[PATCH v2] gdb: ADI support
Wei-min Pan
weimin.pan@oracle.com
Tue Jun 20 17:50:00 GMT 2017
Yao Qi wrote:
> Weimin Pan <weimin.pan@oracle.com> writes:
>
>
>> Tested in sparc64-linux-gnu. No regressions.
>>
>>
>
> Could you build a cross gdb with your patch applied? That is, host and
> build is x86_64-linux and target is sparc64-linux, for example.
>
OK, will do the cross build and address your concern of <adi.h> likely
breaking it as you commented below.
>> gdb/ChangeLog:
>> 2017-06-16 Weimin Pan <weimin.pan@oracle.com>
>>
>> * sparc64-tdep.h: (adi_normalize_address): New export.
>> * sparc64-adi-tdep.c: New file.
>> * Makefile.in (ALLDEPFILES): Add sparc64-adi-tdep.c.
>> * configure.tgt: Add sparc64-adi-tdep.o.
>> * sparc64-linux-nat.c:
>> (sparc64_linux_watchpoint_addr_within_range): New function.
>> (_initialize_sparc64_linux_nat): Register sparc64_linux_watchpoint_addr_within_range.
>>
>
> This line is too long.
> See https://sourceware.org/gdb/wiki/ContributionChecklist#Properly_Formatted_GNU_ChangeLog
>
Will split it.
>> * sparc64-linux-tdep.c: Include <adi.h>.
>> (sparc64_linux_handle_segmentation_fault): New function.
>> (sparc64_linux_init_abi): Register sparc64_linux_handle_segmentation_fault
>>
>
> Likewise.
>
OK.
>
>> * sparc64-tdep.c: (sparc64_pstate_type): Replace PID1 with MCDE.
>> gdb/doc/ChangeLog:
>> 2017-06-16 Weimin Pan <weimin.pan@oracle.com>
>> * gdb.texinfo (Architectures): Add new Sparc64 section to document ADI support.
>> ---
>> gdb/ChangeLog | 14 ++
>> gdb/Makefile.in | 2 +
>> gdb/configure.tgt | 1 +
>> gdb/doc/ChangeLog | 4 +
>> gdb/doc/gdb.texinfo | 85 +++++++++
>> gdb/sparc64-adi-tdep.c | 460 ++++++++++++++++++++++++++++++++++++++++++++++
>> gdb/sparc64-linux-nat.c | 15 ++
>> gdb/sparc64-linux-tdep.c | 49 +++++
>> gdb/sparc64-tdep.c | 2 +-
>> gdb/sparc64-tdep.h | 2 +
>> 10 files changed, 633 insertions(+), 1 deletions(-)
>> create mode 100644 gdb/sparc64-adi-tdep.c
>>
>> diff --git a/gdb/ChangeLog b/gdb/ChangeLog
>> index 5ebc7f8..8126a47 100644
>> --- a/gdb/ChangeLog
>> +++ b/gdb/ChangeLog
>> @@ -1,3 +1,17 @@
>> +2017-06-16 Weimin Pan <weimin.pan@oracle.com>
>> +
>> + * sparc64-tdep.h: (adi_normalize_address): New export.
>> + * sparc64-adi-tdep.c: New file.
>> + * Makefile.in (ALLDEPFILES): Add sparc64-adi-tdep.c.
>> + * configure.tgt: Add sparc64-adi-tdep.o.
>> + * sparc64-linux-nat.c:
>> + (sparc64_linux_watchpoint_addr_within_range): New function.
>> + (_initialize_sparc64_linux_nat): Register sparc64_linux_watchpoint_addr_within_range.
>> + * sparc64-linux-tdep.c: Include <adi.h>.
>> + (sparc64_linux_handle_segmentation_fault): New function.
>> + (sparc64_linux_init_abi): Register sparc64_linux_handle_segmentation_fault
>> + * sparc64-tdep.c: (sparc64_pstate_type): Replace PID1 with MCDE.
