Commit f4eb07c1 authored by Heiko Carstens's avatar Heiko Carstens Committed by Martin Schwidefsky
Browse files

[S390] Virtual memmap for s390.



Virtual memmap support for s390. Inspired by the ia64 implementation.

Unlike ia64 we need a mechanism which allows us to dynamically attach
shared memory regions.
These memory regions are accessed via the dcss device driver. dcss
implements the 'direct_access' operation, which requires struct pages
for every single shared page.
Therefore this implementation provides an interface to attach/detach
shared memory:

int add_shared_memory(unsigned long start, unsigned long size);
int remove_shared_memory(unsigned long start, unsigned long size);

The purpose of the add_shared_memory function is to add the given
memory range to the 1:1 mapping and to make sure that the
corresponding range in the vmemmap is backed with physical pages.
It also initialises the new struct pages.

remove_shared_memory in turn only invalidates the page table
entries in the 1:1 mapping. The page tables and the memory used for
struct pages in the vmemmap are currently not freed. They will be
reused when the next segment will be attached.
Given that the maximum size of a shared memory region is 2GB and
in addition all regions must reside below 2GB this is not too much of
a restriction, but there is room for improvement.
Signed-off-by: default avatarHeiko Carstens <heiko.carstens@de.ibm.com>
Signed-off-by: default avatarMartin Schwidefsky <schwidefsky@de.ibm.com>
parent 7f090145
......@@ -235,6 +235,9 @@ config WARN_STACK_SIZE
source "mm/Kconfig"
config HOLES_IN_ZONE
def_bool y
comment "I/O subsystem configuration"
config MACHCHK_WARNING
......
......@@ -64,7 +64,7 @@ unsigned int console_devno = -1;
unsigned int console_irq = -1;
unsigned long machine_flags = 0;
struct mem_chunk memory_chunk[MEMORY_CHUNKS];
struct mem_chunk __initdata memory_chunk[MEMORY_CHUNKS];
volatile int __cpu_logical_map[NR_CPUS]; /* logical cpu to cpu address */
unsigned long __initdata zholes_size[MAX_NR_ZONES];
static unsigned long __initdata memory_end;
......
......@@ -2,6 +2,6 @@
# Makefile for the linux s390-specific parts of the memory manager.
#
obj-y := init.o fault.o ioremap.o extmem.o mmap.o
obj-y := init.o fault.o ioremap.o extmem.o mmap.o vmem.o
obj-$(CONFIG_CMM) += cmm.o
......@@ -16,6 +16,7 @@
#include <linux/bootmem.h>
#include <linux/ctype.h>
#include <asm/page.h>
#include <asm/pgtable.h>
#include <asm/ebcdic.h>
#include <asm/errno.h>
#include <asm/extmem.h>
......@@ -237,65 +238,6 @@ query_segment_type (struct dcss_segment *seg)
return rc;
}
/*
* check if the given segment collides with guest storage.
* returns 1 if this is the case, 0 if no collision was found
*/
static int
segment_overlaps_storage(struct dcss_segment *seg)
{
int i;
for (i = 0; i < MEMORY_CHUNKS && memory_chunk[i].size > 0; i++) {
if (memory_chunk[i].type != CHUNK_READ_WRITE)
continue;
if ((memory_chunk[i].addr >> 20) > (seg->end >> 20))
continue;
if (((memory_chunk[i].addr + memory_chunk[i].size - 1) >> 20)
< (seg->start_addr >> 20))
continue;
return 1;
}
return 0;
}
/*
