VMware balloon: add batching to the vmw_balloon.
authorXavier Deguillard <xdeguillard@vmware.com>
Thu, 6 Aug 2015 22:17:58 +0000 (15:17 -0700)
committerGreg Kroah-Hartman <gregkh@linuxfoundation.org>
Sun, 4 Oct 2015 11:59:11 +0000 (12:59 +0100)
Introduce a new capability to the driver that allow sending 512 pages in
one hypervisor call. This reduce the cost of the driver when reclaiming
memory.

Signed-off-by: Xavier Deguillard <xdeguillard@vmware.com>
Acked-by: Dmitry Torokhov <dtor@vmware.com>
Signed-off-by: Philip P. Moltmann <moltmann@vmware.com>
Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
drivers/misc/vmw_balloon.c

index ffb56340d0c7c28bf2daaf0f512ae835fa659f38..64f275e330700bd09e37a3c64bc71679d56b911b 100644 (file)
@@ -1,7 +1,7 @@
 /*
  * VMware Balloon driver.
  *
- * Copyright (C) 2000-2010, VMware, Inc. All Rights Reserved.
+ * Copyright (C) 2000-2013, 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
@@ -37,6 +37,7 @@
 #include <linux/types.h>
 #include <linux/kernel.h>
 #include <linux/mm.h>
+#include <linux/vmalloc.h>
 #include <linux/sched.h>
 #include <linux/module.h>
 #include <linux/workqueue.h>
@@ -46,7 +47,7 @@
 
 MODULE_AUTHOR("VMware, Inc.");
 MODULE_DESCRIPTION("VMware Memory Control (Balloon) Driver");
-MODULE_VERSION("1.3.0.0-k");
+MODULE_VERSION("1.3.1.0-k");
 MODULE_ALIAS("dmi:*:svnVMware*:*");
 MODULE_ALIAS("vmware_vmmemctl");
 MODULE_LICENSE("GPL");
@@ -120,13 +121,26 @@ enum vmwballoon_capabilities {
        VMW_BALLOON_BATCHED_CMDS        = (1 << 2)
 };
 
-#define VMW_BALLOON_CAPABILITIES       (VMW_BALLOON_BASIC_CMDS)
+#define VMW_BALLOON_CAPABILITIES       (VMW_BALLOON_BASIC_CMDS \
+                                       | VMW_BALLOON_BATCHED_CMDS)
 
+/*
+ * Backdoor commands availability:
+ *
+ * START, GET_TARGET and GUEST_ID are always available,
+ *
+ * VMW_BALLOON_BASIC_CMDS:
+ *     LOCK and UNLOCK commands,
+ * VMW_BALLOON_BATCHED_CMDS:
+ *     BATCHED_LOCK and BATCHED_UNLOCK commands.
+ */
 #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
 
 /* error codes */
 #define VMW_BALLOON_SUCCESS                    0
@@ -142,18 +156,63 @@ enum vmwballoon_capabilities {
 
 #define VMW_BALLOON_SUCCESS_WITH_CAPABILITIES  (0x03000000)
 
-#define VMWARE_BALLOON_CMD(cmd, data, result)                  \
+/* Batch page description */
+
+/*
+ * Layout of a page in the batch page:
+ *
+ * +-------------+----------+--------+
+ * |             |          |        |
+ * | Page number | Reserved | Status |
+ * |             |          |        |
+ * +-------------+----------+--------+
+ * 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))
+#define VMW_BALLOON_BATCH_STATUS_MASK  ((1UL << 5) - 1)
+#define VMW_BALLOON_BATCH_PAGE_MASK    (~((1UL << PAGE_SHIFT) - 1))
+
+struct vmballoon_batch_page {
+       u64 pages[VMW_BALLOON_BATCH_MAX_PAGES];
+};
+
+static u64 vmballoon_batch_get_pa(struct vmballoon_batch_page *batch, int idx)
+{
+       return batch->pages[idx] & VMW_BALLOON_BATCH_PAGE_MASK;
+}
+
+static int vmballoon_batch_get_status(struct vmballoon_batch_page *batch,
+                               int idx)
+{
+       return (int)(batch->pages[idx] & VMW_BALLOON_BATCH_STATUS_MASK);
+}
+
+static void vmballoon_batch_set_pa(struct vmballoon_batch_page *batch, int idx,
+                               u64 pa)
+{
+       batch->pages[idx] = pa;
+}
+
+
+#define VMWARE_BALLOON_CMD(cmd, arg1, arg2, result)            \
 ({                                                             \
-       unsigned long __status, __dummy1, __dummy2;             \
+       unsigned long __status, __dummy1, __dummy2, __dummy3;   \
        __asm__ __volatile__ ("inl %%dx" :                      \
                "=a"(__status),                                 \
                "=c"(__dummy1),                                 \
                "=d"(__dummy2),                                 \
-               "=b"(result) :                                  \
+               "=b"(result),                                   \
+               "=S" (__dummy3) :                               \
                "0"(VMW_BALLOON_HV_MAGIC),                      \
                "1"(VMW_BALLOON_CMD_##cmd),                     \
                "2"(VMW_BALLOON_HV_PORT),                       \
-               "3"(data) :                                     \
+               "3"(arg1),                                      \
+               "4" (arg2) :                                    \
                "memory");                                      \
        if (VMW_BALLOON_CMD_##cmd == VMW_BALLOON_CMD_START)     \
                result = __dummy1;                              \
@@ -192,6 +251,14 @@ struct vmballoon_stats {
 #define STATS_INC(stat)
 #endif
 
