Skip to contentSkip to conversation

Coverage support

RFC PATCH Superseded Frediano Ziglio 3 participants 12 messages Feb 1, 2013 – Feb 4, 2013

v1 of 10 Superseded Latest is v10 →

lore
Share & tools
lore
#0 lore raw

RFC PATCH 1/1 Coverage support

Updated set of patches for coverage.
Changes:
- implemented way of getting information
- use only 2 parameter in hypercall (still not sure it's better to use sysctl)
- use "coverage" for configuration option as suggested by Ian Campbell
- split header in include/public/gcov.h and include/xen/gcov.h to allow tools to include public header
- use weak to avoid defining compat_coverage_op

I manage to write 2 tools, one that call the new hypercall and dump the blob in a file, one that split this file into required .gcda files. Both very hack ones and dirty but works and prove the hypervisor code is fine. Should I just send them anyway?

The forth patch allow to consume just 0 bytes if coverage is disabled.

#1 lore raw

PATCH 1/4 Reserve hypercall number for handling coverage informations.

---
 xen/include/public/xen.h |    1 +
 1 file changed, 1 insertion(+)
xen/include/public/xen.h +1 -0
index 5593066..feec70e 100644
@@ -101,6 +101,7 @@ DEFINE_XEN_GUEST_HANDLE(xen_ulong_t);
 #define __HYPERVISOR_kexec_op             37
 #define __HYPERVISOR_tmem_op              38
 #define __HYPERVISOR_xc_reserved_op       39 /* reserved for XenClient */
+#define __HYPERVISOR_coverage             40 /* reserved to get coverage information */
 
 /* Architecture-specific hypercall definitions. */
 #define __HYPERVISOR_arch_0               48
Signature · 1 line
1.7.9.5
#2 lore raw

PATCH 2/4 Adding support for coverage information

In reply to Frediano Ziglio

This patch introduce coverage support to Xen.
Currently it allows to compile Xen with coverage support but there is no way to extract them.

The declarations came from Linux source files (as you can see from file headers).

The idea is to have some operations mainly
- get coverage information size
- read coverage information
- reset coverage counters

The op parameter in the hypercall will be the operation. The uarg parameter is a pointer (used to read size or data).

Linux use a file system to export these information. The information will be a blob to handle with some tools (as usually tools require a bunch of files but Xen does not handle files at all). I'll pack them to make things simpler as possible.

These information cannot be put in a specific section (allowing a safe mapping) as gcc use .rodata, .data, .text and .ctors sections.

I added code to handle constructors used in this case to initialize a linked list of files.

I excluded %.init.o files as they are used before Xen start and should not have section like .text or .data.

I used a "coverage" configuration option to mimic the "debug" one.

---
 .gitignore                         |    2 +
 Config.mk                          |    4 ++
 xen/Rules.mk                       |    5 ++
 xen/arch/x86/setup.c               |    3 ++
 xen/arch/x86/x86_64/compat/entry.S |    4 ++
 xen/arch/x86/x86_64/entry.S        |    4 ++
 xen/arch/x86/xen.lds.S             |    7 +++
 xen/common/Makefile                |    2 +
 xen/common/gcov/Makefile           |    5 ++
 xen/common/gcov/gcov.c             |  105 ++++++++++++++++++++++++++++++++++++
 xen/common/gcov/nogcov.c           |   22 ++++++++
 xen/include/public/gcov.h          |   93 ++++++++++++++++++++++++++++++++
 xen/include/xen/gcov.h             |   32 +++++++++++
 xen/include/xen/hypercall.h        |    6 +++
 14 files changed, 294 insertions(+)
 create mode 100644 xen/common/gcov/Makefile
 create mode 100644 xen/common/gcov/gcov.c
 create mode 100644 xen/common/gcov/nogcov.c
 create mode 100644 xen/include/public/gcov.h
 create mode 100644 xen/include/xen/gcov.h
.gitignore +2 -0
index 462e291..af48492 100644
@@ -13,6 +13,8 @@
 *.tmp
 *.spot
 *.spit
+*.gcno
+*.gcda
 TAGS
 GTAGS
 GRTAGS
Config.mk +4 -0
index 64541c8..a49b7c9 100644
@@ -228,3 +228,7 @@ QEMU_TAG ?= 2a1354d655d816feaad7dbdb8364f40a208439c1
 # doing and are prepared for some pain.
 
 CONFIG_TESTS       ?= y
+
+# Test coverage support
+coverage ?= n
+
xen/Rules.mk +5 -0
index c2db449..c9044f5 100644
@@ -103,6 +103,11 @@ subdir-all := $(subdir-y) $(subdir-n)
 
 $(filter %.init.o,$(obj-y) $(obj-bin-y)): CFLAGS += -DINIT_SECTIONS_ONLY
 
+
+ifeq ($(coverage),y)
+$(filter-out %.init.o,$(obj-y) $(obj-bin-y)): CFLAGS += -fprofile-arcs -ftest-coverage -DTEST_COVERAGE
+endif
+
 ifeq ($(lto),y)
 # Would like to handle all object files as bitcode, but objects made from
 # pure asm are in a different format and have to be collected separately.
xen/arch/x86/setup.c +3 -0
index f4d3788..ddf5c4d 100644
@@ -46,6 +46,7 @@
 #include <asm/setup.h>
 #include <xen/cpu.h>
 #include <asm/nmi.h>
+#include <xen/gcov.h>
 
