Re: [PATCHv2 2/8] module/ftrace: handle patchable-function-entry

[Mark Rutland <mark.rutland@arm.com>]

On Wed, Oct 30, 2019 at 04:03:02PM +0100, Torsten Duwe wrote:
> On Tue, Oct 29, 2019 at 04:58:26PM +0000, Mark Rutland wrote:
> > When using patchable-function-entry, the compiler will record the
> > callsites into a section named "__patchable_function_entries" rather
> > than "__mcount_loc". Let's abstract this difference behind a new
> > FTRACE_CALLSITE_SECTION, so that architectures don't have to handle this
> > explicitly (e.g. with custom module linker scripts).
> > 
> > As parisc currently handles this explicitly, it is fixed up accordingly,
> > with its custom linker script removed. Since FTRACE_CALLSITE_SECTION is
> > only defined when DYNAMIC_FTRACE is selected, the parisc module loading
> > code is updated to only use the definition in that case. When
> > DYNAMIC_FTRACE is not selected, modules shouldn't have this section, so
> > this removes some redundant work in that case.
> > 
> > I built parisc generic-{32,64}bit_defconfig with DYNAMIC_FTRACE enabled,
> > and verified that the section made it into the .ko files for modules.
> 
> This is because of remaining #ifdeffery in include/asm-generic/vmlinux.lds.h:
> 
> #ifdef CC_USING_PATCHABLE_FUNCTION_ENTRY
> #define MCOUNT_REC()    . = ALIGN(8);                           \
>                         __start_mcount_loc = .;                 \
>                         KEEP(*(__patchable_function_entries))   \
>                         __stop_mcount_loc = .;
> #else
> #define MCOUNT_REC()    . = ALIGN(8);                           \
>                         __start_mcount_loc = .;                 \
>                         KEEP(*(__mcount_loc))                   \
>                         __stop_mcount_loc = .;
> #endif

For modules we use a combination of scripts/module-common.lds and an
architecture's own module.lds, not vmlinux.lds.h. So I don't think the above is
relevant for modules.

For modules the kernel's ELF loader looks for the ELF ection, not the
__{start,stop}_mcount_loc symbols that we use for the main kernel image.

FWIW, when building a module, I see the following linker operations:

| [mark@blommer:~/src/linux]% toolchain korg gcc-8.1.0-nolibc make V=1 ARCH=arm64 CROSS_COMPILE=aarch64-linux- arch/arm64/crypto/sha512-ce.ko | grep aarch64-linux-ld
|   aarch64-linux-ld  -EL  -maarch64elf    -r -o arch/arm64/crypto/sha512-ce.o arch/arm64/crypto/sha512-ce-glue.o arch/arm64/crypto/sha512-ce-core.o
|   aarch64-linux-ld -r  -EL  -maarch64elf  --build-id  -T ./scripts/module-common.lds -T ./arch/arm64/kernel/module.lds -o arch/arm64/crypto/sha512-ce.ko arch/arm64/crypto/sha512-ce.o arch/arm64/crypto/sha512-ce.mod.o;  true

> Maybe you want to tackle that as well? I suggest to have at least one
> FTRACE_CALLSITE_SECTION definition without double quotes. Alternatively, my
> earlier solution just kept both sections, in case either one or both are
> present.
> 
>                         KEEP(*(__patchable_function_entries))   \
>                         KEEP(*(__mcount_loc))                   \

I agree that the CC_USING_PATCHABLE_FUNCTION_ENTRY ifdeffery could be
simplified, and that it would be nice to consistently use
FTRACE_CALLSITE_SECTION if we can. However, the generic linker script doesn't
include anything, and I don't see a good location for that to live.

What I could do is add an explicit comment:

/*
 * The ftrace call sites are logged to a section whose name depends on the
 * compiler option used. A given kernel image will only use one, AKA
 * FTRACE_CALLSITE_SECTION. We capture all of them here to avoid header
 * dependencies.
 */
#define MCOUNT_REC()						\
			. = ALIGN(8);				\
			__start_mcount_loc = .;			\
			KEEP(*(__patchable_function_entries))   \
			KEEP(*(__mcount_loc))                   \
			__stop_mcount_loc = .;			

... which should make the dependency clear. Does that sound good to you?

Thanks,
Mark.