Build Framework
===============
The perf build framework was adopted from the kernel build system, hence the
idea and the way how objects are built is the same.
Basically the user provides set of 'Build' files that list objects and
directories to nest for specific target to be build.
Unlike the kernel we don't have a single build object 'obj-y' list that where
we setup source objects, but we support more. This allows one 'Build' file to
carry a sources list for multiple build objects.
Build framework makefiles
-------------------------
The build framework consists of 2 Makefiles:
Build.include
Makefile.build
While the 'Build.include' file contains just some generic definitions, the
'Makefile.build' file is the makefile used from the outside. It's
interface/usage is following:
$ make -f tools/build/Makefile.build srctree=$(KSRC) dir=$(DIR) obj=$(OBJECT)
where:
KSRC - is the path to kernel sources
DIR - is the path to the project to be built
OBJECT - is the name of the build object
When succefully finished the $(DIR) directory contains the final object file
called $(OBJECT)-in.o:
$ ls $(DIR)/$(OBJECT)-in.o
which includes all compiled sources described in 'Build' makefiles.
Build makefiles
---------------
The user supplies 'Build' makefiles that contains a objects list, and connects
the build to nested directories.
Assume we have the following project structure:
ex/a.c
/b.c
/c.c
/d.c
/arch/e.c
/arch/f.c
Out of which you build the 'ex' binary ' and the 'libex.a' library:
'ex' - consists of 'a.o', 'b.o' and libex.a
'libex.a' - consists of 'c.o', 'd.o', 'e.o' and 'f.o'
The build framework does not create the 'ex' and 'libex.a' binaries for you, it
only prepares proper objects to be compiled and grouped together.
To follow the above example, the user provides following 'Build' files:
ex/Build:
ex-y += a.o
ex-y += b.o
ex-y += b.o # duplicates in the lists are allowed
libex-y += c.o
libex-y += d.o
libex-y += arch/
ex/arch/Build:
libex-y += e.o
libex-y += f.o
and runs:
$ make -f tools/build/Makefile.build dir=. obj=ex
$ make -f tools/build/Makefile.build dir=. obj=libex
which creates the following objects:
ex/ex-in.o
ex/libex-in.o
that contain request objects names in Build files.
It's only a matter of 2 single commands to create the final binaries:
$ ar rcs libex.a libex-in.o
$ gcc -o ex ex-in.o libex.a
You can check the 'ex' example in 'tools/build/tests/ex' for more details.
Makefile.include
----------------
The tools/build/Makefile.include makefile could be included
via user makefiles to get usefull definitions.
It defines following interface:
- build macro definition:
build := -f $(srctree)/tools/build/Makefile.build dir=. obj
to make it easier to invoke build like:
make $(build)=ex
Fixdep
------
It is necessary to build the fixdep helper before invoking the build.
The Makefile.include file adds the fixdep target, that could be
invoked by the user.
Rules
-----
The build framework provides standard compilation rules to handle .S and .c
compilation.
It's possible to include special rule if needed (like we do for flex or bison
code generation).
CFLAGS
------
It's possible to alter the standard object C flags in the following way:
CFLAGS_perf.o += '...' - adds CFLAGS for perf.o object
CFLAGS_gtk += '...' - adds CFLAGS for gtk build object
CFLAGS_REMOVE_perf.o += '...' - removes CFLAGS for perf.o object
CFLAGS_REMOVE_gtk += '...' - removes CFLAGS for gtk build object
This C flags changes has the scope of the Build makefile they are defined in.
Dependencies
------------
For each built object file 'a.o' the '.a.cmd' is created and holds:
- Command line used to built that object
(for each object)
- Dependency rules generated by 'gcc -Wp,-MD,...'
(for compiled object)
All existing '.cmd' files are included in the Build process to follow properly
the dependencies and trigger a rebuild when necessary.
Single rules
------------
It's possible to build single object file by choice, like:
$ make util/map.o # objects
$ make util/map.i # preprocessor
$ make util/map.s # assembly
lected'>unified
powerpc/mm: Fix spurrious segfaults on radix with autonuma
When autonuma (Automatic NUMA balancing) marks a PTE inaccessible it
clears all the protection bits but leave the PTE valid.
With the Radix MMU, an attempt at executing from such a PTE will
take a fault with bit 35 of SRR1 set "SRR1_ISI_N_OR_G".
It is thus incorrect to treat all such faults as errors. We should
pass them to handle_mm_fault() for autonuma to deal with. The case
of pages that are really not executable is handled by the existing
test for VM_EXEC further down.
That leaves us with catching the kernel attempts at executing user
pages. We can catch that earlier, even before we do find_vma.
It is never valid on powerpc for the kernel to take an exec fault
to begin with. So fold that test with the existing test for the
kernel faulting on kernel addresses to bail out early.
Fixes: 1d18ad026844 ("powerpc/mm: Detect instruction fetch denied and report")
Signed-off-by: Benjamin Herrenschmidt <benh@kernel.crashing.org>
Reviewed-by: Aneesh Kumar K.V <aneesh.kumar@linux.vnet.ibm.com>
Acked-by: Balbir Singh <bsingharora@gmail.com>
Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>