/*
* VMware Balloon driver.
*
- * Copyright (C) 2000-2013, VMware, Inc. All Rights Reserved.
+ * Copyright (C) 2000-2014, VMware, Inc. All Rights Reserved.
*
* This program is free software; you can redistribute it and/or modify it
* under the terms of the GNU General Public License as published by the
#include <linux/workqueue.h>
#include <linux/debugfs.h>
#include <linux/seq_file.h>
+#include <linux/vmw_vmci_defs.h>
+#include <linux/vmw_vmci_api.h>
#include <asm/hypervisor.h>
MODULE_AUTHOR("VMware, Inc.");
MODULE_DESCRIPTION("VMware Memory Control (Balloon) Driver");
-MODULE_VERSION("1.3.2.0-k");
+MODULE_VERSION("1.5.0.0-k");
MODULE_ALIAS("dmi:*:svnVMware*:*");
MODULE_ALIAS("vmware_vmmemctl");
MODULE_LICENSE("GPL");
* measured in pages.
*/
-/*
- * Rate of allocating memory when there is no memory pressure
- * (driver performs non-sleeping allocations).
- */
-#define VMW_BALLOON_NOSLEEP_ALLOC_MAX 16384U
-
/*
* Rates of memory allocaton when guest experiences memory pressure
* (driver performs sleeping allocations).
#define VMW_BALLOON_RATE_ALLOC_MAX 2048U
#define VMW_BALLOON_RATE_ALLOC_INC 16U
-/*
- * Rates for releasing pages while deflating balloon.
- */
-#define VMW_BALLOON_RATE_FREE_MIN 512U
-#define VMW_BALLOON_RATE_FREE_MAX 16384U
-#define VMW_BALLOON_RATE_FREE_INC 16U
-
/*
* When guest is under memory pressure, use a reduced page allocation
* rate for next several cycles.
/*
* Use __GFP_HIGHMEM to allow pages from HIGHMEM zone. We don't
- * allow wait (__GFP_WAIT) for NOSLEEP page allocations. Use
+ * allow wait (__GFP_RECLAIM) for NOSLEEP page allocations. Use
* __GFP_NOWARN, to suppress page allocation failure warnings.
*/
#define VMW_PAGE_ALLOC_NOSLEEP (__GFP_HIGHMEM|__GFP_NOWARN)
*/
#define VMW_PAGE_ALLOC_CANSLEEP (GFP_HIGHUSER)
-/* Maximum number of page allocations without yielding processor */
-#define VMW_BALLOON_YIELD_THRESHOLD 1024
-
/* Maximum number of refused pages we accumulate during inflation cycle */
#define VMW_BALLOON_MAX_REFUSED 16
/*
* Bit 0 is reserved and not associated to any capability.
*/
- VMW_BALLOON_BASIC_CMDS = (1 << 1),
- VMW_BALLOON_BATCHED_CMDS = (1 << 2)
+ VMW_BALLOON_BASIC_CMDS = (1 << 1),
+ VMW_BALLOON_BATCHED_CMDS = (1 << 2),
+ VMW_BALLOON_BATCHED_2M_CMDS = (1 << 3),
+ VMW_BALLOON_SIGNALLED_WAKEUP_CMD = (1 << 4),
};
#define VMW_BALLOON_CAPABILITIES (VMW_BALLOON_BASIC_CMDS \
- | VMW_BALLOON_BATCHED_CMDS)
+ | VMW_BALLOON_BATCHED_CMDS \
+ | VMW_BALLOON_BATCHED_2M_CMDS \
+ | VMW_BALLOON_SIGNALLED_WAKEUP_CMD)
+
+#define VMW_BALLOON_2M_SHIFT (9)
+#define VMW_BALLOON_NUM_PAGE_SIZES (2)
/*
* Backdoor commands availability:
* LOCK and UNLOCK commands,
* VMW_BALLOON_BATCHED_CMDS:
* BATCHED_LOCK and BATCHED_UNLOCK commands.
+ * VMW BALLOON_BATCHED_2M_CMDS:
+ * BATCHED_2M_LOCK and BATCHED_2M_UNLOCK commands,
+ * VMW VMW_BALLOON_SIGNALLED_WAKEUP_CMD:
+ * VMW_BALLOON_CMD_VMCI_DOORBELL_SET command.
