RE: [PATCH v3 7/8] Windows gdb: Implement AVX register support

["Rohr, Stephan" <stephan.rohr@intel.com>]

Hi Hannes,

please see some feedback inlined below.

Let me know if you have any questions.

Thanks
Stephan

> -----Original Message-----
> From: Hannes Domani <ssbssa@yahoo.de>
> Sent: Saturday, 29 August 2026 16:49
> To: gdb-patches@sourceware.org
> Subject: [PATCH v3 7/8] Windows gdb: Implement AVX register support
> 
> This adds support for the Intel AVX registers on Windows.
> It enables accessing registers $ymm0 - $ymm15 where they are available.
> 
> After this patch gdb.arch/i386-avx.exp passes on windows.
> ---
> v3:
>   - merged gdb+gdbserver parts, and split again AVX/AVX-512 parts
> ---
>  gdb/NEWS                    |  3 ++
>  gdb/nat/windows-nat.c       |  2 +-
>  gdb/x86-windows-nat.c       | 67
> +++++++++++++++++++++++++++++++++++--
>  gdbserver/win32-i386-low.cc | 61 +++++++++++++++++++++++++++++----
>  gdbserver/win32-low.cc      | 15 ++++++---
>  5 files changed, 133 insertions(+), 15 deletions(-)
> 
> diff --git a/gdb/NEWS b/gdb/NEWS
> index 10c182067f9..f7effc822e9 100644
> --- a/gdb/NEWS
> +++ b/gdb/NEWS
> @@ -118,6 +118,9 @@
>    intent to remove it in a future release.
>    The s390 64-bit target (s390x-*) remains supported.
> 
> +* Support for Intel AVX registers on Windows.
> +  Support displaying and modifying Intel AVX registers $ymm0 - $ymm31.
> +

I think this should be registers $ymm0 - $ymm15 ?

>  * Configure changes
> 
>  ** --with-babeltrace has been removed.  The babeltrace library was
> diff --git a/gdb/nat/windows-nat.c b/gdb/nat/windows-nat.c
> index 8930536f3ba..c9a21d7c41f 100644
> --- a/gdb/nat/windows-nat.c
> +++ b/gdb/nat/windows-nat.c
> @@ -1339,7 +1339,7 @@ initialize_loadable ()
>      {
>        /* Available XState features masked with implemented features.  */
>        xstate_features = (GetEnabledXStateFeatures ()
> -			 & X86_XSTATE_SSE_MASK);
> +			 & X86_XSTATE_AVX_MASK);
>        /* The extended XState functions are only needed if the available
>  	 features exceed SSE.  */
>        if ((xstate_features & ~X86_XSTATE_SSE_MASK) == 0)
> diff --git a/gdb/x86-windows-nat.c b/gdb/x86-windows-nat.c
> index 3af5ef4dae0..1cefe6171be 100644
> --- a/gdb/x86-windows-nat.c
> +++ b/gdb/x86-windows-nat.c
> @@ -27,6 +27,9 @@
> 
>  #include "i386-tdep.h"
>  #include "i387-tdep.h"
> +#ifdef __x86_64__
> +#include "amd64-tdep.h"
> +#endif
> 
>  using namespace windows_nat;
> 
> @@ -70,6 +73,8 @@ struct x86_windows_nat_target final : public
> x86_nat_target<windows_nat_target>
>  			   windows_thread_info *th, int r) override;
> 
>    bool is_sw_breakpoint (const EXCEPTION_RECORD *er) const override;
> +
> +  const struct target_desc *read_description () override;
>  };
> 
>  /* The current process.  */
> @@ -109,7 +114,31 @@ x86_windows_per_inferior::fill_thread_context
> (windows_thread_info *th)
>        if (context->ContextFlags == 0)
>  	{
>  	  context->ContextFlags = WindowsContext<decltype(context)>::all;
> +	  if (xstate_features != 0)
> +	    {
> +	      context->ContextFlags |= CONTEXT_XSTATE_FLAG;
> +	      set_xstate_features_mask (context, xstate_features);
> +	    }

We have the same code in "i386_get_thread_context" in "win32-i386-low.cc".
Make a shared function in gdb/nat/windows-nat.h?

>  	  CHECK (get_thread_context (th->h, context));
> +
> +	  if (xstate_features != 0)
> +	    {
> +	      DWORD64 features = 0;
> +	      CHECK (get_xstate_features_mask (context, &features));

Should this be changed to sth. like 

  if (!get_xstate_features_mask (context, &features))
    {
      warning (..)
      return;
    }

The call of "CHECK" only prints a message but doesn't error out.  If this
call fails we may still have features == 0.  This implies
"zeroed_features == xstate_features".  With this, the loop clears all features.
IIUC, this would clear the AVX registers on the next call of
"SetThreadContext".

