Re: [PATCH v3 6/6] gdb/doc: Document AArch64 FPMR support

[Luis <luis.machado.foss@gmail.com>]

Hi Ezra,


On 21/10/2025 16:03, Ezra.Sitorus@arm.com wrote:
> From: Ezra Sitorus <ezra.sitorus@arm.com>
> 
> Add NEWS entry and document new feature in gdb.texinfo.
> 
> ---
> Changes from v1->v3:
> * This is new in v3.
> 
> Ezra
> 
>   gdb/NEWS            | 104 ++++++++++++++++++++++----------------------
>   gdb/doc/gdb.texinfo |  71 ++++++++++++++++++++++++++++++
>   2 files changed, 124 insertions(+), 51 deletions(-)
> 
> diff --git a/gdb/NEWS b/gdb/NEWS
> index 01bd1524c7a..75e979b4e9f 100644
> --- a/gdb/NEWS
> +++ b/gdb/NEWS
> @@ -14,6 +14,8 @@
>     this flag is used gdbserver will not escape special shell characters
>     within the inferior arguments.
>   
> +* Support for Floating Point Mode Register (FPMR) in AArch64.
> +
>   * New targets
>   
>   GNU/Linux/MicroBlaze (gdbserver) microblazeel-*linux*
> @@ -1380,7 +1382,7 @@ AMDGPU amdgcn-*-*
>   * New features in the GDB remote stub, GDBserver
>   
>     ** GDBserver is now supported on LoongArch GNU/Linux.
> -
> +

There seems to be a lot of spurious changes like the above in your NEWS 
patch. Maybe a newline/editor issue?

>     ** GDBserver is now supported on CSKY GNU/Linux.
>   
>   * LoongArch floating-point support
> @@ -2274,7 +2276,7 @@ pipe -d DELIM COMMAND DELIM SHELL_COMMAND
>   
>   define-prefix COMMAND
>     Define or mark a command as a user-defined prefix command.
> -
> +
>   with SETTING [VALUE] [-- COMMAND]
>   w SETTING [VALUE] [-- COMMAND]
>     Temporarily set SETTING, run COMMAND, and restore SETTING.
> @@ -3384,7 +3386,7 @@ show disassembler-options
>          0x40057b <main(int, char**)+25>:
>           callq  0x400536 <Func1(int, char const*)>
>   
> -* Fortran: Support structures with fields of dynamic types and
> +* Fortran: Support structures with fields of dynamic types and
>     arrays of dynamic types.
>   
>   * The symbol dumping maintenance commands have new syntax.
> @@ -3996,7 +3998,7 @@ vforkdone stop reason
>     an exec or exit, allowing the vfork parent to resume execution.
>   
>   fork-events and vfork-events features in qSupported
> -  The qSupported packet allows GDB to request support for fork and
> +  The qSupported packet allows GDB to request support for fork and
>     vfork events using new 'gdbfeatures' fork-events and vfork-events,
>     and the qSupported response can contain the corresponding
>     'stubfeatures'.  Set and show commands can be used to display
> @@ -4065,7 +4067,7 @@ Itanium running HP-UX         ia64-*-hpux*
>     ** You can now add attributes to gdb.Objfile and gdb.Progspace objects.
>     ** New function gdb.lookup_objfile.
>   
> -  New events which are triggered when GDB modifies the state of the
> +  New events which are triggered when GDB modifies the state of the
>     inferior.
>   
>     ** gdb.events.inferior_call_pre: Function call is about to be made.
> @@ -4732,7 +4734,7 @@ Tilera TILE-Gx GNU/Linux	tilegx*-*-linux
>   
>   * New commands (for set/show, see "New options" below)
>   
> -catch signal
> +catch signal
>     Catch signals.  This is similar to "handle", but allows commands and
>     conditions to be attached.
>   
> @@ -4816,14 +4818,14 @@ show debug notification
>        "=memory-changed".
>     ** The data-disassemble command response will include a "fullname" field
>        containing the absolute file name when source has been requested.
> -  ** New optional parameter COUNT added to the "-data-write-memory-bytes"
> +  ** New optional parameter COUNT added to the "-data-write-memory-bytes"
>        command, to allow pattern filling of memory areas.
>     ** New commands "-catch-load"/"-catch-unload" added for intercepting
>        library load/unload events.
>     ** The response to breakpoint commands and breakpoint async records
>        includes an "installed" field containing a boolean state about each
>        non-pending tracepoint location is whether installed on target or not.
