| Summary: | X32 and "gdb_static_assert (sizeof (nat_siginfo_t) == sizeof (siginfo_t))" | ||
|---|---|---|---|
| Product: | gdb | Reporter: | Jeffrey Walton <noloader> |
| Component: | gdb | Assignee: | Not yet assigned to anyone <unassigned> |
| Status: | RESOLVED FIXED | ||
| Severity: | normal | CC: | equationssolver, noloader, pedro, walfred.tedeschi |
| Priority: | P2 | ||
| Version: | 7.11.1 | ||
| Target Milestone: | 7.12 | ||
| Host: | Target: | ||
| Build: | Last reconfirmed: | ||
| Attachments: |
Logs for all components located with a `find ... -name 'config.log'`
Fix |
||
|
Description
Jeffrey Walton
2016-06-21 21:05:27 UTC
Created attachment 9348 [details]
Logs for all components located with a `find ... -name 'config.log'`
I can't use Debian supplied GDB because of http://bugs.debian.org/cgi-bin/bugreport.cgi?bug=799556. It was reported to GDB and promptly fixed, but it appears Debian never applied the upstream patch. I can't find the definition for nat_siginfo_t in /usr/include or /usr/lib.
Here' the definition of siginfo_t in the X32 environment. It may (or may not) be the same as x86_64.
$ cat siginfo_t.txt
typedef struct
{
int si_signo; /* Signal number. */
int si_errno; /* If non-zero, an errno value associated with
this signal, as defined in <errno.h>. */
int si_code; /* Signal code. */
union
{
int _pad[__SI_PAD_SIZE];
/* kill(). */
struct
{
__pid_t si_pid; /* Sending process ID. */
__uid_t si_uid; /* Real user ID of sending process. */
} _kill;
/* POSIX.1b timers. */
struct
{
int si_tid; /* Timer ID. */
int si_overrun; /* Overrun count. */
sigval_t si_sigval; /* Signal value. */
} _timer;
/* POSIX.1b signals. */
struct
{
__pid_t si_pid; /* Sending process ID. */
__uid_t si_uid; /* Real user ID of sending process. */
sigval_t si_sigval; /* Signal value. */
} _rt;
/* SIGCHLD. */
struct
{
__pid_t si_pid; /* Which child. */
__uid_t si_uid; /* Real user ID of sending process. */
int si_status; /* Exit value or signal. */
__sigchld_clock_t si_utime;
__sigchld_clock_t si_stime;
} _sigchld;
/* SIGILL, SIGFPE, SIGSEGV, SIGBUS. */
struct
{
void *si_addr; /* Faulting insn/memory ref. */
short int si_addr_lsb; /* Valid LSB of the reported address. */
struct
{
void *_lower;
void *_upper;
} si_addr_bnd;
} _sigfault;
/* SIGPOLL. */
struct
{
long int si_band; /* Band event for SIGPOLL. */
int si_fd;
} _sigpoll;
/* SIGSYS. */
struct
{
void *_call_addr; /* Calling user insn. */
int _syscall; /* Triggering system call number. */
unsigned int _arch; /* AUDIT_ARCH_* of syscall. */
} _sigsys;
} _sifields;
} siginfo_t __SI_ALIGNMENT;
Pedro, I think the assert has to be changed. We have to guarantee that what we need there is: gdb_static_assert (sizeof (nat_siginfo_t) >= sizeof (siginfo_t)) siginfo have to be anyway back compatible. In this way we get a fault when new siginfo is created, so we need to implement it. Do you agree? Thanks and regards, /Fred -----Original Message----- From: palves at redhat dot com [mailto:sourceware-bugzilla@sourceware.org] Sent: Wednesday, June 22, 2016 10:15 AM To: Tedeschi, Walfred <walfred.tedeschi@intel.com> Subject: [Bug gdb/20287] X32 and "gdb_static_assert (sizeof (nat_siginfo_t) == sizeof (siginfo_t))" https://sourceware.org/bugzilla/show_bug.cgi?id=20287 Pedro Alves <palves at redhat dot com> changed: What |Removed |Added ---------------------------------------------------------------------------- CC| |palves at redhat dot com, | |walfred.tedeschi at intel dot com Why are the sizes different in the first place? What's different between the types? It can be that x32 siginfo doe not implement the MPX specific part in it.
I.e. the _sigfault member could be written as:
struct
{
void *si_addr; /* Faulting insn/memory ref. */
short int si_addr_lsb; /* Valid LSB of the reported address. */
} _sigfault;
This is a first thought that comes into my mind.
Or we have new members not implemented yet in GDB.
It can be that x32 siginfo doe not implement the MPX specific part in it.
I.e. the _sigfault member could be written as:
struct
{
void *si_addr; /* Faulting insn/memory ref. */
short int si_addr_lsb; /* Valid LSB of the reported address. */
} _sigfault;
This is a first thought that comes into my mind.
Or we have new members not implemented yet in GDB.
New members should not matter, because they are added as new fields of a _union_, and the union has a field that ensures the size is fixed: int _pad[__SI_PAD_SIZE]; Looks like the sigsys was missing in the nat_sig_info.
Bellow it is nat_siginfo as in mainline.
48 typedef struct nat_siginfo
49 {
50 int si_signo;
51 int si_errno;
52 int si_code;
53
54 union
55 {
56 int _pad[((128 / sizeof (int)) - 4)];
57 /* kill() */
58 struct
59 {
60 unsigned int _pid;
61 unsigned int _uid;
62 } _kill;
63
64 /* POSIX.1b timers */
65 struct
66 {
67 nat_timer_t _tid;
68 int _overrun;
69 nat_sigval_t _sigval;
70 } _timer;
71
72 /* POSIX.1b signals */
73 struct
74 {
75 unsigned int _pid;
76 unsigned int _uid;
77 nat_sigval_t _sigval;
78 } _rt;
79
80 /* SIGCHLD */
81 struct
82 {
83 unsigned int _pid;
84 unsigned int _uid;
85 int _status;
86 nat_clock_t _utime;
87 nat_clock_t _stime;
88 } _sigchld;
89
90 /* SIGILL, SIGFPE, SIGSEGV, SIGBUS */
91 struct
92 {
93 nat_uptr_t _addr;
94 short int _addr_lsb;
95 struct
96 {
97 nat_uptr_t _lower;
98 nat_uptr_t _upper;
99 } si_addr_bnd;
100 } _sigfault;
101
102 /* SIGPOLL */
103 struct
104 {
105 int _band;
106 int _fd;
107 } _sigpoll;
108 } _sifields;
109 } nat_siginfo_t __attribute__ ((__aligned__ (8)));
sigsys is not there.
