AMD Zen/Ryzen thread

wrc1944 · Posted: Thu May 04, 2017 8:18 pm Post subject:

I finally got my Ryzen 1700, and am still waiting on my ASRock Fatal1ty X370 Gaming K4 AM4 mobo to arrive (had to shop Micro-center- just got tired of waiting on Newegg to restock it), but did get the 16Gb of G.skill Flare X 3200, even though Newegg price was sky high. :cry:

It was a toss-up between the MSI X370 GAMING PRO CARBON AM4 AMD X370, the BIOSTAR X370GT7 AM4 AMD X370, and the Asrock.

I'm running kernels 4.10.13 and the new 4.11.0.
Anyone got some good tips on what the AM4 x370's absolutely need enabled in config? Are we still thinking haswell instead of zenver1, and what about "native"?

I've seen no reports as to gcc-6.3.0 adopting more relevant and improved optimizations/features, or any more related kernel patches other than the ALC1220 audio stuff for 4.11, which I really wanted even though I'm not using the system for gaming. I'm hoping we'll get more good stuff as time goes on, and the x370's will be great mobos for many years of improved and compatible cpu revisions.
_________________
Main box- AsRock x370 Gaming K4
Ryzen 7 3700x, 3.6GHz, 16GB GSkill Flare DDR4 3200mhz
Samsung SATA 1000GB, Radeon HD R7 350 2GB DDR5
OpenRC Gentoo ~amd64 plasma, glibc-2.36-r7, gcc-13.2.1_p20230304
kernel-6.8.4 USE=experimental python3_11

wrc1944 · Posted: Fri May 05, 2017 11:37 pm Post subject:

Here's some good news! http://wccftech.com/amd-ryzen-memory-support-to-be-expanded-with-upcoming-microcode-update

New AMD Ryzen Microcode Update Will Expand Memory Support

drizzt · Guru Joined: 21 Jul 2002 Posts: 428

Unfortunately, even compatible listed RAM doesn't help me with occasional segfaults during "compile":

Naib · Posted: Sat May 06, 2017 4:04 pm Post subject:

what does dmesg say when you get a segfault?
_________________

drizzt · Guru Joined: 21 Jul 2002 Posts: 428

Naib · Posted: Sat May 06, 2017 11:07 pm Post subject:

have you tried rebuilding glibc
what gcc are you using
what -march
what binutils has been selected?
_________________

drizzt · Guru Joined: 21 Jul 2002 Posts: 428

Naib · Posted: Sun May 07, 2017 12:27 pm Post subject:

I have just seen this in my dmesg

[ 9107.142219] traps: bash[22168] trap invalid opcode ip:7effb1245eda sp:7fff967e3910 error:0 in libc-2.24.so[7effb11cc000+189000]

I just upgraded to gcc-6.3, the toolchain was rebuilt twice (libtool,gcc,glibc,binutils). emerge -e @system, kernel rebuilt --> reboot followed by emerge -e @world

overkill I know but I wanted to ensure everything was built with a sane toolchain ... yet "invalid opcode" is present.
-march=native here, which should be ready for zen

--edit--
it was repeatable... I rebuilt bash & then gcc built fine. issue with the bash build? coincidence? hardware?
_________________

drizzt · Guru Joined: 21 Jul 2002 Posts: 428

double check that you are using the correct binutils version. I had 3 versions installed, rebuilt always the latest version but used the oldest version.

I also have packages which trigger the error more often than others. For me it is e.g, gcc and gtk-3. I think it is somehow related to the compile process' structure, Libreoffice for instance never segfaulted for me (built about 10 times for testing speed of system)
_________________
People don't have to earn my respect. I offer my respect to them, but be careful to lose my respect...

Naib · Posted: Sun May 07, 2017 2:11 pm Post subject:

eselect binutils list
[1] x86_64-pc-linux-gnu-2.27 *
_________________

drizzt · Guru Joined: 21 Jul 2002 Posts: 428

Maybe it's related to gcc-6.3 and march=native.
I use gcc-5.4 and march=haswell as I read somewhere that even on gcc-6.3 march=native is worse than march=haswell. At least march=znver1 should be used on gcc-6.3 but even this seems not to be better than march=haswell for now. (see https://wiki.gentoo.org/wiki/User:Maffblaster/Drafts/Ryzen#cite_note-1)
_________________
People don't have to earn my respect. I offer my respect to them, but be careful to lose my respect...

Naib · Posted: Sun May 07, 2017 2:35 pm Post subject:

Oh I know ( I edited that wiki page)
I did try setting march=znver1 but something wasn't right (prob a make.conf typo). Even still march=native was showin znver1 was correctly detected
_________________

wrc1944 · Posted: Mon May 15, 2017 2:29 pm Post subject:

Finally got my AM4/ryzen build done, and all works well, EXCEPT for my two Gentoo amd64 partion installations will not boot, even using "recovery" kernel- goes back to my Mint grub within 10-20 seconds.

