Linux x86/x86_64 Will Now Always Reserve The First 1MB Of RAM

[torsionbar28]

I do not see Microsoft as the issue in this case. It is the fault of mainboard manufacturers who choose proprietary BIOSes. Once manufacturers start using free and open software for BIOSes will it lower the costs and it will improve the quality. But as long as manufacturers do not work together in creating a free and open BIOS software will the manufacturers continue to fix issues on their own, pay licensing fees for it, pass the costs on to the customers, who end up with broken and awkward BIOSes, for which OSes need to implement work-arounds. The only winner here are the makers of these proprietary BIOSes, who are now going as far as implementing graphical user interfaces to distinguish themselves from other makers.

Microsoft is the one driving the UEFI and SecureBoot adoption with the vendors. The lack of FOSS commercial UEFI offerings forces board vendors to stick with their same old suppliers. Coreboot doesn't cut it, unless you're a small niche player like System76. What we need, is a "Red Hat style" approach to motherboard firmware. I.e. a corporate entity who can develop, sell, and support the code base with commercial customers, while releasing the source code to the community.

Another factor is that especially with consumer "peecee gamer" motherboards, having a gagillion BIOS options for overclocking and such is a selling point. The FOSS implementations I've seen have all been pretty basic and stripped down with bare minimum options. You and I might well prefer this, but the market demands flashy animations and RGB LED controls and such.

The graphical users interfaces of BIOSes that I so far have seen are all horrible. One cannot get into a BIOS setup without a keyboard, and the mouse control is worse than that of any OS I have seen in the past 30 years. Not to mention the total boot time that x86 PCs these days spend in the BIOS.

For sure, I'd much prefer something like the ARC firmware on the DEC Alpha, where basically the only thing it does is enumerate the hardware and let you select a device to boot from. Small, lightweight, fast.

[coder]

Give it a couple of decades, and we'll see a commit to reserve the first 1 Gigabyte of RAM.

Modern BIOS/equivalents don't seem to be building the same kind of legacy, though. The software in question has legacy stretching back to the era of DOS systems running on 16-bit 8088 machines that had only the concept of physical addressing.

[coder]

How am I going to run the latest kernel on my 386SX with 1MB of RAM now??!??

I first installed Linux on a 386DX 25 MHz with 8 MB of RAM. In the era of Windows 3.x, you wouldn't really see a 386 with less than 2 MB of RAM.

And what about my 8086 from 1988?

AFAIK, the oldest thing Linux ever properly supported was the 80386. And anything before an 80286 lacked memory protection.

[coder]

"640 kB ought to be enough for anybody." Bill Gates* - now we can't even start to boot without less then 1mb.

*I know it is controversial if he has ever said that.

He did say it, or roughly that. However, it got taken out of context. He didn't mean all computer users for all time -- he was referring specifically to machines running DOS that only had the hardware capability of addressing 1 MB.

The context was when they were creating the original memory map of reserved address ranges. Look here:

https://en.wikipedia.org/wiki/Conventional_memory

[torsionbar28]

Modern BIOS/equivalents don't seem to be building the same kind of legacy, though. The software in question has legacy stretching back to the era of DOS systems running on 16-bit 8088 machines that had only the concept of physical addressing.

You're suggesting that the x86 architecture will either go away, or will be redesigned to drop the legacy support? That's the thing about legacy compatibility - you must either carry it forwards indefinitely, or draw a line in the sand and drop it.

[coder]

One thing that bothers me about this is it basically reserves that first 1 MB for anything. So, a root kit could use it, or maybe malware running on a device on PCI/PCIe that's somehow not going through a IOMMU.

[coder]

You're suggesting that the x86 architecture will either go away, or will be redesigned to drop the legacy support? That's the thing about legacy compatibility - you must either carry it forwards indefinitely, or draw a line in the sand and drop it.

What I'm saying is that I believe BIOS isn't directly using beyond the 1 MB of physical addresses. That's what would have to happen, for them to block off a larger chunk, like you suggest.

[KesZerda]

He did say it, or roughly that. However, it got taken out of context. He didn't mean all computer users for all time -- he was referring specifically to machines running DOS that only had the hardware capability of addressing 1 MB.

The context was when they were creating the original memory map of reserved address ranges. Look here:

https://en.wikipedia.org/wiki/Conventional_memory

Exactly. Even had IBM put most of the reserved area at the beginning, there still would have needed to have been a gap because of processor mandates. Thing like the reset vector are set in stone to be at FFFF:FFF0, with no way of changing it, save breaking backward compatibility, and we all know how well the attempts to break backward compatibility with x86 have gone. 640k seemed very reasonable in an era where the base machine had 16k of RAM, and topped out at 256k.

[torsionbar28]

What I'm saying is that I believe BIOS isn't directly using beyond the 1 MB of physical addresses. That's what would have to happen, for them to block off a larger chunk, like you suggest.

Got it. "thumbs up emoji"

[Mark625]

The entire first megabyte? But that's 1/32768th of my computer's installed memory! This is an outrage.