Fix some problems in SPARSE_MAPPING in the kernel linker.

[markj]

Summary

• Fix some style bugs.
• Ensure that the end of the mapping passed to vm_page_wire() is page-aligned. vm_map_wire() expects this.
• Wire pages before reading data into them to ensure that the mapping is backed by resident pages.
• Apply the protections specified in the segment descriptor using vm_map_protect().
• Ensure that on amd64 we load KLDs above KERNBASE.

[kib]

Is there any way to enable SPARSE_MAPPING except manually adding #define ?

sys/kern/link_elf.c

1101 ↗

(On Diff #62434)

I suspect this is too rough. In principle, the segments need not be page-aligned, and you might remove W or X from a neighbor segment that shares the page with us.

I think the easiest would be to refuse loading such objects.

[markj]

In D21756#474923, @kib wrote:

Is there any way to enable SPARSE_MAPPING except manually adding #define ?

No. I'm considering adding a kernel option for it, but it needs a better name than SPARSE_MAPPING IMO.

[markj]

Reject KLDs with unaligned segment base addrs when SPARSE_MAPPING is defined.

[markj]

sys/kern/link_elf.c

994 ↗

(On Diff #62474)

I am not sure why we cannot just use kernel_object here. OBJT_DEFAULT doesn't make a lot of sense, since the memory will not be pageable.