diff options
author | Heiko Carstens <hca@linux.ibm.com> | 2023-10-27 14:12:39 +0200 |
---|---|---|
committer | Vasily Gorbik <gor@linux.ibm.com> | 2023-11-05 22:34:58 +0100 |
commit | a51324c430db3fcf3e7d77c265491322c251a396 (patch) | |
tree | 8287b8d7b681c66ba708982b3c96cd8b26eb5279 /arch/s390/mm/vmem.c | |
parent | 65d37f163add1c6ead3a63788acb2f9590159f94 (diff) |
s390/cmma: rework no-dat handling
Rework the way physical pages are set no-dat / dat:
The old way is:
- Rely on that all pages are initially marked "dat"
- Allocate page tables for the kernel mapping
- Enable dat
- Walk the whole kernel mapping and set PG_arch_1 bit in all struct pages
that belong to pages of kernel page tables
- Walk all struct pages and test and clear the PG_arch_1 bit. If the bit is
not set, set the page state to no-dat
- For all subsequent page table allocations, set the page state to dat
(remove the no-dat state) on allocation time
Change this rather complex logic to a simpler approach:
- Set the whole physical memory (all pages) to "no-dat"
- Explicitly set those page table pages to "dat" which are part of the
kernel image (e.g. swapper_pg_dir)
- For all subsequent page table allocations, set the page state to dat
(remove the no-dat state) on allocation time
In result the code is simpler, and this also allows to get rid of one
odd usage of the PG_arch_1 bit.
Reviewed-by: Claudio Imbrenda <imbrenda@linux.ibm.com>
Signed-off-by: Heiko Carstens <hca@linux.ibm.com>
Signed-off-by: Vasily Gorbik <gor@linux.ibm.com>
Diffstat (limited to 'arch/s390/mm/vmem.c')
-rw-r--r-- | arch/s390/mm/vmem.c | 4 |
1 files changed, 2 insertions, 2 deletions
diff --git a/arch/s390/mm/vmem.c b/arch/s390/mm/vmem.c index b13990776fae..186a020857cf 100644 --- a/arch/s390/mm/vmem.c +++ b/arch/s390/mm/vmem.c @@ -50,8 +50,7 @@ void *vmem_crst_alloc(unsigned long val) if (!table) return NULL; crst_table_init(table, val); - if (slab_is_available()) - __arch_set_page_dat(table, 1UL << CRST_ALLOC_ORDER); + __arch_set_page_dat(table, 1UL << CRST_ALLOC_ORDER); return table; } @@ -67,6 +66,7 @@ pte_t __ref *vmem_pte_alloc(void) if (!pte) return NULL; memset64((u64 *)pte, _PAGE_INVALID, PTRS_PER_PTE); + __arch_set_page_dat(pte, 1); return pte; } |