> -  ** Output of the "-trace-status" command includes a "trace-file" field
> +  ** Output of the "-trace-status" command includes a "trace-file" field
>        containing the name of the trace file being examined.  This field is
>        optional, and only present when examining a trace file.
>     ** The "fullname" field is now always present along with the "file" field,
> @@ -4880,8 +4882,8 @@ qXfer:btrace:read
>     in <http://sourceware.org/systemtap/>.
>   
>   * GDB now supports reversible debugging on ARM, it allows you to
> -  debug basic ARM and THUMB instructions, and provides
> -  record/replay support.
> +  debug basic ARM and THUMB instructions, and provides
> +  record/replay support.
>   
>   * The option "symbol-reloading" has been deleted as it is no longer used.
>   
> @@ -5193,7 +5195,7 @@ QProgramSignals:
>        replaced it.  Additionally, the default for "print-stack" is
>        now "message", which just prints the error message without
>        the stack trace.
> -
> +
>     ** A prompt substitution hook (prompt_hook) is now available to the
>        Python API.
>   
> @@ -5220,7 +5222,7 @@ QProgramSignals:
>   
>     ** A new class "gdb.FinishBreakpoint" is provided to catch the return
>        of a function.  This class is based on the "finish" command
> -     available in the CLI.
> +     available in the CLI.
>   
>     ** Type objects for struct and union types now allow access to
>        the fields using standard Python dictionary (mapping) methods.
> @@ -5397,7 +5399,7 @@ show trace-stop-notes
>   * New remote packets
>   
>   QTEnable
> -
> +
>     Dynamically enable a tracepoint in a started trace experiment.
>   
>   QTDisable
> @@ -5484,7 +5486,7 @@ Renesas RL78				rl78-*-elf
>   
>     ** Breakpoints can now be sub-classed in Python, and in particular
>        you may implement a 'stop' function that is executed each time
> -     the inferior reaches that breakpoint.
> +     the inferior reaches that breakpoint.
>   
>     ** New function gdb.lookup_global_symbol looks up a global symbol.
>   
> @@ -5639,8 +5641,8 @@ Analog Devices, Inc. Blackfin Processor	bfin-*
>     arguments even if the namespace has not been imported.
>     For example:
>       namespace A
> -      {
> -        class B { };
> +      {
> +        class B { };
>           void foo (B) { }
>         }
>       ...
> @@ -6040,7 +6042,7 @@ source
>   * New commands (for set/show, see "New options" below)
>   
>   record save [<FILENAME>]
> -  Save a file (in core file format) containing the process record
> +  Save a file (in core file format) containing the process record
>     execution log for replay debugging at a later time.
>   
>   record restore <FILENAME>
> @@ -6312,7 +6314,7 @@ qXfer:siginfo:write
>   
>   * GDB now supports multiple function calling conventions according to the
>   DWARF-2 DW_AT_calling_convention function attribute.
> -
> +
>   * The SH target utilizes the aforementioned change to distinguish between gcc
>   and Renesas calling convention.  It also adds the new CLI commands
>   `set/show sh calling-convention'.
> @@ -6568,7 +6570,7 @@ show multiple-symbols
>     The value of this variable can be changed to adjust the debugger behavior
>     when an expression or a breakpoint location contains an ambiguous symbol
>     name (an overloaded function name, for instance).
> -
> +
>   set breakpoint always-inserted
>   show breakpoint always-inserted
>     Keep breakpoints always inserted in the target, as opposed to inserting
> @@ -6811,7 +6813,7 @@ vRun
>   
>   *** Changes in GDB 6.7
>   
> -* Resolved 101 resource leaks, null pointer dereferences, etc. in gdb,
> +* Resolved 101 resource leaks, null pointer dereferences, etc. in gdb,
>   bfd, libiberty and opcodes, as revealed by static analysis donated by
>   Coverity, Inc. (http://scan.coverity.com).
>   
> @@ -6823,7 +6825,7 @@ symbol definition in the current shared library if it was built using the
>   recognize the -tui command-line option and print a message that the TUI
>   is not supported.
>   
> -* The GDB remote stub, gdbserver, now has lower overhead for high
> +* The GDB remote stub, gdbserver, now has lower overhead for high
>   frequency signals (e.g. SIGALRM) via the QPassSignals packet.
>   
>   * GDB for MIPS targets now autodetects whether a remote target provides
> @@ -6861,7 +6863,7 @@ has been rewritten to use the standard GDB remote protocol.
