Linux on the Lenovo Y740 Legion (2019)

It’s easy.

No, really, it is. There’s a couple gotchas and some minor inconveniences (probably self-induced in my case), but provided you didn’t do anything particularly unusual with the system configuration at purchase, it should work.

First, I want to preface this with a brief overview of my configuration. I selected one with an NVMe SSD for its primary drive and a mechanical SATA drive for the secondary. I did not select one with Optane and with good reason, but I’ll get to that in a moment. All things considered, it’s a fairly banal, boring configuration with the exception of some of the features new to the 2019 lineup (notably the Corsair RGB keyboard and configurable fan LEDs). Interestingly, the behavior of this system’s EFI isn’t especially novel or noteworthy. It just works.

Caveat emptor: The configuration I discuss in this post may not work for you. I made decisions specific to how I wanted to use this system and performed some tasks manually to avoid overwriting defaults that shipped with the laptop. I’m also using Arch and other distros may present challenges specific to the software they ship or the utilities they package. Always back up any important files and do not perform these steps if you’re unwilling to lose data. I didn’t, but I’d only just gotten the laptop a couple days prior. Had I been using it for some months, I might have performed these steps differently. Some may also question why I didn’t perform a clean install of Windows; I considered it, but I didn’t feel the need to do so.

Now to the meat of the process: Before I started, I made certain to have available two USB sticks (both bootable): One with the Arch ISO image, and the other with a bootable Windows 10 installation via the Media Creation Toolkit. The latter was in my pile of tools in the event things went south and I needed to reinstall completely.

When booting to Linux on the Y740, you’ll note that the NVMe drive is not visible from Linux. This appears to be due to the storage configuration in BIOS. By default, it’s set to Intel’s Rapid Storage Technology; switching it to AHCI resolves the issue. One of the curious things to note when changing this configuration is the ominous warning that applying a different setting will delete all data on attached disks. I found this isn’t the case, but this is also why I’d recommend against selecting a model with Optane installed. I’m not certain on this point–you should certainly do your own research–but I believe with Optane installed, BIOS transparently configures it as a drive cache. Changing this on such a system may cause data corruption which is possibly what the warning implies. (The help text for this setting also mentions it’s specifically for Optane-equipped systems, hence my speculation.)

Once the drives were configured to use AHCI, the NVMe disk was accessible via Arch, and I proceeded to image it to the mechanical storage drive (using dd, of course). This image was then moved to permanent storage on my other systems in the event I did something incredibly stupid.

I look at it this way: If I didn’t have it available, I can almost guarantee I probably would have done something stupid. Always keep a backup!

(Speaking of stupid: This section is somewhat intentionally out of order but necessary for the story to flow; read below for potential video issues, because you will encounter them.)

Now, at this point, I had two choices. My initial thought was to partition the drives accordingly, and reinstall Windows completely. This would have been the ideal situation, but I wanted to save some time (and use the system in its stock state, minus some annoying cruft like McAfee). So, the next choice was to shrink the volumes on the SSD and the storage drive to roughly half. I gave Windows somewhat more storage, because it’s Windows, and because I plan to use this as a casual gaming system in addition to doing work. Doing so is easy enough: To resize the NTFS volumes, go to Computer Management -> Storage -> Disk Management. Then, you’re only a reboot away from getting started.

This is the easy part.

With the partitions resized, and a USB stick in hand, I feverishly pressed F12 to bring up the boot device selection screen–and promptly noticed that the USB stick wasn’t anywhere to be found. After unplugging it and putting it back in, BIOS appeared satisfied enough to present me with it, and off I went. I didn’t notice this with another stick I was using, and I had some difficulty replicating the problem. I’m not sure if this is a fault with the USB flash drive I had the Arch ISO written to or whether it’s just an idiosyncrasy of this BIOS. Either way, things appeared to work great…

Until the stick booted, that is.

Apparently the nouveau drivers that ship with Arch don’t work with the GeForce 2060 particularly well. I was greeted with what looked like vague video corruption and a line roughly one pixel high that appeared to be the bottom strip of whatever text was printed to the screen. Bummer. Rebooting, pressing F2, and getting into BIOS to examine whatever other configurations might help seemed to be my only salvation. Without much clue what else to pick, I noticed the graphics configuration had two states: Discrete graphics (that is the NVIDIA card) and “switchable graphics.” I knew from helping my girlfriend with her now-crippled Acer that the “switchable graphics” setting likely allowed the system to select between the integrated Intel UHD graphics on the CPU die and the discrete (NVIDIA) card; my theory at this point was to presume that doing so would allow Linux to boot up using the Intel chipset, hopefully avoiding the video corruption immediately after the kernel loaded up.