*/
-#define VMW_BALLOON_CMD_START 0
-#define VMW_BALLOON_CMD_GET_TARGET 1
-#define VMW_BALLOON_CMD_LOCK 2
-#define VMW_BALLOON_CMD_UNLOCK 3
-#define VMW_BALLOON_CMD_GUEST_ID 4
-#define VMW_BALLOON_CMD_BATCHED_LOCK 6
-#define VMW_BALLOON_CMD_BATCHED_UNLOCK 7
+#define VMW_BALLOON_CMD_START 0
+#define VMW_BALLOON_CMD_GET_TARGET 1
+#define VMW_BALLOON_CMD_LOCK 2
+#define VMW_BALLOON_CMD_UNLOCK 3
+#define VMW_BALLOON_CMD_GUEST_ID 4
+#define VMW_BALLOON_CMD_BATCHED_LOCK 6
+#define VMW_BALLOON_CMD_BATCHED_UNLOCK 7
+#define VMW_BALLOON_CMD_BATCHED_2M_LOCK 8
+#define VMW_BALLOON_CMD_BATCHED_2M_UNLOCK 9
+#define VMW_BALLOON_CMD_VMCI_DOORBELL_SET 10
+
/* error codes */
#define VMW_BALLOON_SUCCESS 0
* +-------------+----------+--------+
* 64 PAGE_SHIFT 6 0
*
- * For now only 4K pages are supported, but we can easily support large pages
- * by using bits in the reserved field.
- *
* The reserved field should be set to 0.
*/
#define VMW_BALLOON_BATCH_MAX_PAGES (PAGE_SIZE / sizeof(u64))
#ifdef CONFIG_DEBUG_FS
struct vmballoon_stats {
unsigned int timer;
+ unsigned int doorbell;
/* allocation statistics */
- unsigned int alloc;
- unsigned int alloc_fail;
+ unsigned int alloc[VMW_BALLOON_NUM_PAGE_SIZES];
+ unsigned int alloc_fail[VMW_BALLOON_NUM_PAGE_SIZES];
unsigned int sleep_alloc;
unsigned int sleep_alloc_fail;
- unsigned int refused_alloc;
- unsigned int refused_free;
- unsigned int free;
+ unsigned int refused_alloc[VMW_BALLOON_NUM_PAGE_SIZES];
+ unsigned int refused_free[VMW_BALLOON_NUM_PAGE_SIZES];
+ unsigned int free[VMW_BALLOON_NUM_PAGE_SIZES];
/* monitor operations */
- unsigned int lock;
- unsigned int lock_fail;
- unsigned int unlock;
- unsigned int unlock_fail;
+ unsigned int lock[VMW_BALLOON_NUM_PAGE_SIZES];
+ unsigned int lock_fail[VMW_BALLOON_NUM_PAGE_SIZES];
+ unsigned int unlock[VMW_BALLOON_NUM_PAGE_SIZES];
+ unsigned int unlock_fail[VMW_BALLOON_NUM_PAGE_SIZES];
unsigned int target;
unsigned int target_fail;
unsigned int start;
unsigned int start_fail;
unsigned int guest_type;
unsigned int guest_type_fail;
+ unsigned int doorbell_set;
+ unsigned int doorbell_unset;
};
#define STATS_INC(stat) (stat)++
struct vmballoon_ops {
void (*add_page)(struct vmballoon *b, int idx, struct page *p);
int (*lock)(struct vmballoon *b, unsigned int num_pages,
- unsigned int *target);
+ bool is_2m_pages, unsigned int *target);
int (*unlock)(struct vmballoon *b, unsigned int num_pages,
- unsigned int *target);
+ bool is_2m_pages, unsigned int *target);
};
-struct vmballoon {
-
+struct vmballoon_page_size {
/* list of reserved physical pages */
struct list_head pages;
/* transient list of non-balloonable pages */
struct list_head refused_pages;
unsigned int n_refused_pages;
+};
+
+struct vmballoon {
+ struct vmballoon_page_size page_sizes[VMW_BALLOON_NUM_PAGE_SIZES];
+
+ /* supported page sizes. 1 == 4k pages only, 2 == 4k and 2m pages */
+ unsigned supported_page_sizes;
/* balloon size in pages */
unsigned int size;
/* adjustment rates (pages per second) */
unsigned int rate_alloc;
- unsigned int rate_free;
/* slowdown page allocations for next few cycles */
unsigned int slow_allocation_cycles;
struct sysinfo sysinfo;
struct delayed_work dwork;
+
+ struct vmci_handle vmci_doorbell;
};
static struct vmballoon balloon;
static bool vmballoon_send_start(struct vmballoon *b, unsigned long req_caps)
{
unsigned long status, capabilities, dummy = 0;
+ bool success;
STATS_INC(b->stats.start);
switch (status) {
case VMW_BALLOON_SUCCESS_WITH_CAPABILITIES:
b->capabilities = capabilities;
- return true;
+ success = true;
+ break;
case VMW_BALLOON_SUCCESS:
b->capabilities = VMW_BALLOON_BASIC_CMDS;
- return true;
+ success = true;
+ break;
+ default:
+ success = false;
}
- pr_debug("%s - failed, hv returns %ld\n", __func__, status);
- STATS_INC(b->stats.start_fail);
- return false;
+ if (b->capabilities & VMW_BALLOON_BATCHED_2M_CMDS)
+ b->supported_page_sizes = 2;
+ else
+ b->supported_page_sizes = 1;
+
+ if (!success) {
+ pr_debug("%s - failed, hv returns %ld\n", __func__, status);
+ STATS_INC(b->stats.start_fail);
+ }
+ return success;
}
static bool vmballoon_check_status(struct vmballoon *b, unsigned long status)
return false;
}
+static u16 vmballoon_page_size(bool is_2m_page)
+{
+ if (is_2m_page)
+ return 1 << VMW_BALLOON_2M_SHIFT;
+
+ return 1;
+}
+
/*
* Retrieve desired balloon size from the host.
