The gcc switch -mprofile-kernel defines a new ABI for calling _mcount()
very early in the function with minimal overhead.

Although mprofile-kernel has been available since GCC 3.4, there were
bugs which were only fixed recently. Currently it is known to work in
GCC 4.9, 5 and 6.

Additionally there are two possible code sequences generated by the
flag, the first uses mflr/std/bl and the second is optimised to omit the
std. Currently only gcc 6 has the optimised sequence. This patch
supports both sequences.

Initial work started by Vojtech Pavlik, used with permission.

Key changes:
 - rework _mcount() to work for both the old and new ABIs.
 - implement new versions of ftrace_caller() and ftrace_graph_caller()
   which deal with the new ABI.
 - updates to __ftrace_make_nop() to recognise the new mcount calling
   sequence.
 - updates to __ftrace_make_call() to recognise the nop'ed sequence.
 - implement ftrace_modify_call().
 - updates to the module loader to surpress the toc save in the module
   stub when calling mcount with the new ABI.
Reviewed-by: NBalbir Singh <bsingharora@gmail.com>
Signed-off-by: NTorsten Duwe <duwe@suse.de>
Signed-off-by: NMichael Ellerman <mpe@ellerman.id.au>

In order to support the new -mprofile-kernel ABI, we need to be able to
call from the module back to ftrace_caller() (in the kernel) without
using the module's r2. That is because the function in this module which
is calling ftrace_caller() may not have setup r2, if it doesn't
otherwise need it (ie. it accesses no globals).

To make that work we add a new stub which is used for calling
ftrace_caller(), which uses the kernel toc instead of the module toc.
Reviewed-by: NBalbir Singh <bsingharora@gmail.com>
Reviewed-by: NTorsten Duwe <duwe@suse.de>
Signed-off-by: NMichael Ellerman <mpe@ellerman.id.au>

When a module is loaded, calls out to the kernel go via a stub which is
generated at runtime. One of these stubs is used to call _mcount(),
which is the default target of tracing calls generated by the compiler
with -pg.

If dynamic ftrace is enabled (which it typically is), another stub is
used to call ftrace_caller(), which is the target of tracing calls when
ftrace is actually active.

ftrace then wants to disable the calls to _mcount() at module startup,
and enable/disable the calls to ftrace_caller() when enabling/disabling
tracing - all of these it does by patching the code.

As part of that code patching, the ftrace code wants to confirm that the
branch it is about to modify, is in fact a call to a module stub which
calls _mcount() or ftrace_caller().

Currently it does that by inspecting the instructions and confirming
they are what it expects. Although that works, the code to do it is
pretty intricate because it requires lots of knowledge about the exact
format of the stub.

We can make that process easier by marking the generated stubs with a
magic value, and then looking for that magic value. Altough this is not
as rigorous as the current method, I believe it is sufficient in
practice.
Reviewed-by: NBalbir Singh <bsingharora@gmail.com>
Reviewed-by: NTorsten Duwe <duwe@suse.de>
Signed-off-by: NMichael Ellerman <mpe@ellerman.id.au>

Currently we generate the module stub for ftrace_caller() at the bottom
of apply_relocate_add(). However apply_relocate_add() is potentially
called more than once per module, which means we will try to generate
the ftrace_caller() stub multiple times.

Although the current code deals with that correctly, ie. it only
generates a stub the first time, it would be clearer to only try to
generate the stub once.

Note also on first reading it may appear that we generate a different
stub for each section that requires relocation, but that is not the
case. The code in stub_for_addr() that searches for an existing stub
uses sechdrs[me->arch.stubs_section], ie. the single stub section for
this module.

A cleaner approach is to only generate the ftrace_caller() stub once,
from module_finalize(). Although the original code didn't check to see
if the stub was actually generated correctly, it seems prudent to add a
check, so do that. And an additional benefit is we can clean the ifdefs
up a little.