It worked, and from here we could progress.

The Arch installation was fairly pedestrian from this point: Setup the free space with partitions (I went with /boot and root on the SSD, although I should have left them merge–more on this in a moment–and the mechanical drive got a /storage mount point and swap), format as appropriate, install, configure, and… done!

Just kidding. That last part should be cry while you figure out how you want to setup your boot loader. You see, I’ve rarely used EFI systems outside virtualization. All of my systems are pre-2014-ish, and the one or two I’ve had the (dis)pleasure of poking at were all Windows machines. So, what are we to do?

(Aside: Wiping my girlfriend’s system and installing Linux would probably end with my face on a milk carton.)

First thing’s first: We need to figure out the time. No, I don’t mean how long the whole process had taken up to this point! I quite literally mean the time. One of the pain points dual booting Windows and Linux is how to handle the system clock. Once upon a time (sorry), Linux would begrudgingly chug along with the system clock configured to local time. This is a terrible idea, and I still have no idea why Windows insists, but fortunately, you can change Windows’ behavior easily enough. However, doing so requires the foresight to change this setting before getting started–something I didn’t have because I’m stupid and it only occurred to me when I got to the point of configuring the clock. Perhaps you’ll be luckier. You’re reading this after all!

Now, where were we? Oh, right! The bootloader. This is one of the deficiencies with using Linux on newer systems. Generally speaking, EFI, UEFI, or whatever your motherboard manufacturer has decided to call it, requires special attention that we Linux users haven’t had to give to boot loading since the 2000s. No, I wouldn’t use grub either–it does apparently have EFI support, but I have painful memories getting it working under VirtualBox with EFI enabled. Perhaps this is a VirtualBox-specific issue, but I’m inclined to believe we’re better off using tools designed for a specific purpose. In this case, rEFInd.

I won’t pretend to be a subject matter expert. I’ve never used rEFInd before. The Arch Linux wiki does have fantastic resources that can help you get started, but the thing I noticed with my particular configuration is that special attention had to be placed on configuring the OS declarations in esp/EFI/refind/refind.conf. If you’re following along at home, you should at least read this section on UEFI and the entry on rEFInd.

For my system, I did not follow the automatic installation with refind-install, because I didn’t want to overwrite the default EFI boot entry. Thus, I followed the manual installation process by copying rEFInd into the EFI system partition. Note that this alone is not enough to get rEFInd to work with the Y740’s BIOS. I’m not certain whether this is due to a step I’d skipped earlier in the installation or whether it’s an artifact of the manual install, but I found the only way the Lenovo BIOS would see rEFInd even if it’s in the EFI system partition is to configure the boot order via efibootmgr –bootorder 0004,0009,2001,2002,2003. I assume this should just work, but nothing I did would force the BIOS to recognize rEFInd (with the ID 0009). Changing the boot order did work, however, and with Windows (0004) as the primary–temporarily at least–and rEFInd (0009 on my system) as the secondary, I learned that BIOS had been forcibly configured to see rEFInd allowing me to change the UEFI boot order accordingly.

I also discovered that I could not get rEFInd to recognize the /boot/refind_linux.conf configuration. I’ve not investigated why, but I suspect it’s either due to my choice of partitioning (remember, I’m using a separate /boot and root!) or misunderstanding of rEFInd. However, configuring Arch via esp/EFI/refind/refind.conf has worked just fine. I should also note that by esp in the former example and above paragraphs, I mean the EFI System Partition. I mounted this (/dev/nvme0n1p1 on my system) to /boot/efi for ease of access. I’d suggest doing the same if you plan on going down this route.

Finalizing the Linux installation was relatively painless once the bootloader struggle was resolved, and I had NFS plus KDE working more or less out of the box. The system does quite well under Linux thusfar, although I’ve yet to configure wireless networking. I suspect the wireless NIC will work fine: I can see the card in ip link and the ridiculously named “Killer 1550i” cards are, as far as I know, simply rebranded Intel 9560NGW chips.

There is some unfinished business, and I’ve encountered at least one or two teething problems. First, this configuration doesn’t address secure boot. I was primarily focused on getting Linux working rather than working with secure boot. I’m hopeful this won’t be especially difficult, and from what I’ve read it appears the process is relatively straightforward. I’m planning on going the machine owner’s route with regards to kernel signing, with further plans to automate the process in a pacman hook. I’ve also noticed the wired network didn’t come up automatically in spite of being enabled via systemd (I prefer using systemd network configurations via /etc/systemd/network over NetworkManager), and there’s a large amount of i2c_hid spam in the journal. I suspect this may have something to do with some of the peripherals in the system (touchpad? wireless mouse?).

