Skip to contentSkip to conversation

tools/fuzz fails due build, osstest did not notice

PATCH Superseded Olaf Hering 4 participants 14 messages Sep 3, 2018 – Sep 26, 2018 4 messages not in the archive

lore
Share & tools
lore
#0 lore raw

Since about two months staging fails to build because tools/fuzz can not cope with CFLAGS="-O2 -Wall -fstack-protector-strong -funwind-tables -fasynchronous-unwind-tables -fstack-clash-protection". While I can easily hide the bug by undefining _FORTIFY_SOURCE, I always wonder why osstest does not catch such bugs?

Looking at some random build-amd64/6.ts-xen-build.log output, it seems no CFLAGS at all is set. I'm sure SUSE is not the only one that sets CFLAGS during their package build, and further I think SUSE is not the only one who enforces -D_FORTIFY_SOURCE= globally. So on that ground, shouldn't whatever osstest does match what the consumers of xen use? Running osstest with throw-away-binaries compiled with -D_FORTIFY_SOURCE will likely not hurt or invalidate the overall coverage.

Olaf

_______________________________________________
Xen-devel mailing list
[email protected]
https://lists.xenproject.org/mailman/listinfo/xen-devel

#1 lore raw
Quoted 7 lines
Quoted 1 line
Quoted 1 line · On 03.09.18 at 11:53, <[email protected]> wrote:

Since about two months staging fails to build because tools/fuzz can not cope with CFLAGS="-O2 -Wall -fstack-protector-strong -funwind-tables -fasynchronous-unwind-tables -fstack-clash-protection". While I can easily hide the bug by undefining _FORTIFY_SOURCE, I always wonder why osstest does not catch such bugs?

Quoted 7 lines

Looking at some random build-amd64/6.ts-xen-build.log output, it seems no CFLAGS at all is set. I'm sure SUSE is not the only one that sets CFLAGS during their package build, and further I think SUSE is not the only one who enforces -D_FORTIFY_SOURCE= globally. So on that ground, shouldn't whatever osstest does match what the consumers of xen use? Running osstest with throw-away-binaries compiled with -D_FORTIFY_SOURCE will likely not hurt or invalidate the overall coverage.

Leaving aside the suggestion you make (I'm unconvinced either variant is strictly better than the other) - what is the actual problem? The mere listing of compiler flags passed does not make clear to me where the clash is, or how it would surface.

Jan

_______________________________________________
Xen-devel mailing list
[email protected]
https://lists.xenproject.org/mailman/listinfo/xen-devel

#2 lore raw

Am Mon, 03 Sep 2018 06:35:42 -0600
schrieb "Jan Beulich" <[email protected]>:

Quoted 3 lines

what is the actual problem? The mere
listing of compiler flags passed does not make clear to me where the clash is, or how it would surface.

As I noticed just now, it fails to build only in Tumbleweed. So in this specific case osstest would have caught it only in a few years from now.

[   38s] make -C x86_instruction_emulator install
[   38s] make[6]: Entering directory '/home/abuild/rpmbuild/BUILD/xen-4.12.20180831T120653.6164970942/non-dbg/tools/fuzz/x86_instruction_emulator'
[   38s] [ -L x86-emulate.h ] || ln -sf /home/abuild/rpmbuild/BUILD/xen-4.12.20180831T120653.6164970942/non-dbg/tools/fuzz/x86_instruction_emulator/../../../tools/tests/x86_emulator/x86-emulate.h
[   38s] [ -L x86_emulate ] || ln -sf /home/abuild/rpmbuild/BUILD/xen-4.12.20180831T120653.6164970942/non-dbg/tools/fuzz/x86_instruction_emulator/../../../xen/arch/x86/x86_emulate
[   38s] /usr/bin/gcc  -m64 -DBUILD_ID -fno-strict-aliasing -std=gnu99 -Wall -Wstrict-prototypes -Wdeclaration-after-statement -Wno-unused-but-set-variable -Wno-unused-local-typedefs   -O0 -fno-omit-frame-pointer -D__XEN_INTERFACE_VERSION__=__XEN_LATEST_INTERFACE_VERSION__ -MMD -MF .fuzz-emul.o.d -D_LARGEFILE_SOURCE -D_LARGEFILE64_SOURCE  -O2 -Wall -D_FORTIFY_SOURCE=2 -fstack-protector-strong -funwind-tables -fasynchronous-unwind-tables -fstack-clash-protection -I/home/abuild/rpmbuild/BUILD/xen-4.12.20180831T120653.6164970942/non-dbg/tools/fuzz/x86_instruction_emulator/../../../tools/include -D__XEN_TOOLS__ -I.  -c -o fuzz-emul.o fuzz-emul.c
[   38s] In file included from /usr/include/features.h:428,   
[   38s]                  from /usr/include/assert.h:35,
[   38s]                  from fuzz-emul.c:1:
[   38s] fuzz-emul.c: In function 'input_read':
[   38s] /usr/include/bits/string_fortified.h:31:1: error: inlining failed in call to always_inline 'memcpy': target specific option mismatch
[   38s]  __NTH (memcpy (void *__restrict __dest, const void *__restrict __src,
[   38s]  ^~~~~
[   38s] fuzz-emul.c:67:5: note: called from here
[   38s]      memcpy(dst, &s->corpus->data[s->data_index], size);
[   38s]      ^~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
[   38s] In file included from /usr/include/features.h:428,   
[   38s]                  from /usr/include/assert.h:35,
[   38s]                  from fuzz-emul.c:1:
[   38s] /usr/include/bits/string_fortified.h:31:1: error: inlining failed in call to always_inline 'memcpy': target specific option mismatch
[   38s]  __NTH (memcpy (void *__restrict __dest, const void *__restrict __src,
[   38s]  ^~~~~
[   38s] fuzz-emul.c:67:5: note: called from here
[   38s]      memcpy(dst, &s->corpus->data[s->data_index], size);
[   38s]      ^~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
[   38s] make[6]: *** [/home/abuild/rpmbuild/BUILD/xen-4.12.20180831T120653.6164970942/non-dbg/tools/fuzz/x86_instruction_emulator/../../../tools/Rules.mk:225: fuzz-emul.o] Error 1

