accel/amdxdna: Add iommu_bypass debugfs for aie4 only

[xdavidz]

Add debug only debugfs for aie4.

[Copilot]

Pull request overview

Adds an AIE4-specific debugfs hook to the amdxdna driver so device-specific debugfs entries can be registered (intended for an iommu_bypass debug control on AIE4).

Changes:

• Extend struct amdxdna_dev_ops with a per-device debugfs() callback and invoke it from the common debugfs init path.
• Add AIE4 debugfs support (aie4_debugfs.c) and wire it into AIE4 ops.
• Extend AIE4 mailbox protocol definitions with an ECHO opcode and request/response structs.

Reviewed changes

Copilot reviewed 8 out of 8 changed files in this pull request and generated 4 comments.

Show a summary per file

File	Description
drivers/accel/amdxdna/Kbuild	Adds aie4_debugfs.o to the DEBUG_FS build.
drivers/accel/amdxdna/amdxdna_pci_drv.h	Adds a debugfs callback to device ops.
drivers/accel/amdxdna/amdxdna_pci_drv.c	No functional change (whitespace only in shown hunk).
drivers/accel/amdxdna/amdxdna_debugfs.c	Calls the device-specific debugfs hook if present.
drivers/accel/amdxdna/aie4_pci.h	Declares AIE4 debugfs init entrypoint.
drivers/accel/amdxdna/aie4_pci.c	Wires .debugfs = aie4_debugfs_init into AIE4 ops.
drivers/accel/amdxdna/aie4_msg_priv.h	Adds AIE4 echo opcode and message structs.
drivers/accel/amdxdna/aie4_debugfs.c	Introduces AIE4 debugfs file iommu_bypass and echo-based write path.

---

💡 Add Copilot custom instructions for smarter, more guided reviews. Learn how to get started.

[maxzhen]

Is the bypass mode value differing from system to system? If not, you don't really need a debugfs node for this. You can have a global macro to define the mode and make compile time decision. This is much simpler and easier to upstream, I think.

[xdavidz]

Is the bypass mode value differing from system to system? If not, you don't really need a debugfs node for this. You can have a global macro to define the mode and make compile time decision. This is much simpler and easier to upstream, I think.

I would keep this debug only for bringup. We do not need to upstream this.
@houlz0507 you are ok with this not being up-streamed right? This is bringup only feature, that the very first step is alawys the iommu_bypass mode, but it requires ipulib(jtag) to setup a register when mpnpufw is already loaded by the driver. We are just trying to make this a little user-friendly. Any end-user of repleased product would not need to use this feature.

[xdavidz]

Is the bypass mode value differing from system to system? If not, you don't really need a debugfs node for this. You can have a global macro to define the mode and make compile time decision. This is much simpler and easier to upstream, I think.

I would keep this debug only for bringup. We do not need to upstream this. @houlz0507 you are ok with this not being up-streamed right? This is bringup only feature, that the very first step is alawys the iommu_bypass mode, but it requires ipulib(jtag) to setup a register when mpnpufw is already loaded by the driver. We are just trying to make this a little user-friendly. Any end-user of repleased product would not need to use this feature.

after talking with Max. I think we need to do 2 things:

1. covert debugfs to a ifdef macro to minimize the engineering complicity.
2. using OOT driver for using cma for now. I will update this PR later.