Also refer to the implementation in gdbserver/win32-i386-low.cc:

      DWORD64 features = 0;
      if (xstate_features != 0
	  && get_xstate_features_mask (context, &features))
	{

I think it makes sense to unify those as the rest of the code is basically identical.  Put 
shared function into gdb/nat/windows-nat.h?  This keeps the code consistent.

> +	      DWORD64 zeroed_features = xstate_features & ~features;
> +
> +	      for (int f = X86_XSTATE_AVX_ID; f <= X86_XSTATE_CET_U_ID; f++)
> +		{
> +		  DWORD64 flag = 1ULL << f;
> +		  if ((zeroed_features & flag) != 0)
> +		    {
> +		      DWORD size = 0;
> +		      void *loc = locate_xstate_feature (context, f, &size);
> +		      if (loc != nullptr && size > 0)
> +			memset (loc, 0, size);
> +		    }
> +		}
> +	    }
>  	}
>      });
>  }
> @@ -198,6 +227,14 @@
> x86_windows_nat_target::thread_context_continue (windows_thread_info
> *th,
>  	  if (GetExitCodeThread (th->h, &ec)
>  	      && ec == STILL_ACTIVE)
>  	    {
> +	      DWORD debug_registers =
> WindowsContext<decltype(context)>::debug;
> +	      if (xstate_features != 0
> +		  && (context->ContextFlags & ~debug_registers) != 0)
> +		{
> +		  context->ContextFlags |= CONTEXT_XSTATE_FLAG;
> +		  set_xstate_features_mask (context, xstate_features);
> +		}
> +
>  	      BOOL status = set_thread_context (th->h, context);
> 
>  	      if (!killed)
> @@ -227,7 +264,7 @@ x86_windows_nat_target::thread_context_step
> (windows_thread_info *th,
> 
>  template<typename Context>
>  static char *
> -get_context_reg_ptr (Context *context, int r)
> +get_context_reg_ptr (Context *context, int r, i386_gdbarch_tdep *tdep)
>  {
>    const int *mappings;
>    int mappings_count;
> @@ -247,6 +284,13 @@ get_context_reg_ptr (Context *context, int r)
>    char *context_offset;
>    if (r < mappings_count)
>      context_offset = (char *) context + mappings[r];
> +  else if (I387_YMM0H_REGNUM (tdep) > 0 && r >= I387_YMM0H_REGNUM
> (tdep)
> +	   && r < I387_YMMENDH_REGNUM (tdep))

The implementation on gdbserver side guards against 

  xstate_features & X86_XSTATE_AVX) != 0

I wonder if the same guard would be helpful here, too.  I understand the register
number is initialized to -1, so this should not fire.  I'm not sure if it is possible to
have $ymm0 register number > 0 w/o xstate support, e.g., if the target description
is read from file, see "target_find_description"?