Appending -U_FORTIFY_SOURCE in tools/fuzz fixes it. Not sure why fuzz is different from the rest of tools.

Olaf

_______________________________________________
Xen-devel mailing list
[email protected]
https://lists.xenproject.org/mailman/listinfo/xen-devel

#3 lore raw
Quoted 55 lines
Quoted 1 line
Quoted 1 line · On 04.09.18 at 09:32, <[email protected]> wrote:

Am Mon, 03 Sep 2018 06:35:42 -0600
schrieb "Jan Beulich" <[email protected]>:

Quoted 3 lines

what is the actual problem? The mere
listing of compiler flags passed does not make clear to me where the clash is, or how it would surface.

As I noticed just now, it fails to build only in Tumbleweed. So in this specific case osstest would have caught it only in a few years from now.

[ 38s] make -C x86_instruction_emulator install
[ 38s] make[6]: Entering directory
'/home/abuild/rpmbuild/BUILD/xen-4.12.20180831T120653.6164970942/non-dbg/tool s/fuzz/x86_instruction_emulator'
[ 38s] [ -L x86-emulate.h ] || ln -sf
/home/abuild/rpmbuild/BUILD/xen-4.12.20180831T120653.6164970942/non-dbg/tools /fuzz/x86_instruction_emulator/../../../tools/tests/x86_emulator/x86-emulate.h
[ 38s] [ -L x86_emulate ] || ln -sf
/home/abuild/rpmbuild/BUILD/xen-4.12.20180831T120653.6164970942/non-dbg/tools /fuzz/x86_instruction_emulator/../../../xen/arch/x86/x86_emulate
[ 38s] /usr/bin/gcc -m64 -DBUILD_ID -fno-strict-aliasing -std=gnu99 -Wall

(quoted diff) +0 -3
-Wstrict-prototypes -Wdeclaration-after-statement -Wno-unused-but-set-variable 
-Wno-unused-local-typedefs   -O0 -fno-omit-frame-pointer 
-D__XEN_INTERFACE_VERSION__=__XEN_LATEST_INTERFACE_VERSION__ -MMD -MF 

.fuzz-emul.o.d -D_LARGEFILE_SOURCE -D_LARGEFILE64_SOURCE -O2 -Wall

(quoted diff) +0 -3
-D_FORTIFY_SOURCE=2 -fstack-protector-strong -funwind-tables 
-fasynchronous-unwind-tables -fstack-clash-protection 
-I/home/abuild/rpmbuild/BUILD/xen-4.12.20180831T120653.6164970942/non-dbg/tool

s/fuzz/x86_instruction_emulator/../../../tools/include -D__XEN_TOOLS__ -I. -c -o fuzz-emul.o fuzz-emul.c

[   38s] In file included from /usr/include/features.h:428,   
[   38s]                  from /usr/include/assert.h:35,
[   38s]                  from fuzz-emul.c:1:
[   38s] fuzz-emul.c: In function 'input_read':
[   38s] /usr/include/bits/string_fortified.h:31:1: error: inlining failed 
in call to always_inline 'memcpy': target specific option mismatch
[   38s]  __NTH (memcpy (void *__restrict __dest, const void *__restrict 
__src,
[   38s]  ^~~~~
[   38s] fuzz-emul.c:67:5: note: called from here
[   38s]      memcpy(dst, &s->corpus->data[s->data_index], size);
[   38s]      ^~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
[   38s] In file included from /usr/include/features.h:428,   
[   38s]                  from /usr/include/assert.h:35,
[   38s]                  from fuzz-emul.c:1:
[   38s] /usr/include/bits/string_fortified.h:31:1: error: inlining failed 
in call to always_inline 'memcpy': target specific option mismatch
[   38s]  __NTH (memcpy (void *__restrict __dest, const void *__restrict 
__src,
[   38s]  ^~~~~
[   38s] fuzz-emul.c:67:5: note: called from here
[   38s]      memcpy(dst, &s->corpus->data[s->data_index], size);
[   38s]      ^~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
[   38s] make[6]: *** 
Quoted 6 lines

[/home/abuild/rpmbuild/BUILD/xen-4.12.20180831T120653.6164970942/non-dbg/tool s/fuzz/x86_instruction_emulator/../../../tools/Rules.mk:225: fuzz-emul.o] Error 1

Appending -U_FORTIFY_SOURCE in tools/fuzz fixes it. Not sure why fuzz is different from the rest of tools.

