Mailing List Archive

# HG changeset patch
# User awilliam@xenbuild.aw
# Node ID 953753661a3bf13d079823fc422b4c93c5a1c240
# Parent f0f88d9c4c9ede9c37b15a46f790d81f7648518a
[IA64] make xenLinux/ia64 privcmd mmap not to use dom0 memory

xenLinux/ia64 privcmd mmap uses pseudo physical address space.
it used alloc_pages() to allocate the space.
It wastes dom0 memory and sometimes several hundreds megabytes is
allocated depending on domU memory size.
With this patch xenLinux/ia64 trys to find the region which can be
used safely and uses the reasion.

Signed-off-by: Isaku Yamahata <yamahata@valinux.co.jp>
---
linux-2.6-xen-sparse/arch/ia64/xen/hypervisor.c | 265 +++++++++++++++++-------
1 files changed, 197 insertions(+), 68 deletions(-)

diff -r f0f88d9c4c9e -r 953753661a3b linux-2.6-xen-sparse/arch/ia64/xen/hypervisor.c
--- a/linux-2.6-xen-sparse/arch/ia64/xen/hypervisor.c Tue May 23 09:17:57 2006 -0600
+++ b/linux-2.6-xen-sparse/arch/ia64/xen/hypervisor.c Tue May 23 15:09:21 2006 -0600
@@ -360,49 +360,172 @@ struct address_space xen_ia64_foreign_du

///////////////////////////////////////////////////////////////////////////
// foreign mapping
+#include <linux/efi.h>
+#include <asm/meminit.h> // for IA64_GRANULE_SIZE, GRANULEROUND{UP,DOWN}()
+
+static unsigned long privcmd_resource_min = 0;
+// Xen/ia64 currently can handle pseudo physical address bits up to
+// (PAGE_SHIFT * 3)
+static unsigned long privcmd_resource_max = GRANULEROUNDDOWN((1UL << (PAGE_SHIFT * 3)) - 1);
+static unsigned long privcmd_resource_align = IA64_GRANULE_SIZE;
+
+static unsigned long
+md_end_addr(const efi_memory_desc_t *md)
+{
+ return md->phys_addr + (md->num_pages << EFI_PAGE_SHIFT);
+}
+
+#define XEN_IA64_PRIVCMD_LEAST_GAP_SIZE (1024 * 1024 * 1024UL)
+static int
+xen_ia64_privcmd_check_size(unsigned long start, unsigned long end)
+{
+ return (start < end &&
+ (end - start) > XEN_IA64_PRIVCMD_LEAST_GAP_SIZE);
+}
+
+static int __init
+xen_ia64_privcmd_init(void)
+{
+ void *efi_map_start, *efi_map_end, *p;
+ u64 efi_desc_size;
+ efi_memory_desc_t *md;
+ unsigned long tmp_min;
+ unsigned long tmp_max;
+ unsigned long gap_size;
+ unsigned long prev_end;
+
+ if (!is_running_on_xen())
+ return -1;
+
+ efi_map_start = __va(ia64_boot_param->efi_memmap);
+ efi_map_end = efi_map_start + ia64_boot_param->efi_memmap_size;
+ efi_desc_size = ia64_boot_param->efi_memdesc_size;
+
+ // at first check the used highest address
+ for (p = efi_map_start; p < efi_map_end; p += efi_desc_size) {
+ // nothing
+ }
+ md = p - efi_desc_size;
+ privcmd_resource_min = GRANULEROUNDUP(md_end_addr(md));
+ if (xen_ia64_privcmd_check_size(privcmd_resource_min,
+ privcmd_resource_max)) {
+ goto out;
+ }
+
+ // the used highest address is too large. try to find the largest gap.
+ tmp_min = privcmd_resource_max;
+ tmp_max = 0;
+ gap_size = 0;
+ prev_end = 0;
+ for (p = efi_map_start;
+ p < efi_map_end - efi_desc_size;
+ p += efi_desc_size) {
+ unsigned long end;
+ efi_memory_desc_t* next;
+ unsigned long next_start;
+
+ md = p;
+ end = md_end_addr(md);
+ if (end > privcmd_resource_max) {
+ break;
+ }
+ if (end < prev_end) {
+ // work around.
+ // Xen may pass incompletely sorted memory
+ // descriptors like
+ // [x, x + length]
+ // [x, x]
+ // this order should be reversed.
+ continue;
+ }
+ next = p + efi_desc_size;
+ next_start = next->phys_addr;
+ if (next_start > privcmd_resource_max) {
+ next_start = privcmd_resource_max;
+ }
+ if (end < next_start && gap_size < (next_start - end)) {
+ tmp_min = end;
+ tmp_max = next_start;
+ gap_size = tmp_max - tmp_min;
+ }
+ prev_end = end;
+ }
+
+ privcmd_resource_min = GRANULEROUNDUP(tmp_min);
+ if (xen_ia64_privcmd_check_size(privcmd_resource_min, tmp_max)) {
+ privcmd_resource_max = tmp_max;
+ goto out;
+ }
+
+ privcmd_resource_min = tmp_min;
+ privcmd_resource_max = tmp_max;
+ if (!xen_ia64_privcmd_check_size(privcmd_resource_min,
+ privcmd_resource_max)) {
+ // Any large enough gap isn't found.
+ // go ahead anyway with the warning hoping that large region
+ // won't be requested.
+ printk(KERN_WARNING "xen privcmd: large enough region for privcmd mmap is not found.\n");
+ }
+
+out:
+ printk(KERN_INFO "xen privcmd uses pseudo physical addr range [0x%lx, 0x%lx] (%ldMB)\n",
+ privcmd_resource_min, privcmd_resource_max,
+ (privcmd_resource_max - privcmd_resource_min) >> 20);
+ BUG_ON(privcmd_resource_min >= privcmd_resource_max);
+ return 0;
+}
+late_initcall(xen_ia64_privcmd_init);

