> Am 13.01.2020 um 09:41 schrieb Anshuman Khandual <anshuman.khandual@xxxxxxx>: > > > >> On 01/13/2020 01:07 PM, Mike Rapoport wrote: >>> On Fri, Jan 10, 2020 at 08:39:12AM +0530, Anshuman Khandual wrote: >>> On arm64 platform boot memory should never be hot removed due to certain >>> platform specific constraints. Hence the platform would like to override >>> earlier added arch call back arch_memory_removable() for this purpose. In >>> order to reject boot memory hot removal request, it needs to first track >>> them at runtime. In the future, there might be other platforms requiring >>> runtime boot memory enumeration. Hence lets expand the existing generic >>> memblock framework for this purpose rather then creating one just for >>> arm64 platforms. >>> >>> This introduces a new memblock flag MEMBLOCK_BOOT along with helpers which >>> can be marked by given platform on all memory regions discovered during >>> boot. >> >> We already have MEMBLOCK_HOTPLUG to mark hotpluggable region. Can't we use >> it for your use-case? > > At present MEMBLOCK_HOTPLUG flag helps in identifying parts of boot memory > as hotpluggable as indicated by the firmware. This information is then used > to avoid those regions during standard memblock_alloc_*() API requests and > later marking them as ZONE_MOVABLE when buddy gets initialized. > > Memory hot remove does not check for MEMBLOCK_HOTPLUG flag as a requirement > before initiating the process. We could probably use this flag if generic > hot remove can be changed to check for MEMBLOCK_HOTPLUG as a prerequisite > which will require changes to memblock handling (boot and runtime) on all > existing platforms currently supporting hot remove. But what about handling > the movable boot memory created with movablecore/kernelcore command line, > should generic MM update their memblock regions with MEMBLOCK_HOTPLUG ? As I said in my other mail, just disallow offlining of the affected (boot) memory blocks using a memory notifier and you should be good to go. No changes in memory unplug code required. > >> >>> Cc: Mike Rapoport <rppt@xxxxxxxxxxxxx> >>> Cc: Andrew Morton <akpm@xxxxxxxxxxxxxxxxxxxx> >>> Signed-off-by: Anshuman Khandual <anshuman.khandual@xxxxxxx> >>> --- >>> include/linux/memblock.h | 10 ++++++++++ >>> mm/memblock.c | 37 +++++++++++++++++++++++++++++++++++++ >>> 2 files changed, 47 insertions(+) >>> >>> diff --git a/include/linux/memblock.h b/include/linux/memblock.h >>> index b38bbef..fb04c87 100644 >>> --- a/include/linux/memblock.h >>> +++ b/include/linux/memblock.h >>> @@ -31,12 +31,14 @@ extern unsigned long long max_possible_pfn; >>> * @MEMBLOCK_HOTPLUG: hotpluggable region >>> * @MEMBLOCK_MIRROR: mirrored region >>> * @MEMBLOCK_NOMAP: don't add to kernel direct mapping >>> + * @MEMBLOCK_BOOT: memory received from firmware during boot >>> */ >>> enum memblock_flags { >>> MEMBLOCK_NONE = 0x0, /* No special request */ >>> MEMBLOCK_HOTPLUG = 0x1, /* hotpluggable region */ >>> MEMBLOCK_MIRROR = 0x2, /* mirrored region */ >>> MEMBLOCK_NOMAP = 0x4, /* don't add to kernel direct mapping */ >>> + MEMBLOCK_BOOT = 0x8, /* memory received from firmware during boot */ >>> }; >>> >>> /** >>> @@ -116,6 +118,8 @@ int memblock_reserve(phys_addr_t base, phys_addr_t size); >>> void memblock_trim_memory(phys_addr_t align); >>> bool memblock_overlaps_region(struct memblock_type *type, >>> phys_addr_t base, phys_addr_t size); >>> +int memblock_mark_boot(phys_addr_t base, phys_addr_t