 /* opt_nosmp: If true, secondary processors are ignored. */
 static bool_t __initdata opt_nosmp;
@@ -1313,6 +1314,8 @@ void __init __start_xen(unsigned long mbi_p)
 
     init_trace_bufs();
 
+    init_coverage();
+
     console_endboot();
 
     /* Hide UART from DOM0 if we're using it */
xen/arch/x86/x86_64/compat/entry.S +4 -0
index 7051f90..ebbb6b7 100644
@@ -413,6 +413,8 @@ ENTRY(compat_hypercall_table)
         .quad do_domctl
         .quad compat_kexec_op
         .quad do_tmem_op
+        .quad compat_ni_hypercall
+        .quad compat_coverage_op
         .rept __HYPERVISOR_arch_0-((.-compat_hypercall_table)/8)
         .quad compat_ni_hypercall
         .endr
@@ -461,6 +463,8 @@ ENTRY(compat_hypercall_args_table)
         .byte 1 /* do_domctl                */
         .byte 2 /* compat_kexec_op          */
         .byte 1 /* do_tmem_op               */
+        .byte 0 /* compat_ni_hypercall      */
+        .byte 2 /* compat_coverage_op       */
         .rept __HYPERVISOR_arch_0-(.-compat_hypercall_args_table)
         .byte 0 /* compat_ni_hypercall      */
         .endr
xen/arch/x86/x86_64/entry.S +4 -0
index 408c348..670a2e6 100644
@@ -738,6 +738,8 @@ ENTRY(hypercall_table)
         .quad do_domctl
         .quad do_kexec_op
         .quad do_tmem_op
+        .quad do_ni_hypercall
+        .quad do_coverage_op
         .rept __HYPERVISOR_arch_0-((.-hypercall_table)/8)
         .quad do_ni_hypercall
         .endr
@@ -786,6 +788,8 @@ ENTRY(hypercall_args_table)
         .byte 1 /* do_domctl            */
         .byte 2 /* do_kexec             */
         .byte 1 /* do_tmem_op           */
+        .byte 0 /* do_ni_hypercall      */
+        .byte 2 /* do_coverage_op       */
         .rept __HYPERVISOR_arch_0-(.-hypercall_args_table)
         .byte 0 /* do_ni_hypercall      */
         .endr
xen/arch/x86/xen.lds.S +7 -0
index d324afd..5570389 100644
@@ -108,6 +108,13 @@ SECTIONS
        __trampoline_seg_start = .;
        *(.trampoline_seg)
        __trampoline_seg_stop = .;
+
+       . = ALIGN(8);
+       __CTOR_LIST__ = .;
+       QUAD((__CTOR_END__ - __CTOR_LIST__) / 8 - 2)
+       *(.ctors)
+       QUAD(0)
+       __CTOR_END__ = .;
   } :text
   . = ALIGN(32);
   .init.setup : {
xen/common/Makefile +2 -0
index 1677342..51191d6 100644
@@ -59,5 +59,7 @@ subdir-$(CONFIG_COMPAT) += compat
 
 subdir-$(x86_64) += hvm
 
+subdir-y += gcov
+
 subdir-y += libelf
 subdir-$(HAS_DEVICE_TREE) += libfdt
xen/common/gcov/Makefile new file +5 -0
new file mode 100644
index 0000000..43a692c
@@ -0,0 +1,5 @@
+ifneq ($(coverage),y)
+obj-y += nogcov.o
+endif
+obj-$(coverage) += gcov.o
+
xen/common/gcov/gcov.c new file +105 -0

105 lines not on this page · raw message

xen/common/gcov/nogcov.c new file +22 -0
new file mode 100644
index 0000000..9642c96
@@ -0,0 +1,22 @@
+/*
+ * nogov.c
+ *
+ * Compiled if coverage information are not enabled
+ */
+
+#include <xen/config.h>
+#include <xen/lib.h>
+#include <xen/errno.h>
+#include <public/xen.h>
+
+long do_coverage_op(int op, XEN_GUEST_HANDLE_PARAM(void) uarg)
+{
+    return -ENOSYS;
+}
+
+#ifdef CONFIG_COMPAT
+int compat_coverage_op(int op, XEN_GUEST_HANDLE_PARAM(void) uarg)
+    __attribute__ ((alias ("do_coverage_op")));
+#endif
+
+
xen/include/public/gcov.h new file +93 -0

93 lines not on this page · raw message

xen/include/xen/gcov.h new file +32 -0
new file mode 100644
index 0000000..517ed75
@@ -0,0 +1,32 @@
+/*
+ *  Profiling infrastructure declarations.
+ *
+ *  This file is based on gcc-internal definitions. Data structures are
+ *  defined to be compatible with gcc counterparts. For a better
+ *  understanding, refer to gcc source: gcc/gcov-io.h.
+ *
+ *    Copyright IBM Corp. 2009
+ *    Author(s): Peter Oberparleiter <[email protected]>
+ *
+ *    Uses gcc-internal data definitions.
+ */
+
+#ifndef XEN_GCOV_H
+#define XEN_GCOV_H XEN_GCOV_H
+
+#include <public/gcov.h>
+
+/**
+ * Initialize coverage informations
+ * Currently create a linked list of all files compiled in with
+ * coverage informations.
+ */
+#ifdef TEST_COVERAGE
+void __init init_coverage(void);
+#else
+static inline void init_coverage(void)
+{
+}
+#endif
+
+#endif /* GCOV_H */
xen/include/xen/hypercall.h +6 -0
index 7c3d719..e62525a 100644
@@ -140,6 +140,9 @@ do_tmem_op(
 extern long
 do_xenoprof_op(int op, XEN_GUEST_HANDLE_PARAM(void) arg);
 
