Messages in this thread | | | Date | Tue, 23 Oct 2012 13:54:31 -0700 | Subject | Re: when to switch from early_ioremap to ioremap | From | Yinghai Lu <> |
| |
On Tue, Oct 23, 2012 at 12:42 PM, Olaf Hering <olaf@aepfle.de> wrote: > > To obtain an address to a PFN in init_hypervisor_platform (called from > arch/x86/kernel/setup.c:setup_arch) I use early_ioremap. Later the > address to that PFN needs to be obtained with ioremap, and early_iounmap > needs to be called on the first address. > > When is the time to obtain and use the new pointer, and to release to > old pointer? Right now I do that in the pci-device setup code, which > happens to work. Unfortunately early_ioremap is not mentioned in > Documentation/, hence the question.
usually early_ioremap is used in this way: map it and check it and unmap it. and some time you could back up it before unmap.
you should only use early_iomap and early_memmap during SYSTEM_BOOTING
and check_early_ioremap_leak in late_initcall will check if all early one get unmapped.
for 32bit, there is more requirement: you can not cross init_memory_mapping. early_ioremap_page_table_range_init/early_ioremap_reset will be called in init_memory_mapping.
Also early_ioremap has slot limitation: 4. and you should not hold one slot too long, otherwise other users could have problem to reach the limit.
Yinghai
| |