No idea here either, for the moment, even after looking at gcc's ix86_can_inline_p() (which apparently is the function triggering the diagnostic).

Jan

_______________________________________________
Xen-devel mailing list
[email protected]
https://lists.xenproject.org/mailman/listinfo/xen-devel

#4 lore raw
Quoted 60 lines · Jan Beulich
Quoted 60 lines
Quoted 2 lines
Quoted 2 lines · On 04.09.18 at 09:32, <[email protected]> wrote:

Am Mon, 03 Sep 2018 06:35:42 -0600
schrieb "Jan Beulich" <[email protected]>:

Quoted 3 lines

what is the actual problem? The mere
listing of compiler flags passed does not make clear to me where the clash is, or how it would surface.

As I noticed just now, it fails to build only in Tumbleweed. So in this specific case osstest would have caught it only in a few years from now.

[ 38s] make -C x86_instruction_emulator install
[ 38s] make[6]: Entering directory
'/home/abuild/rpmbuild/BUILD/xen-4.12.20180831T120653.6164970942/non-dbg/tool s/fuzz/x86_instruction_emulator'
[ 38s] [ -L x86-emulate.h ] || ln -sf
/home/abuild/rpmbuild/BUILD/xen-4.12.20180831T120653.6164970942/non-dbg/tools /fuzz/x86_instruction_emulator/../../../tools/tests/x86_emulator/x86-emulate.h
[ 38s] [ -L x86_emulate ] || ln -sf
/home/abuild/rpmbuild/BUILD/xen-4.12.20180831T120653.6164970942/non-dbg/tools /fuzz/x86_instruction_emulator/../../../xen/arch/x86/x86_emulate
[ 38s] /usr/bin/gcc -m64 -DBUILD_ID -fno-strict-aliasing -std=gnu99 -Wall

(quoted diff) +0 -3
-Wstrict-prototypes -Wdeclaration-after-statement -Wno-unused-but-set-variable
-Wno-unused-local-typedefs   -O0 -fno-omit-frame-pointer
-D__XEN_INTERFACE_VERSION__=__XEN_LATEST_INTERFACE_VERSION__ -MMD -MF

.fuzz-emul.o.d -D_LARGEFILE_SOURCE -D_LARGEFILE64_SOURCE -O2 -Wall

(quoted diff) +0 -3
-D_FORTIFY_SOURCE=2 -fstack-protector-strong -funwind-tables
-fasynchronous-unwind-tables -fstack-clash-protection
-I/home/abuild/rpmbuild/BUILD/xen-4.12.20180831T120653.6164970942/non-dbg/tool

s/fuzz/x86_instruction_emulator/../../../tools/include -D__XEN_TOOLS__ -I. -c -o fuzz-emul.o fuzz-emul.c

[   38s] In file included from /usr/include/features.h:428,
[   38s]                  from /usr/include/assert.h:35,
[   38s]                  from fuzz-emul.c:1:
[   38s] fuzz-emul.c: In function 'input_read':
[   38s] /usr/include/bits/string_fortified.h:31:1: error: inlining failed
in call to always_inline 'memcpy': target specific option mismatch
[   38s]  __NTH (memcpy (void *__restrict __dest, const void *__restrict
__src,
[   38s]  ^~~~~
[   38s] fuzz-emul.c:67:5: note: called from here
[   38s]      memcpy(dst, &s->corpus->data[s->data_index], size);
[   38s]      ^~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
[   38s] In file included from /usr/include/features.h:428,
[   38s]                  from /usr/include/assert.h:35,
[   38s]                  from fuzz-emul.c:1:
[   38s] /usr/include/bits/string_fortified.h:31:1: error: inlining failed
in call to always_inline 'memcpy': target specific option mismatch
[   38s]  __NTH (memcpy (void *__restrict __dest, const void *__restrict
__src,
[   38s]  ^~~~~
[   38s] fuzz-emul.c:67:5: note: called from here
[   38s]      memcpy(dst, &s->corpus->data[s->data_index], size);
[   38s]      ^~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
[   38s] make[6]: ***

[/home/abuild/rpmbuild/BUILD/xen-4.12.20180831T120653.6164970942/non-dbg/tool s/fuzz/x86_instruction_emulator/../../../tools/Rules.mk:225: fuzz-emul.o] Error 1

Quoted 6 lines
Quoted 2 lines

Appending -U_FORTIFY_SOURCE in tools/fuzz fixes it. Not sure why fuzz is different from the rest of tools.

No idea here either, for the moment, even after looking at gcc's ix86_can_inline_p() (which apparently is the function triggering the diagnostic).

I've just encountered this problem, building with the master branch of OpenEmbedded and Yocto's poky, with compiler flags that include
_FORTIFY_SOURCE=2 -msse3
with gcc 8.2.0.

Xen's x86_emulator header file does:
#pragma GCC target("no-sse")

The pragma was introduced in Xen commit 79136f26, to prevent the compiler from using registers otherwise used by inline memset and memcpy.

Setting _FORTIFY_SOURCE causes the use of always_inline variants of memset and memcpy which use those instructions, which is now causing the inline to fail.

