SGI Octane (IP30) Patch Available

[Kumba]

can't you tarball your linux kernel sources and send me them in order to help me to test and develop ?
(under the promise I won't release to anybody without your authorization)

elsewhere you are alone

I added 4.2.6 and 4.3.0 mips-sources ebuilds to the Gentoo Repo a few days ago. Patchballs are on the mirrors. Previously, there were 4.1.7 and 4.2.0 ebuilds (SMP was added back in 4.1.x). I've not been holding anything back. Enable USE "ip30" to have the IP30-specific patches apply (typically, what I do is run 'USE="ip30" ebuild </path/to/ebuild> clean install', then copy the tree out of the '/image/' folder in $PORTDIR, and repeat as needed for IP27 and IP28 trees, as well. Omit the USE flag for IP22/IP32 systems. No IP35 yet).

[legacy]

Code: Select all

Ufed, USE "ip30"
emerge  sys-kernel/mips-sources

Emerging (1 of 1) sys-kernel/mips-sources-4.3.0::gentoo ...

It's running on my QEMU/MIPS builder

I will configure the kernel adding my ramrootfs
then I will try on IP30 hw, SMP, 2xR12K
testing everything

• GFX
  • VPRO
  • MGRAS
• IOC3
• HEAT
• XBOW
• MENET
• XIO_PCI
  • XIO_PCI/ShoeHorn
  • XIO_PCI/ShoeBox
• PCI board
  • PCI_EHCI_USB (usb bulk only, storage UBA and FTDI2232)
• RAM
  • 1Gbyte of ram (4x256Mbyte)
  • 1.5Gbyte of ram (6x256Mbyte)
  • 2Gbyte of ram (8x256Mbyte)

[Kumba]

It's running on my QEMU/MIPS builder

I was just messing around with Qemu. Isn't the clock speed for all MIPS targets locked to 100MHz? Not very fast emulation...

• PCI board
  • PCI_EHCI_USB (usb bulk only, storage UBA and FTDI2232)

These two will probably give you the most headaches. Not all PCI works, and PCI-based USB won't work because of the PCI-to-PCI Bridge issue and the whole BRIDGE DMA problem still with Octane. I haven't tested MENET too much. The four Ethernet ports get detected, but not the six serial ports. Did not dig 100% into why.

[legacy]

It's running on my QEMU/MIPS builder

I was just messing around with Qemu. Isn't the clock speed for all MIPS targets locked to 100MHz? Not very fast emulation…

but a modern i7 machine eats less electricity

why don't we use a CI20 ? 1Ghz MIPS board

p.s.
in 4.3.0, I can't see an ip30_defconfig, do you have one on hand ?

Code: Select all

arch/mips/configs/
...
ip22_defconfig             malta_kvm_guest_defconfig  nlm_xlp_defconfig          tb0219_defconfig
bmips_be_defconfig         ip27_defconfig             malta_qemu_32r6_defconfig  nlm_xlr_defconfig          tb0226_defconfig
bmips_stb_defconfig        ip28_defconfig             maltaaprp_defconfig        pistachio_defconfig        tb0287_defconfig
capcella_defconfig         ip32_defconfig 

[armanox]

It's running on my QEMU/MIPS builder

I was just messing around with Qemu. Isn't the clock speed for all MIPS targets locked to 100MHz? Not very fast emulation…

but a modern i7 machine eats less electricity

why don't we use a CI20 ? 1Ghz MIPS board

p.s.
in 4.3.0, I can't see an ip30_defconfig, do you have one on hand ?

Code: Select all

arch/mips/configs/
...
ip22_defconfig             malta_kvm_guest_defconfig  nlm_xlp_defconfig          tb0219_defconfig
bmips_be_defconfig         ip27_defconfig             malta_qemu_32r6_defconfig  nlm_xlr_defconfig          tb0226_defconfig
bmips_stb_defconfig        ip28_defconfig             maltaaprp_defconfig        pistachio_defconfig        tb0287_defconfig
capcella_defconfig         ip32_defconfig 

Probably doesn't have a CI20 (and the CI40 is coming soon!). I'd love to get my hands on one of them, but I've got other areas that need my money at the moment. Also, the instruction set is a bit updated but is 32 bit only (MIPS32r2 vs MIPS IV) so there will be some issues there. There again, it'll use a lot less power then your i7 and probably is faster then any SGI box ever ended up being.

