s390/boot: Use entire page for PTEs

Make boot_pte_alloc() always allocate a full PAGE_SIZE page for
PTE tables, instead of carving two 2K PTE tables out of a single
4K page, similar to commit daa8af80d283 ("s390/mm: Allocate page
table with PAGE_SIZE granularity").

This mirrors the change in the vmem code and ensures that boot page
tables backing the early KASAN shadow can later be fully freed by
the vmem page-table teardown helpers (e.g. when unmapping early
KASAN shadow on memory hotplug).

The leftover-based allocation was originally added to reduce physmem
allocator fragmentation when EDAT was disabled. On current hardware
EDAT1 is available on all production systems, so the complexity is no
longer justified and gets in the way of freeing the shadow mappings.

Signed-off-by: Vasily Gorbik <gor@linux.ibm.com>
Reviewed-by: Heiko Carstens <hca@linux.ibm.com>
Signed-off-by: Heiko Carstens <hca@linux.ibm.com>

1 files changed, 2 insertions, 14 deletions

diff --git a/arch/s390/boot/vmem.c b/arch/s390/boot/vmem.c
index fbe64ffdfb96..7d6cc4c85af0 100644
--- a/arch/s390/boot/vmem.c
+++ b/arch/s390/boot/vmem.c
@@ -244,22 +244,10 @@ static void *boot_crst_alloc(unsigned long val)
 
 static pte_t *boot_pte_alloc(void)
 {
-	static void *pte_leftover;
 	pte_t *pte;
 
-	/*
-	 * handling pte_leftovers this way helps to avoid memory fragmentation
-	 * during POPULATE_KASAN_MAP_SHADOW when EDAT is off
-	 */
-	if (!pte_leftover) {
-		pte_leftover = (void *)physmem_alloc_or_die(RR_VMEM, PAGE_SIZE, PAGE_SIZE);
-		pte = pte_leftover + _PAGE_TABLE_SIZE;
-		__arch_set_page_dat(pte, 1);
-	} else {
-		pte = pte_leftover;
-		pte_leftover = NULL;
-	}
-
+	pte = (void *)physmem_alloc_or_die(RR_VMEM, PAGE_SIZE, PAGE_SIZE);
+	__arch_set_page_dat(pte, 1);
 	memset64((u64 *)pte, _PAGE_INVALID, PTRS_PER_PTE);
 	return pte;
 }