Gentoo on Raspberry Pi 3

[steveL]

I've tried an armv8-hardfloat-linux-gnueabi. don't do it, the /proc/cpuinfo gives (falsely) an armv7.
all python packages would look for a armv7 gcc and fail to build....

Answered above:

If you want to compile for 64 bit, use: aarch64-unknown-linux-gnu
This was recommended on the gcc bugzilla and [post=7900188]verified by Neddy[/post] in the [topic=1041352]Raspi3 64 Bit thread[/topic].

[NeddySeagoon]

dr_wulsen,

The Pi3 reports itself as arm7 in 32 bit mode and

Code: Select all

Pi3 64bit ~ # cat /proc/cpuinfo 
processor	: 0
BogoMIPS	: 38.40
Features	: fp asimd evtstrm crc32
CPU implementer	: 0x41
CPU architecture: 8
CPU variant	: 0x0
CPU part	: 0xd03
CPU revision	: 4

in 64 bit mode.

[dr_wulsen]

Hi das_morph,

sorry for linking the wrong file. Obviously I got a little tired when writing the post, as my kernel would work flawless.
However, thank you for pointing it out, original post has been edited so I won't misguide others.

[keet]

According to the Mesa site, the current Mesa, as well as Linux kernel 4.5 (and the Raspberry Pi Foundation's tree of 4.1) includes support for VC4. Hopefully this will mean less trouble building packages that depend on Mesa and more ease building and running programs that use or depend on hardware video acceleration.[/url]

[Yamakuzure]

According to the Mesa site, the current Mesa, as well as Linux kernel 4.5 (and the Raspberry Pi Foundation's tree of 4.1) includes support for VC4. Hopefully this will mean less trouble building packages that depend on Mesa and more ease building and running programs that use or depend on hardware video acceleration.[/url]

I have a mesa-9999 ebuild in my overlay (seden) that allows you to build with VIDEO_CARDS="vc4".

I am still testing and report my progress in this thread.

[keet]

After building mesa with video_cards="vc4", when I try to start a game using retroarch (inside emulationstation), it shows this error:

Code: Select all

retroarch: symbol lookup error: /usr/lib/opengl/raspberrypi-userland/lib/libOpenVG.so: undefined symbol: client_process_state

I am not sure how to fix this; I'll troubleshoot later when I have time. Does anyone here already know the fix?

[vastchen]

somebody know Pi3 how to config /etc/portage/make.conf
Pi 2 CFLAGS="-02 -pipe -march=armv7-a -mfpu=vfpv3 -mfloat-abi=hard "
can use this ??? Pi 3 CFLAGS="-02 -pipe -march=armv8-a -mfpu=vfpv3 -mfloat-abi=hard " ??

-march= -mtune=cortex-a53 ?????

somebody can put you pi3 /etc/portage/make.conf to see?

[vastchen]

somebody know Pi3 how to config /etc/portage/make.conf
Pi 2 CFLAGS="-02 -pipe -march=armv7-a -mfpu=vfpv3 -mfloat-abi=hard "
can use this ??? Pi 3 CFLAGS="-02 -pipe -march=armv8-a -mfpu=vfpv3 -mfloat-abi=hard " ??

-march= -mtune=cortex-a53 ?????

somebody can put you pi3 /etc/portage/make.conf to see?

I found somebody use armv7 somebody use armv8 what should be use ? I will crazzy , if i want pi2 distccd with pi 3 use crossdev -t “？”

[keet]

I found somebody use armv7 somebody use armv8 what should be use ? I will crazzy , if i want pi2 distccd with pi 3 use crossdev -t “？”

I have been using this with no apparent problems:

Code: Select all

CFLAGS="-march=armv8-a+crc -mtune=cortex-a53 -mfpu=crypto-neon-fp-armv8 -mfloat-abi=hard -ftree-vectorize -funsafe-math-optimizations -O2 -pipe"

[vastchen]

I found somebody use armv7 somebody use armv8 what should be use ? I will crazzy , if i want pi2 distccd with pi 3 use crossdev -t “？”

I have been using this with no apparent problems:

Code: Select all

CFLAGS="-march=armv8-a+crc -mtune=cortex-a53 -mfpu=crypto-neon-fp-armv8 -mfloat-abi=hard -ftree-vectorize -funsafe-math-optimizations -O2 -pipe"

thanks for you help . I got it.