+struct vmballoon;
+
+struct vmballoon_ops {
+       void (*add_page)(struct vmballoon *b, int idx, struct page *p);
+       int (*lock)(struct vmballoon *b, unsigned int num_pages);
+       int (*unlock)(struct vmballoon *b, unsigned int num_pages);
+};
+
 struct vmballoon {
 
        /* list of reserved physical pages */
@@ -215,6 +282,14 @@ struct vmballoon {
        /* slowdown page allocations for next few cycles */
        unsigned int slow_allocation_cycles;
 
+       unsigned long capabilities;
+
+       struct vmballoon_batch_page *batch_page;
+       unsigned int batch_max_pages;
+       struct page *page;
+
+       const struct vmballoon_ops *ops;
+
 #ifdef CONFIG_DEBUG_FS
        /* statistics */
        struct vmballoon_stats stats;
@@ -234,16 +309,22 @@ static struct vmballoon balloon;
  * Send "start" command to the host, communicating supported version
  * of the protocol.
  */
-static bool vmballoon_send_start(struct vmballoon *b)
+static bool vmballoon_send_start(struct vmballoon *b, unsigned long req_caps)
 {
-       unsigned long status, capabilities;
+       unsigned long status, capabilities, dummy = 0;
 
        STATS_INC(b->stats.start);
 
-       status = VMWARE_BALLOON_CMD(START, VMW_BALLOON_CAPABILITIES,
-                               capabilities);
-       if (status == VMW_BALLOON_SUCCESS)
+       status = VMWARE_BALLOON_CMD(START, req_caps, dummy, capabilities);
+
+       switch (status) {
+       case VMW_BALLOON_SUCCESS_WITH_CAPABILITIES:
+               b->capabilities = capabilities;
+               return true;
+       case VMW_BALLOON_SUCCESS:
+               b->capabilities = VMW_BALLOON_BASIC_CMDS;
                return true;
+       }
 
        pr_debug("%s - failed, hv returns %ld\n", __func__, status);
        STATS_INC(b->stats.start_fail);
@@ -273,9 +354,10 @@ static bool vmballoon_check_status(struct vmballoon *b, unsigned long status)
  */
 static bool vmballoon_send_guest_id(struct vmballoon *b)
 {
-       unsigned long status, dummy;
+       unsigned long status, dummy = 0;
 
-       status = VMWARE_BALLOON_CMD(GUEST_ID, VMW_BALLOON_GUEST_ID, dummy);
+       status = VMWARE_BALLOON_CMD(GUEST_ID, VMW_BALLOON_GUEST_ID, dummy,
+                               dummy);
 
        STATS_INC(b->stats.guest_type);
 
@@ -295,6 +377,7 @@ static bool vmballoon_send_get_target(struct vmballoon *b, u32 *new_target)
        unsigned long status;
        unsigned long target;
        unsigned long limit;
+       unsigned long dummy = 0;
        u32 limit32;
 
        /*
@@ -313,7 +396,7 @@ static bool vmballoon_send_get_target(struct vmballoon *b, u32 *new_target)
        /* update stats */
        STATS_INC(b->stats.target);
 