>   layout.  It also supports a TextSeg= and DataSeg= response when only
>   segment base addresses (rather than offsets) are available.
>   
> -* The /i format now outputs any trailing branch delay slot instructions
> +* The /i format now outputs any trailing branch delay slot instructions
>   immediately following the last instruction within the count specified.
>   
>   * The GDB remote protocol "T" stop reply packet now supports a
> @@ -7008,7 +7010,7 @@ target ocd
>   
>   DWARF 1 support
>   
> -	A debug information format.  The predecessor to DWARF 2 and
> +	A debug information format.  The predecessor to DWARF 2 and
>   	DWARF 3, which are still supported.
>   
>   Support for the HP aCC compiler on HP-UX/PA-RISC
> @@ -7127,7 +7129,7 @@ The following commands are presently only implemented for native GNU/Linux:
>   
>   checkpoint			Save a snapshot of the program state.
>   
> -restart	<n>			Return the program state to a
> +restart	<n>			Return the program state to a
>   				previously saved state.
>   
>   info checkpoints		List currently saved checkpoints.
> @@ -7272,8 +7274,8 @@ continued, we're looking forward to our first translation.
>   
>   * Ada
>   
> -Initial support for debugging programs compiled with the GNAT
> -implementation of the Ada programming language has been integrated
> +Initial support for debugging programs compiled with the GNAT
> +implementation of the Ada programming language has been integrated
>   into GDB.  In this release, support is limited to expression evaluation.
>   
>   * New native configurations
> @@ -7740,7 +7742,7 @@ Fujitsu SPARClite 			sparclite-fujitsu-none  or  sparclite
>   
>   * REMOVED configurations and files
>   
> -V850EA ISA				
> +V850EA ISA
>   Motorola Delta 88000 running Sys V		m88k-motorola-sysv  or  delta88
>   IBM AIX PS/2					i[3456]86-*-aix
>   i386 running Mach 3.0				i[3456]86-*-mach3*
> @@ -7775,7 +7777,7 @@ shared libs like mad''.
>   
>   * ``gdbserver'' now supports multi-threaded applications on some targets
>   
> -Support for debugging multi-threaded applications which use
> +Support for debugging multi-threaded applications which use
>   the GNU/Linux LinuxThreads package has been added for
>   arm*-*-linux*-gnu*, i[3456]86-*-linux*-gnu*, mips*-*-linux*-gnu*,
>   powerpc*-*-linux*-gnu*, and sh*-*-linux*-gnu*.
> @@ -7890,11 +7892,11 @@ gdb/439: gdb/291: On some ELF object files, gdb was reporting:
>   dwarf2read.c:1072: gdb-internal-error: sect_index_text not initialize
>   Fix, by Fred Fish, imported from mainline.
>   
> -Dwarf2 .debug_frame & .eh_frame handler improved in many ways.
> +Dwarf2 .debug_frame & .eh_frame handler improved in many ways.
>   Surprisingly enough, it works now.
>   By Michal Ludvig, imported from mainline.
>   
> -i386 hardware watchpoint support:
> +i386 hardware watchpoint support:
>   avoid misses on second run for some targets.
>   By Pierre Muller, imported from mainline.
>   
> @@ -7918,18 +7920,18 @@ hosts.  Argument is core file name (defaults to core.<pid>).
>   
>   * New command line option
>   
> -GDB now accepts --pid or -p followed by a process id.
> +GDB now accepts --pid or -p followed by a process id.
>   
>   * Change in command line behavior -- corefiles vs. process ids.
>   
> -There is a subtle behavior in the way in which GDB handles
> +There is a subtle behavior in the way in which GDB handles
>   command line arguments.  The first non-flag argument is always
>   a program to debug, but the second non-flag argument may either
>   be a corefile or a process id.  Previously, GDB would attempt to
>   open the second argument as a corefile, and if that failed, would
>   issue a superfluous error message and then attempt to attach it as
> -a process.  Now, if the second argument begins with a non-digit,
> -it will be treated as a corefile.  If it begins with a digit,
> +a process.  Now, if the second argument begins with a non-digit,
> +it will be treated as a corefile.  If it begins with a digit,
>   GDB will attempt to attach it as a process, and if no such process
>   is found, will then attempt to open it as a corefile.
>   
> @@ -7989,7 +7991,7 @@ for the inferior from gdb's command line.
>   
>   There is a new `operate-and-get-next' function bound to `C-o'.