It shouldn't make a difference because the _pad field should guarantee a stable size. The whole siginfo_t object should be 128 bytes. Unless of course, the _pad field's definition is wrong , which sounds quite plausible: From above, it's: 56 int _pad[((128 / sizeof (int)) - 4)]; While /usr/include/bits/siginfo.h has: # define __SI_MAX_SIZE 128 # if __WORDSIZE == 64 # define __SI_PAD_SIZE ((__SI_MAX_SIZE / sizeof (int)) - 4) # else # define __SI_PAD_SIZE ((__SI_MAX_SIZE / sizeof (int)) - 3) # endif And I believe __WORDSIZE == 32 for x32. Yes, you are right. So we need to mimic this code in gdb. I can prepare a patch for that. > From above, it's: > > 56 int _pad[((128 / sizeof (int)) - 4)]; > > While /usr/include/bits/siginfo.h has: > > # define __SI_MAX_SIZE 128 > # if __WORDSIZE == 64 > # define __SI_PAD_SIZE ((__SI_MAX_SIZE / sizeof (int)) - 4) > # else > # define __SI_PAD_SIZE ((__SI_MAX_SIZE / sizeof (int)) - 3) > # endif > > And I believe __WORDSIZE == 32 for x32. Good find. Jeff All these different siginfo types give me headaches... :-) I don't know exactly which of the foo_siginfo_t types needs to be adjusted. If gdb is built as x32, isn't the layout of the siginfo_t object that PTRACE_GETSIGINFO returns the full 64-bit layout? If so, then isn't nat_siginfo_t all wrong for x32, since it uses x32 types (pointers, etc.)? Only the nat_siginfo has to be changed, in case we are compiling in x32 native.
All others have the right pad.
48 typedef struct nat_siginfo
49 {
50 int si_signo;
51 int si_errno;
52 int si_code;
53
54 union
55 {
56 int _pad[((128 / sizeof (int)) - 4)];
Here is the compatibility layer i.e. 32bit running on x86_64
138 typedef struct compat_siginfo
139 {
140 int si_signo;
141 int si_errno;
142 int si_code;
143
144 union
145 {
146 int _pad[((128 / sizeof (int)) - 3)];
For X32:
205 typedef struct compat_x32_siginfo
206 {
207 int si_signo;
208 int si_errno;
209 int si_code;
210
211 union
212 {
213 int _pad[((128 / sizeof (int)) - 3)];
214
In fact for x32 native the conversion should not be needed. I have to investigate it. Sof if gdb is built as x32, that's the layout of the siginfo_t object that PTRACE_GETSIGINFO returns? x32 layout, or full 64-bit layout? Bah, too many typos. Trying again: So if gdb is built as x32, what's the layout of the siginfo_t object that PTRACE_GETSIGINFO returns? x32 layout, or full 64-bit layout? I would consider that it would come as x32 layout. Otherwise we shouldn't have got the assert, or? Best would be to test it. No processing for x32 native siginfo should be done. This is in amd64-linux-nat.c. > Otherwise we shouldn't have got the assert, or? No. The assert is about the size of compile-time types. Even if we get the size right, if the kernel actually returns a 64-bit layout, then we'll interpret the object's contents incorrectly. > Best would be to test it. Yes, I think so. It might also be easy to determine from the kernel sources. Installed the Debian x32 and analyzed the headers, it looks like we have a int _pad[((128 / sizeof (int)) - 3)]; I intend to add the details tomorrow. In the Kernel sources we have: 1 #ifndef _ASM_X86_SIGINFO_H 2 #define _ASM_X86_SIGINFO_H 3 4 #ifdef __x86_64__ 5 # ifdef __ILP32__ /* x32 */ 6 typedef long long __kernel_si_clock_t __attribute__((aligned(4))); 7 # define __ARCH_SI_CLOCK_T __kernel_si_clock_t 8 # define __ARCH_SI_ATTRIBUTES __attribute__((aligned(8))) 9 # else /* x86-64 */ 10 # define __ARCH_SI_PREAMBLE_SIZE (4 * sizeof(int)) 11 # endif 12 #endif 13 14 #include <asm-generic/siginfo.h> 15 16 #endif /* _ASM_X86_SIGINFO_H */ http://lxr.free-electrons.com/source/arch/x86/include/uapi/asm/siginfo.h So for fix we could set the _pad conditionally as the glic does. (In reply to Walfred Tedeschi from comment #22) > In the Kernel sources we have: > > 1 #ifndef _ASM_X86_SIGINFO_H > 2 #define _ASM_X86_SIGINFO_H > 3 > 4 #ifdef __x86_64__ > 5 # ifdef __ILP32__ /* x32 */ > 6 typedef long long __kernel_si_clock_t __attribute__((aligned(4))); > 7 # define __ARCH_SI_CLOCK_T __kernel_si_clock_t > 8 # define __ARCH_SI_ATTRIBUTES __attribute__((aligned(8))) > 9 # else /* x86-64 */ > 10 # define __ARCH_SI_PREAMBLE_SIZE (4 * sizeof(int)) > 11 # endif > 12 #endif > 13 > 14 #include <asm-generic/siginfo.h> > 15 > 16 #endif /* _ASM_X86_SIGINFO_H */ > > http://lxr.free-electrons.com/source/arch/x86/include/uapi/asm/siginfo.h > > So for fix we could set the _pad conditionally as the glic does. I've got a fresh directory ready to test this. Where would you like it? Grep is turning up 0 hits: # grep -R '_ASM_X86_SIGINFO_H' * # grep -R '__ARCH_SI_ATTRIBUTES' * # debian-8-x32-vm:gdb-7.11# Can't see how that's related to what the kernel puts in the PTRACE_GET_SIGINFO request. That's the _userspace_ header. Those are _compile_ time #ifdefs, thus they can't affect whether ptrace returns an x32 siginfo or not _at runtime_. kernel/ptrace.c:ptrace_request handles PTRACE_GETSIGINFO: case PTRACE_GETSIGINFO: ret = ptrace_getsiginfo(child, &siginfo); if (!ret) ret = copy_siginfo_to_user(datavp, &siginfo); break; and I can't see any custom copy_siginfo_to_user under arch/x86/. There's a custom copy_siginfo_to_user32 implementation (arch/x86/kernel/signal_compat.c), but I think that that's for real 32-bit ptrace, not x32. This file changes the definition of the siginfo itself. It also change the _pad for 64 bit. the default one is the 32bit. This is what I could understand from the header. I am adding the default Kernel Header in here. The kernel is not compiled as a x32 binary... > and I can't see any custom copy_siginfo_to_user under arch/x86/. There's a
> custom copy_siginfo_to_user32 implementation (arch/x86/kernel
> /signal_compat.c), but I think that that's for real 32-bit ptrace, not x32.
OK, I think I see now.
arch/x86/kernel/ptrace.c has an x32_arch_ptrace function, that I assume is what is called when the _ptracer_ is a x32 process (not the kernel!).
That defers to compat_ptrace_request, which then ends up in copy_siginfo_to_user32, which _does_ handle x32:
int copy_siginfo_to_user32(compat_siginfo_t __user *to, const siginfo_t *from)
{
bool ia32 = test_thread_flag(TIF_IA32);
...
put_user_ex(from->si_utime, &to->si_utime);
put_user_ex(from->si_stime, &to->si_stime);
} else {
put_user_ex(from->si_utime, &to->_sifields._sigchld_x32._utime);
put_user_ex(from->si_stime, &to->_sifields._sigchld_x32._stime);
}
put_user_ex(from->si_status, &to->si_status);
...
So it does look plausible that PTRACE_GETSIGINFO returns a x32 siginfo if the ptracer is a x32 process.
Kernel is then 64bit. With a set of libraries we could also have 64bit applications running on the x32 systems, or? Would that be an use case? (In reply to Walfred Tedeschi from comment #28) > Kernel is then 64bit. > > With a set of libraries we could also have 64bit applications running on the > x32 systems, or? > > Would that be an use case? You will likely experience this issue on multiple platforms, and not just X32. Its a problem incumbent to the 32-bit data models (ILP32), and it should resurface on ARM with Aarch32 execution environments on AArch64. The data model is the reasons applications are advised to use the pair {__ILP32__,__x86_64__}, {__ILP32__,__aarch64__}, etc when determining the execution environment. OK, so PTRACE_GETSIGINFO returns a x32 siginfo_t object when the ptracer is a x32 process.
So the correct fix is indeed to make nat_siginfo_t match the x32 siginfo, when gdb is built as an x32 program.
However, it doesn't look like it's just the padding that is wrong. We have this:
/* For native 64-bit, clock_t in _sigchld is 64bit aligned at 4 bytes. */
typedef long __attribute__ ((__aligned__ (4))) nat_clock_t;
However, /usr/include/bits/siginfo.h has:
# if defined __x86_64__ && __WORDSIZE == 32
/* si_utime and si_stime must be 4 byte aligned for x32 to match the
kernel. We align siginfo_t to 8 bytes so that si_utime and si_stime
are actually aligned to 8 bytes since their offsets are multiple of
8 bytes. */
typedef __clock_t __attribute__ ((__aligned__ (4))) __sigchld_clock_t;
# define __SI_ALIGNMENT __attribute__ ((__aligned__ (8)))
# else
typedef __clock_t __sigchld_clock_t;
# define __SI_ALIGNMENT
# endif
So we're currently forcing 4-byte alignment on clock_t, when it should only be so for x32, not 64-bit?
(In reply to Jeffrey Walton from comment #29) > (In reply to Walfred Tedeschi from comment #28) > > Kernel is then 64bit. > > > > With a set of libraries we could also have 64bit applications running on the > > x32 systems, or? > > > > Would that be an use case? > > You will likely experience this issue on multiple platforms, and not just > X32. Its a problem incumbent to the 32-bit data models (ILP32), and it > should resurface on ARM with Aarch32 execution environments on AArch64. > > The data model is the reasons applications are advised to use the pair > {__ILP32__,__x86_64__}, {__ILP32__,__aarch64__}, etc when determining the > execution environment. One other use case to be aware of is this one... It bit me, and its why I [mostly] know how to detect the execution environment... Sun Studio and Clang always defines __ILP32__ when using a 32-bit data model. It will be defined for i386, i686, etc. Its why __x86_64__, __arm64__, __aarch64__, etc are needed to discriminate. It also diverges from System-V amd64 ABI guides at http://sites.google.com/site/x32abi/documents/abi.pdf. When I asked Clang devs about it, they stated they were not prohibited from defining __ILP32__ on 32-bit machines. I have an idea: - leave nat_siginfo only for 64-bit - add a siginfo-that-ptrace-uses typedef : #if __ILP32__ typedef compat_x32_siginfo_t ptrace_siginfo_t #else typedef nat_siginfo_t ptrace_siginfo_t #endif and then use ptrace_siginfo_t instead of nat_siginfo for the memcpys. Hand on while I try it out. Created attachment 9360 [details]
Fix
Like this. Can you give it a try?
(In reply to Pedro Alves from comment #33) > Created attachment 9360 [details] > Fix > > Like this. Can you give it a try? Applied to HEAD with `patch -p1 < 9360.diff`. It looks like there are some issues, but I _think_ some progress was made since the assert did not fire. However, when I commented out the assert in the 7.11.1 tarball, I did not encounter the issues below. ********** gcc -g -O2 -I. -I. -I./common -I./config -DLOCALEDIR="\"/usr/local/share/locale\"" -DHAVE_CONFIG_H -I./../include/opcode -I./../opcodes/.. -I./../readline/.. -I./../zlib -I../bfd -I./../bfd -I./../include -I../libdecnumber -I./../libdecnumber -I./gnulib/import -Ibuild-gnulib/import -Wall -Wpointer-arith -Wno-unused -Wunused-value -Wunused-function -Wno-switch -Wno-char-subscripts -Wempty-body -Wpointer-sign -Wmissing-prototypes -Wdeclaration-after-statement -Wmissing-parameter-type -Wold-style-declaration -Wold-style-definition -Wformat-nonliteral -c -o amd64-linux-siginfo.o -MT amd64-linux-siginfo.o -MMD -MP -MF .deps/amd64-linux-siginfo.Tpo ./nat/amd64-linux-siginfo.c ./nat/amd64-linux-siginfo.c:297:9: error: unknown type name ‘x32_siginfo_t’ typedef x32_siginfo_t ptrace_siginfo_t; ^ ./nat/amd64-linux-siginfo.c: In function ‘compat_siginfo_from_siginfo’: ./nat/amd64-linux-siginfo.c:313:29: error: request for member ‘si_signo’ in something not a structure or union to->si_signo = from_ptrace.si_signo; ^ ./nat/amd64-linux-siginfo.c:314:29: error: request for member ‘si_errno’ in something not a structure or union to->si_errno = from_ptrace.si_errno; ^ ./nat/amd64-linux-siginfo.c:315:28: error: request for member ‘si_code’ in something not a structure or union to->si_code = from_ptrace.si_code; ^ ./nat/amd64-linux-siginfo.c:319:39: error: request for member ‘_sifields’ in something not a structure or union to->cpt_si_timerid = from_ptrace.cpt_si_timerid; ^ ./nat/amd64-linux-siginfo.c:320:39: error: request for member ‘_sifields’ in something not a structure or union to->cpt_si_overrun = from_ptrace.cpt_si_overrun; ^ ./nat/amd64-linux-siginfo.c:321:46: error: request for member ‘_sifields’ in something not a structure or union to->cpt_si_ptr = (intptr_t) from_ptrace.cpt_si_ptr; ^ ./nat/amd64-linux-siginfo.c:325:35: error: request for member ‘_sifields’ in something not a structure or union to->cpt_si_pid = from_ptrace.cpt_si_pid; ^ ./nat/amd64-linux-siginfo.c:326:35: error: request for member ‘_sifields’ in something not a structure or union to->cpt_si_uid = from_ptrace.cpt_si_uid; ^ ./nat/amd64-linux-siginfo.c:330:35: error: request for member ‘_sifields’ in something not a structure or union to->cpt_si_pid = from_ptrace.cpt_si_pid; ^ ./nat/amd64-linux-siginfo.c:331:35: error: request for member ‘_sifields’ in something not a structure or union to->cpt_si_uid = from_ptrace.cpt_si_uid; ^ ./nat/amd64-linux-siginfo.c:332:46: error: request for member ‘_sifields’ in something not a structure or union to->cpt_si_ptr = (intptr_t) from_ptrace.cpt_si_ptr; ^ ./nat/amd64-linux-siginfo.c:339:32: error: request for member ‘_sifields’ in something not a structure or union to->cpt_si_pid = from_ptrace.cpt_si_pid; ^ ./nat/amd64-linux-siginfo.c:340:32: error: request for member ‘_sifields’ in something not a structure or union to->cpt_si_uid = from_ptrace.cpt_si_uid; ^ ./nat/amd64-linux-siginfo.c:341:35: error: request for member ‘_sifields’ in something not a structure or union to->cpt_si_status = from_ptrace.cpt_si_status; ^ ./nat/amd64-linux-siginfo.c:342:34: error: request for member ‘_sifields’ in something not a structure or union to->cpt_si_utime = from_ptrace.cpt_si_utime; ^ ./nat/amd64-linux-siginfo.c:343:34: error: request for member ‘_sifields’ in something not a structure or union to->cpt_si_stime = from_ptrace.cpt_si_stime; ^ ./nat/amd64-linux-siginfo.c:349:44: error: request for member ‘_sifields’ in something not a structure or union to->cpt_si_addr = (intptr_t) from_ptrace.cpt_si_addr; ^ ./nat/amd64-linux-siginfo.c:352:33: error: request for member ‘_sifields’ in something not a structure or union to->cpt_si_band = from_ptrace.cpt_si_band; ^ ./nat/amd64-linux-siginfo.c:353:31: error: request for member ‘_sifields’ in something not a structure or union to->cpt_si_fd = from_ptrace.cpt_si_fd; ^ ./nat/amd64-linux-siginfo.c:356:32: error: request for member ‘_sifields’ in something not a structure or union to->cpt_si_pid = from_ptrace.cpt_si_pid; ^ ./nat/amd64-linux-siginfo.c:357:32: error: request for member ‘_sifields’ in something not a structure or union to->cpt_si_uid = from_ptrace.cpt_si_uid; ^ ./nat/amd64-linux-siginfo.c:358:43: error: request for member ‘_sifields’ in something not a structure or union to->cpt_si_ptr = (intptr_t) from_ptrace.cpt_si_ptr; ^ ./nat/amd64-linux-siginfo.c: In function ‘siginfo_from_compat_siginfo’: ./nat/amd64-linux-siginfo.c:373:12: error: request for member ‘si_signo’ in something not a structure or union to_ptrace.si_signo = from->si_signo; ^ ./nat/amd64-linux-siginfo.c:374:12: error: request for member ‘si_errno’ in something not a structure or union to_ptrace.si_errno = from->si_errno; ^ ./nat/amd64-linux-siginfo.c:375:12: error: request for member ‘si_code’ in something not a structure or union to_ptrace.si_code = from->si_code; ^ ./nat/amd64-linux-siginfo.c:377:16: error: request for member ‘si_code’ in something not a structure or union if (to_ptrace.si_code == SI_TIMER) ^ ./nat/amd64-linux-siginfo.c:379:16: error: request for member ‘_sifields’ in something not a structure or union to_ptrace.cpt_si_timerid = from->cpt_si_timerid; ^ ./nat/amd64-linux-siginfo.c:380:16: error: request for member ‘_sifields’ in something not a structure or union to_ptrace.cpt_si_overrun = from->cpt_si_overrun; ^ ./nat/amd64-linux-siginfo.c:381:16: error: request for member ‘_sifields’ in something not a structure or union to_ptrace.cpt_si_ptr = (void *) (intptr_t) from->cpt_si_ptr; ^ ./nat/amd64-linux-siginfo.c:383:21: error: request for member ‘si_code’ in something not a structure or union else if (to_ptrace.si_code == SI_USER) ^ ./nat/amd64-linux-siginfo.c:385:16: error: request for member ‘_sifields’ in something not a structure or union to_ptrace.cpt_si_pid = from->cpt_si_pid; ^ ./nat/amd64-linux-siginfo.c:386:16: error: request for member ‘_sifields’ in something not a structure or union to_ptrace.cpt_si_uid = from->cpt_si_uid; ^ ./nat/amd64-linux-siginfo.c:388:16: error: request for member ‘si_code’ in something not a structure or union if (to_ptrace.si_code < 0) ^ ./nat/amd64-linux-siginfo.c:390:16: error: request for member ‘_sifields’ in something not a structure or union to_ptrace.cpt_si_pid = from->cpt_si_pid; ^ ./nat/amd64-linux-siginfo.c:391:16: error: request for member ‘_sifields’ in something not a structure or union to_ptrace.cpt_si_uid = from->cpt_si_uid; ^ ./nat/amd64-linux-siginfo.c:392:16: error: request for member ‘_sifields’ in something not a structure or union to_ptrace.cpt_si_ptr = (void *) (intptr_t) from->cpt_si_ptr; ^ ./nat/amd64-linux-siginfo.c:396:24: error: request for member ‘si_signo’ in something not a structure or union switch (to_ptrace.si_signo) ^ ./nat/amd64-linux-siginfo.c:399:13: error: request for member ‘_sifields’ in something not a structure or union to_ptrace.cpt_si_pid = from->cpt_si_pid; ^ ./nat/amd64-linux-siginfo.c:400:13: error: request for member ‘_sifields’ in something not a structure or union to_ptrace.cpt_si_uid = from->cpt_si_uid; ^ ./nat/amd64-linux-siginfo.c:401:13: error: request for member ‘_sifields’ in something not a structure or union to_ptrace.cpt_si_status = from->cpt_si_status; ^ ./nat/amd64-linux-siginfo.c:402:13: error: request for member ‘_sifields’ in something not a structure or union to_ptrace.cpt_si_utime = from->cpt_si_utime; ^ ./nat/amd64-linux-siginfo.c:403:13: error: request for member ‘_sifields’ in something not a structure or union to_ptrace.cpt_si_stime = from->cpt_si_stime; ^ ./nat/amd64-linux-siginfo.c:409:13: error: request for member ‘_sifields’ in something not a structure or union to_ptrace.cpt_si_addr = (void *) (intptr_t) from->cpt_si_addr; ^ ./nat/amd64-linux-siginfo.c:410:13: error: request for member ‘_sifields’ in something not a structure or union to_ptrace.cpt_si_addr_lsb = (short) from->cpt_si_addr_lsb; ^ ./nat/amd64-linux-siginfo.c:413:13: error: request for member ‘_sifields’ in something not a structure or union to_ptrace.cpt_si_band = from->cpt_si_band; ^ ./nat/amd64-linux-siginfo.c:414:13: error: request for member ‘_sifields’ in something not a structure or union to_ptrace.cpt_si_fd = from->cpt_si_fd; ^ ./nat/amd64-linux-siginfo.c:417:13: error: request for member ‘_sifields’ in something not a structure or union to_ptrace.cpt_si_pid = from->cpt_si_pid; ^ ./nat/amd64-linux-siginfo.c:418:13: error: request for member ‘_sifields’ in something not a structure or union to_ptrace.cpt_si_uid = from->cpt_si_uid; ^ ./nat/amd64-linux-siginfo.c:419:13: error: request for member ‘_sifields’ in something not a structure or union to_ptrace.cpt_si_ptr = (void* ) (intptr_t) from->cpt_si_ptr; ^ ./nat/amd64-linux-siginfo.c: In function ‘compat_x32_siginfo_from_siginfo’: ./nat/amd64-linux-siginfo.c:437:29: error: request for member ‘si_signo’ in something not a structure or union to->si_signo = from_ptrace.si_signo; ^ ./nat/amd64-linux-siginfo.c:438:29: error: request for member ‘si_errno’ in something not a structure or union to->si_errno = from_ptrace.si_errno; ^ ./nat/amd64-linux-siginfo.c:439:28: error: request for member ‘si_code’ in something not a structure or union to->si_code = from_ptrace.si_code; ^ ./nat/amd64-linux-siginfo.c:443:39: error: request for member ‘_sifields’ in something not a structure or union to->cpt_si_timerid = from_ptrace.cpt_si_timerid; ^ ./nat/amd64-linux-siginfo.c:444:39: error: request for member ‘_sifields’ in something not a structure or union to->cpt_si_overrun = from_ptrace.cpt_si_overrun; ^ ./nat/amd64-linux-siginfo.c:445:46: error: request for member ‘_sifields’ in something not a structure or union to->cpt_si_ptr = (intptr_t) from_ptrace.cpt_si_ptr; ^ ./nat/amd64-linux-siginfo.c:449:35: error: request for member ‘_sifields’ in something not a structure or union to->cpt_si_pid = from_ptrace.cpt_si_pid; ^ ./nat/amd64-linux-siginfo.c:450:35: error: request for member ‘_sifields’ in something not a structure or union to->cpt_si_uid = from_ptrace.cpt_si_uid; ^ ./nat/amd64-linux-siginfo.c:454:35: error: request for member ‘_sifields’ in something not a structure or union to->cpt_si_pid = from_ptrace.cpt_si_pid; ^ ./nat/amd64-linux-siginfo.c:455:35: error: request for member ‘_sifields’ in something not a structure or union to->cpt_si_uid = from_ptrace.cpt_si_uid; ^ ./nat/amd64-linux-siginfo.c:456:46: error: request for member ‘_sifields’ in something not a structure or union to->cpt_si_ptr = (intptr_t) from_ptrace.cpt_si_ptr; ^ ./nat/amd64-linux-siginfo.c:463:32: error: request for member ‘_sifields’ in something not a structure or union to->cpt_si_pid = from_ptrace.cpt_si_pid; ^ ./nat/amd64-linux-siginfo.c:464:32: error: request for member ‘_sifields’ in something not a structure or union to->cpt_si_uid = from_ptrace.cpt_si_uid; ^ ./nat/amd64-linux-siginfo.c:465:35: error: request for member ‘_sifields’ in something not a structure or union to->cpt_si_status = from_ptrace.cpt_si_status; ^ ./nat/amd64-linux-siginfo.c:466:43: error: request for member ‘_sifields’ in something not a structure or union memcpy (&to->cpt_si_utime, &from_ptrace.cpt_si_utime, ^ ./nat/amd64-linux-siginfo.c:468:43: error: request for member ‘_sifields’ in something not a structure or union memcpy (&to->cpt_si_stime, &from_ptrace.cpt_si_stime, ^ ./nat/amd64-linux-siginfo.c:475:44: error: request for member ‘_sifields’ in something not a structure or union to->cpt_si_addr = (intptr_t) from_ptrace.cpt_si_addr; ^ ./nat/amd64-linux-siginfo.c:478:33: error: request for member ‘_sifields’ in something not a structure or union to->cpt_si_band = from_ptrace.cpt_si_band; ^ ./nat/amd64-linux-siginfo.c:479:31: error: request for member ‘_sifields’ in something not a structure or union to->cpt_si_fd = from_ptrace.cpt_si_fd; ^ ./nat/amd64-linux-siginfo.c:482:32: error: request for member ‘_sifields’ in something not a structure or union to->cpt_si_pid = from_ptrace.cpt_si_pid; ^ ./nat/amd64-linux-siginfo.c:483:32: error: request for member ‘_sifields’ in something not a structure or union to->cpt_si_uid = from_ptrace.cpt_si_uid; ^ ./nat/amd64-linux-siginfo.c:484:43: error: request for member ‘_sifields’ in something not a structure or union to->cpt_si_ptr = (intptr_t) from_ptrace.cpt_si_ptr; ^ ./nat/amd64-linux-siginfo.c: In function ‘siginfo_from_compat_x32_siginfo’: ./nat/amd64-linux-siginfo.c:501:12: error: request for member ‘si_signo’ in something not a structure or union to_ptrace.si_signo = from->si_signo; ^ ./nat/amd64-linux-siginfo.c:502:12: error: request for member ‘si_errno’ in something not a structure or union to_ptrace.si_errno = from->si_errno; ^ ./nat/amd64-linux-siginfo.c:503:12: error: request for member ‘si_code’ in something not a structure or union to_ptrace.si_code = from->si_code; ^ ./nat/amd64-linux-siginfo.c:505:16: error: request for member ‘si_code’ in something not a structure or union if (to_ptrace.si_code == SI_TIMER) ^ ./nat/amd64-linux-siginfo.c:507:16: error: request for member ‘_sifields’ in something not a structure or union to_ptrace.cpt_si_timerid = from->cpt_si_timerid; ^ ./nat/amd64-linux-siginfo.c:508:16: error: request for member ‘_sifields’ in something not a structure or union to_ptrace.cpt_si_overrun = from->cpt_si_overrun; ^ ./nat/amd64-linux-siginfo.c:509:16: error: request for member ‘_sifields’ in something not a structure or union to_ptrace.cpt_si_ptr = (void *) (intptr_t) from->cpt_si_ptr; ^ ./nat/amd64-linux-siginfo.c:511:21: error: request for member ‘si_code’ in something not a structure or union else if (to_ptrace.si_code == SI_USER) ^ ./nat/amd64-linux-siginfo.c:513:16: error: request for member ‘_sifields’ in something not a structure or union to_ptrace.cpt_si_pid = from->cpt_si_pid; ^ ./nat/amd64-linux-siginfo.c:514:16: error: request for member ‘_sifields’ in something not a structure or union to_ptrace.cpt_si_uid = from->cpt_si_uid; ^ ./nat/amd64-linux-siginfo.c:516:16: error: request for member ‘si_code’ in something not a structure or union if (to_ptrace.si_code < 0) ^ ./nat/amd64-linux-siginfo.c:518:16: error: request for member ‘_sifields’ in something not a structure or union to_ptrace.cpt_si_pid = from->cpt_si_pid; ^ ./nat/amd64-linux-siginfo.c:519:16: error: request for member ‘_sifields’ in something not a structure or union to_ptrace.cpt_si_uid = from->cpt_si_uid; ^ ./nat/amd64-linux-siginfo.c:520:16: error: request for member ‘_sifields’ in something not a structure or union to_ptrace.cpt_si_ptr = (void *) (intptr_t) from->cpt_si_ptr; ^ ./nat/amd64-linux-siginfo.c:524:24: error: request for member ‘si_signo’ in something not a structure or union switch (to_ptrace.si_signo) ^ ./nat/amd64-linux-siginfo.c:527:13: error: request for member ‘_sifields’ in something not a structure or union to_ptrace.cpt_si_pid = from->cpt_si_pid; ^ ./nat/amd64-linux-siginfo.c:528:13: error: request for member ‘_sifields’ in something not a structure or union to_ptrace.cpt_si_uid = from->cpt_si_uid; ^ ./nat/amd64-linux-siginfo.c:529:13: error: request for member ‘_sifields’ in something not a structure or union to_ptrace.cpt_si_status = from->cpt_si_status; ^ ./nat/amd64-linux-siginfo.c:530:22: error: request for member ‘_sifields’ in something not a structure or union memcpy (&to_ptrace.cpt_si_utime, &from->cpt_si_utime, ^ ./nat/amd64-linux-siginfo.c:531:22: error: request for member ‘_sifields’ in something not a structure or union sizeof (to_ptrace.cpt_si_utime)); ^ ./nat/amd64-linux-siginfo.c:532:22: error: request for member ‘_sifields’ in something not a structure or union memcpy (&to_ptrace.cpt_si_stime, &from->cpt_si_stime, ^ ./nat/amd64-linux-siginfo.c:533:22: error: request for member ‘_sifields’ in something not a structure or union sizeof (to_ptrace.cpt_si_stime)); ^ ./nat/amd64-linux-siginfo.c:539:13: error: request for member ‘_sifields’ in something not a structure or union to_ptrace.cpt_si_addr = (void *) (intptr_t) from->cpt_si_addr; ^ ./nat/amd64-linux-siginfo.c:542:13: error: request for member ‘_sifields’ in something not a structure or union to_ptrace.cpt_si_band = from->cpt_si_band; ^ ./nat/amd64-linux-siginfo.c:543:13: error: request for member ‘_sifields’ in something not a structure or union to_ptrace.cpt_si_fd = from->cpt_si_fd; ^ ./nat/amd64-linux-siginfo.c:546:13: error: request for member ‘_sifields’ in something not a structure or union to_ptrace.cpt_si_pid = from->cpt_si_pid; ^ ./nat/amd64-linux-siginfo.c:547:13: error: request for member ‘_sifields’ in something not a structure or union to_ptrace.cpt_si_uid = from->cpt_si_uid; ^ ./nat/amd64-linux-siginfo.c:548:13: error: request for member ‘_sifields’ in something not a structure or union to_ptrace.cpt_si_ptr = (void* ) (intptr_t) from->cpt_si_ptr; ^ In file included from ./common/common-defs.h:71:0, from ./nat/amd64-linux-siginfo.c:21: ./nat/amd64-linux-siginfo.c: At top level: ./common/gdb_assert.h:26:14: error: size of array ‘never_defined_just_used_for_checking’ is negative extern int never_defined_just_used_for_checking[(expr) ? 1 : -1] ^ ./nat/amd64-linux-siginfo.c:597:1: note: in expansion of macro ‘gdb_static_assert’ gdb_static_assert (sizeof (ptrace_siginfo_t) == GDB_SI_SIZE); ^ Makefile:2370: recipe for target 'amd64-linux-siginfo.o' failed make[2]: *** [amd64-linux-siginfo.o] Error 1 make[2]: Leaving directory '/gdb-7.11/gdb' Makefile:9156: recipe for target 'all-gdb' failed make[1]: *** [all-gdb] Error 2 make[1]: Leaving directory '/gdb-7.11' Makefile:846: recipe for target 'all' failed make: *** [all] Error 2 (In reply to Jeffrey Walton from comment #34) > (In reply to Pedro Alves from comment #33) > > Created attachment 9360 [details] > > Fix > > > > Like this. Can you give it a try? > > Applied to HEAD with `patch -p1 < 9360.diff`. > > It looks like there are some issues, but I _think_ some progress was made > since the assert did not fire. > > However, when I commented out the assert in the 7.11.1 tarball, I did not > encounter the issues below. Cancel this... I cd'd into gdb, and not binutils-gdb. That probably because I made a typo. Where it reads: typedef x32_siginfo_t ptrace_siginfo_t; Try changing it to: typedef compat_x32_siginfo_t ptrace_siginfo_t; (In reply to Pedro Alves from comment #33) > Created attachment 9360 [details] > Fix > > Like this. Can you give it a try? (In reply to Pedro Alves from comment #33) > Created attachment 9360 [details] > Fix > > Like this. Can you give it a try? OK, now I'm hitting the MAKEINFO bug briefly discussed at https://sourceware.org/bugzilla/show_bug.cgi?id=18987#c4. Any ideas? *********** # ./configure MAKEINFO=false 2>&1 | tee binutils-gdb.txt ... # make 2>&1 | tee binutils-gdb.txt ... /binutils-gdb/missing: 81: /binutils-gdb/missing: makeinfo: not found WARNING: 'makeinfo' is missing on your system. You should only need it if you modified a '.texi' file, or any other file indirectly affecting the aspect of the manual. You might want to install the Texinfo package: <http://www.gnu.org/software/texinfo/> The spurious makeinfo call might also be the consequence of using a buggy 'make' (AIX, DU, IRIX), in which case you might want to install GNU make: <http://www.gnu.org/software/make/> Makefile:472: recipe for target 'bfd.info' failed make[3]: *** [bfd.info] Error 127 make[3]: Leaving directory '/binutils-gdb/bfd/doc' Making info in po make[3]: Entering directory '/binutils-gdb/bfd/po' ( if test 'x.' != 'x.'; then \ posrcprefix='../'; \ else \ posrcprefix="../"; \ fi; \ rm -f SRC-POTFILES-t SRC-POTFILES \ && (sed -e '/^#/d' \ -e '/^[ ]*$/d' \ -e "s@.*@ $posrcprefix& \\\\@" < ./SRC-POTFILES.in \ | sed -e '$s/\\$//') > SRC-POTFILES-t \ && chmod a-w SRC-POTFILES-t \ && mv SRC-POTFILES-t SRC-POTFILES ) ( rm -f BLD-POTFILES-t BLD-POTFILES \ && (sed -e '/^#/d' \ -e '/^[ ]*$/d' \ -e "s@.*@ ../& \\\\@" < ./BLD-POTFILES.in \ | sed -e '$s/\\$//') > BLD-POTFILES-t \ && chmod a-w BLD-POTFILES-t \ && mv BLD-POTFILES-t BLD-POTFILES ) cd .. \ && CONFIG_FILES=po/Makefile.in:po/Make-in \ CONFIG_HEADERS= /bin/bash ./config.status config.status: creating po/Makefile.in config.status: executing depfiles commands config.status: executing libtool commands config.status: executing default-1 commands config.status: executing bfd_stdint.h commands config.status: executing default commands make[3]: Nothing to be done for 'info'. make[3]: Leaving directory '/binutils-gdb/bfd/po' make[3]: Entering directory '/binutils-gdb/bfd' make[3]: Nothing to be done for 'info-am'. make[3]: Leaving directory '/binutils-gdb/bfd' Makefile:1711: recipe for target 'info-recursive' failed make[2]: *** [info-recursive] Error 1 make[2]: Leaving directory '/binutils-gdb/bfd' Makefile:2727: recipe for target 'all-bfd' failed make[1]: *** [all-bfd] Error 2 make[1]: Leaving directory '/binutils-gdb' Makefile:850: recipe for target 'all' failed make: *** [all] Error 2 (In reply to Jeffrey Walton from comment #37) > (In reply to Pedro Alves from comment #33) > > Created attachment 9360 [details] > > Fix > > > > Like this. Can you give it a try? > > (In reply to Pedro Alves from comment #33) > > Created attachment 9360 [details] > > Fix > > > > Like this. Can you give it a try? > > OK, now I'm hitting the MAKEINFO bug briefly discussed at > https://sourceware.org/bugzilla/show_bug.cgi?id=18987#c4. > Confirmed with a: # make distclean # git reset --hard HEAD # ./configure MAKEINFO=false Confirmed that the patch fixes the x32 issue, or confirmed that you still observe the makeinfo issue? BTW, I think the makeinfo workaround is MAKEINFO=true, not MAKEINFO=false. (In reply to Pedro Alves from comment #39) > Confirmed that the patch fixes the x32 issue, or confirmed that you still > observe the makeinfo issue? My bad... Confirmed the MAKEINFO issue. X32 does not supply it. (Its the path of least resistance; I tried it first). > BTW, I think the makeinfo workaround is MAKEINFO=true, not MAKEINFO=false. That's very non-intuitive if I am parsing it correctly: when you *don't* want to build the info files, you specify `MAKEINFO=true`. And when you do want to build the info files, you say `MAKEINFO=false`. Is that correct? I can't help but feel configure should be doing it since that is it's job. the `makeinfo` program is missing; there no sense trying to build the info files. MAKEINFO is not a boolean flag. What this does is override the path to the "makeinfo" binary, to invoke /usr/bin/true instead. Maybe "MAKEINFO=/usr/bin/true" would have been clearer. (In reply to Pedro Alves from comment #41) > MAKEINFO is not a boolean flag. What this does is override the path to the > "makeinfo" binary, to invoke /usr/bin/true instead. > Maybe "MAKEINFO=/usr/bin/true" would have been clearer. Ah, thanks. That makes more sense. So... # make distclean # git reset --hard HEAD # ls /bin/true /bin/true # ./configure MAKEINFO=/bin/true ... # make ... It ends with: /binutils-gdb/missing: 81: /binutils-gdb/missing: makeinfo: not found WARNING: 'makeinfo' is missing on your system. You should only need it if you modified a '.texi' file, or any other file indirectly affecting the aspect of the manual. You might want to install the Texinfo package: <http://www.gnu.org/software/texinfo/> The spurious makeinfo call might also be the consequence of using a buggy 'make' (AIX, DU, IRIX), in which case you might want to install GNU make: <http://www.gnu.org/software/make/> Makefile:472: recipe for target 'bfd.info' failed make[3]: *** [bfd.info] Error 127 make[3]: Leaving directory '/binutils-gdb/bfd/doc' Making info in po make[3]: Entering directory '/binutils-gdb/bfd/po' ( if test 'x.' != 'x.'; then \ posrcprefix='../'; \ else \ posrcprefix="../"; \ fi; \ rm -f SRC-POTFILES-t SRC-POTFILES \ && (sed -e '/^#/d' \ -e '/^[ ]*$/d' \ -e "s@.*@ $posrcprefix& \\\\@" < ./SRC-POTFILES.in \ | sed -e '$s/\\$//') > SRC-POTFILES-t \ && chmod a-w SRC-POTFILES-t \ && mv SRC-POTFILES-t SRC-POTFILES ) ( rm -f BLD-POTFILES-t BLD-POTFILES \ && (sed -e '/^#/d' \ -e '/^[ ]*$/d' \ -e "s@.*@ ../& \\\\@" < ./BLD-POTFILES.in \ | sed -e '$s/\\$//') > BLD-POTFILES-t \ && chmod a-w BLD-POTFILES-t \ && mv BLD-POTFILES-t BLD-POTFILES ) cd .. \ && CONFIG_FILES=po/Makefile.in:po/Make-in \ CONFIG_HEADERS= /bin/bash ./config.status config.status: creating po/Makefile.in config.status: executing depfiles commands config.status: executing libtool commands config.status: executing default-1 commands config.status: executing bfd_stdint.h commands config.status: executing default commands make[3]: Nothing to be done for 'info'. make[3]: Leaving directory '/binutils-gdb/bfd/po' make[3]: Entering directory '/binutils-gdb/bfd' make[3]: Nothing to be done for 'info-am'. make[3]: Leaving directory '/binutils-gdb/bfd' Makefile:1711: recipe for target 'info-recursive' failed make[2]: *** [info-recursive] Error 1 make[2]: Leaving directory '/binutils-gdb/bfd' Makefile:2727: recipe for target 'all-bfd' failed make[1]: *** [all-bfd] Error 2 make[1]: Leaving directory '/binutils-gdb' Makefile:850: recipe for target 'all' failed make: *** [all] Error 2 Should I open another bug report for the MAKEINFO problems?
> Should I open another bug report for the MAKEINFO problems?
For completeness, 'make distclean' appears to be doing its job:
# grep -IR 'MAKEINFO=false' *
#
# grep -IR 'MAKEINFO=/bin/true' *
config.log: $ ./configure MAKEINFO=/bin/true
config.log:ac_cv_prog_MAKEINFO=/bin/true
config.log:TOPLEVEL_CONFIGURE_ARGUMENTS='./configure MAKEINFO=/bin/true'
config.status: with options \"'MAKEINFO=/bin/true'\"
config.status: set X '/bin/bash' './configure' 'MAKEINFO=/bin/true' $ac_configure_extra_args --no-create --no-recursion
config.status:S["TOPLEVEL_CONFIGURE_ARGUMENTS"]="./configure MAKEINFO=/bin/true"
Makefile:TOPLEVEL_CONFIGURE_ARGUMENTS=./configure MAKEINFO=/bin/true
#
Hi any update on that issue? Waiting for someone to confirm that the patch (with the fix mentioned in comment 36 on top) fixes the x32 issue. Hi, Now applying the patch and changing the typedef x32_siginfo_t ptrace_siginfo_t to typedef compat_x32_siginfo_t ptrace_siginfo_t; I can at least build the cross-complied gdbserver for x32 with build as i386-linux and target as i686-pc-linux-gnu But gdbserver gives following warning, gdbserver: linux_ptrace_test_ret_to_nx: Cannot PTRACE_PEEKUSER: Input/output error and gives segmentation fault during debugging even on single instruction step command. e.g 1> Launch gdbserver ./gdbserver :10000 test gdbserver: linux_ptrace_test_ret_to_nx: Cannot PTRACE_PEEKUSER: Input/output error Process test created; pid = 473 Listening on port 10000 now after connecting to gdb and on si command it gives Remote debugging from host 127.0.0.1 Segmentation fault (core dumped) I am looking at it tomorrow 28.07. I installed Ubuntu 16.04 in a vm this morning to take a look at this. Ubuntu has enough x32 runtime packages (glibc, libgcc, etc.) making it easy to build an x32 gdb. I saw that the patch needs a couple tweaks for (remove casts through void *, for C++). > linux_ptrace_test_ret_to_nx: Cannot PTRACE_PEEKUSER: Input/output error Filed Bug 20413 for this one. The master branch has been updated by Pedro Alves <palves@sourceware.org>: https://sourceware.org/git/gitweb.cgi?p=binutils-gdb.git;h=9cf12d57c58a82cfe3e6fee26d1ea55dfe49f9c4 commit 9cf12d57c58a82cfe3e6fee26d1ea55dfe49f9c4 Author: Pedro Alves <palves@redhat.com> Date: Tue Jul 26 19:35:40 2016 +0100 Fix PR gdb/20287 - x32 and "gdb_static_assert (sizeof (nat_siginfo_t) == sizeof (siginfo_t))" Building an x32 gdb trips on a static assertion: In file included from .../src/gdb/common/common-defs.h:71:0, from .../src/gdb/nat/amd64-linux-siginfo.c:21: .../src/gdb/common/gdb_assert.h:26:66: error: size of array ‘never_defined_just_used_for_checking’ is negative extern int never_defined_just_used_for_checking[(expr) ? 1 : -1] ^ .../src/gdb/nat/amd64-linux-siginfo.c:113:1: note: in expansion of macro ‘gdb_static_assert’ gdb_static_assert (sizeof (nat_siginfo_t) == sizeof (siginfo_t)); ^ The problem is that the way nat_siginfo_t is defined, it can only match the host's siginfo_t object when gdb is built as a 64-bit program. Several bits of nat_siginfo_t are off: - nat_siginfo_t's _pad field's definition is: int _pad[((128 / sizeof (int)) - 4)]; while /usr/include/bits/siginfo.h has: # define __SI_MAX_SIZE 128 # if __WORDSIZE == 64 # define __SI_PAD_SIZE ((__SI_MAX_SIZE / sizeof (int)) - 4) # else # define __SI_PAD_SIZE ((__SI_MAX_SIZE / sizeof (int)) - 3) # endif and __WORDSIZE == 32 for x32. This is what causes the size of nat_siginfo_t to be wrong and the assertion to fail. - the nat_clock_t type is incorrect for 64-bit. We have this: /* For native 64-bit, clock_t in _sigchld is 64bit aligned at 4 bytes. */ typedef long __attribute__ ((__aligned__ (4))) nat_clock_t; however, /usr/include/bits/siginfo.h has: # if defined __x86_64__ && __WORDSIZE == 32 /* si_utime and si_stime must be 4 byte aligned for x32 to match the kernel. We align siginfo_t to 8 bytes so that si_utime and si_stime are actually aligned to 8 bytes since their offsets are multiple of 8 bytes. */ typedef __clock_t __attribute__ ((__aligned__ (4))) __sigchld_clock_t; # define __SI_ALIGNMENT __attribute__ ((__aligned__ (8))) # else typedef __clock_t __sigchld_clock_t; # define __SI_ALIGNMENT # endif So we're currently forcing 4-byte alignment on clock_t, when it should only be so for x32, not 64-bit. The fix: - Leaves nat_siginfo_t strictly for the 64-bit ABI. - Adds a new typedef for the siginfo type that ptrace uses (ptrace_siginfo_t). An x32 gdb always gets/sets an x32 siginfo_t type with PTRACE_GETSIGINFO/PTRACE_SETSIGINFO. - Uses this new ptrace_siginfo_t type instead of nat_siginfo_t as the intermediate conversion type. gdb/ChangeLog: 2016-07-26 Pedro Alves <palves@redhat.com> * amd64-linux-nat.c (amd64_linux_siginfo_fixup): Rename 'native' parameter to 'ptrace'. * nat/amd64-linux-siginfo.c (GDB_SI_SIZE): New define. (nat_uptr_t): New an unsigned long. (nat_clock_t): Remove attribute __aligned__. (struct nat_timeval): Delete. (nat_siginfo_t): Remove attribute __aligned__. (ptrace_siginfo_t): Define. (compat_siginfo_from_siginfo, siginfo_from_compat_siginfo) (compat_x32_siginfo_from_siginfo) (siginfo_from_compat_x32_siginfo): Make 'from' parameter const. Convert through a ptrace_siginfo_t instead of a nat_siginfo_t. Remove casts. (amd64_linux_siginfo_fixup_common): Rename 'native' parameter to 'ptrace'. Remove static assertions. (top level): New static assertions. gdb/gdbserver/ChangeLog: 2016-07-26 Pedro Alves <palves@redhat.com> * linux-x86-low.c (x86_siginfo_fixup): Rename 'native' parameter to 'ptrace'. Fixed in master. (In reply to hmb from comment #46) > and gives segmentation fault during debugging even on single instruction > step command. Filed Bug 20414 for this one. All fixed. Please file separate bugs for further issues you may find. Thanks Pedro for doing that. |