The behaviour is described in the GCC bugzilla here:
https://gcc.gnu.org/bugzilla/show_bug.cgi?id=71991

{quote}
cat fuzz-emul.i
__attribute__((__always_inline__)) a() {}
#pragma GCC target "no-sse"
b() { a(); }

Where 'a' is 'memcpy' and 'b' is a function in xen.

Can you Honza also take a look r251333 where we started to reject such code?
{quote}

links to this GCC change:
https://gcc.gnu.org/viewcvs/gcc?view=revision&revision=251333

Would the below Xen patch, which adds undefs for _FORTIFY_SOURCE just for the two binaries which are affected be acceptable? It allows for use of _FORTIFY_SOURCE with the rest of the build.

Christopher

(quoted diff) +0 -0

b/tools/tests/x86_emulator/Makefile
index dec81c3..f7b9e58 100644

tools/tests/x86_emulator/Makefile +1 -1
@@ -133,7 +133,7 @@ x86_emulate/%: x86_emulate ;
 
 HOSTCFLAGS-x86_64 := -fno-PIE
 $(call cc-option-add,HOSTCFLAGS-x86_64,HOSTCC,-no-pie)
-HOSTCFLAGS += $(CFLAGS_xeninclude) -I. $(HOSTCFLAGS-$(XEN_COMPILE_ARCH))
+HOSTCFLAGS += $(CFLAGS_xeninclude) -I.

$(HOSTCFLAGS-$(XEN_COMPILE_ARCH)) -U_FORTIFY_SOURCE

 x86.h := $(addprefix $(XEN_ROOT)/tools/include/xen/asm/,\
                      x86-vendors.h x86-defns.h msr-index.h)
(quoted diff) +0 -0

b/tools/fuzz/x86_instruction_emulator/Makefile
index eb88f94..83e5633 100644

tools/fuzz/x86_instruction_emulator/Makefile +1 -1
@@ -16,7 +16,7 @@ x86_emulate/%: x86_emulate ;
 x86-emulate.c x86-emulate.h wrappers.c: %:
    [ -L $* ] || ln -sf $(XEN_ROOT)/tools/tests/x86_emulator/$*
 
-CFLAGS += $(CFLAGS_xeninclude) -D__XEN_TOOLS__ -I.
+CFLAGS += $(CFLAGS_xeninclude) -D__XEN_TOOLS__ -I. -U_FORTIFY_SOURCE
 
 GCOV_FLAGS := --coverage
 %-cov.o: %.c

_______________________________________________
Xen-devel mailing list
[email protected]
https://lists.xenproject.org/mailman/listinfo/xen-devel

#5 lore raw
Quoted 30 lines
Quoted 1 line
Quoted 1 line · On 20.09.18 at 20:19, <[email protected]> wrote:

I've just encountered this problem, building with the master branch of OpenEmbedded and Yocto's poky, with compiler flags that include
_FORTIFY_SOURCE=2 -msse3
with gcc 8.2.0.

Xen's x86_emulator header file does:
#pragma GCC target("no-sse")

The pragma was introduced in Xen commit 79136f26, to prevent the compiler from using registers otherwise used by inline memset and memcpy.

Setting _FORTIFY_SOURCE causes the use of always_inline variants of memset and memcpy which use those instructions, which is now causing the inline to fail.

The behaviour is described in the GCC bugzilla here:
https://gcc.gnu.org/bugzilla/show_bug.cgi?id=71991

{quote}
cat fuzz-emul.i
__attribute__((__always_inline__)) a() {}
#pragma GCC target "no-sse"
b() { a(); }

Where 'a' is 'memcpy' and 'b' is a function in xen.

Can you Honza also take a look r251333 where we started to reject such
code?
{quote}

Quoted 6 lines

links to this GCC change:
https://gcc.gnu.org/viewcvs/gcc?view=revision&revision=251333

Would the below Xen patch, which adds undefs for _FORTIFY_SOURCE just for the two binaries which are affected be acceptable? It allows for use of _FORTIFY_SOURCE with the rest of the build.

In a suitably massaged form I think this could be acceptable. It needs to fully describe the reasons for the addition, and if at all possible it should add the -U only for affected compilers (the bugzilla entry you found suggests to me that this has been fixed earlier this year).

That said, I'm certainly feeling a little uneasy about adding such overrides to the build flags "just for the two binaries which are affected". How can we be certain the same issue won't creep in elsewhere?

Furthermore it looks as if _FORTIFY_SOURCE implied __always_inline__ solely because its use takes a minimum level of optimization as a prereq. If that's indeed the case, un-defining that symbol won't help builds which enable sufficient optimization, irrespective of the use of _FORTIFY_SOURCE. Yet I'd prefer to see the root of the issue addressed, not just one particular incarnation.

Jan

_______________________________________________
Xen-devel mailing list
[email protected]
https://lists.xenproject.org/mailman/listinfo/xen-devel

#6 lore raw

PATCH 1/1 fuzz, test x86_emulator: disable sse before including always_inline fns

Compiling with _FORTIFY_SOURCE or higher levels of optimization enabled will always_inline several library fns (memset, memcpy, ...) (with gcc 8.2.0 and glibc 2.28).

In fuzz and x86_emulator test, the compiler is instructed not to generate SSE instructions via: #pragma GCC target("no-sse") because those registers are needed for use by the workload.