-       status = VMWARE_BALLOON_CMD(GET_TARGET, limit, target);
+       status = VMWARE_BALLOON_CMD(GET_TARGET, limit, dummy, target);
        if (vmballoon_check_status(b, status)) {
                *new_target = target;
                return true;
@@ -332,7 +415,7 @@ static bool vmballoon_send_get_target(struct vmballoon *b, u32 *new_target)
 static int vmballoon_send_lock_page(struct vmballoon *b, unsigned long pfn,
                                     unsigned int *hv_status)
 {
-       unsigned long status, dummy;
+       unsigned long status, dummy = 0;
        u32 pfn32;
 
        pfn32 = (u32)pfn;
@@ -341,7 +424,7 @@ static int vmballoon_send_lock_page(struct vmballoon *b, unsigned long pfn,
 
        STATS_INC(b->stats.lock);
 
-       *hv_status = status = VMWARE_BALLOON_CMD(LOCK, pfn, dummy);
+       *hv_status = status = VMWARE_BALLOON_CMD(LOCK, pfn, dummy, dummy);
        if (vmballoon_check_status(b, status))
                return 0;
 
@@ -350,13 +433,30 @@ static int vmballoon_send_lock_page(struct vmballoon *b, unsigned long pfn,
        return 1;
 }
 
+static int vmballoon_send_batched_lock(struct vmballoon *b,
+                                       unsigned int num_pages)
+{
+       unsigned long status, dummy;
+       unsigned long pfn = page_to_pfn(b->page);
+
+       STATS_INC(b->stats.lock);
+
+       status = VMWARE_BALLOON_CMD(BATCHED_LOCK, pfn, num_pages, dummy);
+       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);
+       return 1;
+}
+
 /*
  * Notify the host that guest intends to release given page back into
  * the pool of available (to the guest) pages.
  */
 static bool vmballoon_send_unlock_page(struct vmballoon *b, unsigned long pfn)
 {
-       unsigned long status, dummy;
+       unsigned long status, dummy = 0;
        u32 pfn32;
 
        pfn32 = (u32)pfn;
@@ -365,7 +465,7 @@ static bool vmballoon_send_unlock_page(struct vmballoon *b, unsigned long pfn)
 
        STATS_INC(b->stats.unlock);
 
-       status = VMWARE_BALLOON_CMD(UNLOCK, pfn, dummy);
+       status = VMWARE_BALLOON_CMD(UNLOCK, pfn, dummy, dummy);
        if (vmballoon_check_status(b, status))
                return true;
 
@@ -374,6 +474,23 @@ static bool vmballoon_send_unlock_page(struct vmballoon *b, unsigned long pfn)
        return false;
 }
 
+static bool vmballoon_send_batched_unlock(struct vmballoon *b,
+                                               unsigned int num_pages)
+{
+       unsigned long status, dummy;
+       unsigned long pfn = page_to_pfn(b->page);
+
+       STATS_INC(b->stats.unlock);
+
+       status = VMWARE_BALLOON_CMD(BATCHED_UNLOCK, pfn, num_pages, dummy);
+       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);
+       return false;
+}
+
 /*
  * Quickly release all pages allocated for the balloon. This function is
  * called when host decides to "reset" balloon for one reason or another.
@@ -396,22 +513,13 @@ static void vmballoon_pop(struct vmballoon *b)
                        cond_resched();
                }
        }
-}
 
-/*
- * Perform standard reset sequence by popping the balloon (in case it
- * is not  empty) and then restarting protocol. This operation normally
- * happens when host responds with VMW_BALLOON_ERROR_RESET to a command.
- */
-static void vmballoon_reset(struct vmballoon *b)
-{
-       /* free all pages, skipping monitor unlock */
-       vmballoon_pop(b);
+       if ((b->capabilities & VMW_BALLOON_BATCHED_CMDS) != 0) {
+               if (b->batch_page)
+                       vunmap(b->batch_page);
 
-       if (vmballoon_send_start(b)) {
-               b->reset_required = false;
-               if (!vmballoon_send_guest_id(b))
-                       pr_err("failed to send guest ID to the host\n");
+               if (b->page)
+                       __free_page(b->page);
        }
 }
 
@@ -420,9 +528,10 @@ static void vmballoon_reset(struct vmballoon *b)
  * refuse list, those refused page are then released at the end of the
  * inflation cycle.
  */
-static int vmballoon_lock_page(struct vmballoon *b, struct page *page)
+static int vmballoon_lock_page(struct vmballoon *b, unsigned int num_pages)
 {
        int locked, hv_status;
+       struct page *page = b->page;
 
        locked = vmballoon_send_lock_page(b, page_to_pfn(page), &hv_status);
        if (locked > 0) {
@@ -457,17 +566,68 @@ static int vmballoon_lock_page(struct vmballoon *b, struct page *page)
        return 0;
 }
 
