Date: Mon, 28 Sep 2020 13:08:09 -0700 From: Mark Millard <marklmi@yahoo.com> To: freebsd-arm <freebsd-arm@freebsd.org> Subject: FYI: Summary of what it took to get around 8.5 hr head -r365932 buildworld buildkernel times on a RPi4B Message-ID: <1B998B4B-411E-46D9-981D-3197AB1223A3@yahoo.com> References: <1B998B4B-411E-46D9-981D-3197AB1223A3.ref@yahoo.com>
next in thread | previous in thread | raw e-mail | index | archive | help
The major configuration properties contributing to having smaller RPi4B =
FreeBSD
buildworld buildkernel times have been (not ranked):
Cortex-A72 clock rate 2 GHz:
over_voltage=3D6 and arm_freq=3D2000 (config.txt)
For u-boot based booting: also dev.cpu.0.freq=3D2000 (/etc/sysctl.conf)
Always full speed RAM:
sdram_freq_min=3D3200 (config.txt)
Note: Only effective with u-boot based boot. So uefi/ACPI v1.20 context =
is slower.
Using Cortex-A72 tuned code ( from -mcpu=3Dcortex-a72 ) in the running =
kernel/world
When sdram_freq_min=3D3200 is effective: Use of -j4 for doing =
buildworld buildkernel
When sdram_freq_min=3D3200 is ineffective: Use of -j3 for doing =
buildworld buildkernel
Use of RPi4B's with sufficient (effective) RAM
Heaksinks and fan use, along with 5.1v 3.5A power supply
Note: <3072 MiByte effective RAM: not tested.
Note: Other types of cooling: not tested.
Note: Other types of power supplies: not tested.
File system (UFS) on USB3 SSD (vs. microsd card)
Note: zfs: not tested
Note: Other USB3 media: not tested
Other notes:
Note: Recent tests were based on head -r365932 .
Note: 1024 MiByte RAM would slow things at times (significant paging
for the same -jN).
Note: Using a kernel built with -mcpu=3Dcortex-a72 requires head =
-r365918
( stable/12: -r366182 ; stable/11: -r366183 ) or later to avoid
misconfiguring one of the rings for xHCI.
Note: Use of an appropriate u-boot to boot head -r365677 or later
allows 4 MiBYte and 8MiByte RPI4Bs to handle xHCI/USB3 reliably
(USB2 untested). (No MFC?)
Note: Use of uefi/ACPI from https://github.com/pftf/RPi4/releases/
requires use of the uefi 3072 MiByte limit for USB3 I/O to be
reliable for 4 GiByte and 8GiBYte RPi4Bs (USB2 untested). The
https://reviews.freebsd.org/D25219 patch is involved in having
things operate, even for 3072 MiByte. uefi use also ignores
(overrides?) sdram_freq_min=3D3200 use. There are also tradeoffs
for what devices are supported.
=3D=3D=3D
Mark Millard
marklmi at yahoo.com
( dsl-only.net went
away in early 2018-Mar)
Want to link to this message? Use this URL: <https://mail-archive.FreeBSD.org/cgi/mid.cgi?1B998B4B-411E-46D9-981D-3197AB1223A3>