The combination above causes compilation failure as the inline functions use those instructions. This is resolved by reordering the inclusion of <stdio.h> and <string.h> to after the pragma disabling SSE generation.

Adds compile-time checks for unwanted inclusion of stdio.h, string.h

Adds necessary (previously missing) #include <stdio.h> to x86-emulate.h

---
 tools/fuzz/x86_instruction_emulator/fuzz-emul.c |  4 ++--
 tools/tests/x86_emulator/wrappers.c             |  3 ++-
 tools/tests/x86_emulator/x86-emulate.h          | 13 ++++++++++++-
 3 files changed, 16 insertions(+), 4 deletions(-)
tools/fuzz/x86_instruction_emulator/fuzz-emul.c +2 -2
index 03a2473..d0a02d5 100644
@@ -6,9 +6,7 @@
 #include <stdbool.h>
 #include <stddef.h>
 #include <stdint.h>
-#include <stdio.h>
 #include <stdlib.h>
-#include <string.h>
 #include <sys/types.h>
 #include <sys/stat.h>
 #include <sys/mman.h>
@@ -16,6 +14,8 @@
 #include <xen/xen.h>
 
 #include "x86-emulate.h"
+#include <stdio.h>
+#include <string.h>
 #include "fuzz-emul.h"
 
 #define MSR_INDEX_MAX 16
tools/tests/x86_emulator/wrappers.c +2 -1
index d02013c..349b9de 100644
@@ -1,9 +1,10 @@
 #include <stdarg.h>
-#include <stdio.h>
 
 #define WRAP(x) typeof(x) emul_##x
 #include "x86-emulate.h"
 
+#include <stdio.h>
+
 size_t emul_fwrite(const void *src, size_t sz, size_t n, FILE *f)
 {
     emul_save_fpu_state();
tools/tests/x86_emulator/x86-emulate.h +12 -1
index b249e46..8760bb8 100644
@@ -3,12 +3,23 @@
 #include <stddef.h>
 #include <stdint.h>
 #include <stdlib.h>
-#include <string.h>
 
+/*
+ * Use of sse registers must be disabled prior to the definition of
+ * always_inline functions that would use them (memcpy, memset, etc).
+ */
+#ifdef _STRING_H
+# error "Must not include <string.h> before x86-emulate.h"
+#endif
+#ifdef _STDIO_H
+# error "Must not include <stdio.h> before x86-emulate.h"
+#endif
 #if __GNUC__ >= 6
 #pragma GCC target("no-sse")
 #endif
 
+#include <string.h>
+#include <stdio.h>
 #include <xen/xen.h>
 
 #include <xen/asm/msr-index.h>
Signature · 7 lines
2.1.4


_______________________________________________
Xen-devel mailing list
[email protected]
https://lists.xenproject.org/mailman/listinfo/xen-devel

A message that is not in the archive (sent off-list or not yet received): [email protected]

A message that is not in the archive (sent off-list or not yet received): [email protected]

#9 lore raw

Re: [PATCH] fuzz, test x86_emulator: disable sse before including always_inline fns

Quoted 9 lines
Quoted 1 line
Quoted 1 line · On 21.09.18 at 21:25, <[email protected]> wrote:

Compiling with _FORTIFY_SOURCE or higher levels of optimization enabled will always_inline several library fns (memset, memcpy, ...) (with gcc 8.2.0 and glibc 2.28).

In fuzz and x86_emulator test, the compiler is instructed not to generate SSE instructions via: #pragma GCC target("no-sse") because those registers are needed for use by the workload.

Quoted 7 lines

The combination above causes compilation failure as the inline functions use those instructions. This is resolved by reordering the inclusion of <stdio.h> and <string.h> to after the pragma disabling SSE generation.

Adds compile-time checks for unwanted inclusion of stdio.h, string.h

Adds necessary (previously missing) #include <stdio.h> to x86-emulate.h

Why "necessary (previously missing)"? I can't seem to be able to spot anything in that header that would depend on stdio.h.

Quoted 13 lines
tools/fuzz/x86_instruction_emulator/fuzz-emul.c +0 -2
@@ -6,9 +6,7 @@
 #include <stdbool.h>
 #include <stddef.h>
 #include <stdint.h>
-#include <stdio.h>
 #include <stdlib.h>
-#include <string.h>
 #include <sys/types.h>
 #include <sys/stat.h>
 #include <sys/mman.h>

tools/fuzz/x86_instruction_emulator/fuzz-emul.c line 19

Quoted 6 lines
tools/fuzz/x86_instruction_emulator/fuzz-emul.c +2 -0
@@ -16,6 +14,8 @@
 #include <xen/xen.h>
 
 #include "x86-emulate.h"
+#include <stdio.h>
+#include <string.h>
 #include "fuzz-emul.h"

Such, at the first glance mis-placed, #include-s need a comment.