*/
if (pfn32 != pfn)
return -1;
- STATS_INC(b->stats.lock);
+ STATS_INC(b->stats.lock[false]);
*hv_status = status = VMWARE_BALLOON_CMD(LOCK, pfn, dummy, *target);
if (vmballoon_check_status(b, status))
return 0;
pr_debug("%s - ppn %lx, hv returns %ld\n", __func__, pfn, status);
- STATS_INC(b->stats.lock_fail);
+ STATS_INC(b->stats.lock_fail[false]);
return 1;
}
static int vmballoon_send_batched_lock(struct vmballoon *b,
- unsigned int num_pages, unsigned int *target)
+ unsigned int num_pages, bool is_2m_pages, unsigned int *target)
{
unsigned long status;
unsigned long pfn = page_to_pfn(b->page);
- STATS_INC(b->stats.lock);
+ STATS_INC(b->stats.lock[is_2m_pages]);
+
+ if (is_2m_pages)
+ status = VMWARE_BALLOON_CMD(BATCHED_2M_LOCK, pfn, num_pages,
+ *target);
+ else
+ status = VMWARE_BALLOON_CMD(BATCHED_LOCK, pfn, num_pages,
+ *target);
- status = VMWARE_BALLOON_CMD(BATCHED_LOCK, pfn, num_pages, *target);
if (vmballoon_check_status(b, status))
return 0;
pr_debug("%s - batch ppn %lx, hv returns %ld\n", __func__, pfn, status);
- STATS_INC(b->stats.lock_fail);
+ STATS_INC(b->stats.lock_fail[is_2m_pages]);
return 1;
}
if (pfn32 != pfn)
return false;
- STATS_INC(b->stats.unlock);
+ STATS_INC(b->stats.unlock[false]);
status = VMWARE_BALLOON_CMD(UNLOCK, pfn, dummy, *target);
if (vmballoon_check_status(b, status))
return true;
pr_debug("%s - ppn %lx, hv returns %ld\n", __func__, pfn, status);
- STATS_INC(b->stats.unlock_fail);
+ STATS_INC(b->stats.unlock_fail[false]);
return false;
}
static bool vmballoon_send_batched_unlock(struct vmballoon *b,
- unsigned int num_pages, unsigned int *target)
+ unsigned int num_pages, bool is_2m_pages, unsigned int *target)
{
unsigned long status;
unsigned long pfn = page_to_pfn(b->page);
- STATS_INC(b->stats.unlock);
+ STATS_INC(b->stats.unlock[is_2m_pages]);
+
+ if (is_2m_pages)
+ status = VMWARE_BALLOON_CMD(BATCHED_2M_UNLOCK, pfn, num_pages,
+ *target);
+ else
+ status = VMWARE_BALLOON_CMD(BATCHED_UNLOCK, pfn, num_pages,
+ *target);
- status = VMWARE_BALLOON_CMD(BATCHED_UNLOCK, pfn, num_pages, *target);
if (vmballoon_check_status(b, status))
return true;
pr_debug("%s - batch ppn %lx, hv returns %ld\n", __func__, pfn, status);
- STATS_INC(b->stats.unlock_fail);
+ STATS_INC(b->stats.unlock_fail[is_2m_pages]);
return false;
}
+static struct page *vmballoon_alloc_page(gfp_t flags, bool is_2m_page)
+{
+ if (is_2m_page)
+ return alloc_pages(flags, VMW_BALLOON_2M_SHIFT);
+
+ return alloc_page(flags);
+}
+
+static void vmballoon_free_page(struct page *page, bool is_2m_page)
+{
+ if (is_2m_page)
+ __free_pages(page, VMW_BALLOON_2M_SHIFT);
+ else
+ __free_page(page);
+}
+
/*
* Quickly release all pages allocated for the balloon. This function is
* called when host decides to "reset" balloon for one reason or another.
