svn commit: r346598 - head/sys/modules
Greg V
greg at unrelenting.technology
Mon Apr 29 15:23:10 UTC 2019
On Mon, Apr 29, 2019 at 10:08, Ed Maste <emaste at freebsd.org> wrote:
> On Tue, 23 Apr 2019 at 13:26, Rodney W. Grimes
> <freebsd at gndrsh.dnsmgr.net> wrote:
>>
>> Very cool, now how do I get a PCIe slot into a RPI3!!! lol :-)
>
> I know you're joking but the comment does highlight an issue in the
> AArch64 world - there's a lack of good mid-range developer platforms.
> FreeBSD runs on Cavium/Marvell ThunderX and ThunderX2 and now on
> Ampere eMAG with the WIP discussed in PR 237055. These platforms have
> room for lots of memory, very high core/thread counts (32 to 256), and
> a good complement of PCIe interfaces. The specs go far beyond those of
> a typical desktop software development platform, and the price does
> too. We also run on small embedded boards like the RPi, Pine64, etc.
> just fine, but there's not much in the middle. What we really need is
> something like a Mini-ITX form factor 4 to 8 core system that can take
> 8 to 32GB of RAM, has a PCIe slot or two, and is readily available
> selling for well below $1000 US.
SolidRun/Marvell MACCHIATObin is probably the best option available
now, but SolidRun is working on new stuff:
https://www.solid-run.com/nxp-lx2160a-family/clearfog-itx/
NXP LX2160A — SoC with 16 Cortex-A72 cores, dual-channel DDR4 (MCbin
is single channel), 18 lanes of PCIe Gen 4, and as usual a huge
built-in network card we don't have a driver for. Though NXP advertises
that it's possible to reconfigure the SFP+ ports to turn them into more
PCIe.
They even confirmed that overclocking is possible for both CPU and RAM:
https://www.phoronix.com/forums/forum/hardware/motherboards-chipsets/1090102-solidrun-clearfog-a-16-core-arm-itx-workstation-board-aiming-for-500~750-usd?p=1090905#post1090905
Let's hope they implement ACPI fully and correctly :)
>> I am hopeing some of that PCIe WIP might include some of the
>> bits needed or do we already have PCIe slot on RockPro64 code that
>> works?
>
> I don't think this will do anything for RockPro64, it's just a
> workaround for limitations in our current arm64 PCI code for some
> functionality unused by ThunderX* but required for eMAG.
RK3399 seems to need a custom driver, OpenBSD has one:
https://github.com/openbsd/src/blob/master/sys/dev/fdt/rkpcie.c
And I've heard bad things about the controller, apparently no one has
got a GPU working on Linux because the controller doesn't support
enough address space or some other features.
More information about the svn-src-all
mailing list