[PATCH] gdb, amdgpu: handle UNAVAILABLE error when accessing memory

[PATCH] gdb, amdgpu: handle UNAVAILABLE error when accessing memory

[Tankut Baris Aktemur]

In amd_dbgapi_target, accessing memory through dbgapi may fail with
AMD_DBGAPI_STATUS_ERROR_MEMORY_UNAVAILABLE.  Convert this error to
GDB's TARGET_XFER_UNAVAILABLE.  Similarly, in the global memory access
callback function, if the access result is TARGET_XFER_UNAVAILABLE, pass
the result back to dbgapi as AMD_DBGAPI_STATUS_ERROR_MEMORY_UNAVAILABLE.

These conversions in both ways are useful to display to the user an
"<unavailable>" result instead of "Cannot access memory...".  This may
happen in lightweight corefiles, where some chunk of memory may not
have been captured.

Note that if xfer_partial target op replies with TARGET_XFER_UNAVAILABLE,
GDB expects xfered_len to be greater than zero.

Regression-tested in gfx1201.
---
 gdb/amd-dbgapi-target.c | 8 ++++++++
 1 file changed, 8 insertions(+)

diff --git a/gdb/amd-dbgapi-target.c b/gdb/amd-dbgapi-target.c
index 93207a8fdd2..2fd175de6f7 100644
--- a/gdb/amd-dbgapi-target.c
+++ b/gdb/amd-dbgapi-target.c
@@ -786,6 +786,12 @@ amd_dbgapi_target::xfer_partial (enum target_object object, const char *annex,
 				      AMD_DBGAPI_ADDRESS_SPACE_GLOBAL,
 				      offset, &len, writebuf);
 
+  if (status == AMD_DBGAPI_STATUS_ERROR_MEMORY_UNAVAILABLE)
+    {
+      *xfered_len = requested_len;
+      return TARGET_XFER_UNAVAILABLE;
+    }
+
   if (status != AMD_DBGAPI_STATUS_SUCCESS)
     return TARGET_XFER_E_IO;
 
@@ -2570,6 +2576,8 @@ amd_dbgapi_xfer_global_memory_callback
 
   if (status == TARGET_XFER_EOF)
     return AMD_DBGAPI_STATUS_ERROR_PROCESS_EXITED;
+  else if (status == TARGET_XFER_UNAVAILABLE)
+    return AMD_DBGAPI_STATUS_ERROR_MEMORY_UNAVAILABLE;
   else if (status != TARGET_XFER_OK)
     return AMD_DBGAPI_STATUS_ERROR_MEMORY_ACCESS;
 
-- 
2.53.0

Re: [PATCH] gdb, amdgpu: handle UNAVAILABLE error when accessing memory

[Lancelot SIX]

On 14/10/2026 10:08, Tankut Baris Aktemur wrote:
> In amd_dbgapi_target, accessing memory through dbgapi may fail with
> AMD_DBGAPI_STATUS_ERROR_MEMORY_UNAVAILABLE.  Convert this error to
> GDB's TARGET_XFER_UNAVAILABLE.  Similarly, in the global memory access
> callback function, if the access result is TARGET_XFER_UNAVAILABLE, pass
> the result back to dbgapi as AMD_DBGAPI_STATUS_ERROR_MEMORY_UNAVAILABLE.
> 
> These conversions in both ways are useful to display to the user an
> "<unavailable>" result instead of "Cannot access memory...".  This may
> happen in lightweight corefiles, where some chunk of memory may not
> have been captured.
> 
> Note that if xfer_partial target op replies with TARGET_XFER_UNAVAILABLE,
> GDB expects xfered_len to be greater than zero.
> 
> Regression-tested in gfx1201.

Hi,

This looks good to me, thanks.

Approved-by: Lancelot Six <lancelot.six@amd.com> (amdgpu)

Best,
Lancelot.

