Have you tried "pcie_aspm=performance" or "pcie_aspm=off".

I have modified grub.conf with pcie_aspm=performance and rebooted yesterday. For some reason, I cannot modify the policy file in /sys/module/pcie_aspm/parameters directory. I also get operation not permitted if I do an echo "performance" | sudo tee policy 1> /dev/null

Well, at this point I'd try and replace the drive.

I actually modified the grub to force ASPM and was then able to modify the policy. Have yet to reboot it.

Interestingly, after a reboot, the policy changed to [default]

(yes, I agree that the boot/root SSD is mysteriously failing i.e. 'disappearing' off-line)

Can you [easily] get another SSD? (if identical, maybe connect external & simply dd copy [UUID same?])
Or *maybe* 'mirror' [raid/multipath?] the SSD to free space somewhere on other 3 hdd...

I'm wondering if there's a 'better' way to capture console messages (just SEE them; not depend on any disk logging)...
** What 'console'/display do you use, when you find the system frozen?
(Could you leave -like- a serial dumb-terminal console, where displaying last messages isn't dependent on kernel video display. Or maybe remotely, to avoid any screensaver-blanking, need to [load&run] login, ...)

I found something interesting here(middle), (more info and .../power/control):
I tried this with my VirtualBox CentOS7 netinst CD /sr0/ [only; no hdd] (Ctrl-Alt-F1 thru F5)
I got lots of kernel messages on tty4 (Alt-F4: console logs) but [tty2&3] `df /`loop&top kept running ok (new cmds [on tty5] from /bin failed w/msgs of course) Also fdisk -l instead of df catches it gone.

Of course, don't take my 'noob' ramblings as *anywhere near* expert...!
(getting inanely obscure, breakpoint/debug-log [SSD?]driver.c/.ko sending such...)

I assume that /etc and [/usr]/bin&lib are on the SDD, 'mostly' readonly [not /var on SSD]

While the system is down, maybe you could boot single [ro ok for safety] & try the 'delete',
&reset, to see whether a full power-cycle is needed to bring SDD back online (per BIOS)

I wonder if some software could be ['accidentally'] doing something like this...
(blame systemd, *LOL* ... *JUST KIDDING*, sorry)

I'm looking for some way to 'prove' that what is SEEN, fully matches the SDD 'shutting off'.

Well, I had the loop script as suggested by Jjanel running over the weekend. I also had firefox running with Jenkins Web Interface. Came in this morning and it was still running. I have increased the sleep from 2 to 10 seconds for now. I cannot do any further testing without upsetting developers unless it fails again .

As for terminals, it just has an LCD monitor attached via on-board VGA. I normally SSH into it. When it froze, the SSH session is lost.

As luck will have it, it crashed again, even with the script running. Have replaced the Intel 540 240GB M.2 SSD with a standard Intel 540 240GB 2.5" SATA 3 SSD. Will see how things pan out.

@ vladguan. Any update on this issue? How did you eventually fix this? I run into similar issue with CentOS 7 VM, when running bitbake trying to generate a Yocto Project Linux kernel.