ftrace: Fix not to write duplicate header when append to existing ftrace file

[Leo-mediatek]

Currently, tee-supplicant appends to an existing prof file if OP-TEE provides a non-zero id in the rpc call parameter. However, for ftrace data, there is always a header in the buffer.

This patch makes tee-supplicant not to write duplicate header by locating the ftrace magic and discarding data before it when tee-supplicant receives a non-zero ftrace id.

This is required for OP-TEE/optee_os#7657 to produce correct ftrace data.

[jforissier]

This introduces some knowledge on the ftrace format into tee-supplicant (the "FTRACE\0\1" magic). I'd like to avoid that is possible. Could OP-TEE OS be updated to avoid sending the magic twice instead?

[Leo-mediatek]

I cannot remember why I chose to remove the ftrace header here.

I think we can make the header not to be copied to the tee-supplicant by adding an if at optee_os to check whether fbuf->dump_id == 0.

https://github.com/OP-TEE/optee_os/blob/ce964642bf4580d8ab1914b6c4d4bcf7bd055928/ldelf/ftrace.c#L109-L110

[jforissier]

I cannot remember why I chose to remove the ftrace header here.

😁

I think we can make the header not to be copied to the tee-supplicant by adding an if at optee_os to check whether fbuf->dump_id == 0.

https://github.com/OP-TEE/optee_os/blob/ce964642bf4580d8ab1914b6c4d4bcf7bd055928/ldelf/ftrace.c#L109-L110

It should work, yes. Can you please try it and push an update to OP-TEE/optee_os#7657 if it works well? Thanks!

[Leo-mediatek]

I remembered that why I handle the header in tee-supplicant.

Since the header is written at many location:

• https://github.com/OP-TEE/optee_os/blob/2ff02bfa78ff26550259ddccc4d968f75aa00897/core/kernel/user_ta.c#L328
• https://github.com/OP-TEE/optee_os/blob/2ff02bfa78ff26550259ddccc4d968f75aa00897/ldelf/main.c#L106
• https://github.com/OP-TEE/optee_os/blob/2ff02bfa78ff26550259ddccc4d968f75aa00897/ldelf/ftrace.c#L109-L110

Therefore, handle this in tee-supplicant is easier. But I think it is ok to add an if statement to all of them.

[jforissier]

Therefore, handle this in tee-supplicant is easier. But I think it is ok to add an if statement to all of them.

Yes please fix in optee_os, it makes more sense.

[Leo-mediatek]

Okay, I will try to modify that in optee_os and test if the ftrace output is correct.

[Leo-mediatek]

I tested handling the header in optee_os and it works as expected now.

I think this PR is no longer needed.