>> +
>> 2017-06-08 Sergio Durigan Junior <sergiodj@redhat.com>
>>
>> * common/common-utils.c (stringify_argv): Check for "arg[0] !=
>> diff --git a/gdb/Makefile.in b/gdb/Makefile.in
>> index 5e5fcaa..a6b0d99 100644
>> --- a/gdb/Makefile.in
>> +++ b/gdb/Makefile.in
>> @@ -772,6 +772,7 @@ ALL_64_TARGET_OBS = \
>> ia64-tdep.o \
>> ia64-vms-tdep.o \
>> mips64-obsd-tdep.o \
>> + sparc64-adi-tdep.o \
>> sparc64-fbsd-tdep.o \
>> sparc64-linux-tdep.o \
>> sparc64-nbsd-tdep.o \
>> @@ -2660,6 +2661,7 @@ ALLDEPFILES = \
>> sparc-sol2-nat.c \
>> sparc-sol2-tdep.c \
>> sparc-tdep.c \
>> + sparc64-adi-tdep.c \
>> sparc64-fbsd-nat.c \
>> sparc64-fbsd-tdep.c \
>> sparc64-linux-nat.c \
>> diff --git a/gdb/configure.tgt b/gdb/configure.tgt
>> index fdcb7b1..0519c62 100644
>> --- a/gdb/configure.tgt
>> +++ b/gdb/configure.tgt
>> @@ -548,6 +548,7 @@ sparc64-*-linux*)
>> gdb_target_obs="sparc64-tdep.o sparc64-sol2-tdep.o sol2-tdep.o \
>> sparc64-linux-tdep.o sparc-tdep.o sparc-sol2-tdep.o \
>> sparc-linux-tdep.o solib-svr4.o linux-tdep.o \
>> + sparc64-adi-tdep.o \
>> ravenscar-thread.o sparc-ravenscar-thread.o"
>> build_gdbserver=yes
>> ;;
>> diff --git a/gdb/doc/ChangeLog b/gdb/doc/ChangeLog
>> index 868cde9..a411d64 100644
>> --- a/gdb/doc/ChangeLog
>> +++ b/gdb/doc/ChangeLog
>> @@ -1,3 +1,7 @@
>> +2017-06-16 Weimin Pan <weimin.pan@oracle.com>
>> + * gdb.texinfo (Architectures): Add new Sparc64 section to document ADI support.
>> + * NEWS: Add "adi examine" and "adi assign" commands.
>> +
>> 2017-06-07 Sergio Durigan Junior <sergiodj@redhat.com>
>>
>> * gdb.texinfo (Starting) <startup-with-shell>: Add @anchor.
>> diff --git a/gdb/doc/gdb.texinfo b/gdb/doc/gdb.texinfo
>> index 8d7a1c9..843c1bf 100644
>> --- a/gdb/doc/gdb.texinfo
>> +++ b/gdb/doc/gdb.texinfo
>> @@ -22465,6 +22465,7 @@ all uses of @value{GDBN} with the architecture, both native and cross.
>> * SPU:: Cell Broadband Engine SPU architecture
>> * PowerPC::
>> * Nios II::
>> +* Sparc64::
>> @end menu
>>
>> @node AArch64
>> @@ -22849,6 +22850,90 @@ target code in @value{GDBN}.
>> Show the current setting of Nios II debugging messages.
>> @end table
>>
>> +@node Sparc64
>> +@subsection Sparc64
>> +@cindex Sparc64 support
>> +@cindex Application Data Integrity
>> +@subsubsection ADI Support
>> +
>> +The M7 processor supports an Application Data Integrity (ADI) feature that
>> +detects invalid data accesses. When software allocates memory and enables
>> +ADI on the allocated memory, it chooses a 4-bit version number, sets the
>> +version in the upper 4 bits of the 64-bit pointer to that data, and stores
>> +the 4-bit version in every cacheline of that data. Hardware saves the latter
>> +in spare bits in the cache and memory hierarchy. On each load and store,
>> +the processor compares the upper 4 VA (virtual address) bits to the
>> +cacheline's version. If there is a mismatch, the processor generates a
>> +version mismatch trap which can be either precise or disrupting. The trap
>> +is an error condition which the kernel delivers to the process as a SIGSEGV
>> +signal.
>> +
>> +Note that only 64-bit applications can use ADI and need to be built with
>> +ADI-enabled.
>> +
>> +Values of the ADI version tags, which are in granularity of a
>> +cacheline (64 bytes), can be viewed or modified.
