Josh, my current git version (with gcc 5.3.1) makes objtool warn about "duplicate frame pointer save" in drivers/gpu/drm/vmwgfx/vmwgfx_msg.c for both vmw_send_msg() and vmw_host_get_guestinfo().
The reason is that VMW_PORT_HB_OUT() uses a magic instruction sequence (a "rep outsb") to communicate with the hypervisor (it's a virtual GPU driver for vmware), and %rbp is part of the communication. So the inline asm does a save-and-restore of the frame pointer around the instruction sequence.
I actually find the objtool warning to be quite reasonable, so it's not exactly a false positive, since in this case it actually does point out that the frame pointer won't be reliable over that instruction sequence.
But in this particular case it just ends up being the wrong thing - the code is what it is, and %rbp just can't have the frame information due to annoying magic calling conventions.
Is there some way to override objtool for this situation? I hate seeing warnings that I need to ignore, it has just too often caused me to mistakenly ignore warnings I *should* have reacted to.
I guess a STACK_FRAME_NON_STANDARD will shut objtool up (or just disabling it entirely for the whole file), but I was wondering about something more targeted that could be marked in the inline asm itself (rather than have to mark the functions that use it)
Linus
On Wed, May 25, 2016 at 10:14:24AM -0700, Linus Torvalds wrote:
Josh, my current git version (with gcc 5.3.1) makes objtool warn about "duplicate frame pointer save" in drivers/gpu/drm/vmwgfx/vmwgfx_msg.c for both vmw_send_msg() and vmw_host_get_guestinfo().
The reason is that VMW_PORT_HB_OUT() uses a magic instruction sequence (a "rep outsb") to communicate with the hypervisor (it's a virtual GPU driver for vmware), and %rbp is part of the communication. So the inline asm does a save-and-restore of the frame pointer around the instruction sequence.
I actually find the objtool warning to be quite reasonable, so it's not exactly a false positive, since in this case it actually does point out that the frame pointer won't be reliable over that instruction sequence.
But in this particular case it just ends up being the wrong thing - the code is what it is, and %rbp just can't have the frame information due to annoying magic calling conventions.
Is there some way to override objtool for this situation? I hate seeing warnings that I need to ignore, it has just too often caused me to mistakenly ignore warnings I *should* have reacted to.
I guess a STACK_FRAME_NON_STANDARD will shut objtool up (or just disabling it entirely for the whole file), but I was wondering about something more targeted that could be marked in the inline asm itself (rather than have to mark the functions that use it)
I used to have a STACKTOOL_IGNORE_INSN macro that would tell the tool to skip warnings for specific instructions in inline asm:
Here's an example of it how it was used:
https://lkml.kernel.org/r/c0c1a92df921961cae5cce208247bb8d8a975d6d.145044227...
And here's the implementation of the macro:
https://lkml.kernel.org/r/cd7778181d2a5251c3dc21811fdbcaa2c16c98c3.145044227...
As it turns out, we didn't need the macro, so I removed it. But I can easily add something like that again to get rid of the warnings.
There were objections to having "OBJTOOL" in the names of macros, so I guess I'll call it STACK_INSN_NON_STANDARD, unless anybody has a better idea.
On Wed, May 25, 2016 at 10:56 AM, Josh Poimboeuf jpoimboe@redhat.com wrote:
I used to have a STACKTOOL_IGNORE_INSN macro that would tell the tool to skip warnings for specific instructions in inline asm:
Here's an example of it how it was used:
Ok, looking at that, I'm starting to suspect that it is simpler to just use STACK_FRAME_NON_STANDARD and mark the two functions that use this particular inline asm with the odd %rbp problem.
It's a rather special case, after all.
Linus
On Wed, May 25, 2016 at 04:51:21PM -0700, Linus Torvalds wrote:
On Wed, May 25, 2016 at 10:56 AM, Josh Poimboeuf jpoimboe@redhat.com wrote:
I used to have a STACKTOOL_IGNORE_INSN macro that would tell the tool to skip warnings for specific instructions in inline asm:
Here's an example of it how it was used:
Ok, looking at that, I'm starting to suspect that it is simpler to just use STACK_FRAME_NON_STANDARD and mark the two functions that use this particular inline asm with the odd %rbp problem.
It's a rather special case, after all.
That's fine with me, it is indeed a rare case. We can always add the per-instruction macro later if needed. Here's a patch.
---
From: Josh Poimboeuf jpoimboe@redhat.com Subject: [PATCH] drm/vmwgfx: fix "duplicate frame pointer save" warning
objtool reports the following warnings:
drivers/gpu/drm/vmwgfx/vmwgfx_msg.o: warning: objtool: vmw_send_msg()+0x107: duplicate frame pointer save drivers/gpu/drm/vmwgfx/vmwgfx_msg.o: warning: objtool: vmw_host_get_guestinfo()+0x252: duplicate frame pointer save
To quote Linus:
"The reason is that VMW_PORT_HB_OUT() uses a magic instruction sequence (a "rep outsb") to communicate with the hypervisor (it's a virtual GPU driver for vmware), and %rbp is part of the communication. So the inline asm does a save-and-restore of the frame pointer around the instruction sequence.
I actually find the objtool warning to be quite reasonable, so it's not exactly a false positive, since in this case it actually does point out that the frame pointer won't be reliable over that instruction sequence.
But in this particular case it just ends up being the wrong thing - the code is what it is, and %rbp just can't have the frame information due to annoying magic calling conventions."
