On 16.04.2024 00:15, Stefano Stabellini wrote: > On Mon, 15 Apr 2024, Andrew Cooper wrote: >> RFC. In theory this is a great way to avoid some of the spiketraps involved >> with C being the official representation. >> >> However, this doesn't build. gnttab_transfer has a layout that requires a >> CONFIG_COMPAT if we want to satisfy -Wpadding for both forms of the >> structure. >> >> Thoughts on whether this cross-check is worthwhile-enough to warrant the >> ifdefary? > > I like this patch and I think we have no choice but going in this > direction and adding all the padding explicitly with any related > necessary ifdefary. > > The only question for me is whether to: > > 1) add -Wpadding > 2) add __packed__ > 3) do both > > I think it is important to add __packed__ to the headers to clear out > any misconceptions about possible hidden paddings and get a > correct-by-default behavior for anyone that would import the headers > into their own projects. > > The only issue is that __packed__ makes -Wpadding not useful. We could > technically add both if we disable __packed__ for the -Wpadding build. > For instance we could use __packed which is defined by Xen, and change > the definition of __packed to nothing for the -Wpadding build.
The other issue is that __packed__ isn't standard C, yet we intend the public header to be plain C89 compatible. Jan