>> +
>> +
>> +@table @code
>> +@kindex adi examine
>> +@item adi examine
>> +
>> +The @code{adi examine} command displays the value of one ADI version tag per
>> +cacheline. It has the following command syntax:
>> +
>> +@table @code
>> +@itemx adi (examine | x) [ / @var{n} ] @var{addr}
>> +@end table
>> +
>> +@var{n}, the byte count
>> +The count is a decimal integer specifying the number in bytes; the default
>> +is 1. It specifies how much ADI version information, at the ratio of 1:ADI
>> +block size, to display.
>> +
>> +@var{addr}, starting display address
>> +@var{addr} is the address in user address space where you want @value{GDBN}
>> +to begin displaying the ADI version tags.
>> +
>> +Below is an example of displaying ADI versions of variable "shmaddr".
>> +
>> +@smallexample
>> +(@value{GDBP}) adi x/100 shmaddr
>> + 0xfff800010002c000: 0 0
>> +@end smallexample
>> +
>> +@kindex adi assign
>> +@item adi assign
>> +
>> +The @code{adi assign} command is used to assign new ADI version tag
>> +to an address. It has the following command syntax:
>> +
>> +@table @code
>> +@itemx adi (assign | a) [ / @var{n} ] @var{addr} = @var{tag}
>> +@end table
>> +
>> +@var{n}, the byte count
>> +The count is a decimal integer specifying the number in bytes; the default
>> +is 1. It specifies how much ADI version information, at the ratio of 1:ADI
>> +block size, to modify.
>> +
>> +@var{addr}, starting display address
>> +@var{addr} is the address in user address space where you want @value{GDBN}
>> +to begin modifying the ADI version tags.
>> +
>> +@var{tag}, the new ADI version tag
>> +
>> +For example, do the following to modify then verify ADI versions of
>> +variable "shmaddr":
>> +
>> +@smallexample
>> +(@value{GDBP}) adi a/100 shmaddr = 7
>> +(@value{GDBP}) adi x/100 shmaddr
>> + 0xfff800010002c000: 7 7
>> +@end smallexample
>> +
>> +@end table
>> +
>> @node Controlling GDB
>> @chapter Controlling @value{GDBN}
>>
>> diff --git a/gdb/sparc64-adi-tdep.c b/gdb/sparc64-adi-tdep.c
>> new file mode 100644
>> index 0000000..9c9d0ae
>> --- /dev/null
>> +++ b/gdb/sparc64-adi-tdep.c
>> @@ -0,0 +1,460 @@
>> +/* Target-dependent code for Sparc64.
>>
>
> The new file sparc64-adi-tdep.c should be merged into the existing
> sparc64-tdep.c. The file name schema in GDB is ARCH-OSABI-tdep.c.
> Since adi is a new hardware feature rather than an OSABI, it should be
> put in sparc64-tdep.c
>
Originally it was part of sparc64-tdep.c. But our engineer thought it's
cleaner to make it a
separate source file. Being new to the gnu world, I believe following
the file name schema
is the way to go. Thanks.
>> +
>> + Copyright (C) 2017 Free Software Foundation, Inc.
>> +
>> + This file is part of GDB.
>> +
>> + 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 3 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, see <http://www.gnu.org/licenses/>. */
>> +
>> +#include <adi.h>
>>
>
> I am afraid this will break the gdb cross build.
>
>
>> +#include "defs.h"
>> +#include "inferior.h"
>> +#include "gdbcmd.h"
>> +#include "auxv.h"
>> +
>> +/* The M7 processor supports an Application Data Integrity (ADI) feature
>> + that detects invalid data accesses. When software allocates memory and
>> + enables ADI on the allocated memory, it chooses a 4-bit version number,
>> + sets the version in the upper 4 bits of the 64-bit pointer to that data,
>> + and stores the 4-bit version in every cacheline of the object. Hardware
>> + saves the latter in spare bits in the cache and memory hierarchy. On each
>> + load and store, the processor compares the upper 4 VA (virtual address) bits
>> + to the cacheline's version. If there is a mismatch, the processor generates
>> + a version mismatch trap which can be either precise or disrupting.
>> + The trap is an error condition which the kernel delivers to the process
>> + as a SIGSEGV signal.
>> +
>> + The upper 4 bits of the VA represent a version and are not part of the
>> + true address. The processor clears these bits and sign extends bit 59
>> + to generate the true address.