Finally we must propagate the const'ness of some of the pointers passed
to module_finalize(), but that is also an improvement.
Reviewed-by: NBalbir Singh <bsingharora@gmail.com>
Reviewed-by: NTorsten Duwe <duwe@suse.de>
Signed-off-by: NMichael Ellerman <mpe@ellerman.id.au>

PowerPC64 uses the symbol .TOC. much as other targets use
_GLOBAL_OFFSET_TABLE_. It identifies the value of the GOT pointer (or in
powerpc parlance, the TOC pointer). Global offset tables are generally
local to an executable or shared library, or in the kernel, module. Thus
it does not make sense for a module to resolve a relocation against
.TOC. to the kernel's .TOC. value. A module has its own .TOC., and
indeed the powerpc64 module relocation processing ignores the kernel
value of .TOC. and instead calculates a module-local value.

This patch removes code involved in exporting the kernel .TOC., tweaks
modpost to ignore an undefined .TOC., and the module loader to twiddle
the section symbol so that .TOC. isn't seen as undefined.

Note that if the kernel was compiled with -msingle-pic-base then ELFv2
would not have function global entry code setting up r2. In that case
the module call stubs would need to be modified to set up r2 using the
kernel .TOC. value, requiring some of this code to be reinstated.

mpe: Furthermore a change in binutils master (not yet released) causes
the current way we handle the TOC to no longer work when building with
MODVERSIONS=y and RELOCATABLE=n. The symptom is that modules can not be
loaded due to there being no version found for TOC.

Cc: stable@vger.kernel.org # 3.16+
Signed-off-by: NAlan Modra <amodra@gmail.com>
Signed-off-by: NMichael Ellerman <mpe@ellerman.id.au>

GCC 6 will include changes to generated code with -mcmodel=large,
which is used to build kernel modules on powerpc64le.  This was
necessary because the large model is supposed to allow arbitrary
sizes and locations of the code and data sections, but the ELFv2
global entry point prolog still made the unconditional assumption
that the TOC associated with any particular function can be found
within 2 GB of the function entry point:

func:
	addis r2,r12,(.TOC.-func)@ha
	addi  r2,r2,(.TOC.-func)@l
	.localentry func, .-func

To remove this assumption, GCC will now generate instead this global
entry point prolog sequence when using -mcmodel=large:

	.quad .TOC.-func
func:
	.reloc ., R_PPC64_ENTRY
	ld    r2, -8(r12)
	add   r2, r2, r12
	.localentry func, .-func

The new .reloc triggers an optimization in the linker that will
replace this new prolog with the original code (see above) if the
linker determines that the distance between .TOC. and func is in
range after all.

Since this new relocation is now present in module object files,
the kernel module loader is required to handle them too.  This
patch adds support for the new relocation and implements the
same optimization done by the GNU linker.

Cc: stable@vger.kernel.org
Signed-off-by: NUlrich Weigand <ulrich.weigand@de.ibm.com>
Signed-off-by: NMichael Ellerman <mpe@ellerman.id.au>

Use pr_fmt to give some context to the error messages in the
module code, and convert open coded debug printk to pr_debug.

Use pr_err for error messages.
Signed-off-by: NAnton Blanchard <anton@samba.org>
Signed-off-by: NMichael Ellerman <mpe@ellerman.id.au>

The commit 71ec7c55 introduced the magic symbol ".TOC." for ELFv2 ABI.
This symbol is built manually and has no CRC value computed. A zero value
is put in the CRC section to avoid modpost complaining about a missing CRC.
Unfortunately, this breaks the kernel module loading when the kernel is
relocated (kdump case for instance) because of the relocation applied to
the kcrctab values.

This patch compute a CRC value for the TOC symbol which will match the one
compute by the kernel when it is relocated - aka '0 - relocate_start' done in
maybe_relocated called by check_version (module.c).
Signed-off-by: NLaurent Dufour <ldufour@linux.vnet.ibm.com>
Cc: Anton Blanchard <anton@samba.org>
Signed-off-by: NBenjamin Herrenschmidt <benh@kernel.crashing.org>

A simple patch which was supposed to swap r12 and r11 also
inexplicably changed the offset by two bytes.  This instruction
(to load r2) isn't used in LE, so it wasn't noticed.