Silence the warnings by telling objtool to ignore the two functions which use the VMW_PORT_HB_{IN,OUT} macros.
Reported-by: Linus Torvalds torvalds@linux-foundation.org Signed-off-by: Josh Poimboeuf jpoimboe@redhat.com --- drivers/gpu/drm/vmwgfx/vmwgfx_msg.c | 4 +++- 1 file changed, 3 insertions(+), 1 deletion(-)
diff --git a/drivers/gpu/drm/vmwgfx/vmwgfx_msg.c b/drivers/gpu/drm/vmwgfx/vmwgfx_msg.c index 6de283c..f0374f9 100644 --- a/drivers/gpu/drm/vmwgfx/vmwgfx_msg.c +++ b/drivers/gpu/drm/vmwgfx/vmwgfx_msg.c @@ -28,6 +28,7 @@ #include <linux/slab.h> #include <linux/module.h> #include <linux/kernel.h> +#include <linux/frame.h> #include <asm/hypervisor.h> #include "drmP.h" #include "vmwgfx_msg.h" @@ -194,7 +195,7 @@ static int vmw_send_msg(struct rpc_channel *channel, const char *msg)
return -EINVAL; } - +STACK_FRAME_NON_STANDARD(vmw_send_msg);
/** @@ -304,6 +305,7 @@ static int vmw_recv_msg(struct rpc_channel *channel, void **msg,
return 0; } +STACK_FRAME_NON_STANDARD(vmw_recv_msg);
/**
On Thu, May 26, 2016 at 11:43 AM, Josh Poimboeuf jpoimboe@redhat.com wrote:
That's fine with me, it is indeed a rare case. We can always add the per-instruction macro later if needed. Here's a patch.
Ingo, I can take this directly, unless you have other things pending that you want to send anyway and would just add this to that existing pile.
Just let me know.
Linus
Commit-ID: 0b0d81e3b7334897da9b2e3ffee860c2046f7bc0 Gitweb: http://git.kernel.org/tip/0b0d81e3b7334897da9b2e3ffee860c2046f7bc0 Author: Josh Poimboeuf jpoimboe@redhat.com AuthorDate: Thu, 26 May 2016 13:43:43 -0500 Committer: Ingo Molnar mingo@kernel.org CommitDate: Wed, 8 Jun 2016 15:36:18 +0200
objtool, drm/vmwgfx: Fix "duplicate frame pointer save" warning
objtool reports the following warnings:
drivers/gpu/drm/vmwgfx/vmwgfx_msg.o: warning: objtool: vmw_send_msg()+0x107: duplicate frame pointer save drivers/gpu/drm/vmwgfx/vmwgfx_msg.o: warning: objtool: vmw_host_get_guestinfo()+0x252: duplicate frame pointer save
To quote Linus:
"The reason is that VMW_PORT_HB_OUT() uses a magic instruction sequence (a "rep outsb") to communicate with the hypervisor (it's a virtual GPU driver for vmware), and %rbp is part of the communication. So the inline asm does a save-and-restore of the frame pointer around the instruction sequence.
I actually find the objtool warning to be quite reasonable, so it's not exactly a false positive, since in this case it actually does point out that the frame pointer won't be reliable over that instruction sequence.
But in this particular case it just ends up being the wrong thing - the code is what it is, and %rbp just can't have the frame information due to annoying magic calling conventions."
Silence the warnings by telling objtool to ignore the two functions which use the VMW_PORT_HB_{IN,OUT} macros.
Reported-by: Linus Torvalds torvalds@linux-foundation.org Signed-off-by: Josh Poimboeuf jpoimboe@redhat.com Acked-by: Linus Torvalds torvalds@linux-foundation.org Cc: DRI dri-devel@lists.freedesktop.org Cc: Peter Zijlstra peterz@infradead.org Cc: Thomas Gleixner tglx@linutronix.de Link: http://lkml.kernel.org/r/20160526184343.fdtjjjg67smmeekt@treble Signed-off-by: Ingo Molnar mingo@kernel.org --- drivers/gpu/drm/vmwgfx/vmwgfx_msg.c | 4 +++- 1 file changed, 3 insertions(+), 1 deletion(-)
diff --git a/drivers/gpu/drm/vmwgfx/vmwgfx_msg.c b/drivers/gpu/drm/vmwgfx/vmwgfx_msg.c index 6de283c..f0374f9 100644 --- a/drivers/gpu/drm/vmwgfx/vmwgfx_msg.c +++ b/drivers/gpu/drm/vmwgfx/vmwgfx_msg.c @@ -28,6 +28,7 @@ #include <linux/slab.h> #include <linux/module.h> #include <linux/kernel.h> +#include <linux/frame.h> #include <asm/hypervisor.h> #include "drmP.h" #include "vmwgfx_msg.h" @@ -194,7 +195,7 @@ static int vmw_send_msg(struct rpc_channel *channel, const char *msg)
return -EINVAL; } - +STACK_FRAME_NON_STANDARD(vmw_send_msg);
/** @@ -304,6 +305,7 @@ static int vmw_recv_msg(struct rpc_channel *channel, void **msg,
return 0; } +STACK_FRAME_NON_STANDARD(vmw_recv_msg);
/**
dri-devel@lists.freedesktop.org
-
Josh Poimboeuf -
Linus Torvalds -
tip-bot for Josh Poimboeuf