static void vmballoon_pop(struct vmballoon *b)
{
struct page *page, *next;
- unsigned int count = 0;
-
- list_for_each_entry_safe(page, next, &b->pages, lru) {
- list_del(&page->lru);
- __free_page(page);
- STATS_INC(b->stats.free);
- b->size--;
-
- if (++count >= b->rate_free) {
- count = 0;
+ unsigned is_2m_pages;
+
+ for (is_2m_pages = 0; is_2m_pages < VMW_BALLOON_NUM_PAGE_SIZES;
+ is_2m_pages++) {
+ struct vmballoon_page_size *page_size =
+ &b->page_sizes[is_2m_pages];
+ u16 size_per_page = vmballoon_page_size(is_2m_pages);
+
+ list_for_each_entry_safe(page, next, &page_size->pages, lru) {
+ list_del(&page->lru);
+ vmballoon_free_page(page, is_2m_pages);
+ STATS_INC(b->stats.free[is_2m_pages]);
+ b->size -= size_per_page;
cond_resched();
}
}
- if ((b->capabilities & VMW_BALLOON_BATCHED_CMDS) != 0) {
- if (b->batch_page)
- vunmap(b->batch_page);
+ if (b->batch_page) {
+ vunmap(b->batch_page);
+ b->batch_page = NULL;
+ }
- if (b->page)
- __free_page(b->page);
+ if (b->page) {
+ __free_page(b->page);
+ b->page = NULL;
}
}
* inflation cycle.
*/
static int vmballoon_lock_page(struct vmballoon *b, unsigned int num_pages,
- unsigned int *target)
+ bool is_2m_pages, unsigned int *target)
{
int locked, hv_status;
struct page *page = b->page;
+ struct vmballoon_page_size *page_size = &b->page_sizes[false];
+
+ /* is_2m_pages can never happen as 2m pages support implies batching */
locked = vmballoon_send_lock_page(b, page_to_pfn(page), &hv_status,
target);
if (locked > 0) {
- STATS_INC(b->stats.refused_alloc);
+ STATS_INC(b->stats.refused_alloc[false]);
if (hv_status == VMW_BALLOON_ERROR_RESET ||
hv_status == VMW_BALLOON_ERROR_PPN_NOTNEEDED) {
- __free_page(page);
+ vmballoon_free_page(page, false);
return -EIO;
}
* and retry allocation, unless we already accumulated
* too many of them, in which case take a breather.
*/
- if (b->n_refused_pages < VMW_BALLOON_MAX_REFUSED) {
- b->n_refused_pages++;
- list_add(&page->lru, &b->refused_pages);
+ if (page_size->n_refused_pages < VMW_BALLOON_MAX_REFUSED) {
+ page_size->n_refused_pages++;
+ list_add(&page->lru, &page_size->refused_pages);
} else {
- __free_page(page);
+ vmballoon_free_page(page, false);
}
return -EIO;
}
/* track allocated page */
- list_add(&page->lru, &b->pages);
+ list_add(&page->lru, &page_size->pages);
/* update balloon size */
b->size++;
}
static int vmballoon_lock_batched_page(struct vmballoon *b,
- unsigned int num_pages, unsigned int *target)
+ unsigned int num_pages, bool is_2m_pages, unsigned int *target)
{
int locked, i;
+ u16 size_per_page = vmballoon_page_size(is_2m_pages);
- locked = vmballoon_send_batched_lock(b, num_pages, target);
+ locked = vmballoon_send_batched_lock(b, num_pages, is_2m_pages,
+ target);
if (locked > 0) {
for (i = 0; i < num_pages; i++) {
u64 pa = vmballoon_batch_get_pa(b->batch_page, i);
struct page *p = pfn_to_page(pa >> PAGE_SHIFT);
- __free_page(p);
+ vmballoon_free_page(p, is_2m_pages);
}
return -EIO;
for (i = 0; i < num_pages; i++) {
u64 pa = vmballoon_batch_get_pa(b->batch_page, i);
struct page *p = pfn_to_page(pa >> PAGE_SHIFT);
+ struct vmballoon_page_size *page_size =
+ &b->page_sizes[is_2m_pages];
locked = vmballoon_batch_get_status(b->batch_page, i);
switch (locked) {
case VMW_BALLOON_SUCCESS:
- list_add(&p->lru, &b->pages);
- b->size++;
+ list_add(&p->lru, &page_size->pages);
+ b->size += size_per_page;
break;
case VMW_BALLOON_ERROR_PPN_PINNED:
case VMW_BALLOON_ERROR_PPN_INVALID:
- if (b->n_refused_pages < VMW_BALLOON_MAX_REFUSED) {
- list_add(&p->lru, &b->refused_pages);
- b->n_refused_pages++;
+ if (page_size->n_refused_pages
+ < VMW_BALLOON_MAX_REFUSED) {
+ list_add(&p->lru, &page_size->refused_pages);
+ page_size->n_refused_pages++;
break;
}
/* Fallthrough */
case VMW_BALLOON_ERROR_RESET:
case VMW_BALLOON_ERROR_PPN_NOTNEEDED:
- __free_page(p);
+ vmballoon_free_page(p, is_2m_pages);
break;
default:
/* This should never happen */
* to use, if needed.
