Hi Alex,
On 19/3/25 19:22, Alex Bennée wrote:
The current helper.h functions rely on hard coded assumptions about
target endianess to use the tswap macros. We also end up double
swapping a bunch of values if the target can run in multiple endianess
modes. Avoid this by getting the target to pass the endianess and size
via a MemOp and fixing up appropriately.
Signed-off-by: Alex Bennée <alex.ben...@linaro.org>
---
include/gdbstub/registers.h | 30 ++++++++++++++++++++++++++++++
gdbstub/gdbstub.c | 22 ++++++++++++++++++++++
2 files changed, 52 insertions(+)
create mode 100644 include/gdbstub/registers.h
+/**
+ * gdb_get_register_value() - get register value for gdb
+ * mo: size and endian MemOp
Since we pass the size, ...
+ * buf: GByteArray to store in target order
+ * val: pointer to value in host order
we don't know the type of the value, ...
+ *
+ * This replaces the previous legacy read functions with a single
+ * function to handle all sizes. Passing @mo allows the target mode to
+ * be taken into account and avoids using hard coded tswap() macros.
+ *
+ * Returns the number of bytes written to the array.
+ */
+int gdb_get_register_value(MemOp op, GByteArray *buf, uint8_t *val);
so I'd rather use a 'void *' type here (also const),
int gdb_get_register_value(MemOp op, GByteArray *gbuf, const void *ptr);
...
+/*
+ * Target helper function to read value into GByteArray, target
+ * supplies the size and target endianess via the MemOp.
+ */
+int gdb_get_register_value(MemOp op, GByteArray *buf, uint8_t *val)
+{
using uint8_t type being one implementation choice for this method:
int gdb_get_register_value(MemOp op, GByteArray *gbuf, const void *ptr)
{
const uint8_t *buf = ptr;
+ size_t bytes = memop_size(op);
+
+ if (op & MO_BSWAP) {
+ for ( int i = bytes ; i > 0; i--) {
+ g_byte_array_append(buf, &val[i - 1], 1);
+ };
+ } else {
+ g_byte_array_append(buf, val, bytes);
+ }
+
+ return bytes;
+}