Quoted 8 lines
tools/tests/x86_emulator/x86-emulate.h +0 -1
@@ -3,12 +3,23 @@
 #include <stddef.h>
 #include <stdint.h>
 #include <stdlib.h>
-#include <string.h>
 

tools/tests/x86_emulator/x86-emulate.h line 16

Quoted 10 lines
tools/tests/x86_emulator/x86-emulate.h +10 -0
+/*
+ * Use of sse registers must be disabled prior to the definition of
+ * always_inline functions that would use them (memcpy, memset, etc).
+ */
+#ifdef _STRING_H
+# error "Must not include <string.h> before x86-emulate.h"
+#endif
+#ifdef _STDIO_H
+# error "Must not include <stdio.h> before x86-emulate.h"
+#endif

And what excludes other headers gaining always-inline functions (or even having them already, when the glibc implementation is not exactly the upstream one)? Along those lines I'm also unconvinced _STRING_H and _STDIO_H are universally suitable here.

tools/tests/x86_emulator/x86-emulate.h line 19

Quoted 3 lines
tools/tests/x86_emulator/x86-emulate.h +0 -0
 #if __GNUC__ >= 6
 #pragma GCC target("no-sse")
 #endif

I think the right approach would be to move this to the top of the file, ahead of all #include-s. However, I vaguely recall this not having worked on at least some systems back when I had introduced it. And really the commit introducing the above explains why. Therefore the latest with the appearance of some problematic inline function in stdlib.h we'd end up in a dead end.

Bottom line - I can accept something along the lines of your patch as a workaround for what appears to be a compiler bug. But this needs to be explained in comments, including limitations we're aware of so that people finding further need to change this can follow what is done why and where.

Jan

_______________________________________________
Xen-devel mailing list
[email protected]
https://lists.xenproject.org/mailman/listinfo/xen-devel

#10 lore raw

Re: [PATCH] fuzz, test x86_emulator: disable sse before including always_inline fns

Quoted 7 lines · Jan Beulich
Quoted 4 lines
Quoted 2 lines
Quoted 2 lines · On 21.09.18 at 21:25, <[email protected]> wrote:

Adds necessary (previously missing) #include <stdio.h> to x86-emulate.h

Why "necessary (previously missing)"? I can't seem to be able to spot anything in that header that would depend on stdio.h.

When WRAP is defined prior to including x86-emulate.h, it introduces the need for stdio.h to be included to avoid:

| x86-emulate.h:83:6: error: 'fwrite' undeclared here (not in a function); did you mean 'free'?
| WRAP(fwrite);

and similar for the other wrapped functions.

It was fine previously as wrapper.c included <stdio.h> prior to "x86-emulate.h" but that gets changed with this commit. I'll rewrite that part of the commit message though, and I'll also made the new <stdio.h> inclusion dependent on WRAP,
since WRAP being defined is what prompts the need for it.

Quoted 9 lines

Review line 19

Quoted 7 lines
(quoted diff) +2 -0
@@ -16,6 +14,8 @@
 #include <xen/xen.h>
 
 #include "x86-emulate.h"
+#include <stdio.h>
+#include <string.h>
 #include "fuzz-emul.h"

Such, at the first glance mis-placed, #include-s need a comment.

ack, will do.

Quoted 9 lines
Quoted 9 lines
tools/tests/x86_emulator/x86-emulate.h +0 -1
@@ -3,12 +3,23 @@
 #include <stddef.h>
 #include <stdint.h>
 #include <stdlib.h>
-#include <string.h>
 
Quoted 14 lines

tools/tests/x86_emulator/x86-emulate.h line 16

Quoted 10 lines
tools/tests/x86_emulator/x86-emulate.h +10 -0
+/*
+ * Use of sse registers must be disabled prior to the definition of
+ * always_inline functions that would use them (memcpy, memset, etc).
+ */
+#ifdef _STRING_H
+# error "Must not include <string.h> before x86-emulate.h"
+#endif
+#ifdef _STDIO_H
+# error "Must not include <stdio.h> before x86-emulate.h"
+#endif

And what excludes other headers gaining always-inline functions (or even having them already, when the glibc implementation is not exactly the upstream one)?

It is possible that could happen; but these two are the headers that are known to be a problem today, and are also the same headers that define the functions already listed within x86-emulate.h as deserving of their own wrappers, so it seems tolerable to introduce checking for just these two.

Quoted 3 lines

Along those lines I'm also
unconvinced _STRING_H and _STDIO_H are universally suitable here.

OK, I'll switch the <stdio.h> check over to look for EOF instead, as that's defined in the C standard, which will make the check portable. I don't have a similar option for <string.h> though - however, I don't think it absolutely has to be universally correct: for false-negatives, GLIBC is common enough that it should be caught relatively quickly if a violation is somehow introduced by a developer not using GLIBC, and false-positives seem unlikely in practice.

Quoted 8 lines

tools/tests/x86_emulator/x86-emulate.h line 19

Quoted 3 lines
tools/tests/x86_emulator/x86-emulate.h +0 -0
 #if __GNUC__ >= 6
 #pragma GCC target("no-sse")
 #endif

I think the right approach would be to move this to the top of the file, ahead of all #include-s. However, I vaguely recall this not having worked on at least some systems back when I had introduced it.

Yes, this is still the case, failing in my development environment at least. Declaration of atof fails with: "SSE register return with SSE disabled".

Quoted 3 lines

And really the commit introducing the above explains why. Therefore the latest with the appearance of some problematic inline function in stdlib.h we'd end up in a dead end.

Yes, that would be an unfortunate (compiler, optimize level, library) tuple to have to handle.

