weirdtreething has quit [Ping timeout: 480 seconds]
weirdtreething has joined #linux-cros-arm
hexdump0815 has joined #linux-cros-arm
hexdump01 has quit [Ping timeout: 480 seconds]
alpernebbi has joined #linux-cros-arm
<janrinze>
dcz[m]: 16k page size seems to work fine too.
<dcz[m]>
did you just recompile the kernel or also the userspace?
<janrinze>
dcz[m]: disk performance is up to 1000MB/s. No real difference noticable.
<janrinze>
dcz[m]: just the kernel and the modules.
<dcz[m]>
IIRC page size has most difference in cache behaviour
<janrinze>
a few years ago this would simply crash most applications..
<dcz[m]>
do you think most apps relied on a specific page size?
<janrinze>
Indeed. there is a lot of assumptions used in many apps regarding the way the kernel/OS.
<janrinze>
The main issue would be 'custom' memory management.
<dcz[m]>
aren't most apps using what libc gives them?
<janrinze>
Tha's an assumption....
<dcz[m]>
more of a guess I'd say :D
<janrinze>
Remember that most apps are written with x86 in mind..
<janrinze>
So there are many 'work-arounds' used to speed up apps but those rely on 'tricks'.
<janrinze>
On ARM those would not work well or sometimes not at all. It's been an uphill battle for many years to get stuff working on ARM. We're not there yet.
<dcz[m]>
anything else broken apart from Chromium? everything I tossed at my L5 worked on 4k page size
<janrinze>
Chromium 131 is broken for both 4k and 16k i.i.r.c.
<janrinze>
dcz[m]: I have not yet had sound working on this laptop. That might be some config issue but we never figured it out.
<janrinze>
dcz[m]: I do have a patch for the keyboard since i like to use function keys and replace the power key with 'delete'. It helps me to ensure i don't switch off my laptop everytime i think i press delete..
<dcz[m]>
lol
<janrinze>
16k pages definitely makes the NVMe faster, not sure why but it may be related to the sector size. Not sure though.
<janrinze>
Anyone ever tried 64k Pages? :-D
<dcz[m]>
what page size is apple using?
<janrinze>
16k i.i.r.c.
<janrinze>
something todo with the chip.
<janrinze>
also if you have lots of memory a bigger page size reduces overhead, i think.
<dcz[m]>
the chip has extra L1 cache which would not be possible with 4k
<dcz[m]>
don't remember why any more
<janrinze>
just read the specs for Kompanio 1380.. 3 GHz Cortex A78. Hmm.. not bad.
weirdtreething has quit [Ping timeout: 480 seconds]
weirdtreething has joined #linux-cros-arm
<dcz[m]>
is Mediatek contributing anything to mainlining, or is it only Google?