[safeness]

If anyone is interested in my custom .config for the raspberrypi-sources-4.4 to have a starting point for their own kernel, let me know - I will then post them here.

I'd be interested, since I can't seem to get the wifi to work on my pi. Might be missing some kernel modules or something. Which sources are you using for the 4.4 kernel? I tried the raspberrypi-sources package, but neither using crossdev nor native compiling with distcc resulted in a working kernel. The github sources work for me.

Hi das_morph,
I'm using the raspberrypi-sources-4.4.9999 in the portage tree, which actually downloads the github sources.
One thing I've stumbled across was that the simple zImage wouldn't work for me, if just renamed to kernel7.img, but had to use the include mkknlimg.
That's my kernel compile script:

Code: Select all

#!/bin/bash
ARCH=arm CROSS_COMPILE="/usr/bin/armv7a-hardfloat-linux-gnueabi-" make -j5 zImage modules dtbs
mkdir /tmp/rpi3
mkdir /tmp/rpi3/overlays
./scripts/mkknlimg arch/arm/boot/zImage /tmp/rpi3/kernel7.img
cp ./arch/arm/boot/dts/*.dtb /tmp/rpi3
cp ./arch/arm/boot/dts/overlays/*.dtb* /tmp/rpi3/overlays
cp ./arch/arm/boot/dts/overlays/README /tmp/rpi3/overlays
cp .config /tmp/rpi3/kernelconfig

I have sent the config to you via PN.
Of course, you will need the two files named in my earlier post.

@dr_wulsen: Can you send me your config as well?

Thanks!

[dr_wulsen]

If anyone is interested in my custom .config for the raspberrypi-sources-4.4 to have a starting point for their own kernel, let me know - I will then post them here.

I'd be interested, since I can't seem to get the wifi to work on my pi. Might be missing some kernel modules or something. Which sources are you using for the 4.4 kernel? I tried the raspberrypi-sources package, but neither using crossdev nor native compiling with distcc resulted in a working kernel. The github sources work for me.

Hi das_morph,
I'm using the raspberrypi-sources-4.4.9999 in the portage tree, which actually downloads the github sources.
One thing I've stumbled across was that the simple zImage wouldn't work for me, if just renamed to kernel7.img, but had to use the include mkknlimg.
That's my kernel compile script:

Code: Select all

#!/bin/bash
ARCH=arm CROSS_COMPILE="/usr/bin/armv7a-hardfloat-linux-gnueabi-" make -j5 zImage modules dtbs
mkdir /tmp/rpi3
mkdir /tmp/rpi3/overlays
./scripts/mkknlimg arch/arm/boot/zImage /tmp/rpi3/kernel7.img
cp ./arch/arm/boot/dts/*.dtb /tmp/rpi3
cp ./arch/arm/boot/dts/overlays/*.dtb* /tmp/rpi3/overlays
cp ./arch/arm/boot/dts/overlays/README /tmp/rpi3/overlays
cp .config /tmp/rpi3/kernelconfig

I have sent the config to you via PN.
Of course, you will need the two files named in my earlier post.

@dr_wulsen: Can you send me your config as well?

Thanks!

Hey safeness, sorry for being lazy recently, just read your post today. I've sent the config to you via PN.
Have fun

[dr_wulsen]

dr_wulsen,

That kernel emulation option generates code to deal with the illegal instruction exceptions programs built with SWP/SWPB instructs may generate when running on a CPU that is missing SWP/SWPB. Rather like the kernel software floating point option for 386/486SX CPUs that had no hardware floating point.
This is slow and horrible as there will be a user space to kernel space context switch when the illegal instruction exception is raised. The kernel will emulate the instruction, then there will be a kernel space to user space context switch back to the program to the instruction following the illegal instruction. Execution in userspace continues as if nothing had happened.
Consider the effect of this when a SWP is used in a loop ...