struct xen_ia64_privcmd_entry {
atomic_t map_count;
- struct page* page;
+#define INVALID_GPFN (~0UL)
+ unsigned long gpfn;
+};
+
+struct xen_ia64_privcmd_range {
+ atomic_t ref_count;
+ unsigned long pgoff; // in PAGE_SIZE
+ struct resource* res;
+
+ unsigned long num_entries;
+ struct xen_ia64_privcmd_entry entries[0];
+};
+
+struct xen_ia64_privcmd_vma {
+ struct xen_ia64_privcmd_range* range;
+
+ unsigned long num_entries;
+ struct xen_ia64_privcmd_entry* entries;
};

static void
xen_ia64_privcmd_init_entry(struct xen_ia64_privcmd_entry* entry)
{
atomic_set(&entry->map_count, 0);
- entry->page = NULL;
-}
-
-//TODO alloc_page() to allocate pseudo physical address space is
-// waste of memory.
-// When vti domain is created, qemu maps all of vti domain pages which
-// reaches to several hundred megabytes at least.
-// remove alloc_page().
+ entry->gpfn = INVALID_GPFN;
+}
+
static int
xen_ia64_privcmd_entry_mmap(struct vm_area_struct* vma,
unsigned long addr,
- struct xen_ia64_privcmd_entry* entry,
+ struct xen_ia64_privcmd_range* privcmd_range,
+ int i,
unsigned long mfn,
pgprot_t prot,
domid_t domid)
{
int error = 0;
- struct page* page;
+ struct xen_ia64_privcmd_entry* entry = &privcmd_range->entries[i];
unsigned long gpfn;

BUG_ON((addr & ~PAGE_MASK) != 0);
BUG_ON(mfn == INVALID_MFN);

- if (entry->page != NULL) {
+ if (entry->gpfn != INVALID_GPFN) {
error = -EBUSY;
goto out;
}
- page = alloc_page(GFP_KERNEL);
- if (page == NULL) {
- error = -ENOMEM;
- goto out;
- }
- gpfn = page_to_pfn(page);
+ gpfn = (privcmd_range->res->start >> PAGE_SHIFT) + i;

error = HYPERVISOR_add_physmap(gpfn, mfn, 0/* prot:XXX */,
domid);
@@ -413,15 +536,13 @@ xen_ia64_privcmd_entry_mmap(struct vm_ar
prot = vma->vm_page_prot;
error = remap_pfn_range(vma, addr, gpfn, 1 << PAGE_SHIFT, prot);
if (error != 0) {
- (void)HYPERVISOR_zap_physmap(gpfn, 0);
- error = HYPERVISOR_populate_physmap(gpfn, 0, 0);
+ error = HYPERVISOR_zap_physmap(gpfn, 0);
if (error) {
BUG();//XXX
}
- __free_page(page);
} else {
atomic_inc(&entry->map_count);
- entry->page = page;
+ entry->gpfn = gpfn;
}