>> +
>> + Note that 32-bit applications cannot use ADI. */
>>
>
> due to lack of hardware support or kernel support? I ask this because
> it is related to my comment to the following function arch_is_64bit.
>
Yes.
>> +
>> +
>> +#define MAX_PROC_NAME_SIZE sizeof("/proc/99999/lwp/9999/adi/lstatus")
>> +
>> +/* ADI command list. */
>> +static struct cmd_list_element *sparc64adilist = NULL;
>> +
>> +/* Current ADI stat settings. */
>> +static struct
>> +{
>> + /* The ADI block size. */
>> + unsigned long blksize;
>> +
>> + /* Number of bits used for an ADI version tag which can be
>> + * used together with the shift value for an ADI version tag
>> + * to encode or extract the ADI version value in a pointer. */
>> + unsigned long nbits;
>> +
>> + /* The maximum ADI version tag value supported. */
>> + int max_version;
>> +
>> + /* ADI version tag file. */
>> + int tag_fd;
>> +
>> + /* Last ADI address examined. */
>> + CORE_ADDR last_vaddr;
>> +
>> + /* Last specified examination count. */
>> + int last_cnt;
>> +
>> + /* ADI availability check has been done. */
>> + bool checked_avail;
>> +
>> + /* ADI is available. */
>> + bool is_avail;
>> +
>> +} adi_stat;
>>
>
> It is a global data, but as I read the rest of the patch, looks like adi
> data should be per-process.
>
You're right. Will investigate.
>> +
>> +
>> +static void
>> +info_adi_command (char *args, int from_tty)
>> +{
>> + printf_unfiltered ("\"adi\" must be followed by \"examine\" "
>> + "or \"assign\".\n");
>> + help_list (sparc64adilist, "adi ", all_commands, gdb_stdout);
>> +}
>> +
>> +static bool
>> +arch_is_64bit (void)
>> +{
>> + struct gdbarch *gdbarch = get_frame_arch (get_current_frame ());
>> + return gdbarch_bfd_arch_info (gdbarch)->bits_per_word == 64;
>> +
>> +}
>> +
>> +static int
>> +parse_byte_count (const char **sptr)
>> +{
>> + const char *p = *sptr;
>> + int cnt = 1;
>> + if (*p == '-')
>> + {
>> + cnt = -1;
>> + p++;
>> + }
>> + if (*p >= '0' && *p <= '9')
>> + cnt *= atoi (p);
>> + while (*p >= '0' && *p <= '9')
>> + p++;
>> + while (*p == ' ' || *p == '\t')
>> + p++;
>> + *sptr = p;
>> +
>> + if (cnt <= 0)
>> + error (_("Byte count should have a positive value."));
>> +
>> + return cnt;
>> +}
>>
>
> You can use get_number or other apis declared in in gdb/cli-utils.h.
>
It's nice to know. Will do.
>> +
>> +/* Read attributes of a maps entry in /proc/[pid]/adi/maps. */
>> +
>> +static void
>> +read_maps_entry (const char *line,
>> + ULONGEST *addr, ULONGEST *endaddr)
>> +{
>> + const char *p = line;
>> +
>> + *addr = strtoulst (p, &p, 16);
>> + if (*p == '-')
>> + p++;
>> +
>> + *endaddr = strtoulst (p, &p, 16);
>> +}
>> +
>> +/* Check if ADI is available. */
>> +
>> +static bool
>> +adi_available (void)
>> +{
>> + if (adi_stat.checked_avail)
>> + return adi_stat.is_avail;
>> +
>> + adi_stat.checked_avail = true;
>> + if (!arch_is_64bit ())
>> + return false;
>> + if (target_auxv_search (¤t_target, AT_ADI_BLKSZ, &adi_stat.blksize) <= 0)
>>
>
> If kernel doesn't support ADI for 32-bit application, why don't we fetch
> AT_ADI_BLKSZ unconditionally?
>
Yes, we could do that. Will get rid of the arch_is_64bit() call.
