> From: kiltz <[email protected]>
> Date: Mon, 13 Jun 2022 14:46:39 +0200
Hi Stefan,
> Dear kind people at OpenBSD.org,
> we want to run OpenBSD as a firewall system on a Gigabyte R152_P30
> with the following specifications:
>
> Ampere Altra Q80-33 processor (80 Cores, 3,3 GHz)
> 512 GB RAM (3200 MHz ECC-reg.)
> 2 x 480 GB SSD SATA 6 Gb/s 2,5''
> Dual-Port 1 GbE (RJ-45)
> IPMI 2.0 Baseboard Management Controller (BMC)
> 1 x PCIe4.0 x16 (FHHL)
> 1 x PCIe3.0 x16 OCP2.0 (belegt)
> 1 x USB 3.0 (front), 3 x USB 3.0 (rear), 1 x VGA (rear)
>
> We tried both:
> - official stable 7.1 (/pub/OpenBSD/7.1/arm64) and
> - snapshot from 6th of June 2022 (/pub/OpenBSD/snapshots/arm64)
>
> The repeatable result is a working install in single CPU/Core
> installation mode, cpu panic after first reboot with mp kernel. We use
> the serial to LAN console provided by the IMPI/BMC card.
> Attached you will find screenshots from:
>
> - the last 49 columns of the reboot into mp kernel
> (Screenshot_boot_after_install_Gigabyte_R152_P30 at 2022-06-13
> 13-51-00.png),
> - the ddb trace output (Screenshot ddb_trace_2022-06-13 14-02-11.png),
> - the ddb ps output (Screenshot ddb_ps_at 2022-06-13 14-03-25.png),
> - the ddb show panic output (Screenshot ddb_show_panic_at 2022-06-13
> 14-04-28.png)
> - the ddb show registers output (Screenshot ddb_show_registers_at
> 2022-06-13 14-06-34.png)
>
> Due to the nature of the early boot panic, the kernel output is not
> accessible to us.
>
> Interestingly, FreeBSD only supports them in their current release,
> the stable fails with a similar panic. They seem to have found a fix
> of sorts. But we very much prefer OpenBSD for the firewalling role of
> aforementioned system.
>
> Of course we support your effort so if you need more info from us
> regarding the circumstances, we will happily try and supply the
> required information.
The immediate problem is that OpenBSD currently supports a maximum of
32 CPUs. That limit is a bit arbitrary, so the diff below bumps it to
128. You could try building a GENERIC.MP kernel with this diff after
booting the GENERIC (bsd.sp) single-processor kernel. I'll see what
my fellow developers think abut bumping MAXCPUS. Depending on the
outcome of that a snapshot with this change may be available in a few
days.
I'm not sure how well OpenBSD/arm64 scales to 80 CPUs. Probably not
very well but I guess there is only one way to find out...
Cheers,
Mark
Index: arch/arm64/include/cpu.h
===================================================================
RCS file: /cvs/src/sys/arch/arm64/include/cpu.h,v
retrieving revision 1.25
diff -u -p -r1.25 cpu.h
--- arch/arm64/include/cpu.h 23 Mar 2022 23:36:35 -0000 1.25
+++ arch/arm64/include/cpu.h 13 Jun 2022 15:09:32 -0000
@@ -184,7 +184,7 @@ extern struct cpu_info *cpu_info_list;
#define CPU_INFO_FOREACH(cii, ci) for (cii = 0, ci = cpu_info_list; \
ci != NULL; ci = ci->ci_next)
#define CPU_INFO_UNIT(ci) ((ci)->ci_dev ? (ci)->ci_dev->dv_unit : 0)
-#define MAXCPUS 32
+#define MAXCPUS 128
extern struct cpu_info *cpu_info[MAXCPUS];