*/
static int vmballoon_unlock_page(struct vmballoon *b, unsigned int num_pages,
- unsigned int *target)
+ bool is_2m_pages, unsigned int *target)
{
struct page *page = b->page;
+ struct vmballoon_page_size *page_size = &b->page_sizes[false];
+
+ /* is_2m_pages can never happen as 2m pages support implies batching */
if (!vmballoon_send_unlock_page(b, page_to_pfn(page), target)) {
- list_add(&page->lru, &b->pages);
+ list_add(&page->lru, &page_size->pages);
return -EIO;
}
/* deallocate page */
- __free_page(page);
- STATS_INC(b->stats.free);
+ vmballoon_free_page(page, false);
+ STATS_INC(b->stats.free[false]);
/* update balloon size */
b->size--;
}
static int vmballoon_unlock_batched_page(struct vmballoon *b,
- unsigned int num_pages, unsigned int *target)
+ unsigned int num_pages, bool is_2m_pages,
+ unsigned int *target)
{
int locked, i, ret = 0;
bool hv_success;
+ u16 size_per_page = vmballoon_page_size(is_2m_pages);
- hv_success = vmballoon_send_batched_unlock(b, num_pages, target);
+ hv_success = vmballoon_send_batched_unlock(b, num_pages, is_2m_pages,
+ target);
if (!hv_success)
ret = -EIO;
for (i = 0; i < num_pages; i++) {
u64 pa = vmballoon_batch_get_pa(b->batch_page, i);
struct page *p = pfn_to_page(pa >> PAGE_SHIFT);
+ struct vmballoon_page_size *page_size =
+ &b->page_sizes[is_2m_pages];
locked = vmballoon_batch_get_status(b->batch_page, i);
if (!hv_success || locked != VMW_BALLOON_SUCCESS) {
* hypervisor, re-add it to the list of pages owned by
* the balloon driver.
*/
- list_add(&p->lru, &b->pages);
+ list_add(&p->lru, &page_size->pages);
} else {
/* deallocate page */
- __free_page(p);
- STATS_INC(b->stats.free);
+ vmballoon_free_page(p, is_2m_pages);
+ STATS_INC(b->stats.free[is_2m_pages]);
/* update balloon size */
- b->size--;
+ b->size -= size_per_page;
}
}
* Release pages that were allocated while attempting to inflate the
* balloon but were refused by the host for one reason or another.
*/
-static void vmballoon_release_refused_pages(struct vmballoon *b)
+static void vmballoon_release_refused_pages(struct vmballoon *b,
+ bool is_2m_pages)
{
struct page *page, *next;
+ struct vmballoon_page_size *page_size =
+ &b->page_sizes[is_2m_pages];
- list_for_each_entry_safe(page, next, &b->refused_pages, lru) {
+ list_for_each_entry_safe(page, next, &page_size->refused_pages, lru) {
list_del(&page->lru);
- __free_page(page);
- STATS_INC(b->stats.refused_free);
+ vmballoon_free_page(page, is_2m_pages);
+ STATS_INC(b->stats.refused_free[is_2m_pages]);
}
- b->n_refused_pages = 0;
+ page_size->n_refused_pages = 0;
}
static void vmballoon_add_page(struct vmballoon *b, int idx, struct page *p)
*/
static void vmballoon_inflate(struct vmballoon *b)
{
- unsigned int rate;
+ unsigned rate;
unsigned int allocations = 0;
unsigned int num_pages = 0;
int error = 0;
gfp_t flags = VMW_PAGE_ALLOC_NOSLEEP;
+ bool is_2m_pages;
pr_debug("%s - size: %d, target %d\n", __func__, b->size, b->target);
* Start with no sleep allocation rate which may be higher
* than sleeping allocation rate.
*/
- rate = b->slow_allocation_cycles ?