+extern long
+do_coverage_op(int op, XEN_GUEST_HANDLE_PARAM(void) uarg);
+
 #ifdef CONFIG_COMPAT
 
 extern int
@@ -163,6 +166,9 @@ extern int
 compat_xenoprof_op(int op, XEN_GUEST_HANDLE_PARAM(void) arg);
 
 extern int
+compat_coverage_op(int op, XEN_GUEST_HANDLE_PARAM(void) uarg);
+
+extern int
 compat_xen_version(
     int cmd,
     XEN_GUEST_HANDLE_PARAM(void) arg);
Signature · 1 line
1.7.9.5
#3 lore raw

Re: [PATCH 2/4] Adding support for coverage information

Config.mk QEMU_TAG ?= 2a1354d655d816feaad7dbdb8364f40a208439c1 · line 234

Quoted 11 lines
Quoted 1 line
Quoted 1 line · Frediano Ziglio
Config.mk +4 -0
@@ -228,3 +228,7 @@ QEMU_TAG ?= 2a1354d655d816feaad7dbdb8364f40a208439c1
 # doing and are prepared for some pain.
 
 CONFIG_TESTS       ?= y
+
+# Test coverage support
+coverage ?= n
+

Alongside debug and debug_symbols please.

xen/Rules.mk subdir-all := $(subdir-y) $(subdir-n) · line 106

Quoted 7 lines
xen/Rules.mk +1 -0
@@ -103,6 +103,11 @@ subdir-all := $(subdir-y) $(subdir-n)
 
 $(filter %.init.o,$(obj-y) $(obj-bin-y)): CFLAGS += -DINIT_SECTIONS_ONLY
 
+

Stray blank line.

xen/Rules.mk subdir-all := $(subdir-y) $(subdir-n) · line 109

Quoted 3 lines
xen/Rules.mk +3 -0
+ifeq ($(coverage),y)
+$(filter-out %.init.o,$(obj-y) $(obj-bin-y)): CFLAGS += -fprofile-arcs -ftest-coverage -DTEST_COVERAGE
+endif

For one - isn't simply using $(obj-y) here sufficient (i.e. without the $(filter-out ...))?

And second, I would thing this then ought to become a single line:

$(obj-$(coverage)): CFLAGS += -fprofile-arcs -ftest-coverage -DTEST_COVERAGE

xen/common/gcov/Makefile line 5

Quoted 8 lines
xen/common/gcov/Makefile new file +5 -0
@@ -0,0 +1,5 @@
+ifneq ($(coverage),y)
+obj-y += nogcov.o
+endif
+obj-$(coverage) += gcov.o
+

How about

obj-y := nogcov.o
obj-$(coverage) := gcov.o

xen/common/gcov/Makefile

Quoted 6 lines
xen/common/gcov/Makefile new file +6 -0
+typedef void (*ctor_func_t)(void);
+extern struct
+{
+    unsigned long count;
+    ctor_func_t funcs[1];
+} __CTOR_LIST__;

const?

xen/common/gcov/Makefile

Quoted 2 lines
xen/common/gcov/Makefile new file +2 -0
+
+void init_coverage(void)

__init

Quoted 26 lines
xen/include/public/gcov.h new file +23 -0
@@ -0,0 +1,93 @@
+/*
+ *  Profiling infrastructure declarations.
+ *
+ *  This file is based on gcc-internal definitions. Data structures are
+ *  defined to be compatible with gcc counterparts. For a better
+ *  understanding, refer to gcc source: gcc/gcov-io.h.
+ *
+ *    Copyright IBM Corp. 2009
+ *    Author(s): Peter Oberparleiter <[email protected]>
+ *
+ *    Uses gcc-internal data definitions.
+ */
+
+#ifndef XEN_PUBLIC_GCOV_H
+#define XEN_PUBLIC_GCOV_H XEN_PUBLIC_GCOV_H
+
+/*
+ * Profiling data types used for gcc 3.4 and above - these are defined by
+ * gcc and need to be kept as close to the original definition as possible to
+ * remain compatible.
+ */
+#define GCOV_COUNTERS         5
+#define GCOV_DATA_MAGIC       ((unsigned int) 0x67636461)

xen/include/public/gcov.h line 33

Quoted 10 lines
xen/include/public/gcov.h new file +10 -0
+#define GCOV_TAG_FUNCTION     ((unsigned int) 0x01000000)
+#define GCOV_TAG_COUNTER_BASE ((unsigned int) 0x01a10000)
+#define GCOV_TAG_FOR_COUNTER(count) \
+    (GCOV_TAG_COUNTER_BASE + ((unsigned int) (count) << 17))
+
+#if BITS_PER_LONG >= 64
+typedef long gcov_type;
+#else
+typedef long long gcov_type;
+#endif

What's this???