I have the ASRock Fatal1ty X370 Gaming K4, R7 1700, and 16GB G.skill Flare 3200 ddr4, which came with bios 1.60 which I flashed to 2.40. No problems I can see.

On this box I have mint-kde on sda1, and use its grub to multi-boot all the other distros, such as mint-cinnamon, kde-Neon, Kaos linux (Arch based), Ubuntu-Studio, ubuntu-Mate, all of which boot right up on the new AM4 platform and function normally. I first had installed 4.11.0 kernels on all of them, and the new alc1220 sound works fine.

On sdb1 I have a Windows 7 install which I wished to save with all my Pro audio apps and audio backups that I'm currently working on, which also booted to low resolution mode, and I updated everything to AM4/Ryzen with the included ASrock driver cd- worked perfectly with all the usb 3.1 and etc., etc. stuff. The fact that the Fatal1ty X370 Gaming K4 has both ps/2 mouse and keyboard ports made it easy.

However, my two Main usage Gentoo installations on sda6 and sda11 simply refuse to boot. What did I miss? I'm using the same video card, so that's not it. Having SMT set in bios, or no, doesn't matter either.

I was confident than the AM3+ ~amd64 gcc-6.3.0 systems with 4.11.0 and 4.10.x kernels would at least boot and I could further update to AM4 from there, but guess I was wrong. What's the next best thing to try- I assume chroot into them and build a new kernel, or is there some bios setting I haven't yet tried? Seeing as how all those other distros AND even win7 work normally, it has to be my gentoo kernel config, or at least that's my current theory.

Any advice is greatly appreciated, as I spend 95% of my time on my Gentoo installs- just keep the others around so i keep up with what they are doing.

EDIT: One thing I just saw while in Mint-kde- I need to enable the igb ethernet driver in my gentoo kernels:

wrc1944 · Posted: Mon May 15, 2017 8:01 pm Post subject:

More problems: I can't chroot into my gentoo installs on sda6 or sda11 either from a new amd64 gentoo minimal install cd, or a brand new SystemRescueCD-5.0.

Keep getting right after

ct85711 · Veteran Joined: 27 Sep 2005 Posts: 1791

ok, say said you had other linux distro's installed, so why can't you chroot from one of those distros? It does not matter what linux distro you use to chroot into, as the base OS has no effect Gentoo distro...

NeddySeagoon · Posted: Mon May 15, 2017 8:55 pm Post subject:

wrc1944,

bash in the chroot, or one of the libs it needs, uses instructions that your CPU does not know how to execute.

wrc1944 · Posted: Mon May 15, 2017 9:53 pm Post subject:

Neddy,
Thanks much for the info!
Not sure if it helps much, for without chroot with Ryzen cpus how can we get into gentoo installs and fix them, or even do a fresh install off the current minimal installation cd?

What am I missing- there must be a way?
_________________
Main box- AsRock x370 Gaming K4
Ryzen 7 3700x, 3.6GHz, 16GB GSkill Flare DDR4 3200mhz
Samsung SATA 1000GB, Radeon HD R7 350 2GB DDR5
OpenRC Gentoo ~amd64 plasma, glibc-2.36-r7, gcc-13.2.1_p20230304
kernel-6.8.4 USE=experimental python3_11

NeddySeagoon · Posted: Mon May 15, 2017 10:04 pm Post subject:

wrc1944,

The boot is fine, or you would get illegal instruction errors well before the chroot step. Its the chroot that's broken.
Did you fetch the wrong stage3 tarball?
_________________
Regards,

NeddySeagoon

Computer users fall into two groups:-
those that do backups
those that have never had a hard drive fail.

wrc1944 · Posted: Tue May 16, 2017 8:10 am Post subject:

Neddy,
Thanks for the interest.

Guess I'm not explaining the situation clearly enough.
I moved my multi-boot drives (sda and sdb from an AM3+ board to the ASrock x370 AM4/Ryzen 1700 platform. All my distros on sda as well as win7 on sdb1 are working normally after the move to AM4, except my two Gentoo installs on sda7 and sda11, which will not boot. They return to my grub2 on sda1 Mint, which handles all the multi-boot functions.

I'm not at this point trying to re-install Gentoo, but to chroot into them and build a new kernel with Ryzen support. Both Gentoos were synced immediately before I moved to the x370 board.

They are (were) on a amd piledriver core cpu, built with kernel config march=native for the processor, and running kernels 4.11.0 and some 4.10.x's. Both are just updated gcc-6.3.0 ~amd64 systems, which were all depcleaned, and consistent.

Should I have build specific zenver1 or haswell kernels before making the move? Is that even possible?