- b->rate_alloc : VMW_BALLOON_NOSLEEP_ALLOC_MAX;
+ if (b->slow_allocation_cycles) {
+ rate = b->rate_alloc;
+ is_2m_pages = false;
+ } else {
+ rate = UINT_MAX;
+ is_2m_pages =
+ b->supported_page_sizes == VMW_BALLOON_NUM_PAGE_SIZES;
+ }
- pr_debug("%s - goal: %d, no-sleep rate: %d, sleep rate: %d\n",
+ pr_debug("%s - goal: %d, no-sleep rate: %u, sleep rate: %d\n",
__func__, b->target - b->size, rate, b->rate_alloc);
- while (b->size < b->target && num_pages < b->target - b->size) {
+ while (!b->reset_required &&
+ b->size + num_pages * vmballoon_page_size(is_2m_pages)
+ < b->target) {
struct page *page;
if (flags == VMW_PAGE_ALLOC_NOSLEEP)
- STATS_INC(b->stats.alloc);
+ STATS_INC(b->stats.alloc[is_2m_pages]);
else
STATS_INC(b->stats.sleep_alloc);
- page = alloc_page(flags);
+ page = vmballoon_alloc_page(flags, is_2m_pages);
if (!page) {
+ STATS_INC(b->stats.alloc_fail[is_2m_pages]);
+
+ if (is_2m_pages) {
+ b->ops->lock(b, num_pages, true, &b->target);
+
+ /*
+ * ignore errors from locking as we now switch
+ * to 4k pages and we might get different
+ * errors.
+ */
+
+ num_pages = 0;
+ is_2m_pages = false;
+ continue;
+ }
+
if (flags == VMW_PAGE_ALLOC_CANSLEEP) {
/*
* CANSLEEP page allocation failed, so guest
STATS_INC(b->stats.sleep_alloc_fail);
break;
}
- STATS_INC(b->stats.alloc_fail);
/*
* NOSLEEP page allocation failed, so the guest is
b->ops->add_page(b, num_pages++, page);
if (num_pages == b->batch_max_pages) {
- error = b->ops->lock(b, num_pages, &b->target);
+ error = b->ops->lock(b, num_pages, is_2m_pages,
+ &b->target);
num_pages = 0;
if (error)
break;
}
- if (++allocations > VMW_BALLOON_YIELD_THRESHOLD) {
- cond_resched();
- allocations = 0;
- }
+ cond_resched();
if (allocations >= rate) {
/* We allocated enough pages, let's take a break. */
}
if (num_pages > 0)
- b->ops->lock(b, num_pages, &b->target);
+ b->ops->lock(b, num_pages, is_2m_pages, &b->target);
/*
* We reached our goal without failures so try increasing
VMW_BALLOON_RATE_ALLOC_MAX);
}
- vmballoon_release_refused_pages(b);
+ vmballoon_release_refused_pages(b, true);
+ vmballoon_release_refused_pages(b, false);
}
/*
*/
static void vmballoon_deflate(struct vmballoon *b)
{
- struct page *page, *next;
- unsigned int i = 0;
- unsigned int num_pages = 0;
- int error;
+ unsigned is_2m_pages;
- pr_debug("%s - size: %d, target %d, rate: %d\n", __func__, b->size,
- b->target, b->rate_free);
+ pr_debug("%s - size: %d, target %d\n", __func__, b->size, b->target);
/* free pages to reach target */
- list_for_each_entry_safe(page, next, &b->pages, lru) {
- list_del(&page->lru);
- b->ops->add_page(b, num_pages++, page);
+ for (is_2m_pages = 0; is_2m_pages < b->supported_page_sizes;
+ is_2m_pages++) {
+ struct page *page, *next;
+ unsigned int num_pages = 0;
+ struct vmballoon_page_size *page_size =
+ &b->page_sizes[is_2m_pages];
+
+ list_for_each_entry_safe(page, next, &page_size->pages, lru) {
+ if (b->reset_required ||
+ (b->target > 0 &&
+ b->size - num_pages
+ * vmballoon_page_size(is_2m_pages)
+ < b->target + vmballoon_page_size(true)))
+ break;
- if (num_pages == b->batch_max_pages) {
- error = b->ops->unlock(b, num_pages, &b->target);
- num_pages = 0;
- if (error) {
- /* quickly decrease rate in case of error */
- b->rate_free = max(b->rate_free / 2,
- VMW_BALLOON_RATE_FREE_MIN);
- return;
+ list_del(&page->lru);
+ b->ops->add_page(b, num_pages++, page);
+
+ if (num_pages == b->batch_max_pages) {
+ int error;
+
