IRC log of #cubox of Tue 25 Jul 2017. All times are in CEST < Back to index

06:17 jnettlet[m]> So the startech expansion card works well in the 8040 board. It seems like the nvme slot has 2x PCI lanes, and the other 2 lanes are used 1 each for the additional sata ports. which is not bad, my samsung 960 EVO gets about 1.3-1.4GB/s seq read.
06:17 jnettlet[m]> I haven't profiled it yet to see where the bottlenecks are.
11:02 topi`_> jnettlet[m]: somebody said on #linaro that the pcie on macchiatobin is somehow broken
11:02 topi`_> but obviously it works for you since the startech card must be using x4 signals?
11:02 rabeeh> i think there is an issue when using GPUs
11:02 topi`_> how come it doesn't affect NVMe boards then?
11:03 jnettlet[m]> to use a GPU you need to use uefi not u-boot
11:03 topi`_> yes, most probably
11:03 jnettlet[m]> my NVME signal is only using x2, and then 2 x1 for sata ports
11:03 topi`_> but then again, it's not a macchiatobin specific problem ;)
11:03 rabeeh> the reset signal is not held low on power up; so this confuses part of the GPUs; the workaround that we suggest for Marvell (for now) is to add a pull down resistor on the reset signal and it solved the issue for them.
11:04 topi`_> can I try to add such a pulldown myself?
11:04 topi`_> probably at least 10K ohms needed
11:04 rabeeh> yes
11:04 rabeeh> i can send you a picture where to deploy
11:05 topi`_> I probably don't have time for that now, since gotta work on the HB :)
11:05 topi`_> which brings me to the point... jnettlet: do you have any "new" patches for mainline for imx6?
11:05 topi`_> I'm using some patchset that you sent me in April
11:05 topi`_> for 4.9.21
11:07 jnettlet[m]> topi`_: no. mainline wants me to actually minimize my patches for our boards. We have too many variants. So full board support will actually be using u-boot with fdt overlays to support all the boards fully.
11:08 topi`_> 4.9 LTS seems to have received a lot of patchlevels, now it is at 4.9.39
11:09 jnettlet[m]> yep, lots of churn
11:09 topi`_> we try to keep patching the deployed systems remotely so that the worst kernel exploits get patches :)
11:10 topi`_> I'll try to rebase my custom patchset on top of that 4.9.39
11:11 topi`_> I think the next thing I'll test the HB is using it as an accesspoint (base station)
11:11 topi`_> the Wilink seems to be rather decent
11:11 topi`_> I had lots of problems on the older HB> with the broadcom wifi
11:11 topi`_> it would die after a while of transferring photos etc
11:11 topi`_> would only recover after systemctl stop hostapd && systemctl start hostapd
11:14 jnettlet[m]> yes the wilink8 is much better.
12:36 topi`_> rabeeh: any hints for enabling PWM for the macchiatobin fan?
12:36 topi`_> I've enabled PWM on some OMAP3 chip in historical times, it wasn't that hard...
12:36 topi`_> but that was for a LED
12:53 rabeeh> the GPIO is CP0_MPP43
12:53 rabeeh> the GPIO is CP0_MPP48
12:53 rabeeh> it drives a buffer that drives the fan's pwm line
12:54 topi`_> hmm, maybe I can try that semi-manually
12:54 topi`_> can that gpio be controlled via /sys/class/gpio interface?
12:54 rabeeh> i assume that you don't have specs; right?
12:54 topi`_> yeah I'm working on the HB, not the macchiato ;)
12:54 topi`_> but it'd be nice to reduce the fan noise of the latter
12:55 rabeeh> i have a 7 CFM fan connected to it
12:56 rabeeh> basic fan that blows minor air to it
12:56 rabeeh> the temperature rise with the big heatsink is ~20c
12:56 rabeeh> that's when running cpuburn on the 4 processors at 2GHz
12:56 topi`_> are there different heatsinks? mine came fitted with a heatsink with 2 cm cylindrical fins
12:57 rabeeh> silver finish?
12:57 rabeeh> i.e. you see the aluminium color? (i.e. not black)?
12:57 topi`_> yeah alu
12:57 rabeeh> yeah; so that's the bigger heatsink that we have today
12:58 topi`_> it still gets hot... if I compile linux with all cores, it at some point resets the board
12:58 topi`_> the temp reading goes to 93C and above
12:58 topi`_> this was at the 1.3 ghz core setting
12:58 topi`_> 2.0 ghz runs the cores somewhat hotter still
12:59 topi`_> where can I get the right gpio number from linux's point of view?
12:59 topi`_> the GPIO pin <-> linux_gpio mapping was somewhat easy to find in the Hummingboard docs
12:59 topi`_> but here...
13:06 jnettlet[m]> topi`_: I am not sure if the 8040 pwm's work the same as the other armada devices. If so then pwm is just a property of the gpios...basically it has a blink setting in the hardware.
13:08 topi`_> oh?
13:08 topi`_> never heard of that
13:10 jnettlet[m]> here is the armada support. https://lwn.net/Articles/718211/
13:12 jnettlet[m]> the 8040 uses the same gpio driver, so there is a good chance that it does pwm the same
13:16 topi`_> hmm, too much work, maybe I'll just buy a quiet fan for this .D
13:16 topi`_> .D
13:16 topi`_> wife expects me to come home
13:17 topi`_> this is the kind of work Marvell should be doing!