The problem here is

Code: Select all

{standard input}:134: Error: swp{b} use is obsoleted for ARMv8 and later 

that gcc is generating these instructions and passing them to the assembler, which then objects.

There may be a gcc/assembler option (CFLAG) to allow these instructions to be compiled. If so, and this works, you will need that kernel option if SWP/SWPB is not implemented in the Pi 3 CPU.

Neddy,

I just got thinking...the SWP/SWPB is required in the RPi3 kernel.
Cause when compiling sys-libs/db with the ARMv8 arch, the linker will (as you explained perfectly clear, thank you) complain about the SWP/SWPB being unsupported on this arch.
But if we now compile it with an ARMv7 march, it will generate the SWP/SWPB code as well - only the linker is not going to complain, as it was supported on an ARMv7.
So actually, what should happen is the context switches you have predicted.

Do you know of any testing method I could use to have sys-libs/db do actual work - so I should be able to see the context switching happening frequently (and too often) by usage of vmstat?
Thanks!

[NeddySeagoon]

dr_wulsen,

The Pi CPU can operate in either 32 bit or 64 bit mode. Its even possible to switch from 64 bit mode to 32 bit mode and back again.
In 32 bit mode, the Pi is a arm7 CPU - complete with all the instructions that are obsolete on arm8.
Note that these instructions are obsolete, that's not the same as not implemented.

I don't know what uses sys-libs/db. On my desktop, I get

Code: Select all

 $ equery d sys-libs/db 
 * These packages depend on sys-libs/db:
app-office/orage-4.12.1 (berkdb ? >=sys-libs/db-4)
dev-lang/perl-5.22.2 (berkdb ? sys-libs/db)
dev-lang/python-2.7.11-r2 (berkdb ? sys-libs/db:5.3)
                          (berkdb ? sys-libs/db:5.2)
                          (berkdb ? sys-libs/db:5.1)
                          (berkdb ? sys-libs/db:5.0)
                          (berkdb ? sys-libs/db:4.8)
                          (berkdb ? sys-libs/db:4.7)
                          (berkdb ? sys-libs/db:4.6)
                          (berkdb ? sys-libs/db:4.5)
                          (berkdb ? sys-libs/db:4.4)
                          (berkdb ? sys-libs/db:4.3)
                          (berkdb ? sys-libs/db:4.2)
dev-libs/redland-1.0.17-r1 (berkdb ? sys-libs/db)
net-nds/openldap-2.4.44 (berkdb ? <sys-libs/db-6.0)
                        (berkdb ? sys-libs/db:5.3)
                        (berkdb ? sys-libs/db:5.1)
                        (berkdb ? sys-libs/db:4.8)
                        (berkdb ? sys-libs/db:4.7)
                        (berkdb ? sys-libs/db:4.6)
                        (berkdb ? sys-libs/db:4.5)
                        (berkdb ? sys-libs/db:4.4)
sys-apps/iproute2-4.5.0 (berkdb ? sys-libs/db)
sys-libs/pam-1.3.0 (berkdb ? >=sys-libs/db-4.8.30-r1[abi_x86_32(-)?,abi_x86_64(-)?,abi_x86_x32(-)?,abi_mips_n32(-)?,\
abi_mips_n64(-)?,abi_mips_o32(-)?,abi_ppc_32(-)?,abi_ppc_64(-)?,abi_s390_32(-)?,abi_s390_64(-)?])

[Zucca]

I guess this will improve things a bit...

[KAMIKAZE_]

Could someone share a fully built stage with configs or even an image file, please?

[chithanh]

What fully built stage do you want? The normal stage3 for armv7 and arm64 work fine.
Other than that, you will find images in the [post=7898580]64 bit thread[/post]:

You will find Pi3_64_bit_root.tbz2 and Pi3_64_bit_boot.tbz2 at http://grytpype-thynne.org/~roy/Pi3_64bit/
They are the partition2 (ext4) and partition1 (vfat) copied from my SDCard.