* check if segment collides with other segments that are currently loaded
* returns 1 if this is the case, 0 if no collision was found
*/
static int
segment_overlaps_others (struct dcss_segment *seg)
{
struct list_head *l;
struct dcss_segment *tmp;
BUG_ON(!mutex_is_locked(&dcss_lock));
list_for_each(l, &dcss_list) {
tmp = list_entry(l, struct dcss_segment, list);
if ((tmp->start_addr >> 20) > (seg->end >> 20))
continue;
if ((tmp->end >> 20) < (seg->start_addr >> 20))
continue;
if (seg == tmp)
continue;
return 1;
}
return 0;
}
/*
* check if segment exceeds the kernel mapping range (detected or set via mem=)
* returns 1 if this is the case, 0 if segment fits into the range
*/
static inline int
segment_exceeds_range (struct dcss_segment *seg)
{
int seg_last_pfn = (seg->end) >> PAGE_SHIFT;
if (seg_last_pfn > max_pfn)
return 1;
return 0;
}
/*
* get info about a segment
* possible return values:
......@@ -341,24 +283,26 @@ __segment_load (char *name, int do_nonshared, unsigned long *addr, unsigned long
rc = query_segment_type (seg);
if (rc < 0)
goto out_free;
if (segment_exceeds_range(seg)) {
PRINT_WARN ("segment_load: not loading segment %s - exceeds"
" kernel mapping range\n",name);
rc = -ERANGE;
rc = add_shared_memory(seg->start_addr, seg->end - seg->start_addr + 1);
switch (rc) {
case 0:
break;
case -ENOSPC:
PRINT_WARN("segment_load: not loading segment %s - overlaps "
"storage/segment\n", name);
goto out_free;
}
if (segment_overlaps_storage(seg)) {
PRINT_WARN ("segment_load: not loading segment %s - overlaps"
" storage\n",name);
rc = -ENOSPC;
case -ERANGE:
PRINT_WARN("segment_load: not loading segment %s - exceeds "
"kernel mapping range\n", name);
goto out_free;
}
if (segment_overlaps_others(seg)) {
PRINT_WARN ("segment_load: not loading segment %s - overlaps"
" other segments\n",name);
rc = -EBUSY;
default:
PRINT_WARN("segment_load: not loading segment %s (rc: %d)\n",
name, rc);
goto out_free;
}
if (do_nonshared)
dcss_command = DCSS_LOADNSR;
else
......@@ -372,7 +316,7 @@ __segment_load (char *name, int do_nonshared, unsigned long *addr, unsigned long
rc = dcss_diag_translate_rc (seg->end);
dcss_diag(DCSS_PURGESEG, seg->dcss_name,
&seg->start_addr, &seg->end);
goto out_free;
goto out_shared;
}
seg->do_nonshared = do_nonshared;
atomic_set(&seg->ref_count, 1);
......@@ -391,6 +335,8 @@ __segment_load (char *name, int do_nonshared, unsigned long *addr, unsigned long
(void*)seg->start_addr, (void*)seg->end,
segtype_string[seg->vm_segtype]);
goto out;
out_shared:
remove_shared_memory(seg->start_addr, seg->end - seg->start_addr + 1);
out_free:
kfree(seg);
out:
......@@ -530,12 +476,12 @@ segment_unload(char *name)
"please report to linux390@de.ibm.com\n",name);
goto out_unlock;
}
if (atomic_dec_return(&seg->ref_count) == 0) {
list_del(&seg->list);
dcss_diag(DCSS_PURGESEG, seg->dcss_name,
&dummy, &dummy);
kfree(seg);
}
if (atomic_dec_return(&seg->ref_count) != 0)
goto out_unlock;
remove_shared_memory(seg->start_addr, seg->end - seg->start_addr + 1);
list_del(&seg->list);
dcss_diag(DCSS_PURGESEG, seg->dcss_name, &dummy, &dummy);
kfree(seg);
out_unlock:
mutex_unlock(&dcss_lock);
}
......