> ---
>   gdb/amd-dbgapi-target.c | 8 ++++++++
>   1 file changed, 8 insertions(+)
> 
> diff --git a/gdb/amd-dbgapi-target.c b/gdb/amd-dbgapi-target.c
> index 93207a8fdd2..2fd175de6f7 100644
> --- a/gdb/amd-dbgapi-target.c
> +++ b/gdb/amd-dbgapi-target.c
> @@ -786,6 +786,12 @@ amd_dbgapi_target::xfer_partial (enum target_object object, const char *annex,
>   				      AMD_DBGAPI_ADDRESS_SPACE_GLOBAL,
>   				      offset, &len, writebuf);
>   
> +  if (status == AMD_DBGAPI_STATUS_ERROR_MEMORY_UNAVAILABLE)
> +    {
> +      *xfered_len = requested_len;
> +      return TARGET_XFER_UNAVAILABLE;
> +    }
> +
>     if (status != AMD_DBGAPI_STATUS_SUCCESS)
>       return TARGET_XFER_E_IO;
>   
> @@ -2570,6 +2576,8 @@ amd_dbgapi_xfer_global_memory_callback
>   
>     if (status == TARGET_XFER_EOF)
>       return AMD_DBGAPI_STATUS_ERROR_PROCESS_EXITED;
> +  else if (status == TARGET_XFER_UNAVAILABLE)
> +    return AMD_DBGAPI_STATUS_ERROR_MEMORY_UNAVAILABLE;
>     else if (status != TARGET_XFER_OK)
>       return AMD_DBGAPI_STATUS_ERROR_MEMORY_ACCESS;
>   

Re: [PATCH] gdb, amdgpu: handle UNAVAILABLE error when accessing memory

[Tom Tromey]

>>>>> Tankut Baris Aktemur <tankutbaris.aktemur@amd.com> writes:

> +  if (status == AMD_DBGAPI_STATUS_ERROR_MEMORY_UNAVAILABLE)
> +    {

I get this:

../../binutils-gdb/gdb/amd-dbgapi-target.c:789:17: error: ‘AMD_DBGAPI_STATUS_ERROR_MEMORY_UNAVAILABLE’ was not declared in this scope; did you mean ‘AMD_DBGAPI_STATUS_ERROR_NOT_AVAILABLE’?

Not sure I have the right thing but:

bapiya. rpm -q rocdbgapi-devel
rocdbgapi-devel-7.1.0-4.fc44.x86_64

which seems to be the latest on Fedora 43.

Meanwhile I'm going to build with this feature disabled.

Tom

RE: [PATCH] gdb, amdgpu: handle UNAVAILABLE error when accessing memory

[Six, Lancelot]

AMD General

> -----Original Message-----
> From: Tom Tromey <tom@tromey.com>
> Sent: 15 October 2026 16:40
> To: Aktemur, Baris <TankutBaris.Aktemur@amd.com>
> Cc: gdb-patches@sourceware.org; Six, Lancelot <Lancelot.Six@amd.com>
> Subject: Re: [PATCH] gdb, amdgpu: handle UNAVAILABLE error when
> accessing memory
>
> Caution: This message originated from an External Source. Use proper caution
> when opening attachments, clicking links, or responding.
>
>
> >>>>> Tankut Baris Aktemur <tankutbaris.aktemur@amd.com> writes:
>
> > +  if (status == AMD_DBGAPI_STATUS_ERROR_MEMORY_UNAVAILABLE)
> > +    {
>
> I get this:
>
> ../../binutils-gdb/gdb/amd-dbgapi-target.c:789:17: error:
> ‘AMD_DBGAPI_STATUS_ERROR_MEMORY_UNAVAILABLE’ was not declared in
> this scope; did you mean ‘AMD_DBGAPI_STATUS_ERROR_NOT_AVAILABLE’?
>
> Not sure I have the right thing but:
>
> bapiya. rpm -q rocdbgapi-devel
> rocdbgapi-devel-7.1.0-4.fc44.x86_64
>
> which seems to be the latest on Fedora 43.
>
> Meanwhile I'm going to build with this feature disabled.
>
> Tom

Oh, sorry about that.

I think this comes from version in rocm-10.  Will add the configury checks and guards to avoid the breakage.

Best,
Lancelot.