>   
> -*** Changes in GDB 5.1.1
> +*** Changes in GDB 5.1.1
>   
>   Fix compile problem on DJGPP.
>   
> @@ -8021,7 +8023,7 @@ UltraSparc running GNU/Linux			sparc64-*-linux*
>   
>   * OBSOLETE configurations and files
>   
> -x86 FreeBSD before 2.2				i[3456]86*-freebsd{1,2.[01]}*,
> +x86 FreeBSD before 2.2				i[3456]86*-freebsd{1,2.[01]}*,
>   Harris/CXUX m88k				m88*-harris-cxux*
>   Most ns32k hosts and targets			ns32k-*-mach3* ns32k-umax-*
>   						ns32k-utek-sysv* ns32k-utek-*
> @@ -8902,7 +8904,7 @@ There is a new hardware breakpoint for the watch command for sparclite
>   targets.  See gdb/sparclite/hw_breakpoint.note.
>   
>   Hardware watchpoints are also now supported under GNU/Linux.
> -
> +
>   * Annotations
>   
>   Annotations have been added.  These are for use with graphical interfaces,
> @@ -8983,7 +8985,7 @@ for details).
>   
>   * Improvements in C++ mangling/demangling.
>   
> -This release has much better g++ debugging, specifically in name
> +This release has much better g++ debugging, specifically in name
>   mangling/demangling, virtual function calls, print virtual table,
>   call methods, ...etc.
>   
> @@ -9045,7 +9047,7 @@ Z8000 simulator		  		z8k-zilog-none 	  or z8ksim
>   IDT MIPS board over serial line		mips-idt-ecoff
>   
>   Cross-debugging to GO32 targets is supported.  It requires a custom
> -version of the i386-stub.c module which is integrated with the
> +version of the i386-stub.c module which is integrated with the
>   GO32 memory extender.
>   
>    * New remote protocols
> @@ -9666,7 +9668,7 @@ encapulation, but the details have not been worked out yet.
>    *  Improved configuration
>   
>   Only one copy of `configure' exists now, and it is not self-modifying.
> -Porting BFD is simpler.
> +Porting BFD is simpler.
>   
>    *  Stepping improved
>   
> @@ -9768,7 +9770,7 @@ stub on the target system.
>   
>   New CPUs supported include the AMD 29000 and Intel 960.
>   
> -GDB now reads object files and symbol tables via a ``binary file''
> +GDB now reads object files and symbol tables via a ``binary file''
>   library, which allows a single copy of GDB to debug programs of multiple
>   object file types such as a.out and coff.
>   
> @@ -9794,27 +9796,27 @@ confirm on/off:  Enables warning questions for operations that are
>   		 hard to recover from, e.g. rerunning the program while
>   		 it is already running.  Default is ON.
>   
> -editing on/off:  Enables EMACS style command line editing
> -                 of input.  Previous lines can be recalled with
> +editing on/off:  Enables EMACS style command line editing
> +                 of input.  Previous lines can be recalled with
>   		 control-P, the current line can be edited with control-B,
>   		 you can search for commands with control-R, etc.
>   		 Default is ON.
>   
> -history filename NAME:  NAME is where the gdb command history
> +history filename NAME:  NAME is where the gdb command history
>   			will be stored.  The default is .gdb_history,
>   			or the value of the environment variable
>   			GDBHISTFILE.
>   
> -history size N:  The size, in commands, of the command history.  The
> +history size N:  The size, in commands, of the command history.  The
>   		 default is 256, or the value of the environment variable
>   		 HISTSIZE.
>   
>   history save on/off: If this value is set to ON, the history file will
> -		      be saved after exiting gdb.  If set to OFF, the
> +		      be saved after exiting gdb.  If set to OFF, the
>   		      file will not be saved.  The default is OFF.
>   
> -history expansion on/off: If this value is set to ON, then csh-like
> -			  history expansion will be performed  on
> +history expansion on/off: If this value is set to ON, then csh-like
> +			  history expansion will be performed  on
>   			  command line input.  The default is OFF.
>   
>   radix N:  Sets the default radix for input and output.  It can be set
> @@ -9839,7 +9841,7 @@ print address on/off:  Print memory addresses in various command displays,
>   		      more ``symbolic'' if you turn this off; it looks more
>   		      ``machine level'' with it on.  Default is ON.
>   
> -print array on/off:  Prettyprint arrays.  New convenient format!  Default
> +print array on/off:  Prettyprint arrays.  New convenient format!  Default
>                       is OFF.
