Re: [PATCH v1] arch/x86: port I/O tracing on x86

[Peter Zijlstra <peterz@infradead.org>]

On Mon, Sep 18, 2023 at 11:55:10AM -0600, Dan Raymond wrote:
> Add support for port I/O tracing on x86.  Memory mapped I/O tracing is
> available on x86 via CONFIG_MMIOTRACE but that relies on page faults
> so it doesn't work with port I/O.  This feature uses tracepoints in a
> similar manner as CONFIG_TRACE_MMIO_ACCESS.
> 
> Signed-off-by: Dan Raymond <draymond@foxvalley.net>
> Suggested-by: Andy Shevchenko <andriy.shevchenko@linux.intel.com>
> ---
>  arch/x86/include/asm/shared/io.h | 11 +++++++
>  arch/x86/lib/Makefile            |  1 +
>  arch/x86/lib/trace_portio.c      | 18 ++++++++++++
>  include/trace/events/portio.h    | 49 ++++++++++++++++++++++++++++++++
>  4 files changed, 79 insertions(+)
>  create mode 100644 arch/x86/lib/trace_portio.c
>  create mode 100644 include/trace/events/portio.h
> 
> diff --git a/arch/x86/include/asm/shared/io.h b/arch/x86/include/asm/shared/io.h
> index c0ef921c0586..e7ef4212e00b 100644
> --- a/arch/x86/include/asm/shared/io.h
> +++ b/arch/x86/include/asm/shared/io.h
> @@ -2,13 +2,23 @@
>  #ifndef _ASM_X86_SHARED_IO_H
>  #define _ASM_X86_SHARED_IO_H
>  
> +#include <linux/instruction_pointer.h>
>  #include <linux/types.h>
>  
> +#if defined(CONFIG_TRACEPOINTS) && !defined(BOOT_COMPRESSED_MISC_H) && !defined(BOOT_BOOT_H)
> +extern void do_trace_portio_read(u32 value, u16 port, char width, long ip_addr);
> +extern void do_trace_portio_write(u32 value, u16 port, char width, long ip_addr);
> +#else
> +static inline void do_trace_portio_read(u32 value, u16 port, char width, long ip_addr) {}
> +static inline void do_trace_portio_write(u32 value, u16 port, char width, long ip_addr) {}
> +#endif
> +
>  #define BUILDIO(bwl, bw, type)						\
>  static inline void __out##bwl(type value, u16 port)			\
>  {									\
>  	asm volatile("out" #bwl " %" #bw "0, %w1"			\
>  		     : : "a"(value), "Nd"(port));			\
> +	do_trace_portio_write(value, port, #bwl[0], _THIS_IP_);		\
>  }									\
>  									\
>  static inline type __in##bwl(u16 port)					\
> @@ -16,6 +26,7 @@ static inline type __in##bwl(u16 port)					\
>  	type value;							\
>  	asm volatile("in" #bwl " %w1, %" #bw "0"			\
>  		     : "=a"(value) : "Nd"(port));			\
> +	do_trace_portio_read(value, port, #bwl[0], _THIS_IP_);		\
>  	return value;							\
>  }

No, very much no.

This sticks tracing in the very rawest of raw output paths. This means I
can no longer use early_console->write() to print to my
early_serial_console.

That is the one and only fully reliably output path we have. You're not
sticking tracing in it.