vm_page: Stop handling VM_ALLOC_NOOBJ in vm_page_alloc_domain_after()

[markj]

Summary

Also modify the function to unconditionally preserve PG_ZERO, so
VM_ALLOC_ZERO is effectively ignored.

MFC after: never

[kib]

sys/vm/vm_page.c

2143

I do not think () are needed.

[kib]

sys/vm/vm_page.c

2073–2079

I wonder if it makes sense to add initial assert that checks that req contains only flags that the function can handle. For this and all other vm_page_alloc* functions.

[markj]

sys/vm/vm_page.c

2073–2079

Yes, I think it's a good idea.

[markj]

• Add assertions to exclude all unexpected flags.
• Add VM_ALLOC_COUNT_MASK.
• Ensure that vm_page_grab() does not pass VM_ALLOC_IGN_SBUSY to allocator functions.

[markj]

Mask off VM_ALLOC_IGN_SBUSY in vm_page_grab_valid() as well.

[kib]

sys/vm/vm_page.h

553

Could this be VM_ALLOC_COUNT(VM_ALLOC_COUNT_MAX - 1) and count < VM_ALLOC_COUNT_MAX could be used for assert in VM_ALLOC_COUNT?

[markj]

sys/vm/vm_page.h

553

Are you suggesting using a statement expression there, e.g.,

#define VM_ALLOC_COUNT(count) ({ MPASS(count < VM_ALLOC_COUNT_MAX); count << VM_ALLOC_COUNT_SHIFT; })

?

[kib]

sys/vm/vm_page.h

553

Yes something along this line for VM_ALLOC_COUNT(). If vm/vm_page.h is used in usermode (even rarely) then __extension__ would be useful/needed as well.

[markj]

Add an assertion in VM_ALLOC_COUNT().

[markj]

sys/vm/vm_page.h

553

This chunk is kernel-only.

I am not 100% sure about adding the assertion, it means that VM_ALLOC_COUNT() cannot be assigned to a global var. But I don't think this is likely to arise...

[alc]

sys/vm/vm_page.c

2089

We really don't need this variable any longer.

[markj]

Drop the "pool" variable.