From: Tom Rini <trini@konsulko.com>
Date: Wed, 26 Mar 2025 20:07:09 +0000 (-0600)
Subject: Merge tag 'efi-next-26032025' of https://source.denx.de/u-boot/custodians/u-boot... 
X-Git-Tag: v2025.07-rc1~18^2~13
X-Git-Url: http://git.openpandora.org/cgi-bin/gitweb.cgi?a=commitdiff_plain;h=df11ac859d366d96f749f2b9c3d2d5c564b325bc;p=pandora-u-boot.git

Merge tag 'efi-next-26032025' of https://source.denx.de/u-boot/custodians/u-boot-tpm into next

When trying to boot an OS installer or a live image via EFI HTTP the
following happens
- U-Boot downloads the image and mounts it in memory
- The EFI subsystem is invoked and the image is started
- The OS calls ExitBootServices and the memory that holds the mounted
  image might get overwritten

This results in installers complaining that they can't find installer
medium or live images complaining they can't find the root filesystem.

ACPI already deals with it by having NFIT and NVDIMM to provide ramdisks
that need to be preserved by the OS. Linux and device trees have support
for persistent memory devices (pmem).

We can use them and inject a pmem node in the DT to preserve memory across the
entire boot sequence. Linux will just create a block device over the reserved
memory and installers/images can re-discover it.

This is what it looks like from the OS perspective:

nd_pmem namespace0.0: unable to guarantee persistence of writes
pmem0: p1 p2 p3
EXT4-fs (pmem0p3): mounted filesystem f40f64a4-5b41-4828-856e-caaae2c1c2a0 r/w with ordered data mode. Quota mode: disabled.
EXT4-fs (pmem0p3): re-mounted f40f64a4-5b41-4828-856e-caaae2c1c2a0 r/w. Quota mode: disabled.
Adding 45052k swap on /dev/pmem0p2.  Priority:-2 extents:1 across:45052k SS
root@genericarm64:~# mount | grep pmem
/dev/pmem0p3 on / type ext4 (rw,relatime)
/dev/pmem0p1 on /boot type vfat (rw,relatime,fmask=0022,dmask=0022,codepage=437,iocharset=iso8859-1,shortname=mixed,errors=remount-ro)

It's worth noting that Linux behaves differently with reserved memory
(at least on arm64) and that depends on kernel config options.
CONFIG_ZONE_DEVICES and CONFIG_ARM64_PMEM are such options. It boils down to
how the kernel tries to map pages. If devm_memremap_pages() gets called instead
of devm_memremap() mapping the memory fails.

The only safe way is to remove the memory from the EFI memory map,
rather than defining it as /reserved no-map;/ in the DT.
---

df11ac859d366d96f749f2b9c3d2d5c564b325bc