sun4i_defconfig (et al) suggestions about linux-sunxi HOT 15 OPEN

Disabling HIGHMEM might be interesting too, but for devices with >512M RAM (and maybe <512 too?) it causes large quantities of RAM to remain unused due to the way allocations are handled. HIGHMEM works around this issue.

from linux-sunxi.

@romanrm can you make a paste with the desired CONFIG_NETFILTER=m values?

from linux-sunxi.

9638828 enables IPV6 as built-in (doesn't compile as m), and tun and usb network stuff as m too.

pending is still netfilter and no_hz.

edit: did I mention I really prefer patches?

from linux-sunxi.

@amery I got a better idea than copying from debian or inventing our own desired values: why not merge the netfilter section from the defconfig of x86?

from linux-sunxi.

OK while we are at it;

CONFIG_ANDROID=y
CONFIG_ANDROID_BINDER_IPC=y
CONFIG_ANDROID_LOGGER=y
CONFIG_ANDROID_RAM_CONSOLE=y
CONFIG_ANDROID_RAM_CONSOLE_ENABLE_VERBOSE=y
CONFIG_ANDROID_RAM_CONSOLE_EARLY_INIT=y
CONFIG_ANDROID_RAM_CONSOLE_EARLY_ADDR=0
CONFIG_ANDROID_RAM_CONSOLE_EARLY_SIZE=0
CONFIG_ANDROID_TIMED_OUTPUT=y
CONFIG_ANDROID_LOW_MEMORY_KILLER=y

can be replaced by;

'# CONFIG_ANDROID is not set

from linux-sunxi.

I personally would keep CONFIG_ANDROID_RAM_CONSOLE=y; it's always nice to have, even if you're not running Android, and can help catch unexpected kernel panics.

I would also set CONFIG_PANIC_TIMEOUT=1 to reboot on kernel panic.

from linux-sunxi.

@turl

All above android is in the normal defconfig file and should only be in the crane version if you ask me.

I am not aware you can use the ram console under linux? Is that true? Might help MK802 users who can't solder that good (<-Me soon)

@amery; Exactly.

from linux-sunxi.

Here are my suggestions (sorry for not posting a real patch)... :)

CONFIG_ZRAM is not set
zRam (compressed RAM block device) is extremely useful for maximising available memory, this should be built as a module.

CONFIG_CC_OPTIMIZE_FOR_SIZE is not set
Optimising for size. This could be a win (must be benchmarked, though) due to better cache usage.

CONFIG_KERNEL_LZO is not set
This should be enabled. Although the resulting image will be larger, boot times will be shorter (LZO decompresses much faster than gzip).

CONFIG_HIGHMEM=y
Is it really needed? IIRC, only Cortex-A15 supports LPAE.

CONFIG_DNOTIFY=y
dnotify is obsolete, no?

from linux-sunxi.

@j1nx theorically you should be able to, as far as I'm aware it has no userspace dependencies.

@rsalvaterra look at my first comment on here regarding HIGHMEM

from linux-sunxi.

@turl sorry, I missed it. Anyway, wouldn't such issue qualify as an upstream bug? Or is there something that makes ARMv7 special as far as memory allocation is concerned?

from linux-sunxi.

@turl
Nice, but should we then not better change the source to reflect; CONFIG_RAM_CONSOLE=y (without ANDROID)

from linux-sunxi.

23e5456 enables zram as m on all *_defconfigs, and disables android in sun4i_defconfig.

pending discussion: no_hz, highmem, default optimizations (profiling needed), lzo, dnotify,
pending patch: netfilter, ram_console without android (@turl).

from linux-sunxi.

Few things which caught my eye, feel free to ignore

CONFIG_SPARSE_IRQ=y
Can be disabled since it is only relevant for multi-core systems, which sun4i is not (e.g. Mele has total 96 interrupts, compared to average x86 system which has around 1000)

CONFIG_COMPACTION=y
CONFIG_MIGRATION=y
These two are related to hugepages support, which is not present in 3.0.x for ARM arch. Probably does nothing, still safer to disable.

from linux-sunxi.

@j1nx I didn't invent it, it's upstream code. I don't think changing the name is a good idea.

from linux-sunxi.

@turl I dug a little deeper, it seems the HIGHMEM workaround is related to DMA mappings reserved at boot. I believe the DMA mapping changes (and the contiguous memory allocator) in Linux 3.5 will fix the problem, but I could be wrong...

from linux-sunxi.