Quoted 5 lines
xen/include/public/gcov.h new file +5 -0
+/**
+ * struct gcov_ctr_info - profiling data per counter type
+ * @num: number of counter values for this type
+ * @values: array of counter values for this type
+ * @merge: merge function for counter values of this type (unused)

xen/include/public/gcov.h line 65

Quoted 10 lines
xen/include/public/gcov.h new file +10 -0
+ *
+ * This data is generated by gcc during compilation and doesn't change
+ * at run-time with the exception of the values array.
+ */
+struct gcov_ctr_info
+{
+    unsigned int num;
+    gcov_type *values;
+    void (*merge)(gcov_type *, unsigned int);
+};

Pointers, and even more so function ones, can't be part of the public interface.

Jan

#4 lore raw

Re: [PATCH 2/4] Adding support for coverage information

Quoted 6 lines · Jan Beulich
Quoted 6 lines
Quoted 1 line
Quoted 1 line · Frediano Ziglio
Config.mk +0 -0
@@ -228,3 +228,7 @@ QEMU_TAG ?= 2a1354d655d816feaad7dbdb8364f40a208439c1
 # doing and are prepared for some pain.
 
Quoted 7 lines

Config.mk QEMU_TAG ?= 2a1354d655d816feaad7dbdb8364f40a208439c1 · line 234

Quoted 5 lines
Config.mk +4 -0
 CONFIG_TESTS       ?= y
+
+# Test coverage support
+coverage ?= n
+

Alongside debug and debug_symbols please.

Ok.

Quoted 17 lines

xen/Rules.mk subdir-all := $(subdir-y) $(subdir-n) · line 106

Quoted 7 lines
xen/Rules.mk +1 -0
@@ -103,6 +103,11 @@ subdir-all := $(subdir-y) $(subdir-n)
 
 $(filter %.init.o,$(obj-y) $(obj-bin-y)): CFLAGS += -DINIT_SECTIONS_ONLY
 
+

Stray blank line.

xen/Rules.mk subdir-all := $(subdir-y) $(subdir-n) · line 109

Quoted 3 lines
xen/Rules.mk +3 -0
+ifeq ($(coverage),y)
+$(filter-out %.init.o,$(obj-y) $(obj-bin-y)): CFLAGS += -fprofile-arcs -ftest-coverage -DTEST_COVERAGE
+endif

For one - isn't simply using $(obj-y) here sufficient (i.e. without the $(filter-out ...))?

Quoted 4 lines

And second, I would thing this then ought to become a single line:

$(obj-$(coverage)): CFLAGS += -fprofile-arcs -ftest-coverage -DTEST_COVERAGE

Yes, it works fine and seems to compile all proper files but I don't understand why.
What's the difference between $(obj-bin-y) and $(obj-y) ??

Quoted 11 lines

xen/common/gcov/Makefile line 5

Quoted 8 lines
xen/common/gcov/Makefile new file +5 -0
@@ -0,0 +1,5 @@
+ifneq ($(coverage),y)
+obj-y += nogcov.o
+endif
+obj-$(coverage) += gcov.o
+

How about

Quoted 3 lines

obj-y := nogcov.o
obj-$(coverage) := gcov.o

Using sysctl is even simpler, just

subdir-$(coverage) += gcov

in common/Makefile and

obj-y += gcov.o

in common/gcov/Makefile

Quoted 10 lines

xen/common/gcov/Makefile

Quoted 7 lines
xen/common/gcov/Makefile new file +6 -0
+typedef void (*ctor_func_t)(void);
+extern struct
+{
+    unsigned long count;
+    ctor_func_t funcs[1];
+} __CTOR_LIST__;

const?

Fine.

Quoted 5 lines

xen/common/gcov/Makefile

Quoted 2 lines
xen/common/gcov/Makefile new file +2 -0
+
+void init_coverage(void)

__init

I removed from previous and put in the header as suggestion.

Quoted 30 lines
Quoted 30 lines
xen/include/public/gcov.h new file +27 -0
@@ -0,0 +1,93 @@
+/*
+ *  Profiling infrastructure declarations.
+ *
+ *  This file is based on gcc-internal definitions. Data structures are
+ *  defined to be compatible with gcc counterparts. For a better
+ *  understanding, refer to gcc source: gcc/gcov-io.h.
+ *
+ *    Copyright IBM Corp. 2009
+ *    Author(s): Peter Oberparleiter <[email protected]>
+ *
+ *    Uses gcc-internal data definitions.
+ */
+
+#ifndef XEN_PUBLIC_GCOV_H
+#define XEN_PUBLIC_GCOV_H XEN_PUBLIC_GCOV_H
+
+/*
+ * Profiling data types used for gcc 3.4 and above - these are defined by
+ * gcc and need to be kept as close to the original definition as possible to
+ * remain compatible.
+ */
+#define GCOV_COUNTERS         5
+#define GCOV_DATA_MAGIC       ((unsigned int) 0x67636461)
+#define GCOV_TAG_FUNCTION     ((unsigned int) 0x01000000)
+#define GCOV_TAG_COUNTER_BASE ((unsigned int) 0x01a10000)
+#define GCOV_TAG_FOR_COUNTER(count) \
+    (GCOV_TAG_COUNTER_BASE + ((unsigned int) (count) << 17))
Quoted 9 lines

xen/include/public/gcov.h line 33

Quoted 6 lines
xen/include/public/gcov.h new file +6 -0
+
+#if BITS_PER_LONG >= 64
+typedef long gcov_type;
+#else
+typedef long long gcov_type;
+#endif

What's this???