> +    {
> +      context_offset = (char *) locate_xstate_feature
> +	(context, X86_XSTATE_AVX_ID, NULL);
> +      context_offset += 16 * (r - I387_YMM0H_REGNUM (tdep));
> +    }
>    else
>      gdb_assert_not_reached ("invalid register number %d", r);
> 
> @@ -267,7 +311,7 @@ x86_windows_nat_target::fetch_one_register (struct
> regcache *regcache,
>    char *context_offset
>      = x86_windows_process.with_context (th, [&] (auto *context)
>      {
> -      return get_context_reg_ptr (context, r);
> +      return get_context_reg_ptr (context, r, tdep);
>      });
> 
>    gdb_assert (!gdbarch_read_pc_p (gdbarch));
> @@ -333,7 +377,7 @@ x86_windows_nat_target::store_one_register (const
> struct regcache *regcache,
>      = x86_windows_process.with_context (th, [&] (auto *context)
>      {
>        gdb_assert (context->ContextFlags != 0);
> -      return get_context_reg_ptr (context, r);
> +      return get_context_reg_ptr (context, r, tdep);
>      });
> 
>    /* GDB treats some registers as 32-bit, where they are in fact only
> @@ -368,6 +412,23 @@ x86_windows_nat_target::is_sw_breakpoint (const
> EXCEPTION_RECORD *er) const
>  	  || er->ExceptionCode == STATUS_WX86_BREAKPOINT);
>  }
> 
> +const struct target_desc *
> +x86_windows_nat_target::read_description ()
> +{
> +  if (inferior_ptid == null_ptid)
> +    return this->beneath ()->read_description ();
> +
> +  if (xstate_features == 0)
> +    return nullptr;
> +
> +#ifdef __x86_64__
> +  if (!x86_windows_process.wow64_process)
> +    return amd64_target_description (xstate_features, false);
> +  else
> +#endif
> +    return i386_target_description (xstate_features, false);
> +}
> +
>  /* Hardware watchpoint support, adapted from go32-nat.c code.  */
> 
>  /* Pass the address ADDR to the inferior in the I'th debug register.
> diff --git a/gdbserver/win32-i386-low.cc b/gdbserver/win32-i386-low.cc
> index b77f6adc6ed..a7e83c0239c 100644
> --- a/gdbserver/win32-i386-low.cc
> +++ b/gdbserver/win32-i386-low.cc
> @@ -253,6 +253,11 @@ i386_get_thread_context (windows_thread_info
> *th)
>  			       | WindowsContext<decltype(context)>::floating
>  			       | WindowsContext<decltype(context)>::debug
>  			       | extended_registers);
> +      if (xstate_features != 0)
> +	{
> +	  context->ContextFlags |= CONTEXT_XSTATE_FLAG;
> +	  set_xstate_features_mask (context, xstate_features);
> +	}
> 
>        BOOL ret = get_thread_context (th->h, context);
>        if (!ret)
> @@ -267,6 +272,24 @@ i386_get_thread_context (windows_thread_info
> *th)
> 
>  	  error (_("GetThreadContext failure %ld\n"), (long) e);
>  	}
> +
> +      DWORD64 features = 0;
> +      if (xstate_features != 0
> +	  && get_xstate_features_mask (context, &features))
> +	{
> +	  DWORD64 zeroed_features = xstate_features & ~features;
> +	  for (int f = X86_XSTATE_AVX_ID; f <= X86_XSTATE_CET_U_ID; f++)
> +	    {
> +	      DWORD64 flag = 1ULL << f;
> +	      if ((zeroed_features & flag) != 0)
> +		{
> +		  DWORD size = 0;
> +		  void *loc = locate_xstate_feature (context, f, &size);
> +		  if (loc != nullptr && size > 0)
> +		    memset (loc, 0, size);
> +		}
> +	    }
> +	}
>      });
>  }
> 
> @@ -292,6 +315,17 @@ i386_prepare_to_resume (windows_thread_info
> *th)
> 
>        th->debug_registers_changed = false;
>      }
> +
> +  windows_process.with_context (th, [&] (auto *context)
> +    {
> +      DWORD debug_registers = WindowsContext<decltype(context)>::debug;
> +      if (xstate_features != 0
> +	  && (context->ContextFlags & ~debug_registers) != 0)
> +	{
> +	  context->ContextFlags |= CONTEXT_XSTATE_FLAG;
> +	  set_xstate_features_mask (context, xstate_features);
> +	}
> +    });
>  }
> 
>  static void
> @@ -477,7 +511,7 @@ is_segment_register (int r)
> 
>  template<typename Context>
>  static char *
> -get_context_reg_ptr (Context *context, int r)
> +get_context_reg_ptr (Context *context, int r, const target_desc *tdesc)
>  {
>    const int *mappings;
>    int mappings_count;
> @@ -494,9 +528,21 @@ get_context_reg_ptr (Context *context, int r)
>        mappings_count = sizeof (i386_mappings) / sizeof (i386_mappings[0]);
>      }
> 
> +  bool amd64 = register_size (tdesc, 0) == 8;

There is already an " if (!windows_process.wow64_process)" a few lines above.
Wouldn't it make sense to move the "bool amd64" in the ifdef blocks and
assign accordingly?

