Does the altstack attribute require compiler support to be used?

[violin0847]

The alternate stack, altstack, is used for QA_BUILD RM builds. But when I define QA_BUILD and __use_altstack__, 'altstack' attribute directive ignored when compiling, and the compiled kernel modules do not work.

Does the altstack attribute require compiler support to be used?
Have you customized the compiler?

For example:

• macros NV_ENTER_RM_RUNTIME, NV_EXIT_RM_RUNTIME, in src/nvidia/arch/nvalloc/unix/include/osapi.h
• #define NV_API_CALL __attribute__((altstack(0))) in src/common/sdk/nvidia/inc/nv-kernel-interface-api.h

[aritger]

Yes, altstack requires a custom compiler. We don't use altstack in the open-gpu-kernel-module builds and are in general trying to phase out altstack.

And to be clear: I don't believe QA_BUILD RM builds require altstack. The code in osapi.h you mention just gets enabled if both of QA_BUILD and altstack are enabled.