out:
@@ -429,30 +550,28 @@ out:
}

static void
-xen_ia64_privcmd_entry_munmap(struct xen_ia64_privcmd_entry* entry)
-{
- struct page* page = entry->page;
- unsigned long gpfn = page_to_pfn(page);
+xen_ia64_privcmd_entry_munmap(struct xen_ia64_privcmd_range* privcmd_range,
+ int i)
+{
+ struct xen_ia64_privcmd_entry* entry = &privcmd_range->entries[i];
+ unsigned long gpfn = entry->gpfn;
+ //gpfn = (privcmd_range->res->start >> PAGE_SHIFT) +
+ // (vma->vm_pgoff - privcmd_range->pgoff);
int error;

error = HYPERVISOR_zap_physmap(gpfn, 0);
if (error) {
BUG();//XXX
}
-
- error = HYPERVISOR_populate_physmap(gpfn, 0, 0);
- if (error) {
- BUG();//XXX
- }
-
- entry->page = NULL;
- __free_page(page);
+ entry->gpfn = INVALID_GPFN;
}

static int
-xen_ia64_privcmd_entry_open(struct xen_ia64_privcmd_entry* entry)
-{
- if (entry->page != NULL) {
+xen_ia64_privcmd_entry_open(struct xen_ia64_privcmd_range* privcmd_range,
+ int i)
+{
+ struct xen_ia64_privcmd_entry* entry = &privcmd_range->entries[i];
+ if (entry->gpfn != INVALID_GPFN) {
atomic_inc(&entry->map_count);
} else {
BUG_ON(atomic_read(&entry->map_count) != 0);
@@ -460,27 +579,15 @@ xen_ia64_privcmd_entry_open(struct xen_i
}

static int
-xen_ia64_privcmd_entry_close(struct xen_ia64_privcmd_entry* entry)
-{
- if (entry->page != NULL && atomic_dec_and_test(&entry->map_count)) {
- xen_ia64_privcmd_entry_munmap(entry);
- }
-}
-
-struct xen_ia64_privcmd_range {
- atomic_t ref_count;
- unsigned long pgoff; // in PAGE_SIZE
-
- unsigned long num_entries;
- struct xen_ia64_privcmd_entry entries[0];
-};
-
-struct xen_ia64_privcmd_vma {
- struct xen_ia64_privcmd_range* range;
-
- unsigned long num_entries;
- struct xen_ia64_privcmd_entry* entries;
-};
+xen_ia64_privcmd_entry_close(struct xen_ia64_privcmd_range* privcmd_range,
+ int i)
+{
+ struct xen_ia64_privcmd_entry* entry = &privcmd_range->entries[i];
+ if (entry->gpfn != INVALID_GPFN &&
+ atomic_dec_and_test(&entry->map_count)) {
+ xen_ia64_privcmd_entry_munmap(privcmd_range, i);
+ }
+}

static void xen_ia64_privcmd_vma_open(struct vm_area_struct* vma);
static void xen_ia64_privcmd_vma_close(struct vm_area_struct* vma);
@@ -507,7 +614,7 @@ __xen_ia64_privcmd_vma_open(struct vm_ar
privcmd_vma->entries = &privcmd_range->entries[entry_offset];
vma->vm_private_data = privcmd_vma;
for (i = 0; i < privcmd_vma->num_entries; i++) {
- xen_ia64_privcmd_entry_open(&privcmd_vma->entries[i]);
+ xen_ia64_privcmd_entry_open(privcmd_range, entry_offset + i);
}

vma->vm_private_data = privcmd_vma;
@@ -533,10 +640,11 @@ xen_ia64_privcmd_vma_close(struct vm_are
struct xen_ia64_privcmd_vma* privcmd_vma =
(struct xen_ia64_privcmd_vma*)vma->vm_private_data;
struct xen_ia64_privcmd_range* privcmd_range = privcmd_vma->range;
+ unsigned long entry_offset = vma->vm_pgoff - privcmd_range->pgoff;
unsigned long i;

