Opened 12 years ago
Closed 12 years ago
#406 closed defect (fixed)
ia32: init_e820_memory() might be broken
| Reported by: | Martin Decky | Owned by: | Jakub Jermář |
|---|---|---|---|
| Priority: | major | Milestone: | 0.5.0 |
| Component: | helenos/kernel/ia32 | Version: | mainline |
| Keywords: | Cc: | ||
| Blocker for: | Depends on: | ||
| See also: |
Description
On machines with more than 3 GB of physical memory, the physical memory above 3 GB is usually remapped above 4 GB to create a "hole" in the physical address space between 3 GB and 4 GB to map hardware I/O registers to it.
As the current implementation of init_e820_memory() only truncates the 64bit physical addresses and sizes to 32bits by typecasting, the creation of the zones can be confused on IA-32 (the code would try to create zones below 4 GB for physical memory regions above 4 GB as the higher bits are simply thrown away).
This is a regression since changeset mainline,1352 as previously the 64bit addresses and sizes were explicitly truncated to sane values because of the limitations described in ticket #3.
The changeset mainline,1355 only reduces some of the symptoms of this bug, but does not address is completely.
If the ia32 kernel is not PAE enabled, we could somehow instruct init_e820_memory() to ignore memory above 4G as it is unusable anyway.
If, one day, we support PAE on ia32, we will need to introduce a new type for handling physical addresses and let the physical memory subsystem work with it.
I assume amd64 is ok even now.