Quoted 5 lines

Bottom line - I can accept something along the lines of your patch as a workaround for what appears to be a compiler bug. But this needs to be explained in comments, including limitations we're aware of so that people finding further need to change this can follow what is done why and where.

Thanks for the review. I will send an updated version of the patch aiming to address these comments shortly.

Christopher

_______________________________________________
Xen-devel mailing list
[email protected]
https://lists.xenproject.org/mailman/listinfo/xen-devel

#11 lore raw

PATCH v2 1/1 fuzz, test x86_emulator: disable sse before including always_inline fns

Workaround for compiler rejection of SSE-using always_inlines defined before SSE is disabled.

Compiling with _FORTIFY_SOURCE or higher levels of optimization enabled will always_inline several library fns (memset, memcpy, ...) (with gcc 8.2.0 and glibc 2.28).

In fuzz and x86_emulator test, the compiler is instructed not to generate SSE instructions via: #pragma GCC target("no-sse") because those registers are needed for use by the workload.

The combination above causes compilation failure as the inline functions use those instructions. This is resolved by reordering the inclusion of <stdio.h> and <string.h> to after the pragma disabling SSE generation.

It would be preferable to locate the no-sse pragma within x86-emulate.h at the top of the file, prior to including any other headers; unfortunately doing so before <stdlib.h> causes compilation failure due to declaration of 'atof' with:
"SSE register return with SSE disabled".
Fortunately there is no (known) current dependency on any always_inline SSE-inclined function declared in <stdlib.h> or any of its dependencies, so the pragma is therefore issued immediately after inclusion of <stdlib.h> with a comment introduced to explain its location there.

Add compile-time checks for unwanted prior inclusion of <string.h> and <stdio.h>, which are the two headers that provide the library functions that are handled with wrappers and listed within "x86-emulate.h" as ones "we think might access any of the FPU state".
* Use standard-defined "EOF" macro to detect prior <stdio.h> inclusion.
* Use "_STRING_H" (non-standardized guard macro) as best-effort for detection of prior <string.h> inclusion. This is non-universally viable but will provide error output on common GLIBC systems, so provides some defensive coverage.

Adds conditional #include <stdio.h> to x86-emulate.h because fwrite, printf, etc. are referenced when WRAP has been defined.

---
Changes in v2:

* Added comment to explain the deliberate ordering of includes that might otherwise be believed to be incorrect.

* Added comment to explain the positioning of the no-sse pragma and the ordering of includes around it.

* Put the defensive header-inclusion compile-time checks right next to the inclusion of the headers that they are checking for.

* Narrowed the inclusion of stdio.h in x86-emulate.h to be conditional on WRAP being defined, since that condition determines the need to include it. Removed the include of stdio.h from wrappers.c as no longer required and to obviate the need for an explanatory comment about include ordering there.

* Use definition of 'EOF' to detect prior inclusion of <stdio.h>, since this is defined in the standard and increases portability to other C libraries.

* Added context information to the commit message.

 tools/fuzz/x86_instruction_emulator/fuzz-emul.c | 10 +++++++--
 tools/tests/x86_emulator/wrappers.c             |  1 -
 tools/tests/x86_emulator/x86-emulate.h          | 28 +++++++++++++++++++++++--
 3 files changed, 34 insertions(+), 5 deletions(-)
tools/fuzz/x86_instruction_emulator/fuzz-emul.c +8 -2
index 03a2473..0ffd0fb 100644
@@ -6,9 +6,7 @@
 #include <stdbool.h>
 #include <stddef.h>
 #include <stdint.h>
-#include <stdio.h>
 #include <stdlib.h>
-#include <string.h>
 #include <sys/types.h>
 #include <sys/stat.h>
 #include <sys/mman.h>
@@ -16,6 +14,14 @@
 #include <xen/xen.h>
 
 #include "x86-emulate.h"
+/*
+ * include "x86-emulate.h" prior to <stdio.h> and <string.h>:
+ * x86-emulate.h disables use of SSE registers, while <stdio.h> and <string.h>
+ * declare functions that may be always_inline and use those registers
+ * unless they have been disabled earlier, which can fail to compile.
+ */
+#include <stdio.h>
+#include <string.h>
 #include "fuzz-emul.h"
 
 #define MSR_INDEX_MAX 16
tools/tests/x86_emulator/wrappers.c +0 -1
index d02013c..eba7cc9 100644
@@ -1,5 +1,4 @@
 #include <stdarg.h>
-#include <stdio.h>
 
 #define WRAP(x) typeof(x) emul_##x
 #include "x86-emulate.h"
tools/tests/x86_emulator/x86-emulate.h +26 -2
index b249e46..07ea1e8 100644
@@ -3,11 +3,35 @@
 #include <stddef.h>
 #include <stdint.h>
 #include <stdlib.h>
-#include <string.h>
-
+/*
+ * Use of sse registers must be disabled prior to the definition of
+ * always_inline functions that would use them (memcpy, memset, etc),
+ * so do this as early as possible, aiming to be before any always_inline
+ * functions that are used are declared.
+ * Unfortunately, this cannot be done prior to inclusion of <stdlib.h>
+ * due to functions such as 'atof' that have SSE register return declared,
+ * so do so here, immediately after that.
+ */
 #if __GNUC__ >= 6
 #pragma GCC target("no-sse")
 #endif
+ /*
+ * Attempt detection of unwanted prior inclusion of some headers known to use
+ * always_inline with SSE registers in some library / compiler / optimization
+ * combinations.
+ */
+#ifdef _STRING_H
+# error "Must not include <string.h> before x86-emulate.h"
+#endif
+#include <string.h>
+
+/* EOF is a standard macro defined in <stdio.h> so use it for detection */
+#ifdef EOF
+# error "Must not include <stdio.h> before x86-emulate.h"
+#endif
+#ifdef WRAP
+#include <stdio.h>
+#endif
 