>> + return false;
>> + target_auxv_search (¤t_target, AT_ADI_NBITS, &adi_stat.nbits);
>> + adi_stat.max_version = (1 << adi_stat.nbits) - 2;
>> + adi_stat.is_avail = true;
>> +
>> + return adi_stat.is_avail;
>> +}
>> +
>> +#if 0
>> +/* Extract version tag from a versioned address. */
>> +
>> +int adi_extract_version (CORE_ADDR versioned_addr)
>> +{
>> + return ((unsigned long)versioned_addr) >> (64 - adi_stat.nbits);
>> +}
>> +#endif
>> +
>> +/* Normalize a versioned address - a VA with ADI bits (63-60) set. */
>> +
>> +CORE_ADDR
>> +adi_normalize_address (CORE_ADDR addr)
>> +{
>> + if (adi_stat.nbits)
>> + return ((CORE_ADDR)(((long)addr << adi_stat.nbits) >> adi_stat.nbits));
>> + return addr;
>> +}
>>
>
> I think the right approach is to implement gdbarch_addr_bits_remove for
> sparc64-linux, which remove these unused bits.
>
I suspect keeping this little function whose name implies clearly to
normalize a VA might be OK. Will look into it.
>> +
>> +/* Align a normalized address - a VA with bit 59 sign extended into ADI bits. */
>> +
>> +static CORE_ADDR
>> +adi_align_address (CORE_ADDR naddr)
>> +{
>> + return (naddr - (naddr % adi_stat.blksize)) / adi_stat.blksize;
>>
>
> Indentation is wrong.
>
Will check/correct.
>> +}
>> +
>> +/* Convert a byte count to count at a ratio of 1:adi_blksz. */
>> +
>> +static int
>> +adi_convert_byte_count (CORE_ADDR naddr, int nbytes, CORE_ADDR locl)
>> +{
>> + return ((naddr + nbytes + adi_stat.blksize - 1) / adi_stat.blksize) - locl;
>>
>
> This line is too long.
>
>
Will correct.
>> +}
>> +
>> +/* The /proc/[pid]/adi/tag file, which allows gdb to get/set ADI
>> + version in a target process, maps linearly to the address space
>> + of the target process at a ratio of 1:adi_blksz.
>> +
>> + A read (or write) at offset K in the file returns (or modifies)
>> + the ADI version tag stored in the cacheline containing address
>> + K * adi_blksz, encoded as 1 version tag per byte. The allowed
>> + version tag values are between 0 and adi_stat.max_version. */
>> +
>> +static int
>> +adi_tag_fd (void)
>> +{
>> + if (adi_stat.tag_fd != 0)
>> + return adi_stat.tag_fd;
>> +
>> + char cl_name[MAX_PROC_NAME_SIZE];
>> + pid_t pid = ptid_get_pid (inferior_ptid);
>> + snprintf (cl_name, sizeof(cl_name), "/proc/%d/adi/tag", pid);
>> + adi_stat.tag_fd = open (cl_name, O_RDWR|O_EXCL|O_CLOEXEC);
>>
>
> You can't access /proc in *-tdep.c file, because it is also compiled for
> cross-debugger. The rule in general is to move it to sparc64-linux-nat.c.
>
It's nice to know. Will look into it. BTW is there any document that
specifies/defines these rules for
*-tdep.c and *-nat.c files?
>> +
>> + return adi_stat.tag_fd;
>> +}
>> +
>> +/* Check if an address set is ADI enabled, using /proc/[pid]/adi/maps
>> + which was exported by the kernel and contains the currently ADI
>> + mapped memory regions and their access permissions. */
>> +
>> +static bool
>> +adi_is_addr_mapped (CORE_ADDR vaddr, size_t cnt)
>> +{
>> + char filename[MAX_PROC_NAME_SIZE];
>> + size_t i = 0;
>> +
>> + pid_t pid = ptid_get_pid (inferior_ptid);
>> + snprintf (filename, sizeof filename, "/proc/%d/adi/maps", pid);
>> + char *data = target_fileio_read_stralloc (NULL, filename);
>>
>
> It is better to define adi as a qXfer object, which can be read and
> written through target layer.
>
OK, will take a look.
