linuxkpi: Assert that "order" is sane

[markj]

Summary

Suggested by: hselasky

[bz]

I wonder if sizeof(uintptr_t) rather than sizeof(size_t) would be better?

sys/compat/linuxkpi/common/src/linux_page.c

189

We should probably move the assignment below the KASSERT?

[markj]

In D40398#919802, @bz wrote:

I wonder if sizeof(uintptr_t) rather than sizeof(size_t) would be better?

Well, we want 1 << order to fit in a type used to represent sizes, so I think size_t is correct. Consider that on CHERI, sizeof(uintptr_t) is 16, but we still want order <= 63.

sys/compat/linuxkpi/common/src/linux_page.c

189

Good catch, thank you.

[markj]

Assert on "order" before using it.

[• hselasky]

sys/compat/linuxkpi/common/src/linux_page.c

192

Should we subtract PAGE_SHIFT ?
order < sizeof(size_t) * NBBY - PAGE_SHIFT

Then it would never overflow?

[markj]

Tighten the check in one place that shifts PAGE_MASK instead of 1.

[• hselasky]

sys/compat/linuxkpi/common/src/linux_page.c

106

Same here.

164

Same here.

208

Can you verify the actual value of PAGE_MASK here? Because we have this:

/usr/src/sys/compat/linuxkpi/common/include/linux/page.h:#undef	PAGE_MASK
/usr/src/sys/compat/linuxkpi/common/include/linux/page.h:#define	PAGE_MASK	(~(PAGE_SIZE-1))

Linux's PAGE_MASK is not FreeBSD's PAGE_MASK and maybe not use it?

[markj]

sys/compat/linuxkpi/common/src/linux_page.c

106

Why? There is no code here which computes (1ul << order) * PAGE_SIZE. npages is passed directly to the page allocator.

164

Same as above.

208

Oof, thank you.

[• hselasky]

sys/compat/linuxkpi/common/src/linux_page.c

106

I was just thinking that if you cannot allocate a certain amount, you cannot free it either. Maybe just define the limit as a macro / enum ?