| Message ID | 1435773182-23074-1-git-send-email-m-karicheri2@ti.com (mailing list archive) |
|---|---|
| State | New, archived |
| Headers | show |
On Wed, Jul 01, 2015 at 01:53:02PM -0400, Murali Karicheri wrote: > Currently for Keystone devices, user can't change the > value of CONFIG_FORCE_MAX_ZONEORDER option in defconfig. > Users require capability to tune the value of this option on a target > board. So this patch adds this capability No they shouldn't. If we do permit this, it should not be unrestricted - it's a power-of-2 of the page size, so specifying something like 32768 is insane. In any case, it's well known that the Linux MM system fragments memory, and the higher order allocations will fail soon after boot - and the larger the order, the greater chance of it failing. The only case that you want large allocations is for things like DMA, and we have a separate allocator for that called CMA, which is able to grab large chunks of memory, provided it's configured with a large enough zone. Please check whether CMA can be used _before_ considering using this option. If you need to increase the order, it should be justified, and it should be done on a per SoC basis in a static way, not left to the user to dream up some power-of-2 figure. Most importantly, please explain in the commit message why CMA doesn't provide you with what you need.
Russell, On 07/01/2015 02:48 PM, Russell King - ARM Linux wrote: > On Wed, Jul 01, 2015 at 01:53:02PM -0400, Murali Karicheri wrote: >> Currently for Keystone devices, user can't change the >> value of CONFIG_FORCE_MAX_ZONEORDER option in defconfig. >> Users require capability to tune the value of this option on a target >> board. So this patch adds this capability > > No they shouldn't. If we do permit this, it should not be unrestricted - > it's a power-of-2 of the page size, so specifying something like 32768 is > insane. > Thanks for your comments This was present in out internal version of the kernel. Only thing I can recall is that this patch was added to fix an isue in a DEBUG buid. Memory allocation for a large data structures used by a driver during probe (internal version) was failing as the data structure gets expanded as a result of DEBUG option related variables. There is no hurry to add this for now. > In any case, it's well known that the Linux MM system fragments memory, > and the higher order allocations will fail soon after boot - and the > larger the order, the greater chance of it failing.Ok > Agree > The only case that you want large allocations is for things like DMA, and > we have a separate allocator for that called CMA, which is able to grab > large chunks of memory, provided it's configured with a large enough zone. > > Please check whether CMA can be used _before_ considering using this > option. If you need to increase the order, it should be justified, and > it should be done on a per SoC basis in a static way, not left to the > user to dream up some power-of-2 figure. I will explore these options when we upstream the internal driver. For now I will drop this patch. Thanks
diff --git a/arch/arm/Kconfig b/arch/arm/Kconfig index a750c14..9b5f8bc 100644 --- a/arch/arm/Kconfig +++ b/arch/arm/Kconfig @@ -1731,7 +1731,7 @@ config ARM_MODULE_PLTS source "mm/Kconfig" config FORCE_MAX_ZONEORDER - int "Maximum zone order" if ARCH_SHMOBILE_LEGACY + int "Maximum zone order" if ARCH_SHMOBILE_LEGACY || ARCH_KEYSTONE range 11 64 if ARCH_SHMOBILE_LEGACY default "12" if SOC_AM33XX default "9" if SA1111 || ARCH_EFM32
Currently for Keystone devices, user can't change the value of CONFIG_FORCE_MAX_ZONEORDER option in defconfig. Users require capability to tune the value of this option on a target board. So this patch adds this capability Signed-off-by: Murali Karicheri <m-karicheri2@ti.com> --- arch/arm/Kconfig | 2 +- 1 file changed, 1 insertion(+), 1 deletion(-)