>> + if (data)
>> + {
>> + struct cleanup *cleanup = make_cleanup (xfree, data);
>> + char *line;
>> + for (line = strtok (data, "\n"); line; line = strtok (NULL, "\n"))
>> + {
>> + ULONGEST addr, endaddr;
>> +
>> + read_maps_entry (line, &addr, &endaddr);
>> +
>> + while (((vaddr + i) * adi_stat.blksize) >= addr
>> + && ((vaddr + i) * adi_stat.blksize) < endaddr)
>> + {
>> + if (++i == cnt)
>> + {
>> + do_cleanups (cleanup);
>> + return true;
>> + }
>> + }
>> + }
>> + do_cleanups (cleanup);
>> + }
>> + else
>> + warning (_("unable to open /proc file '%s'"), filename);
>> +
>> + return false;
>> +}
>> +
>>
>
>
>> +
>> +/* Print ADI version tag value in "tags" for memory locations starting
>> + at "vaddr" with number of "cnt". */
>> +
>> +static void
>> +adi_print_versions (CORE_ADDR vaddr, size_t cnt, unsigned char *tags)
>> +{
>> + int v_idx = 0;
>> + const int maxelts = 8; /* # of elements per line */
>> +
>> + while (cnt > 0)
>> + {
>> + QUIT;
>> + printf_filtered ("0x%016lx:\t", vaddr * adi_stat.blksize);
>> + for (int i = maxelts; i > 0 && cnt > 0; i--, cnt--)
>> + {
>> + if (tags[v_idx] == 0xff) /* no version tag */
>> + printf_filtered ("- ");
>> + else
>> + printf_filtered ("%1X ", tags[v_idx]);
>> + ++v_idx;
>> + }
>> + printf_filtered ("\n");
>> + gdb_flush (gdb_stdout);
>> + vaddr += maxelts;
>> + }
>> +}
>> +
>> +static void
>> +do_examine (CORE_ADDR start, int bcnt)
>> +{
>> + CORE_ADDR vaddr = adi_normalize_address (start);
>> + struct cleanup *cleanup;
>> +
>> + CORE_ADDR vstart = adi_align_address (vaddr);
>> + int cnt = adi_convert_byte_count (vaddr, bcnt, vstart);
>> + unsigned char *buf = (unsigned char *) xmalloc (cnt);
>> + cleanup = make_cleanup (xfree, buf);
>> + int read_cnt = adi_read_versions (vstart, cnt, buf);
>> + if (read_cnt == -1)
>> + error (_("No ADI information"));
>> + else if (read_cnt < cnt)
>> + error(_("No ADI information at 0x%lx"), vaddr);
>> +
>> + adi_print_versions(vstart, cnt, buf);
>> +
>> + do_cleanups (cleanup);
>> +}
>> +
>> +static void
>> +do_assign (CORE_ADDR start, size_t bcnt, int version)
>> +{
>> + CORE_ADDR vaddr = adi_normalize_address (start);
>> +
>> + CORE_ADDR vstart = adi_align_address (vaddr);
>> + int cnt = adi_convert_byte_count (vaddr, bcnt, vstart);
>> + unsigned char *buf = (unsigned char *) xmalloc (cnt);
>> + memset(buf, version, cnt);
>> +
>> + int set_cnt = adi_write_versions (vstart, cnt, buf);
>> + xfree (buf);
>> +
>> + if (set_cnt == -1)
>> + error (_("No ADI information"));
>> + else if (set_cnt < cnt)
>> + error(_("No ADI information at 0x%lx"), vaddr);
>> +
>> +}
>> +
>> +/* ADI examine version tag command.
>> +
>> + Command syntax:
>> +
>> + adi (examine|x)/count <addr> */
>> +
>> +static void
>> +adi_examine_command (char *args, int from_tty)
>> +{
>> + /* make sure program is active and adi is available */
>> + if (!target_has_execution)
>> + error (_("ADI command requires a live process/thread"));
>> +
>> + if (!adi_available ())
>> + error (_("No ADI information"));
>> +
>> + int cnt = adi_stat.last_cnt? adi_stat.last_cnt : 1;
>> + const char *p = args;
>> + if (p && *p == '/')
>> + {
>> + p++;
>> + cnt = parse_byte_count (&p);
>> + }
>> +
>> + CORE_ADDR next_address = adi_stat.last_vaddr ? adi_stat.last_vaddr : 0;
>> + if (p != 0 && *p != 0)
>> + next_address = parse_and_eval_address (p);
>> + else if (!adi_stat.last_cnt)
>> + error (_("Usage: adi examine|x[/count] <addr>"));
>> +
>> + if (next_address)
>> + do_examine(next_address, cnt);
>> +
>> + adi_stat.last_cnt = cnt;
>> + adi_stat.last_vaddr = next_address + adi_stat.blksize * cnt;
>> +}
>> +
>> +/* ADI assign version tag command.