> +  int ymm0h_regnum;
> +  const int num_xmm_registers = amd64 ? 16 : 8;
> +
>    char *context_offset;
>    if (r < mappings_count)
>      context_offset = (char *) context + mappings[r];
> +  else if ((xstate_features & X86_XSTATE_AVX) != 0
> +	   && r >= (ymm0h_regnum = find_regno (tdesc, "ymm0h"))
> +	   && r < ymm0h_regnum + num_xmm_registers)
> +    {
> +      context_offset = (char *) locate_xstate_feature
> +	(context, X86_XSTATE_AVX_ID, NULL);
> +      context_offset += 16 * (r - ymm0h_regnum);
> +    }
>    else
>      gdb_assert_not_reached ("invalid register number %d", r);
> 
> @@ -510,7 +556,7 @@ i386_fetch_inferior_register (struct regcache
> *regcache,
>  {
>    char *context_offset = windows_process.with_context (th, [&] (auto
> *context)
>      {
> -      return get_context_reg_ptr (context, r);
> +      return get_context_reg_ptr (context, r, regcache->tdesc);
>      });
> 
>    /* GDB treats some registers as 32-bit, where they are in fact only
> @@ -538,7 +584,7 @@ i386_store_inferior_register (struct regcache
> *regcache,
>  {
>    char *context_offset = windows_process.with_context (th, [&] (auto
> *context)
>      {
> -      return get_context_reg_ptr (context, r);
> +      return get_context_reg_ptr (context, r, regcache->tdesc);
>      });
> 
>    /* GDB treats some registers as 32-bit, where they are in fact only
> @@ -571,14 +617,17 @@ i386_arch_setup (void)
>  {
>   target_desc_up tdesc;
> 
> +  DWORD64 xcr0 = xstate_features;
> +  if (xcr0 == 0)
> +    xcr0 = X86_XSTATE_SSE_MASK;
> +
>  #ifdef __x86_64__
> -  tdesc = amd64_create_target_description (X86_XSTATE_SSE_MASK, false,
> -					   false, false);
> +  tdesc = amd64_create_target_description (xcr0, false, false, false);
>    init_target_desc (tdesc.get (), amd64_expedite_regs, WINDOWS_OSABI);
>    win32_tdesc = std::move (tdesc);
>  #endif
> 
> -  tdesc = i386_create_target_description (X86_XSTATE_SSE_MASK, false,
> false);
> +  tdesc = i386_create_target_description (xcr0, false, false);
>    init_target_desc (tdesc.get (), i386_expedite_regs, WINDOWS_OSABI);
>  #ifdef __x86_64__
>    wow64_win32_tdesc = std::move (tdesc);
> diff --git a/gdbserver/win32-low.cc b/gdbserver/win32-low.cc
> index 7629beca213..5ccdc89a7ef 100644
> --- a/gdbserver/win32-low.cc
> +++ b/gdbserver/win32-low.cc
> @@ -33,6 +33,7 @@
>  #include <process.h>
>  #include "gdbsupport/gdb_tilde_expand.h"
>  #include "gdbsupport/common-inferior.h"
> +#include "tdesc.h"
> 
>  using namespace windows_nat;
> 
> @@ -426,8 +427,9 @@ child_fetch_inferior_registers (struct regcache
> *regcache, int r)
>    int regno;
>    windows_thread_info *th = windows_process.find_thread (current_thread-
> >id);
>    win32_require_context (th);
> -  if (r == -1 || r > NUM_REGS)
> -    child_fetch_inferior_registers (regcache, NUM_REGS);

IIUC this was the only use of the NUM_REGS define.  We can remove it.
Same for " i386_win32_num_regs (void)" and "aarch64_win32_num_regs ()".
This allows removing the num_regs hook in win32_target_ops.

> +  if (r == -1)
> +    child_fetch_inferior_registers (regcache,
> +				    regcache->tdesc->reg_defs.size ());
>    else
>      for (regno = 0; regno < r; regno++)
>        (*the_low_target.fetch_inferior_register) (regcache, th, regno);
> @@ -441,8 +443,9 @@ child_store_inferior_registers (struct regcache
> *regcache, int r)
>    int regno;
>    windows_thread_info *th = windows_process.find_thread (current_thread-
> >id);
>    win32_require_context (th);
> -  if (r == -1 || r == 0 || r > NUM_REGS)
> -    child_store_inferior_registers (regcache, NUM_REGS);
> +  if (r == -1)
> +    child_store_inferior_registers (regcache,
> +				    regcache->tdesc->reg_defs.size ());
>    else
>      for (regno = 0; regno < r; regno++)
>        (*the_low_target.store_inferior_register) (regcache, th, regno);
> @@ -1349,7 +1352,9 @@ void
>  initialize_low (void)
>  {
>    set_target_ops (&the_win32_target);
> -  the_low_target.arch_setup ();
> 
>    initialize_loadable ();
> +  /* Has to be done after initialize_loadable, because it uses the xstate
> +     functions if available.  */
> +  the_low_target.arch_setup ();
>  }
> --
> 2.54.0

________________________________________
Intel Deutschland GmbH 

Registered Address: Dornacher Strasse 1, 85622 Feldkirchen, Germany 

Tel: +49 (89) 99143-0 

www.intel.de 

Managing Directors: Candice Moore, Jeffrey Schneiderman, Ramachandran Sitaraman

Chairperson of the Supervisory Board: Sonja Pierer

Registered Seat: Munich Commercial Register B: Amtsgericht Munich HRB 186928

This e-mail and any attachments may contain confidential material for
the sole use of the intended recipient(s). Any review or distribution
by others is strictly prohibited. If you are not the intended
recipient, please contact the sender and delete all copies.