+ error = b->ops->unlock(b, num_pages,
+ is_2m_pages, &b->target);
+ num_pages = 0;
+ if (error)
+ return;
}
+
+ cond_resched();
}
- if (++i >= b->size - b->target)
- break;
+ if (num_pages > 0)
+ b->ops->unlock(b, num_pages, is_2m_pages, &b->target);
}
-
- if (num_pages > 0)
- b->ops->unlock(b, num_pages, &b->target);
-
- /* slowly increase rate if there were no errors */
- if (error == 0)
- b->rate_free = min(b->rate_free + VMW_BALLOON_RATE_FREE_INC,
- VMW_BALLOON_RATE_FREE_MAX);
}
static const struct vmballoon_ops vmballoon_basic_ops = {
return true;
}
+/*
+ * Receive notification and resize balloon
+ */
+static void vmballoon_doorbell(void *client_data)
+{
+ struct vmballoon *b = client_data;
+
+ STATS_INC(b->stats.doorbell);
+
+ mod_delayed_work(system_freezable_wq, &b->dwork, 0);
+}
+
+/*
+ * Clean up vmci doorbell
+ */
+static void vmballoon_vmci_cleanup(struct vmballoon *b)
+{
+ int error;
+
+ VMWARE_BALLOON_CMD(VMCI_DOORBELL_SET, VMCI_INVALID_ID,
+ VMCI_INVALID_ID, error);
+ STATS_INC(b->stats.doorbell_unset);
+
+ if (!vmci_handle_is_invalid(b->vmci_doorbell)) {
+ vmci_doorbell_destroy(b->vmci_doorbell);
+ b->vmci_doorbell = VMCI_INVALID_HANDLE;
+ }
+}
+
+/*
+ * Initialize vmci doorbell, to get notified as soon as balloon changes
+ */
+static int vmballoon_vmci_init(struct vmballoon *b)
+{
+ int error = 0;
+
+ if ((b->capabilities & VMW_BALLOON_SIGNALLED_WAKEUP_CMD) != 0) {
+ error = vmci_doorbell_create(&b->vmci_doorbell,
+ VMCI_FLAG_DELAYED_CB,
+ VMCI_PRIVILEGE_FLAG_RESTRICTED,
+ vmballoon_doorbell, b);
+
+ if (error == VMCI_SUCCESS) {
+ VMWARE_BALLOON_CMD(VMCI_DOORBELL_SET,
+ b->vmci_doorbell.context,
+ b->vmci_doorbell.resource, error);
+ STATS_INC(b->stats.doorbell_set);
+ }
+ }
+
+ if (error != 0) {
+ vmballoon_vmci_cleanup(b);
+
+ return -EIO;
+ }
+
+ return 0;
+}
+
/*
* Perform standard reset sequence by popping the balloon (in case it
* is not empty) and then restarting protocol. This operation normally
*/
static void vmballoon_reset(struct vmballoon *b)
{
+ int error;
+
+ vmballoon_vmci_cleanup(b);
+
/* free all pages, skipping monitor unlock */
vmballoon_pop(b);
}
b->reset_required = false;
+
+ error = vmballoon_vmci_init(b);
+ if (error)
+ pr_err("failed to initialize vmci doorbell\n");
+
if (!vmballoon_send_guest_id(b))
pr_err("failed to send guest ID to the host\n");
}
if (b->slow_allocation_cycles > 0)
b->slow_allocation_cycles--;
- if (vmballoon_send_get_target(b, &target)) {
+ if (!b->reset_required && vmballoon_send_get_target(b, &target)) {
/* update target, adjust size */
b->target = target;
if (b->size < target)
vmballoon_inflate(b);
- else if (b->size > target)
+ else if (target == 0 ||
+ b->size > target + vmballoon_page_size(true))
vmballoon_deflate(b);
}
struct vmballoon *b = f->private;
struct vmballoon_stats *stats = &b->stats;
+ /* format capabilities info */
+ seq_printf(f,
+ "balloon capabilities: %#4x\n"
+ "used capabilities: %#4lx\n"
+ "is resetting: %c\n",
+ VMW_BALLOON_CAPABILITIES, b->capabilities,
+ b->reset_required ? 'y' : 'n');
+
/* format size info */
seq_printf(f,
"target: %8d pages\n"
/* format rate info */
seq_printf(f,
- "rateNoSleepAlloc: %8d pages/sec\n"
- "rateSleepAlloc: %8d pages/sec\n"
- "rateFree: %8d pages/sec\n",
- VMW_BALLOON_NOSLEEP_ALLOC_MAX,
- b->rate_alloc, b->rate_free);
+ "rateSleepAlloc: %8d pages/sec\n",
+ b->rate_alloc);
seq_printf(f,
"\n"
"timer: %8u\n"
+ "doorbell: %8u\n"
"start: %8u (%4u failed)\n"
"guestType: %8u (%4u failed)\n"
