It Turns Out Windows Unconditionally Reserves The First 1MB Of RAM, Linux Was Just Late To Do So

[lamka02sk]

Any motherboard manufacturers to avoid? If they can't fix their shit, they certainly don't deserve my money.

[rene]

Turns out PC BIOS' still are utter garbage, and engineers always hated it like the plague for good reasons. It's hilarious that the good old BIOS was replaced with even more crap code running and corrupting things. Heck, in 2021 not even the bloody settings work halfway decent: https://www.youtube.com/watch?v=gHHogyokkRQ

[ms178]

Setting up a Sandy Bridge HP laptop with 5.13-RC was a nightmare. The problems started with several distros refusing to boot from an USB stick, crashing right after GRUB initialization (only openSUSE worked, Ubuntu and Manjaro did not). The 6770M did not work at all (dmesg showed radeon related errors) and 3 of four threads were not showing up or used at all which made it painfully slow. It turned out a BIOS option could at least get me back the lost three threads. But unfortunately it felt less responsive with that setting.

[AJenbo]

The correct way would be for both Windows and Linux to refuse to boot if the BIOS corrupts the memory. That would lead to an actual fix of the root cause....

It's not detectable, that's why they apply this unconditionally in the first place.

[Guest]

This is yet another case of when dealing with low level system interfaces, it's best to do whatever Windows does (as that's all firmware programmers tend to test) rather than what the spec says.

There was another similar issue a few years ago where doing what Windows does fixed reboot hangs on some systems: https://mjg59.livejournal.com/137313.html

Oh god. That's just so cringy (the things they have to do to get reboot working). Ew.

[Guest]

I wonder how many hundreds of bugs will magically get fixed now..........

[Guest]

Any motherboard manufacturers to avoid? If they can't fix their shit, they certainly don't deserve my money.

They're probably all crap, but I haven't had any problems with MSI so far on two different boards. ThinkPad seems to work fine too.

Then again, it might just be due to kernel devs adding work around which is why it works so well for me.

[rene]

It's not detectable, that's why they apply this unconditionally in the first place.

well, normally we denylisted defects like this. It is also detectable, IIRC the linux kernel even has (or had) some periodical memory consistency checking option. If one wanted to be overzealous, e.g. for vintage hardware, one could periodically test pattern check it for corruption before use and allocate that area last ;-)

[lolren]

AI wonder if this was the reason Lenovo laptops use to break just by booting Linux: https://www.omgubuntu.co.uk/2017/12/...vo-laptop-bios

[jabl]

The correct way would be for both Windows and Linux to refuse to boot if the BIOS corrupts the memory. That would lead to an actual fix of the root cause....

Unfortunately idealism doesn't take you very far when dealing with hardware. Particularly so for Linux which doesn't have the market share to dictate squat in the consumer PC market (servers, embedded and android are different). Specs are poorly written, overspecify some parts and underspecify others, and contain bugs. Hardware implements those specs poorly if at all, and then it's up to software to make it work somehow. If you've followed Linux kernel (or any other kernel) development or looked at driver code, you'll see it's chock full of WTF code for dealing with various hardware quirks.

This particular issue, unfortunately, is but one in a very long list of things done to deal with HW bugs or outright stupidity.

The marvel here is that the entire house of cards works as well as it does, despite all this.