 #include <xen/xen.h>
 
Signature · 7 lines
2.1.4


_______________________________________________
Xen-devel mailing list
[email protected]
https://lists.xenproject.org/mailman/listinfo/xen-devel

A message that is not in the archive (sent off-list or not yet received): [email protected]

#15 lore raw

Re: [PATCH] fuzz, test x86_emulator: disable sse before including always_inline fns

Quoted 17 lines
Quoted 1 line
Quoted 1 line · On 25.09.18 at 01:17, <[email protected]> wrote:
Quoted 7 lines · Jan Beulich
Quoted 4 lines
Quoted 2 lines
Quoted 2 lines · On 21.09.18 at 21:25, <[email protected]> wrote:

Adds necessary (previously missing) #include <stdio.h> to x86-emulate.h

Why "necessary (previously missing)"? I can't seem to be able to spot anything in that header that would depend on stdio.h.

When WRAP is defined prior to including x86-emulate.h, it introduces the need for stdio.h to be included to avoid:

| x86-emulate.h:83:6: error: 'fwrite' undeclared here (not in a function); did you mean 'free'?
| WRAP(fwrite);

Quoted 8 lines

and similar for the other wrapped functions.

It was fine previously as wrapper.c included <stdio.h> prior to "x86-emulate.h"
but that gets changed with this commit. I'll rewrite that part of the commit message though, and I'll also made the new <stdio.h> inclusion dependent on WRAP,
since WRAP being defined is what prompts the need for it.

But that's (in my opinion) a requirement towards the file defining WRAP then, not towards the header.

Quoted 11 lines
Quoted 8 lines
Quoted 3 lines

#if __GNUC__ >= 6
#pragma GCC target("no-sse")
#endif

I think the right approach would be to move this to the top of the file, ahead of all #include-s. However, I vaguely recall this not having worked on at least some systems back when I had introduced it.

Yes, this is still the case, failing in my development environment at least. Declaration of atof fails with: "SSE register return with SSE disabled".

Hmm - I (now) wonder whether -mfpmath= could help here.

Jan

_______________________________________________
Xen-devel mailing list
[email protected]
https://lists.xenproject.org/mailman/listinfo/xen-devel

A message that is not in the archive (sent off-list or not yet received): [email protected]

Reviewed-by
#17 lore raw

Re: [PATCH v2] fuzz, test x86_emulator: disable sse before including always_inline fns

tools/tests/x86_emulator/wrappers.c line 4

Quoted 9 lines
Quoted 1 line
Quoted 1 line · On 25.09.18 at 01:22, <[email protected]> wrote:
tools/tests/x86_emulator/wrappers.c +0 -1
@@ -1,5 +1,4 @@
 #include <stdarg.h>
-#include <stdio.h>
 
 #define WRAP(x) typeof(x) emul_##x
 #include "x86-emulate.h"

Ah, I see now why you can't take care of the inclusion here.

with one cosmetic remark (which I may take care of while committing):

Quoted 27 lines
tools/tests/x86_emulator/x86-emulate.h +16 -2
@@ -3,11 +3,35 @@
 #include <stddef.h>
 #include <stdint.h>
 #include <stdlib.h>
-#include <string.h>
-
+/*
+ * Use of sse registers must be disabled prior to the definition of
+ * always_inline functions that would use them (memcpy, memset, etc),
+ * so do this as early as possible, aiming to be before any always_inline
+ * functions that are used are declared.
+ * Unfortunately, this cannot be done prior to inclusion of <stdlib.h>
+ * due to functions such as 'atof' that have SSE register return declared,
+ * so do so here, immediately after that.
+ */
 #if __GNUC__ >= 6
 #pragma GCC target("no-sse")
 #endif
+ /*
+ * Attempt detection of unwanted prior inclusion of some headers known to use
+ * always_inline with SSE registers in some library / compiler / optimization
+ * combinations.
+ */
+#ifdef _STRING_H
+# error "Must not include <string.h> before x86-emulate.h"

tools/tests/x86_emulator/x86-emulate.h line 34

Quoted 10 lines
tools/tests/x86_emulator/x86-emulate.h +10 -0
+#endif
+#include <string.h>
+
+/* EOF is a standard macro defined in <stdio.h> so use it for detection */
+#ifdef EOF
+# error "Must not include <stdio.h> before x86-emulate.h"
+#endif
+#ifdef WRAP
+#include <stdio.h>
+#endif

There's some inconsistency with directive indentation: The two "error" ones you indent, but the "include" you leave unindented (matching the pre-existing "pragma"). At least all your additions should be consistent with one another.

Jan

_______________________________________________
Xen-devel mailing list
[email protected]
https://lists.xenproject.org/mailman/listinfo/xen-devel