This came from Linux, probably for Xen a simple

typedef long gcov_type

is ok.

Quoted 10 lines
Quoted 10 lines
xen/include/public/gcov.h new file +10 -0
+/**
+ * struct gcov_ctr_info - profiling data per counter type
+ * @num: number of counter values for this type
+ * @values: array of counter values for this type
+ * @merge: merge function for counter values of this type (unused)
+ *
+ * This data is generated by gcc during compilation and doesn't change
+ * at run-time with the exception of the values array.
+ */
+struct gcov_ctr_info
Quoted 9 lines

xen/include/public/gcov.h line 65

Quoted 5 lines
xen/include/public/gcov.h new file +5 -0
+{
+    unsigned int num;
+    gcov_type *values;
+    void (*merge)(gcov_type *, unsigned int);
+};

Pointers, and even more so function ones, can't be part of the public interface.

Yes, probably there is no need to expose this to public.

Probably defining only the exported blob would be ok and have no licence issues.

Quoted 2 lines

Jan

Frediano

#5 lore raw

Re: [PATCH 2/4] Adding support for coverage information

Quoted 10 lines
Quoted 1 line
Quoted 1 line · Frediano Ziglio
Quoted 8 lines · Jan Beulich

xen/Rules.mk subdir-all := $(subdir-y) $(subdir-n) · line 109

Quoted 4 lines
Quoted 1 line
Quoted 1 line · Frediano Ziglio
xen/Rules.mk +3 -0
+ifeq ($(coverage),y)
+$(filter-out %.init.o,$(obj-y) $(obj-bin-y)): CFLAGS += -fprofile-arcs -ftest-coverage -DTEST_COVERAGE
+endif

For one - isn't simply using $(obj-y) here sufficient (i.e. without the $(filter-out ...))?

Quoted 8 lines
Quoted 4 lines

And second, I would thing this then ought to become a single line:

$(obj-$(coverage)): CFLAGS += -fprofile-arcs -ftest-coverage -DTEST_COVERAGE

Yes, it works fine and seems to compile all proper files but I don't understand why.
What's the difference between $(obj-bin-y) and $(obj-y) ??

The former is the set of all .init.o files, which (when building with clang instead of gcc) must not undergo certain transformations (and aren't worth it, as init-time only code isn't really performance critical).

Quoted 7 lines
Quoted 5 lines

Review

Quoted 2 lines
(quoted diff) +2 -0
+
+void init_coverage(void)

__init

I removed from previous and put in the header as suggestion.

Not sure what you mean here, but just in case - __init annotations do specifically not belong on declarations, but only on definitions.

Quoted 9 lines
Quoted 8 lines

xen/include/public/gcov.h line 33

Quoted 5 lines
xen/include/public/gcov.h new file +5 -0
+#if BITS_PER_LONG >= 64
+typedef long gcov_type;
+#else
+typedef long long gcov_type;
+#endif

What's this???

Quoted 5 lines

This came from Linux, probably for Xen a simple

typedef long gcov_type

is ok.

No. Your goal - just to reiterate - must be not to use types the size of which depends on the guest word size.

Jan

#6 lore raw

PATCH 3/4 Implement code to read coverage informations

In reply to Frediano Ziglio

Implement 4 operations
- check if coverage is compiled in
- read size of coverage information
- read coverage information
- reset coverage counters

Information are stored in a single blob in a format specific to Xen designed to make code that generate coverage as small as possible.

Check does not check for any permission while other operations require a privileged domain. This cause these statistics can lead to security problems.

---
 xen/common/gcov/gcov.c    |  157 +++++++++++++++++++++++++++++++++++++++++++--
 xen/include/public/gcov.h |   24 +++++++
 2 files changed, 177 insertions(+), 4 deletions(-)
xen/common/gcov/gcov.c +153 -4
index e208596..d5d74d0 100644
@@ -19,10 +19,10 @@
 #include <xen/hypercall.h>
 #include <xen/gcov.h>
 #include <xen/errno.h>
+#include <xen/guest_access.h>
 #include <public/xen.h>
 
 static struct gcov_info *info_list;
-static unsigned num_info = 0;
 
 /*
  * __gcov_init is called by gcc-generated constructor code for each object
@@ -37,7 +37,6 @@ void __gcov_init(struct gcov_info *info)
     /* add new profiling data structure to list */
     info->next = info_list;
     info_list = info;
-    ++num_info;
 }
 
 /*
@@ -78,15 +77,165 @@ void init_coverage(void)
         __CTOR_LIST__.funcs[n]();
 
 #ifndef NDEBUG
-    printk(XENLOG_INFO "Initialized %u coverage strucures\n", num_info);
     if ( info_list )
         printk(XENLOG_INFO "First coverage file %s\n", info_list->filename);
 #endif
 }
 