[Yamakuzure]

For a while now my RPi3 is a brick.

I can use the normal terminal, and I can use xterm from within LxQT, but not from within Lumina.

Other applications segfault after a few seconds. The Backtrace in gdb is always the same with frame 0 being:

Code: Select all

#0  malloc_consolidate (av=av@entry=0x76f5d79c <main_arena>) at malloc.c:4123

The address, of course, changes, but everything else is the same.

The very same applications ran fine until 2 weeks ago. I even recompiled my whole system.

Did anybody have this, too? Maybe my PI is broken?

Edit: My PI is fine. Switched the card with another one running Raspbian, and the issue is the same on the other PI.

Edit 2: Interesting. From within LxQT I can run xterm. From within Fluxbox, Openbox or Lumina I can not even get xterm to start.
But: If I do a startx as root firing up a twm without lightdm bein in front of it, xterm, Konsole, and everything else works just fine.

Edit 3: Wow. As a regular user TWM doesn't even start. Could it be my user profile? That would be really weird, no?
... but it *is* the profile. Dammit! Why didn't I try it out sooner? Everything, no matter from which DM/WM works just fine if I am root.
And now everything works fine again after I removed ~/.??*

[Zucca]

Hi. What filesystems you prefer to use on your Pi?
I plan to move some directories to an external USB stick (plug rather).

I wonder what would be the best fs for all the mmc partitions. Maybe some filesystem designed for flash storage? And what for the USB mass storage?

[NeddySeagoon]

Zucca,

You must use vfat for /boot or it won't
On the SDCard I use ext4 with journaling off. You don't need the extra writes and I keep mine backed up as from the original Pi days, mine often got trashed.

For a USB HDD, I use a mix of ext4 and ext4 with journaling off, depending on the filesystem content.
Throwaway content gets journaling off.

[Zucca]

Throwaway content gets journaling off.

I plan to use my USB stick fot temporary files. Mainly for portage temp files and maybe swap. Although swap has its "own filesystem".

I wish there was RAM-based USB sticks. I don't mind the slowness of USB2.0 but rather the endurance.

[NeddySeagoon]

Zucca,

A USB stick will be slow as its still FLASH.
I suspect a USB3 stick would move the bottleneck to the USB2 I/O but I've not tested.
There is another complication with a USB3 stick on the Pi. The USB2 may not power it.

I use a USB HDD that has its own PSU (not USB bus powered)

[axl]

i tried to use an external sata / usb thingie with a ssd drive as storage. the result was not as good as expected. on a pc it would do great. it would work usb3/sata3. on the pi it would only go usb2.

and did aweful. really.

it reminds me of the dreaded "southbridge" on intel platforms. the dreaded IO bridge where everything worked bad. so happy my nvme disk now resides with ram/cpu/videocard on the northbridge.

anyway... u can't speed up a pi. it's own cardreader is the fastest storage it has. you can only buy a really good card thingie and hope for the best. or use google to look for the wiki page with benchmarks which card works best.


what i do know, usb is not a viable option for pi and storage.

[NeddySeagoon]

axl,

I manage 5MB/sec on the SD card and 30MB/sec over USB.

Tested on a Pi 3 in 64 bit mode with hdparm -tT /dev/.
That's tested with a selection of SD cards but only one USB HDD.

[Zucca]

Interesting...
I thought the microSD slot with it's own bus would be fastest storage but no.
My tests tell the same story.

Code: Select all

/dev/mmcblk0:
 Timing cached reads:   1172 MB in  2.00 seconds = 585.37 MB/sec
 Timing buffered disk reads:  52 MB in  3.09 seconds =  16.84 MB/sec

Code: Select all

/dev/sda:
 Timing cached reads:   1196 MB in  2.00 seconds = 597.35 MB/sec
 Timing buffered disk reads: 100 MB in  3.05 seconds =  32.83 MB/sec

My USB stick is Sandisk Ultra Fit USB3.0 16GB. So it's confirmed a working USB3.0 USB flash stick. :)