Difference between dts and ACPI

问题

We can declare platform device information in dts file, rather than hard coding every data into operating system. Taking "arm" architecture as example. it supports dts and we will take dts from arch/arm/boot/dts/xx.dts. Convert this xx.dts into xx.dtb and loaded with kernel Image. I recently came across ACPI, when i came across x86 architecture, from the documents, what i understood is ACPI is similar to device tree. We can declare platform device information information in ACPI tables, my doubt is where exactly these ACPI tables present. How can i load this info to linux. What is the advantage of using ACPI over dts. Please correct me if i am wrong. Thanks in advance

回答1:

IMHO

ACPI and DT are used for similar purpose. but they have there unique functionalities. Now a days the effort of defining ACPI configs in DT.

both used to solve different issues.

ACPI purpose was to improve power efficiency.
DT purpose to remove platform files out side kernel.

Device tree is mostly passed to linux kernel before it boots up. mostly ACPI is loaded while linux kernel is booting (check Documentation/acpi/enumeration.txt for more info)

for any other thing just comment.

回答2:

Not completely correct:

ACPI started as an interface between firmware (formerly BIOS) and OS for things like power management, but also things like platform device probing
DT was always (even long before ACPI existed) about declarative platform device descriptions (probing and configuration), so the OS can properly initialize all drivers, configure operation points, etc, etc.

ACPI was always very limited in scope and depends on firmware, while DT stands on its own (just requires the bootloader to pass the right dtb to the kernel).

ACPI is the unprofessional, hackish attempt of bios and board vendors to solve a small subset of the problems that DT already solved long ago. A major pro argument for those gallows-wearing folks probably is that ACPI/BIOS hides lots of low level configuration stuff (up to runtime device programming, eg. for power management) in the firmware blob, thus preventing the OS kernel to have full control over the machine. (which finally leads to things like broken machines by broken BIOS, etc). We, the kernel developers, often have to work around crappy BIOSes.

My strong advice: get of ACPI when you can.

回答3:

Breathe with lungs or gills? Depends on where you live.

A rough classification of architectures is

x86 - Server/PC - ACPI table 
ARM - embedded systems - Device Tree

On server/PC motherboards, the ACPI table is a part of the UEFI firmware, which resides on the flash chip. The OS would be installed later somewhere else (hard drive or so). The OS parses the ACPI table, but OS developers don't control what is already written in the firmware; or they don't even know the internal design of the board. The board vendor (firmware provider) needs to support whatever OS to be installed, not only Linux, so they have to follow standards (UEFI), instead of focusing on a Linux thing, such as device tree.

On embedded systems, OS and everything else are programmed once by the vendor and never again by the user. The OS is part of the firmware. So no need to worry about the OS support matrix, and just have a 1-to-1 relationship between the board and your OS image. U-Boot, kernel, initramfs, device tree blob reside on the same flash storage (i.e. NAND). So developers have access and control to what to be deployed as the device tree (must match real hardware though).

Hardware designers should be able to provide both an ACPI table and a device tree. Depending on the receiver, one will be preferred.

References: