* (patch) hpjyg21: FAT_FREE_PSYMTABS
@ 1999-11-11 12:53 Jimmy Guo
1999-11-11 15:10 ` Andrew Cagney
0 siblings, 1 reply; 2+ messages in thread
From: Jimmy Guo @ 1999-11-11 12:53 UTC (permalink / raw)
To: gdb-patches
***
Patch dependecies: hpjyg05 (config/pa/tm-hppa.h)
hpjyg16 (symtab.c)
hpjyg20 (symtab.c)
***
This patch contains:
- addition of codes ifdef'd by FAT_FREE_PSYMTABS to rely on linker
symbol table entries to determine which translation unit's partial
symbol table must be expanded. Here is a more complete overview of
this feature:
We use the partial symbol tables to identify which TU's partial
symbol table must be expanded to get further information about a
symbol. This decision can also be made by consulting the minimal
symbol table, determining the address of the symbol therefrom
and finding out which partial symbol table's address range
(texthigh & textlow) envelops the symbol's address.
If we are successful in using the minimal symbol table to zoom
in on which psymtab to expand, then we could eliminate the
"kernels" of these N partial symbol tables and build just the shell.
ChangeLog:
1999-11-11 Jimmy Guo <guo@cup.hp.com>
* symtab.c: FAT_FREE_PSYMTABS support.
(find_pc_sect_psymtab): If find_pc_sect_psymbol failed to find a
psymtab, check textlow and texthigh and return tpst if it houses
pc + 4.
(find_main_psymtab): Fall back to linker symbol table to find
the psymtab housing psymbol.
(find_functions): Iterate over linker symbol table to expand the
psymtab containing psymbol.
(expand_containing_psymtab): New function.
* minsyms.c (foreach_text_minsym): New function.
* config/pa/tm-hppa.h (FAT_FREE_PSYMTABS): Define.
Index: gdb/symtab.c
/opt/gnu/bin/diff -r -c -N /view/guo.wdb.c//CLO/Components/WDB/Src/gnu/gdb/symtab.c gdb/symtab.c
*** /view/guo.wdb.c//CLO/Components/WDB/Src/gnu/gdb/symtab.c Thu Nov 11 12:45:24 1999
--- gdb/symtab.c Thu Nov 11 12:34:12 1999
***************
*** 449,454 ****
--- 449,477 ----
p = find_pc_sect_psymbol (tpst, pc, section);
+ #ifdef FAT_FREE_PSYMTABS
+
+ /* srikanth, in the case of the Wildebeest, the psymtabs could be empty
+ and hence the call to find_pc_sect_psymbol() could fail to find anything.
+ Due potentially to a combination of bugs in compilers, linkers, loaders,
+ editors and debugger engineers :-), sometimes there is an overlap of
+ textlow and texthigh values between and multiple psymtabs. As a result,
+ if we are handed an address equal to the texthigh or textlow, we could
+ end up identifying the wrong psymtab.
+
+ To be absolutely sure of the identity of the psymtab, let us verify that
+ the one we narrow down on, also houses pc + 4.
+ */
+
+ if (p == NULL)
+ {
+ CORE_ADDR pc_plus4 = pc + 4;
+ if (pc_plus4 >= tpst->textlow &&
+ pc_plus4 <= tpst->texthigh)
+ return tpst;
+ }
+ #endif
+
if (p != NULL
&& SYMBOL_VALUE_ADDRESS (p)
== SYMBOL_VALUE_ADDRESS (msymbol))
***************
*** 1367,1372 ****
--- 1390,1408 ----
}
}
+ #ifdef FAT_FREE_PSYMTABS
+
+ /* srikanth, under the new world order, psymtabs could be empty
+ for the Wildebeest. Fall back on the linker symbol table, use the
+ address therefrom to find out which psymtab would have housed the
+ psymbol were one to exist.
+ */
+
+ if ((m = lookup_minimal_symbol_text (default_main, NULL, NULL)))
+ return find_pc_psymtab (SYMBOL_VALUE_ADDRESS (m));
+
+ #endif
+
return (NULL);
}
***************
*** 2725,2730 ****
--- 2761,2785 ----
return (i1);
}
+ #ifdef FAT_FREE_PSYMTABS
+
+ static void
+ expand_containing_psymtab (m)
+ struct minimal_symbol *m;
+ {
+ struct objfile *objfile;
+ struct partial_symtab *pst = NULL;
+ CORE_ADDR pc;
+
+ if (m)
+ pst = find_pc_psymtab (SYMBOL_VALUE_ADDRESS (m));
+
+ if (pst && !pst->readin)
+ PSYMTAB_TO_SYMTAB (pst);
+ }
+
+ #endif
+
/* Helper function for decode_line_1.
Look for functions named NAME in all the symbol tables.
Return number of matches.
***************
*** 2812,2817 ****
--- 2867,2887 ----
/* If we found match(es) in the current source file we are done */
if (i1)
goto lookup_over;
+
+ #ifdef FAT_FREE_PSYMTABS
+
+ /* srikanth, in the case of the Wildebeest, the psymtabs are no
+ longer populated with psymbols. We need to rely on the linker
+ symbol table to lookup the symbol, use its address to decide
+ which psymtab would have housed the psymbol, were one to exist.
+ The function `foreach_text_minsym' is an iterator, which would
+ call its second argument for each matched minimal symbol
+ (mst_text and mst_file_text) and pass the match.
+ */
+
+ foreach_text_minsym (name, expand_containing_psymtab);
+
+ #endif
/* Now search all the global symbols. Do the symtab's first, then
check the psymtab's. If a psymtab indicates the existence
Index: gdb/minsyms.c
/opt/gnu/bin/diff -r -c -N /view/guo.wdb.c//CLO/Components/WDB/Src/gnu/gdb/minsyms.c gdb/minsyms.c
*** /view/guo.wdb.c//CLO/Components/WDB/Src/gnu/gdb/minsyms.c Wed Nov 10 12:30:56 1999
--- gdb/minsyms.c Thu Nov 11 09:25:44 1999
***************
*** 80,85 ****
--- 80,108 ----
static int
compact_minimal_symbols PARAMS ((struct minimal_symbol *, int));
+ /* srikanth, an iterator for functions in the linker symbol table.
+ Looks up `name' and for each matching symbol, calls `action' and
+ passes the selected symbol. */
+
+ void
+ foreach_text_minsym (name, action)
+ char *name;
+ void (*action) (struct minimal_symbol *);
+ {
+
+ struct objfile *objfile;
+ struct minimal_symbol m, *msymbol = &m;
+ char *mangled_prefix;
+
+ ALL_MSYMBOLS (objfile, msymbol)
+ {
+ if (MSYMBOL_TYPE (msymbol) == mst_text ||
+ MSYMBOL_TYPE (msymbol) == mst_file_text)
+ if (SYMBOL_MATCHES_NAME (msymbol, name))
+ action (msymbol);
+ }
+ }
+
/* Look through all the current minimal symbol tables and find the
first minimal symbol that matches NAME. If OBJF is non-NULL, limit
the search to that objfile. If SFILE is non-NULL, limit the search
Index: gdb/config/pa/tm-hppa.h
/opt/gnu/bin/diff -r -c -N /view/guo.wdb.c//CLO/Components/WDB/Src/gnu/gdb/config/pa/tm-hppa.h gdb/config/pa/tm-hppa.h
*** /view/guo.wdb.c//CLO/Components/WDB/Src/gnu/gdb/config/pa/tm-hppa.h Tue Nov 2 16:04:12 1999
--- gdb/config/pa/tm-hppa.h Wed Nov 10 12:06:11 1999
***************
*** 146,152 ****
top byte of the address for all 1's. Sigh.
*/
#define PC_REQUIRES_RUN_BEFORE_USE(pc) \
! (! target_has_stack && (pc & 0xFF000000))
/* return instruction is bv r0(rp) or bv,n r0(rp) */
--- 146,152 ----
top byte of the address for all 1's. Sigh.
*/
#define PC_REQUIRES_RUN_BEFORE_USE(pc) \
! (! target_has_stack && ((pc & 0xFF000000) == 0xFF000000))
/* return instruction is bv r0(rp) or bv,n r0(rp) */
***************
*** 331,337 ****
else \
memcpy ((VALBUF), \
(char *)(REGBUF) + REGISTER_BYTE (28) + \
! (TYPE_LENGTH (TYPE) > 4 ? (8 - TYPE_LENGTH (TYPE)) : (4 - TYPE_LENGTH (TYPE))), \
TYPE_LENGTH (TYPE)); \
}
--- 331,337 ----
else \
memcpy ((VALBUF), \
(char *)(REGBUF) + REGISTER_BYTE (28) + \
! (TYPE_LENGTH (TYPE) > 4 ? (8 - TYPE_LENGTH (TYPE)) : (REGISTER_SIZE - TYPE_LENGTH (TYPE))), \
TYPE_LENGTH (TYPE)); \
}
***************
*** 810,815 ****
--- 810,817 ----
|| (strncmp ((symname), "$PIC", 4) == 0) \
|| ((symname)[0] == '$' && isdigit ((symname)[1])) \
))
+
+ #define FAT_FREE_PSYMTABS
/* Here's how to step off a permanent breakpoint. */
#define SKIP_PERMANENT_BREAKPOINT (hppa_skip_permanent_breakpoint)
^ permalink raw reply [flat|nested] 2+ messages in thread
* Re: (patch) hpjyg21: FAT_FREE_PSYMTABS
1999-11-11 12:53 (patch) hpjyg21: FAT_FREE_PSYMTABS Jimmy Guo
@ 1999-11-11 15:10 ` Andrew Cagney
0 siblings, 0 replies; 2+ messages in thread
From: Andrew Cagney @ 1999-11-11 15:10 UTC (permalink / raw)
To: gdb-patches; +Cc: Jimmy Guo
Maintainers,
I've another meta maintainer question on this :-)
Is it better to use #ifdef or have the compiler eliminate optional
code? For instance:
if (FAT_FREE_PSYMTABS)
{
}
If the compiler does the elimination (instead of CPP) the code is always
being checked and hence less likely to suffer bit rot.
Yes I know it's something of a radical departure from the existing
conventions.
Andrew
> + #ifdef FAT_FREE_PSYMTABS
> +
> + /* srikanth, in the case of the Wildebeest, the psymtabs could be empty
> + and hence the call to find_pc_sect_psymbol() could fail to find anything.
> + Due potentially to a combination of bugs in compilers, linkers, loaders,
> + editors and debugger engineers :-), sometimes there is an overlap of
> + textlow and texthigh values between and multiple psymtabs. As a result,
> + if we are handed an address equal to the texthigh or textlow, we could
> + end up identifying the wrong psymtab.
> +
> + To be absolutely sure of the identity of the psymtab, let us verify that
> + the one we narrow down on, also houses pc + 4.
> + */
> +
> + if (p == NULL)
> + {
> + CORE_ADDR pc_plus4 = pc + 4;
> + if (pc_plus4 >= tpst->textlow &&
> + pc_plus4 <= tpst->texthigh)
> + return tpst;
> + }
> + #endif
From shebs@cygnus.com Thu Nov 11 15:41:00 1999
From: Stan Shebs <shebs@cygnus.com>
To: cagney@cygnus.com
Cc: gdb-patches@sourceware.cygnus.com, guo@cup.hp.com
Subject: Re: (patch) hpjyg21: FAT_FREE_PSYMTABS
Date: Thu, 11 Nov 1999 15:41:00 -0000
Message-id: <199911112341.PAA29252@andros.cygnus.com>
References: <382B4CD6.4B13DFA1@cygnus.com>
X-SW-Source: 1999-q4/msg00238.html
Content-length: 1094
Date: Fri, 12 Nov 1999 10:10:14 +1100
From: Andrew Cagney <ac131313@cygnus.com>
I've another meta maintainer question on this :-)
Is it better to use #ifdef or have the compiler eliminate optional
code? For instance:
if (FAT_FREE_PSYMTABS)
{
}
I favor this over ifdefs. Not only does it facilitate maintenance,
but it leaves open options for cross-debugging and for making
decisions at runtime. The only reasonable exception would be provably
time-critical code where ifdefs would allow compiler optimizations not
available if the code were merely "if (0)". (There might be examples
of this in the stabs reader, although I don't know of any myself.)
If the compiler does the elimination (instead of CPP) the code is always
being checked and hence less likely to suffer bit rot.
Yes I know it's something of a radical departure from the existing
conventions.
A departure from habit, true, but there's not a rule that says "use
more ifdefs". Ifdefs are the easy way out, but we want to go the
right way, not the expedient way.
Stan
From ac131313@cygnus.com Thu Nov 11 15:52:00 1999
From: Andrew Cagney <ac131313@cygnus.com>
To: Jimmy Guo <guo@cup.hp.com>
Cc: gdb-patches@sourceware.cygnus.com
Subject: Re: (patch) hpjyg15: hppah-nat.c & related
Date: Thu, 11 Nov 1999 15:52:00 -0000
Message-id: <382B5609.A4548C6D@cygnus.com>
References: <Pine.LNX.4.10.9911081520150.2309-100000@hpcll168.cup.hp.com>
X-SW-Source: 1999-q4/msg00239.html
Content-length: 1445
Jimmy Guo wrote:
>
> This patch updates hppah-nat.c and related files target.[ch] and
> infrun.c. See ChangeLog below.
>
> - Jimmy Guo, guo@cup.hp.com
>
> ChangeLog
>
> 1999-11-08 Jimmy Guo <guo@cup.hp.com>
>
> * hppah-nat.c (child_xfer_memory): when reading, also try the
> other way.
> (require_notification_of_events): start by ignoring all
> signals and then adding back in ones we're interested in.
> (child_has_execd): cache pathname info, used for the 2nd exec
> event reported by ptrace.
> (child_pid_to_exec_file): make sure ptrace request for
> pathname is made after an exec.
>
> * target.c (target_signal_exists): New function. Given a GDB
> target signal enum, return 1 if it's available on the host.
>
> * target.h (target_signal_exists): Declare.
I'd prefer the name (target_signal_p()) (Well actually I don't prefer
that name but its more consistent with the general naming convetion :-).
I'll check in a change to target.[ch] that adds the target_singal_p().
Hmm, I might do it slightly differently - were getting too many separate
places where there is code like:
#if SIG...
case TARGET_SIGNAL...
#endif
With that in, can the hppa-nat.c be re-submitted? BTW, you might also
split it, I think it contains two changes - the code that uses this new
function and something else.
enjoy,
Andrew
From ac131313@cygnus.com Thu Nov 11 16:03:00 1999
From: Andrew Cagney <ac131313@cygnus.com>
To: Jimmy Guo <guo@cup.hp.com>
Cc: gdb-patches@sourceware.cygnus.com
Subject: Re: (patch) hpjyg15: hppah-nat.c & related
Date: Thu, 11 Nov 1999 16:03:00 -0000
Message-id: <382B585D.421E98F3@cygnus.com>
References: <Pine.LNX.4.10.9911081520150.2309-100000@hpcll168.cup.hp.com>
X-SW-Source: 1999-q4/msg00240.html
Content-length: 25465
Jimmy Guo wrote:
>
> This patch updates hppah-nat.c and related files target.[ch] and
> infrun.c. See ChangeLog below.
Jim,
Don't forget that the GDB's coding style is defined by the output of
indent. Looking more closely I noticed that some of the changes are to
``fix'' the indentation. Please just grit your teath and ignore it.
It's a good idea to carefully review the diffs and revert/remove any
changes not specific to the problem at hand.
Doing this makes the reviewers life so much easier.
enjoy,
Andrew
> ***************
> *** 74,80 ****
> len = REGISTER_RAW_SIZE (regno);
>
> /* Requests for register zero actually want the save_state's
> ! ss_flags member. As RM says: "Oh, what a hack!" */
> if (regno == 0)
> {
> save_state_t ss;
> --- 74,80 ----
> len = REGISTER_RAW_SIZE (regno);
>
> /* Requests for register zero actually want the save_state's
> ! ss_flags member. As RM says: "Oh, what a hack!" */
> if (regno == 0)
> {
> save_state_t ss;
> ***************
> *** 90,109 ****
>
> /* Floating-point registers come from the ss_fpblock area. */
> else if (regno >= FP0_REGNUM)
> ! addr = (HPPAH_OFFSETOF (save_state_t, ss_fpblock)
> + (REGISTER_BYTE (regno) - REGISTER_BYTE (FP0_REGNUM)));
>
> /* Wide registers come from the ss_wide area.
> ! I think it's more PC to test (ss_flags & SS_WIDEREGS) to select
> ! between ss_wide and ss_narrow than to use the raw register size.
> ! But checking ss_flags would require an extra ptrace call for
> ! every register reference. Bleah. */
> else if (len == 8)
> ! addr = (HPPAH_OFFSETOF (save_state_t, ss_wide)
> + REGISTER_BYTE (regno));
>
> /* Narrow registers come from the ss_narrow area. Note that
> ! ss_narrow starts with gr1, not gr0. */
> else if (len == 4)
> addr = (HPPAH_OFFSETOF (save_state_t, ss_narrow)
> + (REGISTER_BYTE (regno) - REGISTER_BYTE (1)));
> --- 90,109 ----
>
> /* Floating-point registers come from the ss_fpblock area. */
> else if (regno >= FP0_REGNUM)
> ! addr = (HPPAH_OFFSETOF (save_state_t, ss_fpblock)
> + (REGISTER_BYTE (regno) - REGISTER_BYTE (FP0_REGNUM)));
>
> /* Wide registers come from the ss_wide area.
> ! I think it's more PC to test (ss_flags & SS_WIDEREGS) to select
> ! between ss_wide and ss_narrow than to use the raw register size.
> ! But checking ss_flags would require an extra ptrace call for
> ! every register reference. Bleah. */
> else if (len == 8)
> ! addr = (HPPAH_OFFSETOF (save_state_t, ss_wide)
> + REGISTER_BYTE (regno));
>
> /* Narrow registers come from the ss_narrow area. Note that
> ! ss_narrow starts with gr1, not gr0. */
> else if (len == 4)
> addr = (HPPAH_OFFSETOF (save_state_t, ss_narrow)
> + (REGISTER_BYTE (regno) - REGISTER_BYTE (1)));
> ***************
> *** 112,131 ****
>
> #ifdef GDB_TARGET_IS_HPPA_20W
> /* Unbelieveable. The PC head and tail must be written in 64bit hunks
> ! or we will get an error. Worse yet, the oddball ptrace/ttrace
> ! layering will not allow us to perform a 64bit register store.
>
> ! What a crock. */
> if (regno == PCOQ_HEAD_REGNUM || regno == PCOQ_TAIL_REGNUM && len == 8)
> {
> CORE_ADDR temp;
>
> ! temp = *(CORE_ADDR *)®isters[REGISTER_BYTE (regno)];
>
> /* Set the priv level (stored in the low two bits of the PC. */
> temp |= 0x3;
>
> ! ttrace_write_reg_64 (inferior_pid, (CORE_ADDR)addr, (CORE_ADDR)&temp);
>
> /* If we fail to write the PC, give a true error instead of
> just a warning. */
> --- 112,131 ----
>
> #ifdef GDB_TARGET_IS_HPPA_20W
> /* Unbelieveable. The PC head and tail must be written in 64bit hunks
> ! or we will get an error. Worse yet, the oddball ptrace/ttrace
> ! layering will not allow us to perform a 64bit register store.
>
> ! What a crock. */
> if (regno == PCOQ_HEAD_REGNUM || regno == PCOQ_TAIL_REGNUM && len == 8)
> {
> CORE_ADDR temp;
>
> ! temp = *(CORE_ADDR *) ®isters[REGISTER_BYTE (regno)];
>
> /* Set the priv level (stored in the low two bits of the PC. */
> temp |= 0x3;
>
> ! ttrace_write_reg_64 (inferior_pid, (CORE_ADDR) addr, (CORE_ADDR) &temp);
>
> /* If we fail to write the PC, give a true error instead of
> just a warning. */
> ***************
> *** 134,150 ****
> char *err = safe_strerror (errno);
> char *msg = alloca (strlen (err) + 128);
> sprintf (msg, "writing `%s' register: %s",
> ! REGISTER_NAME (regno), err);
> perror_with_name (msg);
> }
> return;
> }
>
> /* Another crock. HPUX complains if you write a nonzero value to
> ! the high part of IPSW. What will it take for HP to catch a
> ! clue about building sensible interfaces? */
> ! if (regno == IPSW_REGNUM && len == 8)
> ! *(int *)®isters[REGISTER_BYTE (regno)] = 0;
> #endif
>
> for (i = 0; i < len; i += sizeof (int))
> --- 134,150 ----
> char *err = safe_strerror (errno);
> char *msg = alloca (strlen (err) + 128);
> sprintf (msg, "writing `%s' register: %s",
> ! REGISTER_NAME (regno), err);
> perror_with_name (msg);
> }
> return;
> }
>
> /* Another crock. HPUX complains if you write a nonzero value to
> ! the high part of IPSW. What will it take for HP to catch a
> ! clue about building sensible interfaces? */
> ! if (regno == IPSW_REGNUM && len == 8)
> ! *(int *) ®isters[REGISTER_BYTE (regno)] = 0;
> #endif
>
> for (i = 0; i < len; i += sizeof (int))
> ***************
> *** 155,167 ****
> if (errno != 0)
> {
> /* Warning, not error, in case we are attached; sometimes
> ! the kernel doesn't let us at the registers. */
> char *err = safe_strerror (errno);
> char *msg = alloca (strlen (err) + 128);
> sprintf (msg, "writing `%s' register: %s",
> ! REGISTER_NAME (regno), err);
> /* If we fail to write the PC, give a true error instead of
> ! just a warning. */
> if (regno == PCOQ_HEAD_REGNUM || regno == PCOQ_TAIL_REGNUM)
> perror_with_name (msg);
> else
> --- 155,167 ----
> if (errno != 0)
> {
> /* Warning, not error, in case we are attached; sometimes
> ! the kernel doesn't let us at the registers. */
> char *err = safe_strerror (errno);
> char *msg = alloca (strlen (err) + 128);
> sprintf (msg, "writing `%s' register: %s",
> ! REGISTER_NAME (regno), err);
> /* If we fail to write the PC, give a true error instead of
> ! just a warning. */
> if (regno == PCOQ_HEAD_REGNUM || regno == PCOQ_TAIL_REGNUM)
> perror_with_name (msg);
> else
> ***************
> *** 197,212 ****
> len = sizeof (ss.ss_flags);
>
> /* Note that ss_flags is always an int, no matter what
> ! REGISTER_RAW_SIZE(0) says. Assuming all HP-UX PA machines
> ! are big-endian, put it at the least significant end of the
> ! value, and zap the rest of the buffer. */
> offset = REGISTER_RAW_SIZE (0) - len;
> memset (buf, 0, sizeof (buf));
> }
>
> /* Floating-point registers come from the ss_fpblock area. */
> else if (regno >= FP0_REGNUM)
> ! addr = (HPPAH_OFFSETOF (save_state_t, ss_fpblock)
> + (REGISTER_BYTE (regno) - REGISTER_BYTE (FP0_REGNUM)));
>
> /* Wide registers come from the ss_wide area.
> --- 197,212 ----
> len = sizeof (ss.ss_flags);
>
> /* Note that ss_flags is always an int, no matter what
> ! REGISTER_RAW_SIZE(0) says. Assuming all HP-UX PA machines
> ! are big-endian, put it at the least significant end of the
> ! value, and zap the rest of the buffer. */
> offset = REGISTER_RAW_SIZE (0) - len;
> memset (buf, 0, sizeof (buf));
> }
>
> /* Floating-point registers come from the ss_fpblock area. */
> else if (regno >= FP0_REGNUM)
> ! addr = (HPPAH_OFFSETOF (save_state_t, ss_fpblock)
> + (REGISTER_BYTE (regno) - REGISTER_BYTE (FP0_REGNUM)));
>
> /* Wide registers come from the ss_wide area.
> ***************
> *** 215,221 ****
> But checking ss_flags would require an extra ptrace call for
> every register reference. Bleah. */
> else if (len == 8)
> ! addr = (HPPAH_OFFSETOF (save_state_t, ss_wide)
> + REGISTER_BYTE (regno));
>
> /* Narrow registers come from the ss_narrow area. Note that
> --- 215,221 ----
> But checking ss_flags would require an extra ptrace call for
> every register reference. Bleah. */
> else if (len == 8)
> ! addr = (HPPAH_OFFSETOF (save_state_t, ss_wide)
> + REGISTER_BYTE (regno));
>
> /* Narrow registers come from the ss_narrow area. Note that
> ***************
> *** 232,240 ****
> errno = 0;
> /* Copy an int from the U area to buf. Fill the least
> significant end if len != raw_size. */
> ! * (int *) &buf[offset + i] =
> ! call_ptrace (PT_RUREGS, inferior_pid,
> ! (PTRACE_ARG3_TYPE) addr + i, 0);
> if (errno != 0)
> {
> /* Warning, not error, in case we are attached; sometimes
> --- 232,240 ----
> errno = 0;
> /* Copy an int from the U area to buf. Fill the least
> significant end if len != raw_size. */
> ! *(int *) &buf[offset + i] =
> ! call_ptrace (PT_RUREGS, inferior_pid,
> ! (PTRACE_ARG3_TYPE) addr + i, 0);
> if (errno != 0)
> {
> /* Warning, not error, in case we are attached; sometimes
> ***************
> *** 277,283 ****
> {
> register int i;
> /* Round starting address down to longword boundary. */
> ! register CORE_ADDR addr = memaddr & - (CORE_ADDR)(sizeof (int));
> /* Round ending address up; get number of longwords that makes. */
> register int count
> = (((memaddr + len) - addr) + sizeof (int) - 1) / sizeof (int);
> --- 277,283 ----
> {
> register int i;
> /* Round starting address down to longword boundary. */
> ! register CORE_ADDR addr = memaddr & -(CORE_ADDR) (sizeof (int));
> /* Round ending address up; get number of longwords that makes. */
> register int count
> = (((memaddr + len) - addr) + sizeof (int) - 1) / sizeof (int);
> ***************
> *** 363,370 ****
> inferior_pid, (PTRACE_ARG3_TYPE) addr, 0);
> if (errno)
> {
> ! free (buffer);
> ! return 0;
> }
> QUIT;
> }
> --- 363,379 ----
> inferior_pid, (PTRACE_ARG3_TYPE) addr, 0);
> if (errno)
> {
> ! /* As above, we may have guessed wrong about code vs. data, so
> ! try it the other way. */
> ! errno = 0;
> ! buffer[i] = call_ptrace (addr >= text_end ?
> ! PT_RIUSER : PT_RDUSER,
> ! inferior_pid,
> ! (PTRACE_ARG3_TYPE) addr, 0);
> ! if (errno) {
> ! free(buffer);
> ! return 0;
> ! }
> }
> QUIT;
> }
> ***************
> *** 697,702 ****
> --- 706,713 ----
> {
> #if defined(PT_SET_EVENT_MASK)
> int pt_status;
> + int nsigs;
> + int signum;
> ptrace_event_t ptrace_events;
>
> /* Instruct the kernel as to the set of events we wish to be
> ***************
> *** 709,715 ****
> the kernel to keep certain signals hidden from us, we do it
> by calling sigdelset (ptrace_events.pe_signals, signal) for
> each such signal here, before doing PT_SET_EVENT_MASK. */
> ! sigemptyset (&ptrace_events.pe_signals);
>
> ptrace_events.pe_set_event = 0;
>
> --- 720,749 ----
> the kernel to keep certain signals hidden from us, we do it
> by calling sigdelset (ptrace_events.pe_signals, signal) for
> each such signal here, before doing PT_SET_EVENT_MASK. */
> ! /* RM: The above comment is no longer true. We start with ignoring
> ! * all signals, and then add the ones we are interested in. We could
> ! * do it the other way: start by looking at all signals and then
> ! * deleting the ones that we aren't interested in, except that
> ! * multiple gdb signals may be mapped to the same host signal
> ! * (eg. TARGET_SIGNAL_IO and TARGET_SIGNAL_POLL both get mapped to
> ! * signal 22 on HPUX 10.20) We want to be notified if we are
> ! * interested in either signal.
> ! */
> ! sigfillset (&ptrace_events.pe_signals);
> !
> ! /* RM: Let's not bother with signals we don't care about */
> ! nsigs = (int) TARGET_SIGNAL_LAST;
> ! for (signum = nsigs; signum > 0; signum--)
> ! {
> ! if ((signal_stop_state (signum)) ||
> ! (signal_print_state (signum)) ||
> ! (!signal_pass_state (signum)))
> ! {
> ! if (target_signal_exists (signum))
> ! sigdelset (&ptrace_events.pe_signals,
> ! target_signal_to_host (signum));
> ! }
> ! }
>
> ptrace_events.pe_set_event = 0;
>
> ***************
> *** 724,730 ****
> errno = 0;
> pt_status = call_ptrace (PT_SET_EVENT_MASK,
> pid,
> ! (PTRACE_ARG3_TYPE) & ptrace_events,
> sizeof (ptrace_events));
> if (errno)
> perror_with_name ("ptrace");
> --- 758,764 ----
> errno = 0;
> pt_status = call_ptrace (PT_SET_EVENT_MASK,
> pid,
> ! (PTRACE_ARG3_TYPE) &ptrace_events,
> sizeof (ptrace_events));
> if (errno)
> perror_with_name ("ptrace");
> ***************
> *** 763,769 ****
> errno = 0;
> pt_status = call_ptrace (PT_SET_EVENT_MASK,
> pid,
> ! (PTRACE_ARG3_TYPE) & ptrace_events,
> sizeof (ptrace_events));
> if (errno)
> perror_with_name ("ptrace");
> --- 797,803 ----
> errno = 0;
> pt_status = call_ptrace (PT_SET_EVENT_MASK,
> pid,
> ! (PTRACE_ARG3_TYPE) &ptrace_events,
> sizeof (ptrace_events));
> if (errno)
> perror_with_name ("ptrace");
> ***************
> *** 913,919 ****
> errno = 0;
> pt_status = call_ptrace (PT_GET_PROCESS_STATE,
> pid,
> ! (PTRACE_ARG3_TYPE) & ptrace_state,
> sizeof (ptrace_state));
> if (errno)
> perror_with_name ("ptrace");
> --- 947,953 ----
> errno = 0;
> pt_status = call_ptrace (PT_GET_PROCESS_STATE,
> pid,
> ! (PTRACE_ARG3_TYPE) &ptrace_state,
> sizeof (ptrace_state));
> if (errno)
> perror_with_name ("ptrace");
> ***************
> *** 947,953 ****
> errno = 0;
> pt_status = call_ptrace (PT_GET_PROCESS_STATE,
> pid,
> ! (PTRACE_ARG3_TYPE) & ptrace_state,
> sizeof (ptrace_state));
> if (errno)
> perror_with_name ("ptrace");
> --- 981,987 ----
> errno = 0;
> pt_status = call_ptrace (PT_GET_PROCESS_STATE,
> pid,
> ! (PTRACE_ARG3_TYPE) &ptrace_state,
> sizeof (ptrace_state));
> if (errno)
> perror_with_name ("ptrace");
> ***************
> *** 1008,1013 ****
> --- 1042,1049 ----
> int pid;
> char **execd_pathname;
> {
> + static char saved_pathname[1024];
> +
> /* This request is only available on HPUX 10.0 and later. */
> #if !defined(PT_GET_PROCESS_STATE)
> *execd_pathname = NULL;
> ***************
> *** 1020,1026 ****
> errno = 0;
> pt_status = call_ptrace (PT_GET_PROCESS_STATE,
> pid,
> ! (PTRACE_ARG3_TYPE) & ptrace_state,
> sizeof (ptrace_state));
> if (errno)
> perror_with_name ("ptrace");
> --- 1056,1062 ----
> errno = 0;
> pt_status = call_ptrace (PT_GET_PROCESS_STATE,
> pid,
> ! (PTRACE_ARG3_TYPE) &ptrace_state,
> sizeof (ptrace_state));
> if (errno)
> perror_with_name ("ptrace");
> ***************
> *** 1029,1036 ****
>
> if (ptrace_state.pe_report_event & PTRACE_EXEC)
> {
> ! char *exec_file = target_pid_to_exec_file (pid);
> ! *execd_pathname = savestring (exec_file, strlen (exec_file));
> return 1;
> }
>
> --- 1065,1083 ----
>
> if (ptrace_state.pe_report_event & PTRACE_EXEC)
> {
> ! char *exec_file;
> !
> ! if (ptrace_state.pe_path_len > 0)
> ! {
> ! /* RM: ptrace reports 2 exec events per exec. Only the first one
> ! allows us to get a valid pathname. For the second one simply
> ! return the previously cached pathname */
> ! exec_file = target_pid_to_exec_file (pid);
> ! *execd_pathname = savestring (exec_file, strlen (exec_file));
> ! strncpy (saved_pathname, exec_file, sizeof (saved_pathname) - 1);
> ! }
> ! else
> ! *execd_pathname = saved_pathname;
> return 1;
> }
>
> ***************
> *** 1064,1069 ****
> --- 1111,1117 ----
> {
> static char exec_file_buffer[1024];
> int pt_status;
> + ptrace_state_t ptrace_state;
> CORE_ADDR top_of_stack;
> char four_chars[4];
> int name_index;
> ***************
> *** 1073,1084 ****
>
> #ifdef PT_GET_PROCESS_PATHNAME
> /* As of 10.x HP-UX, there's an explicit request to get the pathname. */
> ! pt_status = call_ptrace (PT_GET_PROCESS_PATHNAME,
> pid,
> ! (PTRACE_ARG3_TYPE) exec_file_buffer,
> ! sizeof (exec_file_buffer) - 1);
> ! if (pt_status == 0)
> ! return exec_file_buffer;
> #endif
>
> /* It appears that this request is broken prior to 10.30.
> --- 1121,1146 ----
>
> #ifdef PT_GET_PROCESS_PATHNAME
> /* As of 10.x HP-UX, there's an explicit request to get the pathname. */
> ! /* RM: According to the documentation, this request only works right
> ! * after an exec */
> ! errno = 0;
> ! pt_status = call_ptrace (PT_GET_PROCESS_STATE,
> pid,
> ! (PTRACE_ARG3_TYPE) &ptrace_state,
> ! sizeof (ptrace_state));
> ! if (errno)
> ! perror_with_name ("ptrace");
> !
> ! if (ptrace_state.pe_report_event & PTRACE_EXEC)
> ! {
> ! pt_status = call_ptrace (PT_GET_PROCESS_PATHNAME,
> ! pid,
> ! (PTRACE_ARG3_TYPE) exec_file_buffer,
> ! sizeof (exec_file_buffer) - 1);
> ! /* RM: ??? pt_status < 0 indicates failure, I think */
> ! if (pt_status >= 0)
> ! return exec_file_buffer;
> ! }
> #endif
>
> /* It appears that this request is broken prior to 10.30.
> Index: gdb/target.c
> /opt/gnu/bin/diff -r -c -N /view/guo.wdb.c//CLO/Components/WDB/Src/gnu/gdb/target.c gdb/target.c
> *** /view/guo.wdb.c//CLO/Components/WDB/Src/gnu/gdb/target.c Fri Nov 5 17:36:01 1999
> --- gdb/target.c Fri Nov 5 17:37:12 1999
> ***************
> *** 1915,1920 ****
> --- 1915,2066 ----
> }
> }
>
> + int
> + target_signal_exists (oursig)
> + enum target_signal oursig;
> + {
> + switch (oursig)
> + {
> + case TARGET_SIGNAL_0:
> + #if defined (SIGHUP)
> + case TARGET_SIGNAL_HUP:
> + #endif
> + #if defined (SIGINT)
> + case TARGET_SIGNAL_INT:
> + #endif
> + #if defined (SIGQUIT)
> + case TARGET_SIGNAL_QUIT:
> + #endif
> + #if defined (SIGILL)
> + case TARGET_SIGNAL_ILL:
> + #endif
> + #if defined (SIGTRAP)
> + case TARGET_SIGNAL_TRAP:
> + #endif
> + #if defined (SIGABRT)
> + case TARGET_SIGNAL_ABRT:
> + #endif
> + #if defined (SIGEMT)
> + case TARGET_SIGNAL_EMT:
> + #endif
> + #if defined (SIGFPE)
> + case TARGET_SIGNAL_FPE:
> + #endif
> + #if defined (SIGKILL)
> + case TARGET_SIGNAL_KILL:
> + #endif
> + #if defined (SIGBUS)
> + case TARGET_SIGNAL_BUS:
> + #endif
> + #if defined (SIGSEGV)
> + case TARGET_SIGNAL_SEGV:
> + #endif
> + #if defined (SIGSYS)
> + case TARGET_SIGNAL_SYS:
> + #endif
> + #if defined (SIGPIPE)
> + case TARGET_SIGNAL_PIPE:
> + #endif
> + #if defined (SIGALRM)
> + case TARGET_SIGNAL_ALRM:
> + #endif
> + #if defined (SIGTERM)
> + case TARGET_SIGNAL_TERM:
> + #endif
> + #if defined (SIGUSR1)
> + case TARGET_SIGNAL_USR1:
> + #endif
> + #if defined (SIGUSR2)
> + case TARGET_SIGNAL_USR2:
> + #endif
> + #if defined (SIGCHLD) || defined (SIGCLD)
> + case TARGET_SIGNAL_CHLD:
> + #endif /* SIGCLD or SIGCHLD */
> + #if defined (SIGPWR)
> + case TARGET_SIGNAL_PWR:
> + #endif
> + #if defined (SIGWINCH)
> + case TARGET_SIGNAL_WINCH:
> + #endif
> + #if defined (SIGURG)
> + case TARGET_SIGNAL_URG:
> + #endif
> + #if defined (SIGIO)
> + case TARGET_SIGNAL_IO:
> + #endif
> + #if defined (SIGPOLL)
> + case TARGET_SIGNAL_POLL:
> + #endif
> + #if defined (SIGSTOP)
> + case TARGET_SIGNAL_STOP:
> + #endif
> + #if defined (SIGTSTP)
> + case TARGET_SIGNAL_TSTP:
> + #endif
> + #if defined (SIGCONT)
> + case TARGET_SIGNAL_CONT:
> + #endif
> + #if defined (SIGTTIN)
> + case TARGET_SIGNAL_TTIN:
> + #endif
> + #if defined (SIGTTOU)
> + case TARGET_SIGNAL_TTOU:
> + #endif
> + #if defined (SIGVTALRM)
> + case TARGET_SIGNAL_VTALRM:
> + #endif
> + #if defined (SIGPROF)
> + case TARGET_SIGNAL_PROF:
> + #endif
> + #if defined (SIGXCPU)
> + case TARGET_SIGNAL_XCPU:
> + #endif
> + #if defined (SIGXFSZ)
> + case TARGET_SIGNAL_XFSZ:
> + #endif
> + #if defined (SIGWIND)
> + case TARGET_SIGNAL_WIND:
> + #endif
> + #if defined (SIGPHONE)
> + case TARGET_SIGNAL_PHONE:
> + #endif
> + #if defined (SIGLOST)
> + case TARGET_SIGNAL_LOST:
> + #endif
> + #if defined (SIGWAITING)
> + case TARGET_SIGNAL_WAITING:
> + #endif
> + #if defined (SIGLWP)
> + case TARGET_SIGNAL_LWP:
> + #endif
> + #if defined (SIGDANGER)
> + case TARGET_SIGNAL_DANGER:
> + #endif
> + #if defined (SIGGRANT)
> + case TARGET_SIGNAL_GRANT:
> + #endif
> + #if defined (SIGRETRACT)
> + case TARGET_SIGNAL_RETRACT:
> + #endif
> + #if defined (SIGMSG)
> + case TARGET_SIGNAL_MSG:
> + #endif
> + #if defined (SIGSOUND)
> + case TARGET_SIGNAL_SOUND:
> + #endif
> + #if defined (SIGSAK)
> + case TARGET_SIGNAL_SAK:
> + #endif
> + #if defined (SIGPRIO)
> + case TARGET_SIGNAL_PRIO:
> + #endif
> + return 1;
> +
> + default:
> + return 0;
> + }
> + }
> +
> /* Helper function for child_wait and the Lynx derivatives of child_wait.
> HOSTSTATUS is the waitstatus from wait() or the equivalent; store our
> translation of that in OURSTATUS. */
> Index: gdb/target.h
> /opt/gnu/bin/diff -r -c -N /view/guo.wdb.c//CLO/Components/WDB/Src/gnu/gdb/target.h gdb/target.h
> *** /view/guo.wdb.c//CLO/Components/WDB/Src/gnu/gdb/target.h Fri Nov 5 17:37:37 1999
> --- gdb/target.h Fri Nov 5 17:38:12 1999
> ***************
> *** 1327,1332 ****
> --- 1327,1333 ----
> /* Convert between host signal numbers and enum target_signal's. */
> extern enum target_signal target_signal_from_host PARAMS ((int));
> extern int target_signal_to_host PARAMS ((enum target_signal));
> + extern int target_signal_exists PARAMS ((enum target_signal));
>
> /* Convert from a number used in a GDB command to an enum target_signal. */
> extern enum target_signal target_signal_from_command PARAMS ((int));
> Index: gdb/infrun.c
> /opt/gnu/bin/diff -r -c -N /view/guo.wdb.c//CLO/Components/WDB/Src/gnu/gdb/infrun.c gdb/infrun.c
> *** /view/guo.wdb.c//CLO/Components/WDB/Src/gnu/gdb/infrun.c Fri Nov 5 18:37:38 1999
> --- gdb/infrun.c Fri Nov 5 18:38:41 1999
> ***************
> *** 3621,3626 ****
> --- 3621,3633 ----
> }
> }
>
> +
> + #ifdef GDB_TARGET_IS_HPPA
> + /* RM: Use OS interface to ignore signals we don't care about */
> + if (target_has_execution)
> + require_notification_of_events (inferior_pid);
> + #endif
> +
> do_cleanups (old_chain);
> }
>
From ac131313@cygnus.com Thu Nov 11 16:07:00 1999
From: Andrew Cagney <ac131313@cygnus.com>
To: Stan Shebs <shebs@cygnus.com>
Cc: cagney@cygnus.com, gdb-patches@sourceware.cygnus.com, guo@cup.hp.com
Subject: Re: (patch) hpjyg21: FAT_FREE_PSYMTABS
Date: Thu, 11 Nov 1999 16:07:00 -0000
Message-id: <382B5A0D.31585A7A@cygnus.com>
References: <199911112341.PAA29252@andros.cygnus.com>
X-SW-Source: 1999-q4/msg00241.html
Content-length: 967
Stan Shebs wrote:
>
> Date: Fri, 12 Nov 1999 10:10:14 +1100
> From: Andrew Cagney <ac131313@cygnus.com>
>
> I've another meta maintainer question on this :-)
>
> Is it better to use #ifdef or have the compiler eliminate optional
> code? For instance:
>
> if (FAT_FREE_PSYMTABS)
> {
> }
>
> I favor this over ifdefs. Not only does it facilitate maintenance,
> but it leaves open options for cross-debugging and for making
> decisions at runtime. The only reasonable exception would be provably
> time-critical code where ifdefs would allow compiler optimizations not
> available if the code were merely "if (0)". (There might be examples
> of this in the stabs reader, although I don't know of any myself.)
``Trust me'', there are no cases where this can occure :-)
Compiling with -O eliminates all of them (well it does for GCC, if other
compilers can't eliminate that then they are broken :-).
Andrew
From guo@cup.hp.com Thu Nov 11 16:31:00 1999
From: Jimmy Guo <guo@cup.hp.com>
To: Andrew Cagney <ac131313@cygnus.com>
Cc: gdb-patches@sourceware.cygnus.com
Subject: Re: (patch) hpjyg15: hppah-nat.c & related
Date: Thu, 11 Nov 1999 16:31:00 -0000
Message-id: <Pine.LNX.4.10.9911111610120.6426-100000@hpcll168.cup.hp.com>
References: <382B585D.421E98F3@cygnus.com>
X-SW-Source: 1999-q4/msg00242.html
Content-length: 1348
>Don't forget that the GDB's coding style is defined by the output of
>indent. Looking more closely I noticed that some of the changes are to
>``fix'' the indentation. Please just grit your teath and ignore it.
>
>It's a good idea to carefully review the diffs and revert/remove any
>changes not specific to the problem at hand.
>
>Doing this makes the reviewers life so much easier.
Andrew,
I'm aware of this problem and I'm trying to resolve this. Interestly,
yes I'm reversing what indent did, _not_ to create noises, but to remove
them. If you run indent on, say hppah-nat.c, straight from the 19991108
snapshot, you will see that it's not in GDB's coding style (since indent
is not happy with the way the code is!)
I _ran_ indent on source files before I make a patch (ok, in some cases
I might forgot to do so which might have turned out to be a good thing)
... and that's how most diff noises are introduced. I came to realize
this and am starting to do the reserve -- change back what indent changed
that's not relevant to the patch, to make the changes in my patches
conform to GDB's coding style. Given how much indent changes a source
file, it could be very time consuming and 'non-productive', but I'm
doing my share to help the situation.
BTW, I'm using indent 1.9.1 on Linux, with no arguments.
- Jimmy Guo, guo@cup.hp.com
^ permalink raw reply [flat|nested] 2+ messages in thread
end of thread, other threads:[~1999-11-11 15:10 UTC | newest]
Thread overview: 2+ messages (download: mbox.gz / follow: Atom feed)
-- links below jump to the message on this page --
1999-11-11 12:53 (patch) hpjyg21: FAT_FREE_PSYMTABS Jimmy Guo
1999-11-11 15:10 ` Andrew Cagney
This is a public inbox, see mirroring instructions
for how to clone and mirror all data and code used for this inbox