......@@ -69,6 +69,8 @@ void show_mem(void)
printk("Free swap: %6ldkB\n", nr_swap_pages<<(PAGE_SHIFT-10));
i = max_mapnr;
while (i-- > 0) {
if (!pfn_valid(i))
continue;
page = pfn_to_page(i);
total++;
if (PageReserved(page))
......@@ -84,67 +86,53 @@ void show_mem(void)
printk("%d pages swap cached\n",cached);
}
static void __init setup_ro_region(void)
{
pgd_t *pgd;
pmd_t *pmd;
pte_t *pte;
pte_t new_pte;
unsigned long address, end;
address = ((unsigned long)&__start_rodata) & PAGE_MASK;
end = PFN_ALIGN((unsigned long)&__end_rodata);
for (; address < end; address += PAGE_SIZE) {
pgd = pgd_offset_k(address);
pmd = pmd_offset(pgd, address);
pte = pte_offset_kernel(pmd, address);
new_pte = mk_pte_phys(address, __pgprot(_PAGE_RO));
set_pte(pte, new_pte);
}
}
extern unsigned long __initdata zholes_size[];
extern void vmem_map_init(void);
/*
* paging_init() sets up the page tables
*/
#ifndef CONFIG_64BIT
void __init paging_init(void)
{
pgd_t * pg_dir;
pte_t * pg_table;
pte_t pte;
int i;
unsigned long tmp;
unsigned long pfn = 0;
unsigned long pgdir_k = (__pa(swapper_pg_dir) & PAGE_MASK) | _KERNSEG_TABLE;
static const int ssm_mask = 0x04000000L;
unsigned long ro_start_pfn, ro_end_pfn;
pgd_t *pg_dir;
int i;
unsigned long pgdir_k;
static const int ssm_mask = 0x04000000L;
unsigned long zones_size[MAX_NR_ZONES];
unsigned long dma_pfn, high_pfn;
ro_start_pfn = PFN_DOWN((unsigned long)&__start_rodata);
ro_end_pfn = PFN_UP((unsigned long)&__end_rodata);
memset(zones_size, 0, sizeof(zones_size));
zones_size[ZONE_DMA] = max_low_pfn;
free_area_init_node(0, &contig_page_data, zones_size,
__pa(PAGE_OFFSET) >> PAGE_SHIFT,
zholes_size);
/* unmap whole virtual address space */
pg_dir = swapper_pg_dir;
pg_dir = swapper_pg_dir;
#ifdef CONFIG_64BIT
pgdir_k = (__pa(swapper_pg_dir) & PAGE_MASK) | _KERN_REGION_TABLE;
for (i = 0; i < PTRS_PER_PGD; i++)
pmd_clear((pmd_t *) pg_dir++);
/*
* map whole physical memory to virtual memory (identity mapping)
*/
pg_dir = swapper_pg_dir;
while (pfn < max_low_pfn) {
/*
* pg_table is physical at this point
*/
pg_table = (pte_t *) alloc_bootmem_pages(PAGE_SIZE);
pmd_populate_kernel(&init_mm, (pmd_t *) pg_dir, pg_table);
pg_dir++;
for (tmp = 0 ; tmp < PTRS_PER_PTE ; tmp++,pg_table++) {
if (pfn >= ro_start_pfn && pfn < ro_end_pfn)
pte = pfn_pte(pfn, __pgprot(_PAGE_RO));
else
pte = pfn_pte(pfn, PAGE_KERNEL);
if (pfn >= max_low_pfn)
pte_val(pte) = _PAGE_TYPE_EMPTY;
set_pte(pg_table, pte);
pfn++;
}
}
pgd_clear(pg_dir + i);
#else
pgdir_k = (__pa(swapper_pg_dir) & PAGE_MASK) | _KERNSEG_TABLE;
for (i = 0; i < PTRS_PER_PGD; i++)
pmd_clear((pmd_t *)(pg_dir + i));
#endif
vmem_map_init();
setup_ro_region();
S390_lowcore.kernel_asce = pgdir_k;
......@@ -154,31 +142,9 @@ void __init paging_init(void)
__ctl_load(pgdir_k, 13, 13);
__raw_local_irq_ssm(ssm_mask);
local_flush_tlb();
}
#else /* CONFIG_64BIT */
void __init paging_init(void)
{
pgd_t * pg_dir;
pmd_t * pm_dir;
pte_t * pt_dir;
pte_t pte;
int i,j,k;
unsigned long pfn = 0;
unsigned long pgdir_k = (__pa(swapper_pg_dir) & PAGE_MASK) |
_KERN_REGION_TABLE;
static const int ssm_mask = 0x04000000L;
unsigned long zones_size[MAX_NR_ZONES];
unsigned long dma_pfn, high_pfn;
unsigned long ro_start_pfn, ro_end_pfn;
memset(zones_size, 0, sizeof(zones_size));
dma_pfn = MAX_DMA_ADDRESS >> PAGE_SHIFT;
high_pfn = max_low_pfn;
ro_start_pfn = PFN_DOWN((unsigned long)&__start_rodata);
ro_end_pfn = PFN_UP((unsigned long)&__end_rodata);
if (dma_pfn > high_pfn)
zones_size[ZONE_DMA] = high_pfn;
......@@ -190,56 +156,7 @@ void __init paging_init(void)
/* Initialize mem_map[]. */