+static inline int counter_active(const struct gcov_info *info, unsigned int type)
+{
+    return (1 << type) & info->ctr_mask;
+}
+
+DEFINE_XEN_GUEST_HANDLE(uint8_t);
+
+typedef struct write_iter_t
+{
+    XEN_GUEST_HANDLE(uint8_t) ptr;
+    int real;
+    unsigned write_offset;
+} write_iter_t;
+
+static int write_raw(struct write_iter_t *iter, const void *data,
+                     size_t data_len)
+{
+    if ( iter->real &&
+        copy_to_guest_offset(iter->ptr, iter->write_offset,
+                             (const unsigned char *) data, data_len) )
+        return -EFAULT;
+
+    iter->write_offset =
+        (iter->write_offset + data_len + (sizeof(uint32_t) - 1)) &
+        -sizeof(uint32_t);
+    return 0;
+}
+
+#define chk(v) do { ret=(v); if ( ret ) return ret; } while(0)
+
+static inline int write32(write_iter_t *iter, uint32_t val)
+{
+    return write_raw(iter, &val, sizeof(val));
+}
+
+static int write_string(write_iter_t *iter, const char *s)
+{
+    int ret;
+    size_t len = strlen(s);
+
+    chk(write32(iter, len));
+    return write_raw(iter, s, len);
+}
+
+static inline int next_type(const struct gcov_info *info, int *type)
+{
+    while ( ++*type < GCOV_COUNTERS && !counter_active(info, *type) )
+        continue;
+    return *type;
+}
+
+static int write_gcov(write_iter_t *iter)
+{
+    struct gcov_info *info;
+    int ret;
+
+    /* reset offset */
+    iter->write_offset = 0;
+
+    /* dump all files */
+    for ( info = info_list ; info; info = info->next )
+    {
+        const struct gcov_ctr_info *ctr;
+        int type;
+        size_t size_fn = sizeof(struct gcov_fn_info);
+
+        chk(write32(iter, GCOV_DATA_MAGIC));
+        chk(write32(iter, info->version));
+        chk(write32(iter, info->stamp));
+        chk(write_string(iter, info->filename));
+
+        /* dump counters */
+        ctr = info->counts;
+        for ( type = -1; next_type(info, &type) < GCOV_COUNTERS; ++ctr )
+        {
+            chk(write32(iter, GCOV_TAG_FOR_COUNTER(type)));
+            chk(write32(iter, ctr->num));
+            chk(write_raw(iter, ctr->values,
+                          ctr->num * sizeof(ctr->values[0])));
+
+            size_fn += sizeof(unsigned);
+        }
+
+        /* dump all functions together */
+        chk(write32(iter, GCOV_TAG_FUNCTION));
+        chk(write32(iter, info->n_functions));
+        chk(write_raw(iter, info->functions, info->n_functions * size_fn));
+    }
+
+    /* stop tag */
+    chk(write32(iter, 0));
+    return 0;
+}
+
+static int reset_counters(void)
+{
+    struct gcov_info *info;
+
+    for ( info = info_list ; info; info = info->next )
+    {
+        const struct gcov_ctr_info *ctr;
+        int type;
+
+        /* reset counters */
+        ctr = info->counts;
+        for ( type = -1; next_type(info, &type) < GCOV_COUNTERS; ++ctr )
+            memset(ctr->values, 0, ctr->num * sizeof(ctr->values[0]));
+    }
+
+    return 0;
+}
+
 long do_coverage_op(int op, XEN_GUEST_HANDLE_PARAM(void) uarg)
 {
-    return -EINVAL;
+    long ret = -EINVAL;
+    uint32_t len;
+    write_iter_t iter;
+
+    /*
+     * just return success to check for coverage support,
+     * no permission check
+     */
+    if ( op == XEN_COVERAGE_enabled )
+        return 0;
+
+    /*
+     * all other information although statistical one
+     * are too accurate and could lead to security issues
+     */
+    if ( !IS_PRIV(current->domain) )
+        return -EPERM;
+
+    switch ( op )
+    {
+    case XEN_COVERAGE_get_total_size:
+        iter.real = 0;
+
+        write_gcov(&iter);
+        len = iter.write_offset;
+        ret = copy_to_guest(uarg, &len, 1) ? -EFAULT : 0;
+        break;
+
+    case XEN_COVERAGE_read:
+        iter.ptr = guest_handle_cast(uarg, uint8_t);
+        iter.real = 1;
+
+        ret = write_gcov(&iter);
+        break;
+
+    case XEN_COVERAGE_reset:
+        ret = reset_counters();
+        break;
+    }
+    return ret;
 }
 
 #ifdef CONFIG_COMPAT
xen/include/public/gcov.h +24 -0
index 67aedf6..a2863f7 100644
@@ -90,4 +90,28 @@ struct gcov_info
     struct gcov_ctr_info      counts[0];
 };
 
+
+/*
+ * Check if coverage informations are available
+ * return just success or error, no parameters
+ */
+#define XEN_COVERAGE_enabled        0
+
+/*
+ * Get total size of information, to help allocate
+ * the buffer. The pointer points to a 32 bit value.
+ */
+#define XEN_COVERAGE_get_total_size 1
+
+/*
+ * Read coverage information in a single run
+ * You must use a tool to split them
+ */
+#define XEN_COVERAGE_read           2
+
+/*
+ * Reset all the coverage counters to 0
+ */
+#define XEN_COVERAGE_reset          3
+
 #endif /* XEN_PUBLIC_GCOV_H */
Signature · 1 line
1.7.9.5
#7 lore raw

PATCH 4/4 Use weak symbol to avoid bad trick in Makefiles

In reply to Frediano Ziglio

Define do_coverage_op to point to do_ni_hypercall as weak symbol. This allow the linked to redirect coverage operations to void function if coverage are not enabled.
This save some space and compression bytes if coverage is disabled.