size); >>> +int memblock_clear_boot(phys_addr_t base, phys_addr_t size); >>> int memblock_mark_hotplug(phys_addr_t base, phys_addr_t size); >>> int memblock_clear_hotplug(phys_addr_t base, phys_addr_t size); >>> int memblock_mark_mirror(phys_addr_t base, phys_addr_t size); >>> @@ -216,6 +220,11 @@ static inline bool memblock_is_nomap(struct memblock_region *m) >>> return m->flags & MEMBLOCK_NOMAP; >>> } >>> >>> +static inline bool memblock_is_boot(struct memblock_region *m) >>> +{ >>> + return m->flags & MEMBLOCK_BOOT; >>> +} >>> + >>> #ifdef CONFIG_HAVE_MEMBLOCK_NODE_MAP >>> int memblock_search_pfn_nid(unsigned long pfn, unsigned long *start_pfn, >>> unsigned long *end_pfn); >>> @@ -449,6 +458,7 @@ void memblock_cap_memory_range(phys_addr_t base, phys_addr_t size); >>> void memblock_mem_limit_remove_map(phys_addr_t limit); >>> bool memblock_is_memory(phys_addr_t addr); >>> bool memblock_is_map_memory(phys_addr_t addr); >>> +bool memblock_is_boot_memory(phys_addr_t addr); >>> bool memblock_is_region_memory(phys_addr_t base, phys_addr_t size); >>> bool memblock_is_reserved(phys_addr_t addr); >>> bool memblock_is_region_reserved(phys_addr_t base, phys_addr_t size); >>> diff --git a/mm/memblock.c b/mm/memblock.c >>> index 4bc2c7d..e10207f 100644 >>> --- a/mm/memblock.c >>> +++ b/mm/memblock.c >>> @@ -865,6 +865,30 @@ static int __init_memblock memblock_setclr_flag(phys_addr_t base, >>> } >>> >>> /** >>> + * memblock_mark_bootmem - Mark boot memory with flag MEMBLOCK_BOOT. >>> + * @base: the base phys addr of the region >>> + * @size: the size of the region >>> + * >>> + * Return: 0 on success, -errno on failure. >>> + */ >>> +int __init_memblock memblock_mark_boot(phys_addr_t base, phys_addr_t size) >>> +{ >>> + return memblock_setclr_flag(base, size, 1, MEMBLOCK_BOOT); >>> +} >>> + >>> +/** >>> + * memblock_clear_bootmem - Clear flag MEMBLOCK_BOOT for a specified region. >>> + * @base: the base phys addr of the region >>> + * @size: the size of the region >>> + * >>> + * Return: 0 on success, -errno on failure. >>> + */ >>> +int __init_memblock memblock_clear_boot(phys_addr_t base, phys_addr_t size) >>> +{ >>> + return memblock_setclr_flag(base, size, 0, MEMBLOCK_BOOT); >>> +} >>> + >>> +/** >>> * memblock_mark_hotplug - Mark hotpluggable memory with flag MEMBLOCK_HOTPLUG. >>> * @base: the base phys addr of the region >>> * @size: the size of the region >>> @@ -974,6 +998,10 @@ static bool should_skip_region(struct memblock_region *m, int nid, int flags) >>> if ((flags & MEMBLOCK_MIRROR) && !memblock_is_mirror(m)) >>> return true; >>> >>> + /* if we want boot memory skip non-boot memory regions */ >>> + if ((flags & MEMBLOCK_BOOT) && !memblock_is_boot(m)) >>> + return true; >>> + >>> /* skip nomap memory unless we were asked for it explicitly */ >>> if (!(flags & MEMBLOCK_NOMAP) && memblock_is_nomap(m)) >>> return true; >>> @@ -1785,6 +1813,15 @@ bool __init_memblock memblock_is_map_memory(phys_addr_t addr) >>> return !memblock_is_nomap(&memblock.memory.regions[i]); >>> } >>> >>> +bool __init_memblock memblock_is_boot_memory(phys_addr_t addr) >>> +{ >>> + int i = memblock_search(&memblock.memory, addr); >>> + >>> + if (i == -1) >>> + return false; >>> + return memblock_is_boot(&memblock.memory.regions[i]); >>> +} >>> + >>> #ifdef CONFIG_HAVE_MEMBLOCK_NODE_MAP >>> int __init_memblock memblock_search_pfn_nid(unsigned long pfn, >>> unsigned long *start_pfn, unsigned long *end_pfn) >>> -- >>> 2.7.4 >>> >> >