free_area_init_node(0, &contig_page_data, zones_size,
__pa(PAGE_OFFSET) >> PAGE_SHIFT, zholes_size);
/*
* map whole physical memory to virtual memory (identity mapping)
*/
pg_dir = swapper_pg_dir;
for (i = 0 ; i < PTRS_PER_PGD ; i++,pg_dir++) {
if (pfn >= max_low_pfn) {
pgd_clear(pg_dir);
continue;
}
pm_dir = (pmd_t *) alloc_bootmem_pages(PAGE_SIZE * 4);
pgd_populate(&init_mm, pg_dir, pm_dir);
for (j = 0 ; j < PTRS_PER_PMD ; j++,pm_dir++) {
if (pfn >= max_low_pfn) {
pmd_clear(pm_dir);
continue;
}
pt_dir = (pte_t *) alloc_bootmem_pages(PAGE_SIZE);
pmd_populate_kernel(&init_mm, pm_dir, pt_dir);
for (k = 0 ; k < PTRS_PER_PTE ; k++,pt_dir++) {
if (pfn >= ro_start_pfn && pfn < ro_end_pfn)
pte = pfn_pte(pfn, __pgprot(_PAGE_RO));
else
pte = pfn_pte(pfn, PAGE_KERNEL);
if (pfn >= max_low_pfn)
pte_val(pte) = _PAGE_TYPE_EMPTY;
set_pte(pt_dir, pte);
pfn++;
}
}
}
S390_lowcore.kernel_asce = pgdir_k;
/* enable virtual mapping in kernel mode */
__ctl_load(pgdir_k, 1, 1);
__ctl_load(pgdir_k, 7, 7);
__ctl_load(pgdir_k, 13, 13);
__raw_local_irq_ssm(ssm_mask);
local_flush_tlb();
}
#endif /* CONFIG_64BIT */
void __init mem_init(void)
{
......@@ -269,6 +186,8 @@ void __init mem_init(void)
printk("Write protected kernel read-only data: %#lx - %#lx\n",
(unsigned long)&__start_rodata,
PFN_ALIGN((unsigned long)&__end_rodata) - 1);
printk("Virtual memmap size: %ldk\n",
(max_pfn * sizeof(struct page)) >> 10);
}
void free_initmem(void)
......
/*
* arch/s390/mm/vmem.c
*
* Copyright IBM Corp. 2006
* Author(s): Heiko Carstens <heiko.carstens@de.ibm.com>
*/
#include <linux/bootmem.h>
#include <linux/pfn.h>
#include <linux/mm.h>
#include <linux/module.h>
#include <linux/list.h>
#include <asm/pgalloc.h>
#include <asm/pgtable.h>
#include <asm/setup.h>
#include <asm/tlbflush.h>
unsigned long vmalloc_end;
EXPORT_SYMBOL(vmalloc_end);
static struct page *vmem_map;
static DEFINE_MUTEX(vmem_mutex);
struct memory_segment {
struct list_head list;
unsigned long start;
unsigned long size;
};
static LIST_HEAD(mem_segs);
void memmap_init(unsigned long size, int nid, unsigned long zone,
unsigned long start_pfn)
{
struct page *start, *end;
struct page *map_start, *map_end;
int i;
start = pfn_to_page(start_pfn);
end = start + size;
for (i = 0; i < MEMORY_CHUNKS && memory_chunk[i].size > 0; i++) {
unsigned long cstart, cend;
cstart = PFN_DOWN(memory_chunk[i].addr);
cend = cstart + PFN_DOWN(memory_chunk[i].size);
map_start = mem_map + cstart;
map_end = mem_map + cend;
if (map_start < start)
map_start = start;
if (map_end > end)
map_end = end;
map_start -= ((unsigned long) map_start & (PAGE_SIZE - 1))
/ sizeof(struct page);
map_end += ((PFN_ALIGN((unsigned long) map_end)
- (unsigned long) map_end)
/ sizeof(struct page));
if (map_start < map_end)
memmap_init_zone((unsigned long)(map_end - map_start),
nid, zone, page_to_pfn(map_start));
}
}
static inline void *vmem_alloc_pages(unsigned int order)
{
if (slab_is_available())
return (void *)__get_free_pages(GFP_KERNEL, order);
return alloc_bootmem_pages((1 << order) * PAGE_SIZE);
}
static inline pmd_t *vmem_pmd_alloc(void)
{
pmd_t *pmd;
int i;
pmd = vmem_alloc_pages(PMD_ALLOC_ORDER);
if (!pmd)
return NULL;
for (i = 0; i < PTRS_PER_PMD; i++)
pmd_clear(pmd + i);
return pmd;
}
static inline pte_t *vmem_pte_alloc(void)
{
pte_t *pte;
pte_t empty_pte;
int i;
pte = vmem_alloc_pages(PTE_ALLOC_ORDER);
if (!pte)
return NULL;
pte_val(empty_pte) = _PAGE_TYPE_EMPTY;
for (i = 0; i < PTRS_PER_PTE; i++)
set_pte(pte + i, empty_pte);
return pte;
}
/*
* Add a physical memory range to the 1:1 mapping.
