From 97a70e548bd97d5a46ae9d44f24aafcc013fd701 Mon Sep 17 00:00:00 2001 From: "Rafael J. Wysocki" Date: Wed, 12 Nov 2008 23:22:35 +0100 Subject: x86, hibernate: fix breakage on x86_32 with CONFIG_NUMA set Impact: fix crash during hibernation on 32-bit NUMA The NUMA code on x86_32 creates special memory mapping that allows each node's pgdat to be located in this node's memory. For this purpose it allocates a memory area at the end of each node's memory and maps this area so that it is accessible with virtual addresses belonging to low memory. As a result, if there is high memory, these NUMA-allocated areas are physically located in high memory, although they are mapped to low memory addresses. Our hibernation code does not take that into account and for this reason hibernation fails on all x86_32 systems with CONFIG_NUMA=y and with high memory present. Fix this by adding a special mapping for the NUMA-allocated memory areas to the temporary page tables created during the last phase of resume. Signed-off-by: Rafael J. Wysocki Signed-off-by: Ingo Molnar diff --git a/arch/x86/include/asm/mmzone_32.h b/arch/x86/include/asm/mmzone_32.h index 485bdf0..07f1af4 100644 --- a/arch/x86/include/asm/mmzone_32.h +++ b/arch/x86/include/asm/mmzone_32.h @@ -34,10 +34,14 @@ static inline void get_memcfg_numa(void) extern int early_pfn_to_nid(unsigned long pfn); +extern void resume_map_numa_kva(pgd_t *pgd); + #else /* !CONFIG_NUMA */ #define get_memcfg_numa get_memcfg_numa_flat +static inline void resume_map_numa_kva(pgd_t *pgd) {} + #endif /* CONFIG_NUMA */ #ifdef CONFIG_DISCONTIGMEM diff --git a/arch/x86/mm/numa_32.c b/arch/x86/mm/numa_32.c index 847c164..8518c67 100644 --- a/arch/x86/mm/numa_32.c +++ b/arch/x86/mm/numa_32.c @@ -222,6 +222,41 @@ static void __init remap_numa_kva(void) } } +#ifdef CONFIG_HIBERNATION +/** + * resume_map_numa_kva - add KVA mapping to the temporary page tables created + * during resume from hibernation + * @pgd_base - temporary resume page directory + */ +void resume_map_numa_kva(pgd_t *pgd_base) +{ + int node; + + for_each_online_node(node) { + unsigned long start_va, start_pfn, size, pfn; + + start_va = (unsigned long)node_remap_start_vaddr[node]; + start_pfn = node_remap_start_pfn[node]; + size = node_remap_size[node]; + + printk(KERN_DEBUG "%s: node %d\n", __FUNCTION__, node); + + for (pfn = 0; pfn < size; pfn += PTRS_PER_PTE) { + unsigned long vaddr = start_va + (pfn << PAGE_SHIFT); + pgd_t *pgd = pgd_base + pgd_index(vaddr); + pud_t *pud = pud_offset(pgd, vaddr); + pmd_t *pmd = pmd_offset(pud, vaddr); + + set_pmd(pmd, pfn_pmd(start_pfn + pfn, + PAGE_KERNEL_LARGE_EXEC)); + + printk(KERN_DEBUG "%s: %08lx -> pfn %08lx\n", + __FUNCTION__, vaddr, start_pfn + pfn); + } + } +} +#endif + static unsigned long calculate_numa_remap_pages(void) { int nid; diff --git a/arch/x86/power/hibernate_32.c b/arch/x86/power/hibernate_32.c index f2b6e3f..81197c6 100644 --- a/arch/x86/power/hibernate_32.c +++ b/arch/x86/power/hibernate_32.c @@ -12,6 +12,7 @@ #include #include #include +#include /* Defined in hibernate_asm_32.S */ extern int restore_image(void); @@ -127,6 +128,9 @@ static int resume_physical_mapping_init(pgd_t *pgd_base) } } } + + resume_map_numa_kva(pgd_base); + return 0; } -- cgit v0.10.2 From 604d20554883cf03f888440d58ea7c6d36899839 Mon Sep 17 00:00:00 2001 From: "Rafael J. Wysocki" Date: Wed, 12 Nov 2008 23:26:14 +0100 Subject: x86: make NUMA on 32-bit depend on EXPERIMENTAL again My previous patch to make CONFIG_NUMA on x86_32 depend on BROKEN turned out to be unnecessary, after all, since the source of the hibernation vs CONFIG_NUMA problem turned out to be the fact that we didn't take the NUMA KVA remapping into account in the hibernation code. Signed-off-by: Rafael J. Wysocki Signed-off-by: Ingo Molnar diff --git a/arch/x86/Kconfig b/arch/x86/Kconfig index 93224b5..4cf0ab1 100644 --- a/arch/x86/Kconfig +++ b/arch/x86/Kconfig @@ -957,7 +957,7 @@ config ARCH_PHYS_ADDR_T_64BIT config NUMA bool "Numa Memory Allocation and Scheduler Support (EXPERIMENTAL)" depends on SMP - depends on X86_64 || (X86_32 && HIGHMEM64G && (X86_NUMAQ || X86_BIGSMP || X86_SUMMIT && ACPI) && BROKEN) + depends on X86_64 || (X86_32 && HIGHMEM64G && (X86_NUMAQ || X86_BIGSMP || X86_SUMMIT && ACPI) && EXPERIMENTAL) default n if X86_PC default y if (X86_NUMAQ || X86_SUMMIT || X86_BIGSMP) help -- cgit v0.10.2