Re: linux-next: Tree for May 18 (mm/memory-failure.c)

On Tue, May 19, 2015 at 09:46:36AM -0400, Steven Rostedt wrote:
...

quoted

diff --git a/include/ras/ras_event.h b/include/ras/ras_event.h
index 1443d79e4fe6..43054c0fcf65 100644
--- a/include/ras/ras_event.h
+++ b/include/ras/ras_event.h

@@ -1,6 +1,8 @@
 #undef TRACE_SYSTEM
 #define TRACE_SYSTEM ras
 #define TRACE_INCLUDE_FILE ras_event
+#undef TRACE_INCLUDE_PATH
+#define TRACE_INCLUDE_PATH ../../include/ras

Note, ideally, you want:

#define TRACE_INCLUDE_PATH .

OK, so we had better move include/ras/ras_event.h under include/trace/events.
I'll do this in a separate work.

and change the Makefile to have:

CFLAGS_ras.o := -I$(src)

It seems that if we do both of these, I hit the following error:

    CC      drivers/ras/ras.o
  In file included from include/trace/events/ras_event.h:327,
                   from drivers/ras/ras.c:13:
  include/trace/define_trace.h:83:43: error: ./ras_event.h: No such file or directory

, so I guess it's enough to do either.

...

quoted

 
 #if !defined(_TRACE_HW_EVENT_MC_H) || defined(TRACE_HEADER_MULTI_READ)
 #define _TRACE_HW_EVENT_MC_H

diff --git a/mm/memory-failure.c b/mm/memory-failure.c
index 8cbe23ac1056..e88e14d87571 100644
--- a/mm/memory-failure.c
+++ b/mm/memory-failure.c

@@ -57,6 +57,7 @@
 #include <linux/mm_inline.h>
 #include <linux/kfifo.h>
 #include "internal.h"
+#define CREATE_TRACE_POINTS
 #include "ras/ras_event.h"

Um, you can only define CREATE_TRACE_POINTS for a single instance.
Otherwise you will be making duplicate functions with the same name and
same variables.

That is, you must either pick CREATE_TRACE_POINTS for ras_event.h in
mm/memory-failure.c or drivers/ras/ras.c. Not both.

OK, so it seems that the root cause of the original error is a wrong
dependency among CONFIGs.
CONFIG_RAS should depend on CONFIG_MEMORY_FAILURE, but that is not true
(CONFIG_RAS=n and CONFIG_MEMORY_FAILURE=y in Randy's .config.)
This problem is visible when CONFIG_X86_64=n and CONFIG_SPARSEMEM=n (and all
other dependencies of CONFIG_RAS from ACPI_EXTLOG/PCIEAER/EDAC are false).
I'll fix such dependency.

Thanks,
Naoya Horiguchi