Linux 6.10 Will Print the Number of Populated Memory Slots at Boot Time | Hacker News Reader

Linux 6.10 Will Print the Number of Populated Memory Slots at Boot Time

phoronix.com

99 pointsamalinovic24 comments

I have been yearning for this feature unremittingly since 1992. Now the intolerable void inside me will finally have been filled.

Unlike your spare DIMM slots.

it would be SIMM's in 1992!

I like this improvement. Could have saved me at least one session of trouble shooting a build!

dmidecode -t memory

(Not that the kernel printing it hurts, but at the same time, you're not going to read it real time, it flies past the screen too fast. So either way it's dmesg or dmidecode...)

It will at least also show up in a serial console, which can be buffered by the remote terminal.

> it flies past the screen too fast. So either way it's dmesg or dmidecode...

You won't believe how finicky server GPUs can be. One of our servers can update the screen so slow during booting, not only you can read everything line by line, that 30 second boot takes more than 2 minutes because of it.

Maybe it also writes to (network) serial console?

Though I suppose maybe some graphics-mode consoles can still be slow. I'm not sure if there's much point using them, though, if there is never a graphical interface running.

Nope, I'm sure. I install them. It's the normal boot console. Just KMS changes the resolution during the boot. It's very useful for reading long log lines if the server needs close attention for any reason.

[In Linux] You can use 132 character wide text modes as well, so if the boot slowness disturbs you, you could if they work in your system. (Actually also 160 wide, but I wonder if they are as commonly supported..) In addition there's the netconsole, but I guess sometimes the network interface can be down.

Thanks for the info. Actually it doesn't. We seldom reboot our servers. We use the lights-out infrastructure provided by servers if needed (the BMC thingy).

IPMI serial console has its own buffer, it doesn't usually do flow control all the way to the other side.

I have seen serial console slow things down when configured for flow control on the wrong port though. As well as saving a significant portion of boot up time by reducing logging to vga on servers with particularly slow cards (helped everywhere, but we had a new batch of severs that were much slower than the old ones... too long ago for me to remember details though)

Just curious: what GPU model is it?

IIRC, ServerEngines/Matrox G200e.

Unaccelerated framebuffer can horribly slow, scrolling is actually painful when done memory copies.

Yes, but then for use this, you need the boot to have completed succesfuly. So, it is of no help to you if you want this information to figure out why the boot is crashing.

Most modern UEFI BIOSes also show which sticks are installed in each slot.

Heck, even older pre-EFI systems. You could get clues by which read SPD

This is what serial ports are for.

Memtest86 is a great build companion as well, it will display populated slots and run memory testing, which I've found is a good overall health indicator, and always run it on new machines.

https://www.memtest86.com/

A couple years ago we bought a new Dell server and after 24h of memtest it would fail one of the last few tests, and identified a particular memory module. We went back and forth with their support, they didn't want to replace the module unless their built-in memory test, which only ran for ~30 minutes, said the module was bad. Finally we contacted the sales people and asked "How long do we have to return the server for a refund?" and shortly after that tech support agreed to send us a replacement module, which did end up passing memtest.

Another option is the related memtest86+ (https://www.memtest.org/), which can be found as a package on many Linux distributions (and the package often adds it to the grub menu), and on some Linux distributions it's even on the installation DVD (so you can use it to test the memory before installing).

For a while, it didn't run on UEFI (it required legacy boot), but recent versions now run on both UEFI and legacy BIOS boot.

Wonder if it can print 0.

Could be possible if the device has soldered on ram or the ram is located on the soc. Haven't actually checked out the code.

dmidecode had made this information available for a few decades…

Reply on news.ycombinator.com