I’ll eventually write a part two once I get some of the remaining issues resolved, along with documenting whatever else I may encounter. If you’re using Linux and just bought one of these systems, please don’t feel overwhelmed with the installation process. Just be cautious, think things through, and have a plan of attack. Don’t forget to back things up, too. Oh, and as much as I don’t like Windows 10’s Microsoft account option, I would recommend logging in with one at least once because it ties the software license for that machine to your account. If you decide to reinstall Windows, this is a good thing, in my opinion!

2 comments.
***

Updates Suck (Part 2): KDE Plasma 5.5 breaks task manager minimized application display

If you’ve recently updated to KDE Plasma 5.5, you may have noticed that minimizing your applications no longer produces a noticeable effect in the task manager. (In actuality, the status effect of minimized versus non-minimized applications is limited to a thin, gray outline around the application title–hardly something that conveys application state well enough to process visually at first glance.)

The fault of this change lies with this bug report posted almost 3 years ago. After much discussion, it was decided that 1) the “disabled” state of minimized applications was a mistake, 2) the minimized state makes it too difficult for some people to locate the gray scale application icon, and 3) no one wanted to add a configuration option to enable or disable this feature, so the next best thing was to rip it out. Now, I’ll be honest: Maybe I just happen to have better-than-average pattern matching capabilities in my visual cortex (doubtful), but the gray scale icons for minimized applications really don’t bother me. What does bother me is this change: It takes me twice as long to find minimized applications on the task manager. To say nothing of the fact that two or three people complaining incessantly for 3 years can effect such a substantial change on the rest of us who actually liked the feature…

Humorously, I didn’t notice this change during the initial update. Since I spread my work out across multiple desktops, I don’t often minimize my applications. This explains, at least in part, why it took three days for me to discover this problem and make a post about it (and why I blamed it on plasma-shell’s propensity to crash all the time, although that seemed like a reasonable explanation in lieu of anything more substantive).

The good news is that I traced it down to this commit which turned off the minimized gray scale display. As expected, removing this patch returns the task manager’s display of minimized applications to its previous state. The bad news is that, yes, you do need to recompile plasma-desktop from source. If you happen to use Arch Linux, I’ve put together a PKGBUILD containing the fix based off the PKGBUILD in ABS (hence why I haven’t yet changed any of the credits–I should note that this is not an official package). I’ll keep this updated as best as I can until such time that upstream finally settles on a better idea than a 1 pixel gray outline for non-minimized applications.

The rest of you will have to apply the patch and build the sources yourself. Sorry about that.

No comments.
***

Updates Suck (Part 1): Firefox 43, HTML5 breakage, and more

Sometimes, updates are awful. They break things, they undo configuration options you were sure you had set at some point in the past (either resetting them to defaults or removing them entirely).

I updated my Arch install for the second time in two days in the hopes that some minor bugs would be resolved (more on that in another post), but I hadn’t noticed that Firefox was included in the update list (version 42 to 43). Sadly, not only did this break Firefox’s appearance using oxygen-gtk, but it also broke HTML5 video. Every video on Youtube refused to load following the updates unless Flash was enabled (I have it disabled), and I couldn’t figure out why. A new profile proved somewhat useful: Youtube would work–unless I copied my old profile over in bits and pieces.

After about a half hour of exploring, I found that the culprit was actually in the profile’s prefs.js. In particular, because I had found some instructions on enabling HTML5 under Linux, apparently there were some configuration options that are no longer used and conflict with Firefox 43, which enables many of these by default.

If you encounter a similar problem, the fix should be relatively easy. Simply remove:

user_pref("media.fragmented-mp4.exposed", true);
user_pref("media.fragmented-mp4.gmp.enabled", true);

From your prefs.js, restart Firefox, and verify that it still works. If not, try removing all of the media.* preferences. If that still doesn’t work and you’re a user of FlashBlock, try uninstalling it and removing any residual preferences (also from prefs.js).

I used to recommend using FlashBlock in combination with NoScript to reduce the browser’s attack surface, but as fewer and fewer sites I encounter actively use Flash, I find that disabling the plugin (settings, addons, plugins) works best for me. Unfortunately, the author included a new feature to block HTML5 video by default, and I’ve found that it tends to cause breakage with some sites, most notably Youtube. Since then, I’ve had the plugin disabled, but as I no longer have a need for it, why keep it around?

Hopefully, if you’ve encountered a similar problem with HTML5 video, the suggested fixes here will work for you. If not, feel free to leave a comment or pester me on Twitter (@zancarius).

(I still haven’t gotten Firefox 43 to play nicely with oxygen-gtk3, so I may eventually consider abandoning the theme–though I find breeze-gtk still contains a variety of display bugs that make it unpleasant to use.)

No comments.
***