[PATCH v1] gdb: Fix assignment to TYPE_CODE_FLAGS registers

[PATCH v1] gdb: Fix assignment to TYPE_CODE_FLAGS registers

[srinath.parvathaneni]

From: Srinath Parvathaneni <srinath.parvathaneni@arm.com>

GDB reports "Invalid cast" when assigning small integer literals (32-bit) to
64-bit TYPE_CODE_FLAGS registers. For example, with the existing upstream FPMR
register, set $fpmr = 0x777777777 succeeds, whereas set $fpmr = 0x77 fails with
"Invalid cast" and this patch fixes the issue but makes changes to a generic
function value_cast in gdb/valops.c.

        * testsuite/gdb.arch/aarch64-fpmr.exp : Add "set $fpmr" test.
        * gdb/valops.c (value_cast): Add TYPE_CODE_FLAGS condition to integer
	assignment.

Regression tested on aarch64-none-linux-gnu with no regressions.

Ok for GDB master?

Regards,
Srinath
---
 gdb/testsuite/gdb.arch/aarch64-fpmr.exp | 6 ++++++
 gdb/valops.c                            | 2 +-
 2 files changed, 7 insertions(+), 1 deletion(-)

diff --git a/gdb/testsuite/gdb.arch/aarch64-fpmr.exp b/gdb/testsuite/gdb.arch/aarch64-fpmr.exp
index 72fd977fc67..2eee10fe924 100644
--- a/gdb/testsuite/gdb.arch/aarch64-fpmr.exp
+++ b/gdb/testsuite/gdb.arch/aarch64-fpmr.exp
@@ -97,3 +97,9 @@ gdb_test_multiple "set \$fpmr=0x3f007f4008" "" {
 	pass "Write to FPMR from GDB"
     }
 }
+
+gdb_test_multiple "set \$fpmr=0x1255C041" "" {
+    -re ".*\r\n.*\[ F8S1=E4M3 F8S2=E5M2 F8D=E4M3 OSM=MaxNormal OSC=MaxNormal LSCALE=85 NSCALE=18 LSCALE2=0 \]" {
+	pass "Write 32-bit value to 64-bit register FPMR from GDB"
+    }
+}
diff --git a/gdb/valops.c b/gdb/valops.c
index ab6fd5079e1..170311dc63c 100644
--- a/gdb/valops.c
+++ b/gdb/valops.c
@@ -557,7 +557,7 @@ value_cast (struct type *type, struct value *arg2)
 	return value_from_longest (to_type, value_as_long (arg2));
     }
   else if ((code1 == TYPE_CODE_INT || code1 == TYPE_CODE_ENUM
-	    || code1 == TYPE_CODE_RANGE)
+	    || code1 == TYPE_CODE_RANGE || code1 == TYPE_CODE_FLAGS)
 	   && (scalar || code2 == TYPE_CODE_PTR
 	       || code2 == TYPE_CODE_MEMBERPTR))
     {
-- 
2.43.0

Re: [PATCH v1] gdb: Fix assignment to TYPE_CODE_FLAGS registers

[Simon Marchi]

On 7/13/26 5:11 AM, srinath.parvathaneni@arm.com wrote:
> From: Srinath Parvathaneni <srinath.parvathaneni@arm.com>
> 
> GDB reports "Invalid cast" when assigning small integer literals (32-bit) to
> 64-bit TYPE_CODE_FLAGS registers. For example, with the existing upstream FPMR
> register, set $fpmr = 0x777777777 succeeds, whereas set $fpmr = 0x77 fails with
> "Invalid cast" and this patch fixes the issue but makes changes to a generic
> function value_cast in gdb/valops.c.
> 
>         * testsuite/gdb.arch/aarch64-fpmr.exp : Add "set $fpmr" test.
>         * gdb/valops.c (value_cast): Add TYPE_CODE_FLAGS condition to integer
> 	assignment.
> 
> Regression tested on aarch64-none-linux-gnu with no regressions.
> 
> Ok for GDB master?

This seems fine to me, but I am not as confident with this value_cast
code as I am with other things, so I'll give you the:

Approved-By: Simon Marchi <simon.marchi@efficios.com>

but wait ~1 week before pushing in case someone sees something I don't.

Simon

RE: [PATCH v1] gdb: Fix assignment to TYPE_CODE_FLAGS registers

[Aktemur, Baris]

AMD General

On Monday, July 13, 2026 11:11 AM, srinath.parvathaneni@arm.com wrote:
> From: Srinath Parvathaneni <srinath.parvathaneni@arm.com>
>
> GDB reports "Invalid cast" when assigning small integer literals (32-bit) to
> 64-bit TYPE_CODE_FLAGS registers. For example, with the existing upstream FPMR
> register, set $fpmr = 0x777777777 succeeds, whereas set $fpmr = 0x77 fails with
> "Invalid cast" and this patch fixes the issue but makes changes to a generic
> function value_cast in gdb/valops.c.
>
>         * testsuite/gdb.arch/aarch64-fpmr.exp : Add "set $fpmr" test.
>         * gdb/valops.c (value_cast): Add TYPE_CODE_FLAGS condition to integer
>         assignment.
>
> Regression tested on aarch64-none-linux-gnu with no regressions.
>
> Ok for GDB master?
>
> Regards,
> Srinath
> ---
>  gdb/testsuite/gdb.arch/aarch64-fpmr.exp | 6 ++++++
>  gdb/valops.c                            | 2 +-
>  2 files changed, 7 insertions(+), 1 deletion(-)
>
> diff --git a/gdb/testsuite/gdb.arch/aarch64-fpmr.exp b/gdb/testsuite/gdb.arch/aarch64-
> fpmr.exp
> index 72fd977fc67..2eee10fe924 100644
> --- a/gdb/testsuite/gdb.arch/aarch64-fpmr.exp
> +++ b/gdb/testsuite/gdb.arch/aarch64-fpmr.exp
> @@ -97,3 +97,9 @@ gdb_test_multiple "set \$fpmr=0x3f007f4008" "" {
>         pass "Write to FPMR from GDB"
>      }
>  }
> +
> +gdb_test_multiple "set \$fpmr=0x1255C041" "" {
> +    -re ".*\r\n.*\[ F8S1=E4M3 F8S2=E5M2 F8D=E4M3 OSM=MaxNormal OSC=MaxNormal

It should be possible to remove ".*\r\n.*" from the beginning of the regexp.

Is there a specific reason gdb_test_multiple was used instead of a simpler gdb_test?
If gdb_test_multiple is necessary for some reason, I'd also suggest adding -wrap to avoid
leaving unmatched contents in the buffer.

Regards,
-Baris


> LSCALE=85 NSCALE=18 LSCALE2=0 \]" {
> +       pass "Write 32-bit value to 64-bit register FPMR from GDB"
> +    }
> +}
> diff --git a/gdb/valops.c b/gdb/valops.c
> index ab6fd5079e1..170311dc63c 100644
> --- a/gdb/valops.c
> +++ b/gdb/valops.c
> @@ -557,7 +557,7 @@ value_cast (struct type *type, struct value *arg2)
>         return value_from_longest (to_type, value_as_long (arg2));
>      }
>    else if ((code1 == TYPE_CODE_INT || code1 == TYPE_CODE_ENUM
> -           || code1 == TYPE_CODE_RANGE)
> +           || code1 == TYPE_CODE_RANGE || code1 == TYPE_CODE_FLAGS)
>            && (scalar || code2 == TYPE_CODE_PTR
>                || code2 == TYPE_CODE_MEMBERPTR))
>      {
> --
> 2.43.0