---
 xen/common/Makefile        |    2 +-
 xen/common/compat/kernel.c |    1 +
 xen/common/gcov/Makefile   |    5 +----
 xen/common/gcov/nogcov.c   |   22 ----------------------
 xen/common/kernel.c        |    4 ++++
 5 files changed, 7 insertions(+), 27 deletions(-)
 delete mode 100644 xen/common/gcov/nogcov.c
xen/common/Makefile +1 -1
index 51191d6..8a0c506 100644
@@ -59,7 +59,7 @@ subdir-$(CONFIG_COMPAT) += compat
 
 subdir-$(x86_64) += hvm
 
-subdir-y += gcov
+subdir-$(coverage) += gcov
 
 subdir-y += libelf
 subdir-$(HAS_DEVICE_TREE) += libfdt
xen/common/compat/kernel.c +1 -0
index a2a2751..564c2a1 100644
@@ -42,6 +42,7 @@ CHECK_TYPE(domain_handle);
 #define xennmi_callback_t compat_nmi_callback_t
 
 #define DO(fn) int compat_##fn
+#define NAME(fn) "compat_" #fn
 #define COMPAT
 
 #include "../kernel.c"
xen/common/gcov/Makefile +1 -4
index 43a692c..c9efe6c 100644
@@ -1,5 +1,2 @@
-ifneq ($(coverage),y)
-obj-y += nogcov.o
-endif
-obj-$(coverage) += gcov.o
+obj-y += gcov.o
 
xen/common/gcov/nogcov.c deleted +0 -22
deleted file mode 100644
index 9642c96..0000000
@@ -1,22 +0,0 @@
-/*
- * nogov.c
- *
- * Compiled if coverage information are not enabled
- */
-
-#include <xen/config.h>
-#include <xen/lib.h>
-#include <xen/errno.h>
-#include <public/xen.h>
-
-long do_coverage_op(int op, XEN_GUEST_HANDLE_PARAM(void) uarg)
-{
-    return -ENOSYS;
-}
-
-#ifdef CONFIG_COMPAT
-int compat_coverage_op(int op, XEN_GUEST_HANDLE_PARAM(void) uarg)
-    __attribute__ ((alias ("do_coverage_op")));
-#endif
-
-
xen/common/kernel.c +4 -0
index 55caff6..25ba667 100644
@@ -197,6 +197,7 @@ void __init do_initcalls(void)
 }
 
 # define DO(fn) long do_##fn
+# define NAME(fn) "do_" #fn
 
 #endif
 
@@ -367,6 +368,9 @@ DO(ni_hypercall)(void)
     return -ENOSYS;
 }
 
+DO(coverage_op)(int op, XEN_GUEST_HANDLE_PARAM(void) uarg)
+    __attribute__ ((weak, alias(NAME(ni_hypercall))));
+
 /*
  * Local variables:
  * mode: C
Signature · 1 line
1.7.9.5
#8 lore raw
In reply to Frediano Ziglio
Quoted 10 lines · Frediano Ziglio

Updated set of patches for coverage.
Changes:
- implemented way of getting information
- use only 2 parameter in hypercall (still not sure it's better to use sysctl)
- use "coverage" for configuration option as suggested by Ian Campbell
- split header in include/public/gcov.h and include/xen/gcov.h to allow tools to include public header
- use weak to avoid defining compat_coverage_op

Quoted 7 lines

I manage to write 2 tools, one that call the new hypercall and dump the blob in a file, one that split this file into required .gcda files. Both very hack ones and dirty but works and prove the hypervisor code is fine. Should I just send them anyway?

The forth patch allow to consume just 0 bytes if coverage is disabled.

These are the two utilities. If anybody wants to try.

Frediano

/* get_coverage - program to get coverage informations from Xen
*
* Copyright (C) 2013 - Citrix Systems
* -----
*
* This program is free software; you can redistribute it and/or modify
* it under the terms of the GNU General Public License as published by
* the Free Software Foundation; either version 2 of the License, or
* (at your option) any later version.
*
* This program is distributed in the hope that it will be useful,
* but WITHOUT ANY WARRANTY; without even the implied warranty of
* MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
* GNU General Public License for more details.
*
* You should have received a copy of the GNU General Public License
* along with this program; if not, write to the Free Software
* Foundation, Inc., 675 Mass Ave, Cambridge, MA 02139, USA. */

