On Tue, 2024-10-01 at 21:40 +0000, Huang, Kai wrote:
> On Tue, 2024-10-01 at 08:19 -0700, Dave Hansen wrote:
> > On 10/1/24 03:44, Huang, Kai wrote:
> > > Please let me know if you have any concern? Otherwise I will go with
> > > this route.
> > I still see some long unwieldy #defines in the mail thread. That's my
> > biggest worry.
>
> I suppose you mean the read_sys_metadata_field() macro?
>
> We can split that into two smaller macros by moving BUILD_BUG_ON() out:
>
> /* Don't use this directly, use read_sys_metadata_read() instead. */
> #define __read_sys_metadata_field(_field_id, _valptr) \
> ({ \
> u64 ___tmp; \
> int ___ret; \
> \
> ___ret = tdh_sys_rd(_field_id, &___tmp); \
> *_valptr = ___tmp; \
> \
> ___ret; \
> })
>
> /* @_valptr must be pointer of u8/u16/u32/u64. */
> #define read_sys_metadata_field(_field_id, _valptr) \
> ({ \
> BUILD_BUG_ON(MD_FIELD_ELE_SIZE(_field_id) != \
> sizeof(*_valptr)); \
> __read_sys_metadata_field(_field_id, _valptr); \
> })
>
> Does this look good to you?
Hi Dave,
Would you let me know are you OK with this?
If not, I will revert back to what you suggested:
https://lore.kernel.org/lkml/cover.1727173372.git.kai.huang@xxxxxxxxx/T/#m3874080ef158a4704c4082259d3594aa0a322fc8
