lkml.org 
[lkml]   [2017]   [Nov]   [15]   [last100]   RSS Feed
Views: [wrap][no wrap]   [headers]  [forward] 
 
Messages in this thread
/
Date
From
SubjectRe: [PATCH v4 2/3] powerpc/modules: Don't try to restore r2 after a sibling call
Josh Poimboeuf wrote:
> On Tue, Nov 14, 2017 at 03:59:21PM +0530, Naveen N. Rao wrote:
>> Kamalesh Babulal wrote:
>> > From: Josh Poimboeuf <jpoimboe@redhat.com>
>> >
>> > When attempting to load a livepatch module, I got the following error:
>> >
>> > module_64: patch_module: Expect noop after relocate, got 3c820000
>> >
>> > The error was triggered by the following code in
>> > unregister_netdevice_queue():
>> >
>> > 14c: 00 00 00 48 b 14c <unregister_netdevice_queue+0x14c>
>> > 14c: R_PPC64_REL24 net_set_todo
>> > 150: 00 00 82 3c addis r4,r2,0
>> >
>> > GCC didn't insert a nop after the branch to net_set_todo() because it's
>> > a sibling call, so it never returns. The nop isn't needed after the
>> > branch in that case.
>> >
>> > Signed-off-by: Josh Poimboeuf <jpoimboe@redhat.com>
>> > Signed-off-by: Kamalesh Babulal <kamalesh@linux.vnet.ibm.com>
>> > ---
>> > arch/powerpc/kernel/module_64.c | 4 ++++
>> > 1 file changed, 4 insertions(+)
>> >
>> > diff --git a/arch/powerpc/kernel/module_64.c b/arch/powerpc/kernel/module_64.c
>> > index 39b01fd..9e5391f 100644
>> > --- a/arch/powerpc/kernel/module_64.c
>> > +++ b/arch/powerpc/kernel/module_64.c
>> > @@ -489,6 +489,10 @@ static int restore_r2(u32 *instruction, struct module *me)
>> > if (is_early_mcount_callsite(instruction - 1))
>> > return 1;
>> >
>> > + /* Sibling calls don't return, so they don't need to restore r2 */
>> > + if (instruction[-1] == PPC_INST_BRANCH)
>> > + return 1;
>> > +
>>
>> This looks quite fragile, unless we know for sure that gcc will _always_
>> emit this instruction form for sibling calls with relocations.
>>
>> As an alternative, does it make sense to do the following check instead?
>> if ((instr_is_branch_iform(insn) || instr_is_branch_bform(insn))
>> && !(insn & 0x1))
>
> Yes, good point. How about something like this?

That looks good to me (with a very minor nit below).
Acked-by: Naveen N. Rao <naveen.n.rao@linux.vnet.ibm.com>

>
> (completely untested because I don't have access to a box at the moment)
>
>
> diff --git a/arch/powerpc/include/asm/code-patching.h b/arch/powerpc/include/asm/code-patching.h
> index abef812de7f8..302e4368debc 100644
> --- a/arch/powerpc/include/asm/code-patching.h
> +++ b/arch/powerpc/include/asm/code-patching.h
> @@ -33,6 +33,7 @@ int patch_branch(unsigned int *addr, unsigned long target, int flags);
> int patch_instruction(unsigned int *addr, unsigned int instr);
>
> int instr_is_relative_branch(unsigned int instr);
> +int instr_is_link_branch(unsigned int instr);
> int instr_is_branch_to_addr(const unsigned int *instr, unsigned long addr);
> unsigned long branch_target(const unsigned int *instr);
> unsigned int translate_branch(const unsigned int *dest,
> diff --git a/arch/powerpc/kernel/module_64.c b/arch/powerpc/kernel/module_64.c
> index 9cb007bc7075..b5148a206b4d 100644
> --- a/arch/powerpc/kernel/module_64.c
> +++ b/arch/powerpc/kernel/module_64.c
> @@ -487,11 +487,13 @@ static bool is_early_mcount_callsite(u32 *instruction)
> restore r2. */
> static int restore_r2(u32 *instruction, struct module *me)
> {
> - if (is_early_mcount_callsite(instruction - 1))
> + u32 *prev_insn = instruction - 1;
> +
> + if (is_early_mcount_callsite(prev_insn))
> return 1;
>
> /* Sibling calls don't return, so they don't need to restore r2 */
> - if (instruction[-1] == PPC_INST_BRANCH)
> + if (!instr_is_link_branch(*prev_insn))
> return 1;
>
> if (*instruction != PPC_INST_NOP) {
> diff --git a/arch/powerpc/lib/code-patching.c b/arch/powerpc/lib/code-patching.c
> index c9de03e0c1f1..4727fafd37e4 100644
> --- a/arch/powerpc/lib/code-patching.c
> +++ b/arch/powerpc/lib/code-patching.c
> @@ -304,6 +304,12 @@ int instr_is_relative_branch(unsigned int instr)
> return instr_is_branch_iform(instr) || instr_is_branch_bform(instr);
> }
>
> +int instr_is_link_branch(unsigned int instr)
> +{
> + return (instr_is_branch_iform(instr) || instr_is_branch_bform(instr)) &&
> + (instr & BRANCH_SET_LINK);
> +}
> +

Nitpicking here, but since we're not considering the other branch forms,
perhaps this can be renamed to instr_is_link_relative_branch() (or maybe
instr_is_relative_branch_link()), just so we're clear :)


- Naveen

> static unsigned long branch_iform_target(const unsigned int *instr)
> {
> signed long imm;
> --
> To unsubscribe from this list: send the line "unsubscribe live-patching" in
> the body of a message to majordomo@vger.kernel.org
> More majordomo info at http://vger.kernel.org/majordomo-info.html
>
>

\
 
 \ /
  Last update: 2017-11-15 10:29    [W:0.113 / U:3.108 seconds]
©2003-2020 Jasper Spaans|hosted at Digital Ocean and TransIP|Read the blog|Advertise on this site