On 2021/9/17 15:04, Greg KH wrote:
On Fri, Sep 17, 2021 at 02:55:18PM +0800, Kefeng Wang wrote:
On 2021/9/17 14:24, Greg KH wrote:
On Fri, Sep 10, 2021 at 01:33:53PM +0800, Kefeng Wang wrote:
Percpu embedded first chunk allocator is the firstly option, but it
could fails on ARM64, eg,
"percpu: max_distance=0x5fcfdc640000 too large for vmalloc space 0x781fefff0000"
"percpu: max_distance=0x600000540000 too large for vmalloc space 0x7dffb7ff0000"
"percpu: max_distance=0x5fff9adb0000 too large for vmalloc space 0x5dffb7ff0000"
then we could meet "WARNING: CPU: 15 PID: 461 at vmalloc.c:3087 pcpu_get_vm_areas+0x488/0x838",
even the system could not boot successfully.
Let's implement page mapping percpu first chunk allocator as a fallback
to the embedding allocator to increase the robustness of the system.
Reviewed-by: Catalin Marinas <catalin.marinas@xxxxxxx>
Signed-off-by: Kefeng Wang <wangkefeng.wang@xxxxxxxxxx>
---
arch/arm64/Kconfig | 4 ++
drivers/base/arch_numa.c | 82 +++++++++++++++++++++++++++++++++++-----
2 files changed, 76 insertions(+), 10 deletions(-)
diff --git a/arch/arm64/Kconfig b/arch/arm64/Kconfig
index 077f2ec4eeb2..04cfe1b4e98b 100644
--- a/arch/arm64/Kconfig
+++ b/arch/arm64/Kconfig
@@ -1042,6 +1042,10 @@ config NEED_PER_CPU_EMBED_FIRST_CHUNK
def_bool y
depends on NUMA
+config NEED_PER_CPU_PAGE_FIRST_CHUNK
+ def_bool y
+ depends on NUMA
Why is this a config option at all?
The config is introduced from
commit 08fc45806103e59a37418e84719b878f9bb32540
Author: Tejun Heo <tj@xxxxxxxxxx>
Date: Fri Aug 14 15:00:49 2009 +0900
percpu: build first chunk allocators selectively
There's no need to build unused first chunk allocators in. Define
CONFIG_NEED_PER_CPU_*_FIRST_CHUNK and let archs enable them
selectively.
For now, there are three ARCHs support both PER_CPU_EMBED_FIRST_CHUNK
and PER_CPU_PAGE_FIRST_CHUNK.
arch/powerpc/Kconfig:config NEED_PER_CPU_PAGE_FIRST_CHUNK
arch/sparc/Kconfig:config NEED_PER_CPU_PAGE_FIRST_CHUNK
arch/x86/Kconfig:config NEED_PER_CPU_PAGE_FIRST_CHUNK
and we have a cmdline to choose a alloctor.
percpu_alloc= Select which percpu first chunk allocator to use.
Currently supported values are "embed" and "page".
Archs may support subset or none of the selections.
See comments in mm/percpu.c for details on each
allocator. This parameter is primarily for debugging
and performance comparison.
embed percpu first chunk allocator is the first choice, but it could fails
due to some
memory layout(it does occurs on ARM64 too.), so page mapping percpu first
chunk
allocator is as a fallback, that is what this patch does.
+
source "kernel/Kconfig.hz"
config ARCH_SPARSEMEM_ENABLE
diff --git a/drivers/base/arch_numa.c b/drivers/base/arch_numa.c
index 46c503486e96..995dca9f3254 100644
--- a/drivers/base/arch_numa.c
+++ b/drivers/base/arch_numa.c
@@ -14,6 +14,7 @@
#include <linux/of.h>
#include <asm/sections.h>
+#include <asm/pgalloc.h>
struct pglist_data *node_data[MAX_NUMNODES] __read_mostly;
EXPORT_SYMBOL(node_data);
@@ -168,22 +169,83 @@ static void __init pcpu_fc_free(void *ptr, size_t size)
memblock_free_early(__pa(ptr), size);
}
+#ifdef CONFIG_NEED_PER_CPU_PAGE_FIRST_CHUNK
Ick, no #ifdef in .c files if at all possible please.
The drivers/base/arch_numa.c is shared by RISCV/ARM64, so I add this config
to
no need to build this part on RISCV.
Ok, then you need to get reviews from the mm people as I know nothing
about this at all, sorry. This file ended up in drivers/base/ for some
reason to make it easier for others to use cross-arches, not that it had
much to do with the driver core :(
Ok, I has Cc'ed Andrew and mm list ;)
Hi Catalin and Will, this patchset is mostly changed for arm64,
and the change itself is not too big, could you pick it up from arm64
tree if there are no more comments, many thanks.
thanks,
greg k-h
.