>> +
>> + Command syntax:
>> +
>> + adi (assign|a)/count <addr> = <version> */
>> +
>> +static void
>> +adi_assign_command (char *args, int from_tty)
>> +{
>> + /* make sure program is active and adi is available */
>> + if (!target_has_execution)
>> + error (_("ADI command requires a live process/thread"));
>> +
>> + if (!adi_available ())
>> + error (_("No ADI information"));
>> +
>> + const char *exp = args;
>> + if (exp == 0)
>> + error_no_arg (_("Usage: adi assign|a[/count] <addr> = <version>"));
>> +
>> + char *q = (char *) strchr (exp, '=');
>> + if (q)
>> + *q++ = 0;
>> + else
>> + error (_("Usage: adi assign|a[/count] <addr> = <version>"));
>> +
>> + size_t cnt = 1;
>> + const char *p = exp;
>> + if (exp && *exp == '/')
>> + {
>> + p = exp + 1;
>> + cnt = parse_byte_count (&p);
>> + }
>> +
>> + CORE_ADDR next_address = 0;
>> + if (p != 0 && *p != 0)
>> + next_address = parse_and_eval_address (p);
>> + else
>> + error (_("Usage: adi assign|a[/count] <addr> = <version>"));
>> +
>> + int version = 0;
>> + if (q) /* parse version tag */
>> + {
>> + version = parse_and_eval_long (q);
>> + if (version < 0 || version > adi_stat.max_version)
>> + error (_("Invalid ADI version tag %d"), version);
>> + }
>> +
>> + do_assign(next_address, cnt, version);
>> +}
>> +
>> +extern initialize_file_ftype _initialize_sparc64_adi_tdep; /* -Wmissing-prototypes */
>> +
>>
>
> This line is not needed.
>
I thought "lint" would need this line so it can be added to init.c?
>> +void
>> +_initialize_sparc64_adi_tdep (void)
>> +{
>> +
>> + add_prefix_cmd ("adi", class_support, info_adi_command,
>> + _("ADI version related commands."),
>> + &sparc64adilist, "adi ", 0, &cmdlist);
>> + add_cmd ("examine", class_support, adi_examine_command,
>> + _("Examine ADI versions."), &sparc64adilist);
>> + add_alias_cmd ("x", "examine", no_class, 1, &sparc64adilist);
>> + add_cmd ("assign", class_support, adi_assign_command,
>> + _("Assign ADI versions."), &sparc64adilist);
>> +
>> +}
>> +
>> diff --git a/gdb/sparc64-linux-nat.c b/gdb/sparc64-linux-nat.c
>> index 638aa29..01adcd6 100644
>> --- a/gdb/sparc64-linux-nat.c
>> +++ b/gdb/sparc64-linux-nat.c
>> @@ -69,6 +69,18 @@ fill_fpregset (const struct regcache *regcache,
>> sparc64_collect_fpregset (&sparc64_bsd_fpregmap, regcache, regnum, fpregs);
>> }
>>
>> +/* Implement the "to_watchpoint_addr_within_range" target_ops method. */
>> +
>> +static int
>> +sparc64_linux_watchpoint_addr_within_range (struct target_ops *target,
>> + CORE_ADDR addr,
>> + CORE_ADDR start, int length)
>> +{
>> + start = adi_normalize_address (start);
>> + addr = adi_normalize_address (addr);
>> + return (start <= addr) && (start + length - 1 >= addr);
>>
>
> I suppose once you implement gdbarch_addr_bits_remove, you don't need
> this change.
>
OK. Will check that.