+ "2m-lock: %8u (%4u failed)\n"
"lock: %8u (%4u failed)\n"
+ "2m-unlock: %8u (%4u failed)\n"
"unlock: %8u (%4u failed)\n"
"target: %8u (%4u failed)\n"
+ "prim2mAlloc: %8u (%4u failed)\n"
"primNoSleepAlloc: %8u (%4u failed)\n"
"primCanSleepAlloc: %8u (%4u failed)\n"
+ "prim2mFree: %8u\n"
"primFree: %8u\n"
+ "err2mAlloc: %8u\n"
"errAlloc: %8u\n"
- "errFree: %8u\n",
+ "err2mFree: %8u\n"
+ "errFree: %8u\n"
+ "doorbellSet: %8u\n"
+ "doorbellUnset: %8u\n",
stats->timer,
+ stats->doorbell,
stats->start, stats->start_fail,
stats->guest_type, stats->guest_type_fail,
- stats->lock, stats->lock_fail,
- stats->unlock, stats->unlock_fail,
+ stats->lock[true], stats->lock_fail[true],
+ stats->lock[false], stats->lock_fail[false],
+ stats->unlock[true], stats->unlock_fail[true],
+ stats->unlock[false], stats->unlock_fail[false],
stats->target, stats->target_fail,
- stats->alloc, stats->alloc_fail,
+ stats->alloc[true], stats->alloc_fail[true],
+ stats->alloc[false], stats->alloc_fail[false],
stats->sleep_alloc, stats->sleep_alloc_fail,
- stats->free,
- stats->refused_alloc, stats->refused_free);
+ stats->free[true],
+ stats->free[false],
+ stats->refused_alloc[true], stats->refused_alloc[false],
+ stats->refused_free[true], stats->refused_free[false],
+ stats->doorbell_set, stats->doorbell_unset);
return 0;
}
static int __init vmballoon_init(void)
{
int error;
-
+ unsigned is_2m_pages;
/*
* Check if we are running on VMware's hypervisor and bail out
* if we are not.
if (x86_hyper != &x86_hyper_vmware)
return -ENODEV;
- INIT_LIST_HEAD(&balloon.pages);
- INIT_LIST_HEAD(&balloon.refused_pages);
+ for (is_2m_pages = 0; is_2m_pages < VMW_BALLOON_NUM_PAGE_SIZES;
+ is_2m_pages++) {
+ INIT_LIST_HEAD(&balloon.page_sizes[is_2m_pages].pages);
+ INIT_LIST_HEAD(&balloon.page_sizes[is_2m_pages].refused_pages);
+ }
/* initialize rates */
balloon.rate_alloc = VMW_BALLOON_RATE_ALLOC_MAX;
- balloon.rate_free = VMW_BALLOON_RATE_FREE_MAX;
INIT_DELAYED_WORK(&balloon.dwork, vmballoon_work);
- /*
- * Start balloon.
- */
- if (!vmballoon_send_start(&balloon, VMW_BALLOON_CAPABILITIES)) {
- pr_err("failed to send start command to the host\n");
- return -EIO;
- }
-
- if ((balloon.capabilities & VMW_BALLOON_BATCHED_CMDS) != 0) {
- balloon.ops = &vmballoon_batched_ops;
- balloon.batch_max_pages = VMW_BALLOON_BATCH_MAX_PAGES;
- if (!vmballoon_init_batching(&balloon)) {
- pr_err("failed to init batching\n");
- return -EIO;
- }
- } else if ((balloon.capabilities & VMW_BALLOON_BASIC_CMDS) != 0) {
- balloon.ops = &vmballoon_basic_ops;
- balloon.batch_max_pages = 1;
- }
-
- if (!vmballoon_send_guest_id(&balloon)) {
- pr_err("failed to send guest ID to the host\n");
- return -EIO;
- }
-
error = vmballoon_debugfs_init(&balloon);
if (error)
return error;
+ balloon.vmci_doorbell = VMCI_INVALID_HANDLE;
+ balloon.batch_page = NULL;
+ balloon.page = NULL;
+ balloon.reset_required = true;
+
queue_delayed_work(system_freezable_wq, &balloon.dwork, 0);
return 0;
static void __exit vmballoon_exit(void)
{
+ vmballoon_vmci_cleanup(&balloon);
cancel_delayed_work_sync(&balloon.dwork);
vmballoon_debugfs_exit(&balloon);
* Reset connection before deallocating memory to avoid potential for
* additional spurious resets from guest touching deallocated pages.
*/
- vmballoon_send_start(&balloon, VMW_BALLOON_CAPABILITIES);
+ vmballoon_send_start(&balloon, 0);
vmballoon_pop(&balloon);
}
module_exit(vmballoon_exit);
projects / karo-tx-linux.git / blobdiff