Fixes: b1ce369e ("powerpc: modules: use r12 for stub jump address.)
Reported-by: NAlistair Popple <alistair@popple.id.au>
Signed-off-by: NRusty Russell <rusty@rustcorp.com.au>
Tested-by: NAlistair Popple <alistair@popple.id.au>
Signed-off-by: NBenjamin Herrenschmidt <benh@kernel.crashing.org>

ftrace has way too much knowledge of our kernel module trampoline
layout hidden inside it. Create module_trampoline_target() that gives
the target address of a kernel module trampoline.
Signed-off-by: NAnton Blanchard <anton@samba.org>

ftrace has way too much knowledge of our kernel module trampoline
layout hidden inside it. Create is_module_trampoline() that can
abstract this away inside the module loader code.
Signed-off-by: NAnton Blanchard <anton@samba.org>

ELFv2 doesn't use function descriptors, because it doesn't need to
load a new r2 when calling into a function.  On the other hand, you're
supposed to use a local entry point for R_PPC_REL24 branches.
Signed-off-by: NRusty Russell <rusty@rustcorp.com.au>

ELFv2 doesn't need to set up r2 when calling a function.
Signed-off-by: NRusty Russell <rusty@rustcorp.com.au>

In ELFv2, r12 is supposed to equal to PC on entry to a function.
Our stubs use r11, so change swap that with r12.
Signed-off-by: NRusty Russell <rusty@rustcorp.com.au>

ELFv2 uses a different stack offset (24 vs 40) to save r2.
Signed-off-by: NRusty Russell <rusty@rustcorp.com.au>

ELFv2 doesn't use function descriptors, so we don't expect symbols to
start with ".".  But because depmod and modpost strip ".", and we have
the special symbol ".TOC.", we still need to do it.
Signed-off-by: NRusty Russell <rusty@rustcorp.com.au>

The new ELF ABI tends to use R_PPC64_REL16_LO and R_PPC64_REL16_HA
relocations (PC-relative), so implement them.
Signed-off-by: NRusty Russell <rusty@rustcorp.com.au>

The kernel resolved the '.TOC.' to a fake symbol, so we need to fix it up
to point to our .toc section plus 0x8000.
Signed-off-by: NRusty Russell <rusty@rustcorp.com.au>

Signed-off-by: NRusty Russell <rusty@rustcorp.com.au>

By representing them as words, rather than chars, we can avoid
endian ifdefs.
Signed-off-by: NRusty Russell <rusty@rustcorp.com.au>

Move the few declarations from arch/powerpc/kernel/setup.h
into arch/powerpc/include/asm/setup.h.  This resolves a
sparse warning for arch/powerpc/mm/numa.c which defines
do_init_bootmem() but can't include the setup.h header
in the prior path.

Resolves:
arch/powerpc/mm/numa.c:998:13:
        warning: symbol 'do_init_bootmem' was not declared.
                 Should it be static?
Signed-off-by: NRobert C Jennings <rcj@linux.vnet.ibm.com>
Signed-off-by: NBenjamin Herrenschmidt <benh@kernel.crashing.org>

Signed-off-by: NEugene Surovegin <surovegin@google.com>
Signed-off-by: NBenjamin Herrenschmidt <benh@kernel.crashing.org>

Finally remove the two level TOC and build with -mcmodel=medium.

Unfortunately we can't build modules with -mcmodel=medium due to
the tricks the kernel module loader plays with percpu data:

# -mcmodel=medium breaks modules because it uses 32bit offsets from
# the TOC pointer to create pointers where possible. Pointers into the
# percpu data area are created by this method.
#
# The kernel module loader relocates the percpu data section from the
# original location (starting with 0xd...) to somewhere in the base
# kernel percpu data space (starting with 0xc...). We need a full
# 64bit relocation for this to work, hence -mcmodel=large.