>> +}
>> +
>> /* Provide a prototype to silence -Wmissing-prototypes. */
>> void _initialize_sparc64_linux_nat (void);
>>
>> @@ -86,6 +98,9 @@ _initialize_sparc64_linux_nat (void)
>> t->to_fetch_registers = sparc_fetch_inferior_registers;
>> t->to_store_registers = sparc_store_inferior_registers;
>>
>> + t->to_watchpoint_addr_within_range =
>> + sparc64_linux_watchpoint_addr_within_range;
>> +
>> /* Register the target. */
>> linux_nat_add_target (t);
>>
>> diff --git a/gdb/sparc64-linux-tdep.c b/gdb/sparc64-linux-tdep.c
>> index 814bc29..17eb8c5 100644
>> --- a/gdb/sparc64-linux-tdep.c
>> +++ b/gdb/sparc64-linux-tdep.c
>> @@ -32,6 +32,7 @@
>> #include "tramp-frame.h"
>> #include "xml-syscall.h"
>> #include "linux-tdep.h"
>> +#include <adi.h>
>>
>> /* The syscall's XML filename for sparc 64-bit. */
>> #define XML_SYSCALL_FILENAME_SPARC64 "syscalls/sparc64-linux.xml"
>> @@ -104,6 +105,52 @@ sparc64_linux_sigframe_init (const struct tramp_frame *self,
>> }
>> trad_frame_set_id (this_cache, frame_id_build (base, func));
>> }
>> +
>> +/* sparc64 GNU/Linux implementation of the handle_segmentation_fault
>> + gdbarch hook.
>> + Displays information related to ADI memory corruptions. */
>> +
>> +void
>> +sparc64_linux_handle_segmentation_fault (struct gdbarch *gdbarch,
>> + struct ui_out *uiout)
>> +{
>> + if (gdbarch_bfd_arch_info (gdbarch)->bits_per_word != 64)
>> + return;
>> +
>>
>
> Do you need to return if some sparc64 processor doesn't have adi?
>
No. The code that followed will check for the si_code bit which would
signal if the segfault is caused by adi.
>> + CORE_ADDR addr = 0;
>> + long si_code = 0;
>> +
>> + TRY
>> + {
>> + /* Evaluate si_code to see if the segfault is ADI related. */
>> + si_code = parse_and_eval_long ("$_siginfo.si_code\n");
>> +
>> + if (si_code >= SEGV_ACCADI && si_code <= SEGV_ADIPERR)
>> + addr = parse_and_eval_long ("$_siginfo._sifields._sigfault.si_addr");
>> + }
>> + CATCH (exception, RETURN_MASK_ALL)
>> + {
>> + return;
>> + }
>> + END_CATCH
>> +
>> + /* Print out ADI event based on sig_code value */
>> + switch (si_code)
>> + {
>> + case SEGV_ACCADI: /* adi not enabled */
>> + printf_unfiltered (" (ADI disabled at address 0x%lx)\n", addr);
>> + break;
>> + case SEGV_ADIDERR: /* disrupting mismatch */
>> + printf_unfiltered (" (ADI deferred mismatch at address 0x%lx)\n", addr);
>> + break;
>> + case SEGV_ADIPERR: /* precise mismatch */
>> + printf_unfiltered (" (ADI precise mismatch at address
>> 0x%lx)\n", addr);
>>
>
> Use uiout instead print_unfiltered. See i386_linux_handle_segmentation_fault.
>
OK.
>
>> + break;
>> + default:
>> + break;
>> + }
>> +}
>> +
>>
>> /* Return the address of a system call's alternative return
>> address. */
>> @@ -338,6 +385,8 @@ sparc64_linux_init_abi (struct gdbarch_info info, struct gdbarch *gdbarch)
>> set_xml_syscall_file_name (gdbarch, XML_SYSCALL_FILENAME_SPARC64);
>> set_gdbarch_get_syscall_number (gdbarch,
>> sparc64_linux_get_syscall_number);
>> + set_gdbarch_handle_segmentation_fault (gdbarch,
>> + sparc64_linux_handle_segmentation_fault);
>> }
>>
>>
>> diff --git a/gdb/sparc64-tdep.c b/gdb/sparc64-tdep.c
>> index 9e0e6b5..e8c5a63 100644
>> --- a/gdb/sparc64-tdep.c
>> +++ b/gdb/sparc64-tdep.c
>> @@ -165,7 +165,7 @@ sparc64_pstate_type (struct gdbarch *gdbarch)
>> append_flags_type_flag (type, 8, "TLE");
>> append_flags_type_flag (type, 9, "CLE");
>> append_flags_type_flag (type, 10, "PID0");
>> - append_flags_type_flag (type, 11, "PID1");
>> + append_flags_type_flag (type, 11, "MCDE");
>>
>>
>
> Why do you do this change?
>
This change, which indicated that the TSTATE reigster contained the
"mcde" value, was requested by our kernel engineers.
Thanks very much for your comments.
More information about the Gdb-patches
mailing list