#include <stdio.h>
#include <stdlib.h>
#include <sys/types.h>
#include <sys/stat.h>
#include <fcntl.h>
#include <errno.h>
#include <sys/ioctl.h>
#include <inttypes.h>
#include <sys/mman.h>
#include <err.h>

typedef struct privcmd_hypercall

{
	uint64_t op;
	uint64_t arg[5];
} privcmd_hypercall_t;
#define IOCTL_PRIVCMD_HYPERCALL								 \
	_IOC(_IOC_NONE, 'P', 0, sizeof(privcmd_hypercall_t))

#define PAGE_SIZE 4096u

static int privcmd_fd = -1;

int get_gcov_info(int op, void *ptr)

{
	privcmd_hypercall_t hc[1];
	int ret;

	hc->op = 40; // TODO
	hc->arg[0] = op;
	hc->arg[1] = (intptr_t) ptr;
	ret = ioctl(privcmd_fd, IOCTL_PRIVCMD_HYPERCALL, hc);
	if (ret >= 0)
		return ret;
	return -errno;
}

int main(void)

{
	// open priv command
	privcmd_fd = open("/proc/xen/privcmd", O_RDWR);
	if (privcmd_fd < 0)
		err(1, "opening privcmd");

	// check support
	if (get_gcov_info(0, NULL) < 0)
		err(1, "checking coverage support");

	// allocate
	uint8_t *p = mmap(0, PAGE_SIZE, PROT_WRITE|PROT_READ, MAP_PRIVATE|MAP_ANON|MAP_LOCKED, -1, 0);
	if (p == (uint8_t *) -1)
		err(1, "allocating memory");

	// get total length
	uint32_t *ui = (uint32_t *) p;
	*ui = 0;
	if (get_gcov_info(1, ui) < 0)
		err(1, "getting total length");
	printf("returned %u bytes\n", (unsigned) *ui);

	// safe check
	uint32_t total_len = *ui;
	if (total_len > 16u * 1024u * 1024u)
		errx(1, "coverage size too big %u bytes\n", total_len);

	// reallocate
	munmap(p, PAGE_SIZE);
	size_t size = total_len + PAGE_SIZE;
	size -= (size % PAGE_SIZE);
	p = mmap(0, size, PROT_WRITE|PROT_READ, MAP_PRIVATE|MAP_ANON|MAP_LOCKED, -1, 0);
	if (p == (uint8_t *) -1)
		err(1, "mapping memory for coverage");

	// get data
	if (get_gcov_info(2, p) < 0)
		err(1, "getting coverage information");

	// write to a file
	FILE *f = fopen("gcda.dat", "w");
	if (!f)
		err(1, "opening output file");
	if (fwrite(p, 1, total_len, f) != total_len)
		err(1, "writing coverage to file");
	fclose(f);

	return 0;
}

_______________________________________________
Xen-devel mailing list
[email protected]
http://lists.xen.org/xen-devel

#9 lore raw
Quoted 16 lines · Frediano Ziglio
Quoted 15 lines · Frediano Ziglio

Updated set of patches for coverage.
Changes:
- implemented way of getting information
- use only 2 parameter in hypercall (still not sure it's better to use sysctl)
- use "coverage" for configuration option as suggested by Ian Campbell
- split header in include/public/gcov.h and include/xen/gcov.h to allow tools to include public header
- use weak to avoid defining compat_coverage_op

I manage to write 2 tools, one that call the new hypercall and dump the blob in a file, one that split this file into required .gcda files. Both very hack ones and dirty but works and prove the hypervisor code is fine. Should I just send them anyway?

Quoted 4 lines
Quoted 2 lines

The forth patch allow to consume just 0 bytes if coverage is disabled.

These are the two utilities. If anybody wants to try.

Thanks, are you going to eventually post these for inclusion in the tree? (tools/misc or somewhere perhaps).

Ian.

#10 lore raw
In reply to Frediano Ziglio
Quoted 10 lines
Quoted 1 line
Quoted 1 line · Frediano Ziglio

Updated set of patches for coverage.
Changes:
- implemented way of getting information
- use only 2 parameter in hypercall (still not sure it's better to use sysctl)
- use "coverage" for configuration option as suggested by Ian Campbell
- split header in include/public/gcov.h and include/xen/gcov.h to allow tools to include public header
- use weak to avoid defining compat_coverage_op

While probably not formally documented anywhere, we decided against using weak some time ago.

Furthermore, iirc I already told you that with a suitably defined interface you won't need a compat variant at all (and in particular that would be the case if you went the sysctl route, as was suggested to you).

Jan

#11 lore raw
Quoted 14 lines · Jan Beulich
Quoted 10 lines
Quoted 1 line
Quoted 1 line · Frediano Ziglio

Updated set of patches for coverage.
Changes:
- implemented way of getting information
- use only 2 parameter in hypercall (still not sure it's better to use sysctl)
- use "coverage" for configuration option as suggested by Ian Campbell
- split header in include/public/gcov.h and include/xen/gcov.h to allow tools to include public header
- use weak to avoid defining compat_coverage_op

While probably not formally documented anywhere, we decided against using weak some time ago.

Quoted 7 lines

Furthermore, iirc I already told you that with a suitably defined interface you won't need a compat variant at all (and in particular that would be the case if you went the sysctl route, as was suggested to you).

Jan

I changed today the interface to use sysctl instead of directly hypercall. No weak symbols and easier makefiles. Still to pack updated version (see my mail about possible license issues).

Frediano