On Mon, 9 Mar 2026 at 16:22, Florian Hofhammer
<[email protected]> wrote:
>
> This patch adds docstrings for typedefs and function declarations in
> include/plugins/qemu-plugin.h that were previously missing. This
> resolves inconsistencies in the docs, e.g., the description for
> qemu_plugin_read_register() referring to qemu_plugin_register_flush_cb()
> but code cache flush callbacks not being documented themselves.
>
> Signed-off-by: Florian Hofhammer <[email protected]>
I don't know this API either, so a couple of questions below
for those who do:
> +/**
> + * enum qemu_plugin_mem_rw - type of memory access
> + *
> + * @QEMU_PLUGIN_MEM_R: memory read access only
> + * @QEMU_PLUGIN_MEM_W: memory write access only
> + * @QEMU_PLUGIN_MEM_RW: memory read and write access
> + */
> enum qemu_plugin_mem_rw {
> QEMU_PLUGIN_MEM_R = 1,
> QEMU_PLUGIN_MEM_W,
> QEMU_PLUGIN_MEM_RW,
> };
This define sets up the values such that you can use
the _R and _W values as bit checks (i.e. _RW == _R | _W).
Is that intentional and can code using the API rely on it,
or must it strictly check against _R and _RW separately?
> +/**
> + * enum qemu_plugin_mem_value_type - size of memory value
> + *
> + * @QEMU_PLUGIN_MEM_VALUE_U8: unsigned 8-bit value
> + * @QEMU_PLUGIN_MEM_VALUE_U16: unsigned 16-bit value
> + * @QEMU_PLUGIN_MEM_VALUE_U32: unsigned 32-bit value
> + * @QEMU_PLUGIN_MEM_VALUE_U64: unsigned 64-bit value
> + * @QEMU_PLUGIN_MEM_VALUE_U128: unsigned 128-bit value
> + */
> enum qemu_plugin_mem_value_type {
> QEMU_PLUGIN_MEM_VALUE_U8,
> QEMU_PLUGIN_MEM_VALUE_U16,
> @@ -353,7 +369,13 @@ enum qemu_plugin_mem_value_type {
> QEMU_PLUGIN_MEM_VALUE_U128,
> };
In this enum, the values are set up such that if you have
a qemu_plugin_mem_value_type vt, then the size of the data
is (1 << vt) (i.e., a byte is 0, a short is 1, 32 bits are 2,
and so on up). Is that intentional and can code using the
API rely on it, or must it strictly use the enum only as
something to switch() on to select a length.
I actually ran into that last one in tests/tcg/plugins/mem.c:
update_region_info() has a switch() on this enum which (among
other things) sets val_size in each case, because I didn't
know if it was guaranteed to work to set val_size = 1 << value.type.
thanks
-- PMM