[Kumba]

but a modern i7 machine eats less electricity :D

Electricity costs are irrelevant in America. We'll just burn more coal :P

why don't we use a CI20 ? 1Ghz MIPS board

It is tempting, but as Armonox already pointed out, CI20 is mips32r2 only, and I need to build, at minimum, MIPS-III stages. I'll have to wait until Imgtec somes out with 64-bit class systems. Or someone helps hunt down the IP27 lockup bug, or IP35 gets whipped into better shape. So for now, my Octane, trusty as it is heavy, is my best build box.

p.s.
in 4.3.0, I can't see an ip30_defconfig, do you have one on hand ?

I always keep forgetting to add one to the core patches. I took the one I have and stripped it down to the utter basics, so you'll still want to navigate through the menus and select stuff. I should have the bare minimum hardware options set, meaning stuff like SMP is off, so set things accordingly. And since I don't know how long it's been since you've seen a kernel >2.6.x (</snerk>), use the "nconfig" target instead of "menuconfig". It's a newer, better menu system for selecting kernel options. Just make sure your TERM is set right.

https://dev.gentoo.org/~kumba/ip30/ip30_defconfig-4.3.0

[Kumba]

Probably doesn't have a CI20 (and the CI40 is coming soon!). I'd love to get my hands on one of them, but I've got other areas that need my money at the moment. Also, the instruction set is a bit updated but is 32 bit only (MIPS32r2 vs MIPS IV) so there will be some issues there. There again, it'll use a lot less power then your i7 and probably is faster then any SGI box ever ended up being.

CI20 is only ~$65 on Imgtec's online store. It is tempting to get one, but the deal-killer for me is the lack of a small case. I don't really need permanent access to the circuits, so I'd like to seal machines like that up, without having to Dremel my own case or such out of acrylic. And then there's the 32-bit issue...

And electrical draw isn't that much of an issue. I'm in the Mid-Atlantic (not far from you, actually), and electric rates are, per my utility, ~$0.0807/kWh. The Octane consumes ~298-301W in its current configuration, so very rough math sets me out at ~$24-$31/mo at the present rates (after adding in the transmission charges per kWh and other little fees and the like), assuming I ran it 24/7. As long as I keep it tasked with work, it's worth the cost. That's why I can't afford to run the IP27 machine, because when tasked with work, the chances of the lockup bug happening increase dramatically. If it locks up when I'm not home, then it'll just idle there, sucking ~700W of power. If I can figure that bug out and fix it, then I can put that machine to work and justify the electrical costs.

[legacy]

gentoo-4.3.0.20151126

here the SMP is not properly working, I continuously get the following error

Code: Select all

Instruction bus error, epc == 0000000000443b3c, ra == 0000000000444b88

[legacy]

it seems the SMP has something wrong, I sometimes get this error

Code: Select all

Instruction bus error, epc == a80000002026bbc8, ra == a8000000200c776c
Oops[#1]:
CPU: 1 PID: 35 Comm: sh Not tainted 4.2.6-Hurr-durr-I-ma-sheep-ip30 #10
task: a80000007942b380 ti: a800000079440000 task.ti: a800000079440000
$ 0   : 0000000000000000 a800000079bf4000 0000000000001000 0000000000000000
$ 4   : 0000000000000000 0000000000001000 0000000000001000 a800000079443dc0
$ 8   : fffffffffffffffa 0000000000001000 0000000077891060 0000000000000000
$12   : 0000000000000000 0000000000001000 a8000000200fd3e8 0000000000000000
$16   : 0000000000000000 a800000079443dc0 0000000000001000 a800000079443e10
$20   : a80000007feadd48 0000000077887000 a80000002183b700 0000000000000000
$24   : 0000000000000000 0000000000000000                                  
$28   : a800000079440000 a800000079443cf0 a8000000209c0000 a8000000200c776c
Hi    : ffffffffe0dfebef
Lo    : ffffffff9ccf4dc5
epc   : a80000002026bbc8 0xa80000002026bbc8
ra    : a8000000200c776c 0xa8000000200c776c
Status: 9004dce3        KX SX UX KERNEL EXL IE 
Cause : 00000018 (ExcCode 06)
PrId  : 00000f24 (R14000)
Modules linked in:
Process sh (pid: 35, threadinfo=a800000079440000, task=a80000007942b380, tls=0000000000000000)
Stack : 0000000000000000 0000000000000000 0000000000000000 0000000000001000
          a80000007feadd48 0000000000000000 a800000079bef600 a800000079443de8
          a800000079443dc0 a8000000200c776c a80000002183b700 0000000000001000
          0000000000000000 a80000007feade88 0000000000001000 0000000000000802
          a800000079bef600 a800000079443e70 0000000077887000 a800000079443e70
          0000000000000003 ffffffffffffffff 0000000000000000 0000000077889128
          000000007fedf0c8 a8000000200fc21c 0000000000000802 0000000000000000
          0000000000001000 a800000079443e10 0000000000000001 a800000079bef600
          0000000000000000 0000000000000000 0000000000000000 0000000000000000
          0000000077887000 0000000000001000 0000000000000003 a800000079bef600
          ...
Call Trace:[<a8000000200c776c>] 0xa8000000200c776c
[<a8000000200fc21c>] 0xa8000000200fc21c
[<a8000000200fc9a0>] 0xa8000000200fc9a0
[<a8000000200fd444>] 0xa8000000200fd444
[<a80000002001a298>] 0xa80000002001a298
[<a8000000200fa394>] 0xa8000000200fa394