Not sure I understand your lddtree /bin/bash code comment- wouldn't the just synced and updated ~amd64 Gentoo systems already have the latest versions of those packages available?

@ct85711,
Thanks much for the comment. Seems like a good idea, but what about the basic chroot /mnt/gentoo /bin/bash problem I have with chroot, and Neddy's /bin/bash comment? Why would chroot be expected to work from another distro if both the brand new gentoo minimal install cd and the newest SystemRescueCD don't? Would it be better to install a gentoo based binary distro on another partition, or use my Arch based Kaos Linux, or one of the "buntu" based distros I currently have installed? All those have 4.11.0 kernels installed, and just updated (not sure if they have chroot- will check.

Are Gentoo users now in the position that we need a new minimal install cd built with AM4 and ryzen cpu support? What about trying one of nightly auto-build gentoo install cd's? I'm still not clear on why I'm having a problem with chroot and /bin/bash presenting this problem.

_________________
Main box- AsRock x370 Gaming K4
Ryzen 7 3700x, 3.6GHz, 16GB GSkill Flare DDR4 3200mhz
Samsung SATA 1000GB, Radeon HD R7 350 2GB DDR5
OpenRC Gentoo ~amd64 plasma, glibc-2.36-r7, gcc-13.2.1_p20230304
kernel-6.8.4 USE=experimental python3_11

NeddySeagoon · Posted: Tue May 16, 2017 8:36 am Post subject:

wrc1944,

Ryzen does not support AMDs legacy SIMD instruction sets. That's 3DNow! and 3Dnowext.
Any attempt to execute instructions from those instruction sets will get you an illegal instruction exception.

When you built with -march=native on your amd piledriver, gcc was free to use those instructions.
If you have

wrc1944 · Posted: Tue May 16, 2017 9:35 am Post subject:

Thanks very much, Neddy. I really appreciate your help.
While you were responding I found this, which seems relevant to what you are saying: https://www.reddit.com/r/Gentoo/comments/5rk4kz/chroot_causing_invalid_instruction_set

This makes me think think my problem is that my kernels on the gentoo AM3+ systems were build with march=native (meaning only piledriver), and wouldn't work on AM4/Ryzen hardware, whereas all the other distros that do bot and function normally are using "generic" 64 bit kernels.

If I decide to just reformat the two Gentoo partitions, and just do a fresh install will the chroot problem with the gentoo install cd go away?

Could you please elaborate a little about the "bit of a challenge to use busybox as your shell in Gentoo though" you mention, so I can decide if I'm up to it?

Am I correct in thinking if I can get into the Gentoo installs and build a new 4.11.1 kernel with zen config settings all will be well, and i can boot normally from my Linux Mint grub2, as before?

Edit: Just found a post of your's from 2006- https://forums.gentoo.org/viewtopic-t-524504-start-0.html is it still relevant? I noticed you say "Unfortunately, gcc depends on glibc, so emerge will still be broken". But I could rebuild a kernel, correct? Or download a 4.11.1 kernel, and copy it over to the gentoo partition /usr/src, and compile there? Hmmm... but that needs gcc. :roll:

Edit 2: Oh- that guy had a broken glibc and no gcc, but I don't, so if i can get in i should be abled to do a new kernel. right?
_________________
Main box- AsRock x370 Gaming K4
Ryzen 7 3700x, 3.6GHz, 16GB GSkill Flare DDR4 3200mhz
Samsung SATA 1000GB, Radeon HD R7 350 2GB DDR5
OpenRC Gentoo ~amd64 plasma, glibc-2.36-r7, gcc-13.2.1_p20230304
kernel-6.8.4 USE=experimental python3_11

NeddySeagoon · Posted: Tue May 16, 2017 10:44 am Post subject:

Tony0945 · Posted: Tue May 16, 2017 1:19 pm Post subject:

If I may make a suggestion:

Instead of wiping the partitions and doing a complete reinstall:

1. return the drives to the pile driver system. If the system doesn't exist, just setup the old mobo on a workbench or table with a spare PS.
If this isn't practical, boot into one of the generic linuces and create mount points for the gentoo partitions. Work in Ubuntu or whatever, don't chroot.
2. Download a stage3 tarball and expand it into /mnt/gentoo or at least /bin and /usr/bin, you do not need to chroot.
3. Restore the drives to the new system.

The gentoo partitions should now boot as they are running the generic binaries.

4. Once booted into gentoo, update make.conf and CPUFLAGS.
5. emerge -e @system --keep-going (resume as needed)
6. emerge -e @world --keep-going (resume as needed)

Khayyam recently posted a one line script to emerge -e world excluding what's in system, but unfortunately, while I ran it, I did not save it.
IIRC, emerge -e @system only took a few hours on an Athlon II X3 and should be a lot faster on your Ryzen 1700.

I hate to reinstall Gentoo. All the timezone setup, loss of history, various customizations, ...