*/
static int vmem_add_range(unsigned long start, unsigned long size)
{
unsigned long address;
pgd_t *pg_dir;
pmd_t *pm_dir;
pte_t *pt_dir;
pte_t pte;
int ret = -ENOMEM;
for (address = start; address < start + size; address += PAGE_SIZE) {
pg_dir = pgd_offset_k(address);
if (pgd_none(*pg_dir)) {
pm_dir = vmem_pmd_alloc();
if (!pm_dir)
goto out;
pgd_populate(&init_mm, pg_dir, pm_dir);
}
pm_dir = pmd_offset(pg_dir, address);
if (pmd_none(*pm_dir)) {
pt_dir = vmem_pte_alloc();
if (!pt_dir)
goto out;
pmd_populate_kernel(&init_mm, pm_dir, pt_dir);
}
pt_dir = pte_offset_kernel(pm_dir, address);
pte = pfn_pte(address >> PAGE_SHIFT, PAGE_KERNEL);
set_pte(pt_dir, pte);
}
ret = 0;
out:
flush_tlb_kernel_range(start, start + size);
return ret;
}
/*
* Remove a physical memory range from the 1:1 mapping.
* Currently only invalidates page table entries.
*/
static void vmem_remove_range(unsigned long start, unsigned long size)
{
unsigned long address;
pgd_t *pg_dir;
pmd_t *pm_dir;
pte_t *pt_dir;
pte_t pte;
pte_val(pte) = _PAGE_TYPE_EMPTY;
for (address = start; address < start + size; address += PAGE_SIZE) {
pg_dir = pgd_offset_k(address);
if (pgd_none(*pg_dir))
continue;
pm_dir = pmd_offset(pg_dir, address);
if (pmd_none(*pm_dir))
continue;
pt_dir = pte_offset_kernel(pm_dir, address);
set_pte(pt_dir, pte);
}
flush_tlb_kernel_range(start, start + size);
}
/*
* Add a backed mem_map array to the virtual mem_map array.
*/
static int vmem_add_mem_map(unsigned long start, unsigned long size)
{
unsigned long address, start_addr, end_addr;
struct page *map_start, *map_end;
pgd_t *pg_dir;
pmd_t *pm_dir;
pte_t *pt_dir;
pte_t pte;
int ret = -ENOMEM;
map_start = vmem_map + PFN_DOWN(start);
map_end = vmem_map + PFN_DOWN(start + size);
start_addr = (unsigned long) map_start & PAGE_MASK;
end_addr = PFN_ALIGN((unsigned long) map_end);
for (address = start_addr; address < end_addr; address += PAGE_SIZE) {
pg_dir = pgd_offset_k(address);
if (pgd_none(*pg_dir)) {
pm_dir = vmem_pmd_alloc();
if (!pm_dir)
goto out;
pgd_populate(&init_mm, pg_dir, pm_dir);
}
pm_dir = pmd_offset(pg_dir, address);
if (pmd_none(*pm_dir)) {
pt_dir = vmem_pte_alloc();
if (!pt_dir)
goto out;
pmd_populate_kernel(&init_mm, pm_dir, pt_dir);
}
pt_dir = pte_offset_kernel(pm_dir, address);
if (pte_none(*pt_dir)) {
unsigned long new_page;
new_page =__pa(vmem_alloc_pages(0));
if (!new_page)
goto out;
pte = pfn_pte(new_page >> PAGE_SHIFT, PAGE_KERNEL);
set_pte(pt_dir, pte);
}
}
ret = 0;
out:
flush_tlb_kernel_range(start_addr, end_addr);
return ret;
}
static int vmem_add_mem(unsigned long start, unsigned long size)