>   
>   print demangle on/off:   Print C++ symbols in "source" form if on,
> @@ -9870,7 +9872,7 @@ from dynamically linked programs), gdb reads the symbols from each
>   shared library when you type the ``sharedlibrary'' command.
>   It can be abbreviated ``share''.
>   
> -sharedlibrary REGEXP:  Load shared object library symbols for files
> +sharedlibrary REGEXP:  Load shared object library symbols for files
>                          matching a unix regular expression.  No argument
>   		       indicates to load symbols for all shared libraries.
>   
> @@ -9938,7 +9940,7 @@ find your source file in the current directory, type "dir .".
>   For normal use, type ``./configure host''.  See README or gdb.texinfo
>   for more details.
>   
> -GDB now handles cross debugging.  If you are remotely debugging between
> +GDB now handles cross debugging.  If you are remotely debugging between
>   two different machines, type ``./configure host -target=targ''.
>   Host is the machine where GDB will run; targ is the machine
>   where the program that you are debugging will run.
> diff --git a/gdb/doc/gdb.texinfo b/gdb/doc/gdb.texinfo
> index 676143be416..0ec1b92d7c3 100644
> --- a/gdb/doc/gdb.texinfo
> +++ b/gdb/doc/gdb.texinfo
> @@ -27080,6 +27080,26 @@ Program received signal SIGSEGV, Segmentation fault
>   Guarded Control Stack error.
>   @end smallexample
>   
> +@subsubsection AArch64 Floating Point Mode Register
> +@cindex Floating Point Mode Register, AArch64
> +@cindex FPMR, AArch64
> +
> +When @value{GDBN} is debugging the AArch64 architecture and the Floating Point
> +Mode Register (FPMR) is available, then @value{GDBN} will make the @code{fpmr}
> +available. This register determines the behaviour of FP8 instructions.
> +
> +To aid debugging, @value{GDBN} interprets the mode or value of each field in
> +@code{fpmr}.
> +
> +@smallexample
> +p $fpmr
> +$1 = [ F8S1=E5M2 F8S2=E5M2 F8D=E5M2 OSM=Inf OSC=Inf/NaN LSCALE=0 NSCALE=0 LSCALE2=0 ]
> +@end smallexample
> +
> +For more information about FPMR, please refer to the official
> +@uref{https://developer.arm.com/documentation/ddi0601/latest,ignored,
> +architecture registers documentation}.
> +
>   @node x86
>   @subsection x86
>   
> @@ -49770,6 +49790,57 @@ documentation} in the Linux kernel.
>   Extra registers are allowed in these features, but they will not affect
>   @value{GDBN}.
>   
> +@subsubsection AArch64 FPMR feature
> +
> +The @samp{org.gnu.gdb.aarch64.fpmr} optional feature was introduced so
> +@value{GDBN} could detect support for the Floating Point Mode Register
> +extension.  If present, this feature must have the following register:
> +
> +@itemize @minus
> +@item
> +@samp{fpmr}, the tag control register.  It is represented as if it were a 64-bit
> +register with a custom flags type.
> +@end itemize
> +
> +The custom flags type allows @value{GDBN} to print a human-friendly
> +representation of the contents of @code{fpmr} should contain:
> +
> +@itemize @minus
> +
> +@item
> +@code{F8S1}, which is either @code{E5M2} or @code{E4M3}
> +
> +@item
> +@code{F8S2}, which is either @code{E5M2} or @code{E4M3}
> +
> +@item
> +@code{F8D}, which is either @code{E5M2} or @code{E4M3}
> +
> +@item
> +@code{OSM}, which is either @code{Inf} or @code{MaxNormal}
> +
> +@item
> +@code{OSC}, which is either @code{Inf/NaN} or @code{MaxNormal}
> +
> +@item
> +@code{LSCALE}, which is a 7-bit unsigned value
> +
> +@item
> +@code{NSCALE}, which is actually an 8-bit signed value but represented as an
> +8-bit unsigned value
> +
> +@item
> +@code{LSCALE2}, which is a 6-bit unsigned value
> +
> +@end itemize
> +
> +Extra registers are allowed in this feature, but they will not affect
> +@value{GDBN}.
> +
> +For more information about FPMR, please refer to the official
> +@uref{https://developer.arm.com/documentation/ddi0601/latest,ignored,
> +architecture registers documentation}.
> +
>   @node Alpha Features
>   @subsection Alpha Features
>   @cindex target descriptions, Alpha Features