Code: 0120682d  bfb40000  0060202d <a2a00000> 14800060  0240a02d  6442ffff  2405f000  02a2102d 
---[ end trace 6e6538eab13e747b ]---

sometimes means:
If I reset the machine, I can have two situations
1) the machine works perfectly, with an uptime of 20hours without issuing any error
2) the machine continuously issues the above error and it's forced to reset

it happens randomly

I have removed the GFX, the Kb&mouse
I have 1.5Gbyte of ram installed, without an hard drive


the same happens if I plug a V6

[Kumba]

You'll want to compile debugging symbols into your kernel. Raw addresses aren't very useful, because I don't have your System.map file or anything.

The other thing I can think of....are you using transparent hugepage support by chance? If so, disable that and try again. Known issue with that feature on IP30. Seems the R14000 CPUs make it even worse. This option is in the Kernel settings. Also disable hugetlbfs support under File Systems (might be dependent on hugepages, so if you disable (or have disabled) hugepages, then I don't think hugetlbfs support will be visible).

Also, when you get an IBE, make sure to fully power the machine down and unplug the power cord for ~30 seconds. I had this issue with the DBEs in which the caches weren't being cleared correctly or the hardware not reinitialized correctly with a plain reboot or quick power cycle.

[legacy]

You'll want to compile debugging symbols into your kernel. Raw addresses aren't very useful, because I don't have your System.map file or anything

I have removed in order to reduce the file size, currently we are 11Mbyte of Kernel +ramrootfs

The other thing I can think of....are you using transparent hugepage support by chance?

ummm no

Seems the R14000 CPUs make it even worse

the above machine is 2xR12K
I am using two machines, mine is 2xR12K, my friend is 2xR14
both of them have the same issues

Also, when you get an IBE, make sure to fully power the machine down and unplug the power cord for ~30 seconds. I had this issue with the DBEs in which the caches weren't being cleared correctly or the hardware not reinitialized correctly with a plain reboot or quick power cycle.

have you seen my front plane hack ? it's useful for this purpose, I can remotely force a power down + cable unplug
there is a rele of 20A between the 220V plug and the IP30-PSU

the whole looks similar to the L2 controller used in Terzo!


btw, can't we have a proper piece of assembly code to invalidate the cache ?


do I have to post here my .config ?

[legacy]

news?

[nuts]

Any news?

I've seen a linux (t2sde) port for octane here
It will be awsome if we can see gentoo arrives on the octane

[legacy]

was the last stage3 available for MIPS4/be made in 2014?
anyone has a more recent autobuild stage3?

[legacy]

see gentoo arrives on the octane

you have been asking the same question for eleven years ...
well, in 2008 we *had* gento working on IP (I still have my stage4 in my backups)
as we *had* gentoo in 2014

Currently, there two problems: - lack of testing - and - lack of updated stage3-mips4
There is a kernel 4.*, there is no kernel 5.* (is there?).

I also know there are a few issues with ssh and related. I haven't checked yet by myself, but I will!