On older kernels we fall back to the two level TOC (-mminimal-toc)
Signed-off-by: NAnton Blanchard <anton@samba.org>
Signed-off-by: NBenjamin Herrenschmidt <benh@kernel.crashing.org>

This patch removes all the module loader hook implementations in the
architecture specific code where the functionality is the same as that
now provided by the recently added default hooks.
Signed-off-by: NJonas Bonn <jonas@southpole.se>
Acked-by: NMike Frysinger <vapier@gentoo.org>
Acked-by: NGeert Uytterhoeven <geert@linux-m68k.org>
Tested-by: NMichal Simek <monstr@monstr.eu>
Signed-off-by: NRusty Russell <rusty@rustcorp.com.au>

Create a new header that becomes a single location for defining PowerPC
opcodes used by code that is either generationg instructions
at runtime (fixups, debug, etc.), emulating instructions, or just
compiling instructions old assemblers don't know about.

We currently don't handle the floating point emulation or alignment decode
as both are better handled by the specific decode support they already
have.

Added support for the new dcbzl, dcbal, msgsnd, tlbilx, & wait instructions
since older assemblers don't know about them.
Signed-off-by: NKumar Gala <galak@kernel.crashing.org>
Signed-off-by: NBenjamin Herrenschmidt <benh@kernel.crashing.org>

Impact: Allow 64 bit PowerPC to trace modules with dynamic ftrace

This adds code to handle the PPC64 module trampolines, and allows for
PPC64 to use dynamic ftrace.

Thanks to Paul Mackerras for these updates:

  - fix the mod and rec->arch.mod NULL checks.
  - fix to is_bl_op compare.

Thanks to Milton Miller for:

  - finding the nasty race with using two nops, and recommending
    instead that I use a branch 8 forward.
Signed-off-by: NSteven Rostedt <srostedt@redhat.com>

Commit deac93df ("lib: Correct printk
%pF to work on all architectures") broke the non modular builds by
moving an essential function into modules.c.  Fix this by moving it
out again and into asm/sections.h as an inline.  To do this, the
definition of struct ppc64_opd_entry has been lifted out of modules.c
and put in asm/elf.h where it belongs.
Signed-off-by: NJames Bottomley <James.Bottomley@HansenPartnership.com>
Signed-off-by: NPaul Mackerras <paulus@samba.org>

It was introduced by "vsprintf: add support for '%pS' and '%pF' pointer
formats" in commit 0fe1ef24.  However,
the current way its coded doesn't work on parisc64.  For two reasons: 1)
parisc isn't in the #ifdef and 2) parisc has a different format for
function descriptors

Make dereference_function_descriptor() more accommodating by allowing
architecture overrides.  I put the three overrides (for parisc64, ppc64
and ia64) in arch/kernel/module.c because that's where the kernel
internal linker which knows how to deal with function descriptors sits.
Signed-off-by: NJames Bottomley <James.Bottomley@HansenPartnership.com>
Acked-by: NBenjamin Herrenschmidt <benh@kernel.crashing.org>
Acked-by: NTony Luck <tony.luck@intel.com>
Acked-by: NKyle McMartin <kyle@mcmartin.ca>
Signed-off-by: NLinus Torvalds <torvalds@linux-foundation.org>

A bunch of code has hard-coded the value for a "nop" instruction, it
would be nice to have a #define for it.
Signed-off-by: NMichael Ellerman <michael@ellerman.id.au>
Acked-by: NKumar Gala <galak@kernel.crashing.org>
Signed-off-by: NPaul Mackerras <paulus@samba.org>

Refactor common code between ppc32 and ppc64 module handling into a
shared filed.
Signed-off-by: NKumar Gala <galak@kernel.crashing.org>
Signed-off-by: NPaul Mackerras <paulus@samba.org>

When a module has relocation sections with tens of thousands of
entries, counting the distinct/unique entries only (i.e. no
duplicates) at load time can take tens of seconds and up to minutes.
The sore point is the count_relocs() function which is called as part
of the architecture specific module loading processing path:

	-> load_module()			generic
	   -> module_frob_arch_sections()	arch specific
	      -> get_plt_size()		32-bit
	      -> get_stubs_size()	64-bit
		 -> count_relocs()

Here count_relocs is being called to find out how many distinct
targets of R_PPC_REL24 relocations there are, since each distinct
target needs a PLT entry or a stub created for it.