for (i = 0; i < privcmd_vma->num_entries; i++) {
- xen_ia64_privcmd_entry_close(&privcmd_vma->entries[i]);
+ xen_ia64_privcmd_entry_close(privcmd_range, entry_offset + i);
}
vma->vm_private_data = NULL;
kfree(privcmd_vma);
@@ -547,9 +655,11 @@ xen_ia64_privcmd_vma_close(struct vm_are
struct xen_ia64_privcmd_entry* entry =
&privcmd_range->entries[i];
BUG_ON(atomic_read(&entry->map_count) != 0);
- BUG_ON(entry->page != NULL);
+ BUG_ON(entry->gpfn != INVALID_GPFN);
}
#endif
+ release_resource(privcmd_range->res);
+ kfree(privcmd_range->res);
vfree(privcmd_range);
}
}
@@ -557,13 +667,18 @@ int
int
privcmd_mmap(struct file * file, struct vm_area_struct * vma)
{
- unsigned long num_entries = (vma->vm_end - vma->vm_start) >> PAGE_SHIFT;
- struct xen_ia64_privcmd_range* privcmd_range;
- struct xen_ia64_privcmd_vma* privcmd_vma;
+ int error;
+ unsigned long size = vma->vm_end - vma->vm_start;
+ unsigned long num_entries = size >> PAGE_SHIFT;
+ struct xen_ia64_privcmd_range* privcmd_range = NULL;
+ struct xen_ia64_privcmd_vma* privcmd_vma = NULL;
+ struct resource* res = NULL;
unsigned long i;
BUG_ON(!running_on_xen);

BUG_ON(file->private_data != NULL);
+
+ error = -ENOMEM;
privcmd_range =
vmalloc(sizeof(*privcmd_range) +
sizeof(privcmd_range->entries[0]) * num_entries);
@@ -574,6 +689,18 @@ privcmd_mmap(struct file * file, struct
if (privcmd_vma == NULL) {
goto out_enomem1;
}
+ res = kzalloc(sizeof(*res), GFP_KERNEL);
+ if (res == NULL) {
+ goto out_enomem1;
+ }
+ res->name = "Xen privcmd mmap";
+ error = allocate_resource(&iomem_resource, res, size,
+ privcmd_resource_min, privcmd_resource_max,
+ privcmd_resource_align, NULL, NULL);
+ if (error) {
+ goto out_enomem1;
+ }
+ privcmd_range->res = res;

/* DONTCOPY is essential for Xen as copy_page_range is broken. */
vma->vm_flags |= VM_RESERVED | VM_IO | VM_DONTCOPY | VM_PFNMAP;
@@ -589,10 +716,11 @@ privcmd_mmap(struct file * file, struct
return 0;

out_enomem1:
+ kfree(res);
kfree(privcmd_vma);
out_enomem0:
vfree(privcmd_range);
- return -ENOMEM;
+ return error;
}

int
@@ -605,6 +733,9 @@ direct_remap_pfn_range(struct vm_area_st
{
struct xen_ia64_privcmd_vma* privcmd_vma =
(struct xen_ia64_privcmd_vma*)vma->vm_private_data;
+ struct xen_ia64_privcmd_range* privcmd_range = privcmd_vma->range;
+ unsigned long entry_offset = vma->vm_pgoff - privcmd_range->pgoff;
+
unsigned long i;
unsigned long offset;
int error = 0;
@@ -618,9 +749,7 @@ direct_remap_pfn_range(struct vm_area_st

i = (address - vma->vm_start) >> PAGE_SHIFT;
for (offset = 0; offset < size; offset += PAGE_SIZE) {
- struct xen_ia64_privcmd_entry* entry =
- &privcmd_vma->entries[i];
- error = xen_ia64_privcmd_entry_mmap(vma, (address + offset) & PAGE_MASK, entry, mfn, prot, domid);
+ error = xen_ia64_privcmd_entry_mmap(vma, (address + offset) & PAGE_MASK, privcmd_range, entry_offset + i, mfn, prot, domid);
if (error != 0) {
break;
}

_______________________________________________
Xen-changelog mailing list
Xen-changelog@lists.xensource.com
http://lists.xensource.com/xen-changelog