[PATCH RESEND] ARM: kdump: 2nd kernel should use strict pfn_valid in SPARSEMEM platform

[PATCH RESEND] ARM: kdump: 2nd kernel should use strict pfn_valid in SPARSEMEM platform

[Wang Nan]

This patch removes an irrational limitation for crash dump kernel on ARM
platform with SPARSEMEM enabled.

Without this patch, crash reservation area for a crash dump kernel with
SPARSEMEM selected must occupy a full section plus 1MiB. If not,
elfcorehdr and some memory space used by the first kernel will unable to
get accessed. This is caused by pfn_valid: fast pfn_valid ragards any
pfn in a valid section as valid and prevents it to be ioremapped.

This limitation wastes memory, because sections are always large and
crash dump kernel should be as small as possible.

This patch selects HAVE_ARCH_PFN_VALID for CRASH_DUMP, makes crash dump
kernel to use strict version of pfn_valid().

Signed-off-by: Wang Nan <wangnan0@huawei.com>
---

This is the third time I post this patch. The previous records can be
retrived from:

http://lists.infradead.org/pipermail/linux-arm-kernel/2014-May/256498.html
http://lists.infradead.org/pipermail/linux-arm-kernel/2014-May/257472.html

Different from previous version, this patch select HAVE_ARCH_PFN_VALID
right after CRASH_DUMP config entry.

---

 arch/arm/Kconfig | 1 +
 1 file changed, 1 insertion(+)

diff --git a/arch/arm/Kconfig b/arch/arm/Kconfig
index c57ddcb..d5ffbb1 100644
--- a/arch/arm/Kconfig
+++ b/arch/arm/Kconfig
@@ -2160,6 +2160,7 @@ config ATAGS_PROC
 
 config CRASH_DUMP
 	bool "Build kdump crash kernel (EXPERIMENTAL)"
+	select HAVE_ARCH_PFN_VALID if SPARSEMEM
 	help
 	  Generate crash dump after being started by kexec. This should
 	  be normally only set in special crash dump kernels which are
-- 
1.8.4

[PATCH RESEND] ARM: kdump: 2nd kernel should use strict pfn_valid in SPARSEMEM platform

[AKASHI Takahiro]

Catalin, Will

Can we assume that HAVE_ARCH_PFN_VALID is alway yes on arm64?
Looking at arm64/Kconfig,
config ARCH_HAS_HOLES_MEMORYMODEL
	def_bool y if SPARSEMEM
...
config HAVE_ARCH_PFN_VALID
	def_bool ARCH_HAS_HOLES_MEMORYMODEL || !SPARSEMEM

is this intentional?

-Takahiro AKASHI

On 05/28/2014 05:44 PM, Wang Nan wrote:
> This patch removes an irrational limitation for crash dump kernel on ARM
> platform with SPARSEMEM enabled.
>
> Without this patch, crash reservation area for a crash dump kernel with
> SPARSEMEM selected must occupy a full section plus 1MiB. If not,
> elfcorehdr and some memory space used by the first kernel will unable to
> get accessed. This is caused by pfn_valid: fast pfn_valid ragards any
> pfn in a valid section as valid and prevents it to be ioremapped.
>
> This limitation wastes memory, because sections are always large and
> crash dump kernel should be as small as possible.
>
> This patch selects HAVE_ARCH_PFN_VALID for CRASH_DUMP, makes crash dump
> kernel to use strict version of pfn_valid().
>
> Signed-off-by: Wang Nan <wangnan0@huawei.com>
> ---
>
> This is the third time I post this patch. The previous records can be
> retrived from:
>
> http://lists.infradead.org/pipermail/linux-arm-kernel/2014-May/256498.html
> http://lists.infradead.org/pipermail/linux-arm-kernel/2014-May/257472.html
>
> Different from previous version, this patch select HAVE_ARCH_PFN_VALID
> right after CRASH_DUMP config entry.
>
> ---
>
>   arch/arm/Kconfig | 1 +
>   1 file changed, 1 insertion(+)
>
> diff --git a/arch/arm/Kconfig b/arch/arm/Kconfig
> index c57ddcb..d5ffbb1 100644
> --- a/arch/arm/Kconfig
> +++ b/arch/arm/Kconfig
> @@ -2160,6 +2160,7 @@ config ATAGS_PROC
>
>   config CRASH_DUMP
>   	bool "Build kdump crash kernel (EXPERIMENTAL)"
> +	select HAVE_ARCH_PFN_VALID if SPARSEMEM
>   	help
>   	  Generate crash dump after being started by kexec. This should
>   	  be normally only set in special crash dump kernels which are
>

[PATCH RESEND] ARM: kdump: 2nd kernel should use strict pfn_valid in SPARSEMEM platform

[Will Deacon]

On Thu, May 29, 2014 at 05:54:02AM +0100, AKASHI Takahiro wrote:
> Catalin, Will
> 
> Can we assume that HAVE_ARCH_PFN_VALID is alway yes on arm64?
> Looking at arm64/Kconfig,
> config ARCH_HAS_HOLES_MEMORYMODEL
> 	def_bool y if SPARSEMEM
> ...
> config HAVE_ARCH_PFN_VALID
> 	def_bool ARCH_HAS_HOLES_MEMORYMODEL || !SPARSEMEM
> 
> is this intentional?

Looks like an artifact of the way we constructed those option, so yes, we
could just make that a def_bool y if you like.

Will