The previous counting algorithm has O(n^2) complexity.  Basically two
solutions were proposed on the e-mail list: a hash based approach and
a sort based approach.

The hash based approach is the fastest (O(n)) but the has it needs
additional memory and for certain corner cases it could take lots of
memory due to the degeneration of the hash.  One such proposal was
submitted here:

http://ozlabs.org/pipermail/linuxppc-dev/2007-June/037641.html

The sort based approach is slower (O(n * log n + n)) but if the
sorting is done "in place" it doesn't need additional memory.
This has O(n + n * log n) complexity with no additional memory
requirements.

This commit implements the in-place sort option.
Signed-off-by: NEmil Medve <Emilian.Medve@Freescale.com>
Signed-off-by: NPaul Mackerras <paulus@samba.org>

This makes powerpc use the generic BUG machinery.  The biggest reports the
function name, since it is redundant with kallsyms, and not needed in general.

There is an overall reduction of code, since module_32/64 duplicated several
functions.

Unfortunately there's no way to tell gcc that BUG won't return, so the BUG
macro includes a goto loop.  This will generate a real jmp instruction, which
is never used.

[akpm@osdl.org: build fix]
[paulus@samba.org: remove infinite loop in BUG_ON]
Signed-off-by: NJeremy Fitzhardinge <jeremy@goop.org>
Cc: Andi Kleen <ak@muc.de>
Cc: Hugh Dickens <hugh@veritas.com>
Cc: Michael Ellerman <michael@ellerman.id.au>
Cc: Benjamin Herrenschmidt <benh@kernel.crashing.org>
Cc: Rusty Russell <rusty@rustcorp.com.au>
Signed-off-by: NAndrew Morton <akpm@osdl.org>
Signed-off-by: NPaul Mackerras <paulus@samba.org>

This patch adds support for feature fixups in modules. This involves
adding support for R_PPC64_REL64 relocs to the 64 bits module loader.
It also modifies modpost.c to ignore the powerpc fixup sections (or it
would warn when used in .init.text).
Signed-off-by: NBenjamin Herrenschmidt <benh@kernel.crashing.org>
Acked-by: NOlof Johansson <olof@lixom.net>
Signed-off-by: NPaul Mackerras <paulus@samba.org>

Normally, ppc64 module .ko files contain a table-of-contents (.toc)
section, but if the module doesn't reference any static or external
data or external procedures, it is possible for gcc/binutils to
generate a .ko that doesn't have a .toc.  Currently the module
loader refuses to load such a module, since it needs the address
of the .toc section to use in relocations.

This patch fixes the problem by using the address of the .stubs
section instead, which is an acceptable substitute in this situation.
Signed-off-by: NPaul Mackerras <paulus@samba.org>

Also deletes files in arch/ppc64 that are no longer used now that
we don't compile with ARCH=ppc64 any more.
Signed-off-by: NPaul Mackerras <paulus@samba.org>

Newer gcc's are generating this relocation, so the module loader needs to
handle it.
Signed-off-by: NPeter Bergner <bergner@vnet.ibm.com>
Signed-off-by: NAnton Blanchard <anton@samba.org>
Signed-off-by: NAndrew Morton <akpm@osdl.org>
Signed-off-by: NLinus Torvalds <torvalds@osdl.org>

Initial git repository build. I'm not bothering with the full history,
even though we have it. We can create a separate "historical" git
archive of that later if we want to, and in the meantime it's about
3.2GB when imported into git - space that would just make the early
git days unnecessarily complicated, when we don't have a lot of good
infrastructure for it.

Let it rip!