+#endif
+}
+
 /*
  * arch_mem_init - initialize memory management subsystem
  *
@@ -842,6 +843,8 @@ static void __init request_crashkernel(struct resource *res)
  */
 static void __init arch_mem_init(char **cmdline_p)
 {
+	unsigned int node;
+	unsigned long start_pfn, end_pfn;
 	struct memblock_region *reg;
 	extern void plat_mem_setup(void);
 
@@ -923,6 +972,12 @@ static void __init arch_mem_init(char **cmdline_p)
 				crashk_res.end - crashk_res.start + 1,
 				BOOTMEM_DEFAULT);
 #endif
+	for_each_online_node(node) {
+		get_pfn_range_for_nid(node, &start_pfn, &end_pfn);
+		reserve_crashm_region(node, start_pfn, end_pfn);
+		reserve_oldmem_region(node, start_pfn, end_pfn);
+	}
+
 	device_tree_init();
 	sparse_init();
 	plat_swiotlb_setup();
-- 
2.7.0