+static int vmballoon_lock_batched_page(struct vmballoon *b,
+                               unsigned int num_pages)
+{
+       int locked, i;
+
+       locked = vmballoon_send_batched_lock(b, num_pages);
+       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);
+               }
+
+               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);
+
+               locked = vmballoon_batch_get_status(b->batch_page, i);
+
+               switch (locked) {
+               case VMW_BALLOON_SUCCESS:
+                       list_add(&p->lru, &b->pages);
+                       b->size++;
+                       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++;
+                               break;
+                       }
+                       /* Fallthrough */
+               case VMW_BALLOON_ERROR_RESET:
+               case VMW_BALLOON_ERROR_PPN_NOTNEEDED:
+                       __free_page(p);
+                       break;
+               default:
+                       /* This should never happen */
+                       WARN_ON_ONCE(true);
+               }
+       }
+
+       return 0;
+}
+
 /*
  * Release the page allocated for the balloon. Note that we first notify
  * the host so it can make sure the page will be available for the guest
  * to use, if needed.
  */
-static int vmballoon_release_page(struct vmballoon *b, struct page *page)
+static int vmballoon_unlock_page(struct vmballoon *b, unsigned int num_pages)
 {
-       if (!vmballoon_send_unlock_page(b, page_to_pfn(page)))
-               return -EIO;
+       struct page *page = b->page;
 
-       list_del(&page->lru);
+       if (!vmballoon_send_unlock_page(b, page_to_pfn(page))) {
+               list_add(&page->lru, &b->pages);
+               return -EIO;
+       }
 
        /* deallocate page */
        __free_page(page);
@@ -479,6 +639,41 @@ static int vmballoon_release_page(struct vmballoon *b, struct page *page)
        return 0;
 }
 
+static int vmballoon_unlock_batched_page(struct vmballoon *b,
+                                       unsigned int num_pages)
+{
+       int locked, i, ret = 0;
+       bool hv_success;
+
+       hv_success = vmballoon_send_batched_unlock(b, num_pages);
+       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);
+
+               locked = vmballoon_batch_get_status(b->batch_page, i);
+               if (!hv_success || locked != VMW_BALLOON_SUCCESS) {
+                       /*
+                        * That page wasn't successfully unlocked by the
+                        * hypervisor, re-add it to the list of pages owned by
+                        * the balloon driver.
+                        */
+                       list_add(&p->lru, &b->pages);
+               } else {
+                       /* deallocate page */
+                       __free_page(p);
+                       STATS_INC(b->stats.free);
+
+                       /* update balloon size */
+                       b->size--;
+               }
+       }
+
+       return ret;
+}
+
 /*
  * Release pages that were allocated while attempting to inflate the
  * balloon but were refused by the host for one reason or another.
@@ -496,6 +691,18 @@ static void vmballoon_release_refused_pages(struct vmballoon *b)
        b->n_refused_pages = 0;
 }
 
+static void vmballoon_add_page(struct vmballoon *b, int idx, struct page *p)
+{
+       b->page = p;
+}
+
+static void vmballoon_add_batched_page(struct vmballoon *b, int idx,
+                               struct page *p)
+{
+       vmballoon_batch_set_pa(b->batch_page, idx,
+                       (u64)page_to_pfn(p) << PAGE_SHIFT);
+}
+
 /*
  * Inflate the balloon towards its target size. Note that we try to limit
  * the rate of allocation to make sure we are not choking the rest of the
@@ -507,6 +714,7 @@ static void vmballoon_inflate(struct vmballoon *b)
        unsigned int rate;
        unsigned int i;
        unsigned int allocations = 0;
+       unsigned int num_pages = 0;
        int error = 0;
        gfp_t flags = VMW_PAGE_ALLOC_NOSLEEP;
 
@@ -539,14 +747,13 @@ static void vmballoon_inflate(struct vmballoon *b)
                 __func__, goal, rate, b->rate_alloc);
 
        for (i = 0; i < goal; i++) {
-               struct page *page;
+               struct page *page = alloc_page(flags);
 
                if (flags == VMW_PAGE_ALLOC_NOSLEEP)
                        STATS_INC(b->stats.alloc);
                else
                        STATS_INC(b->stats.sleep_alloc);
 
-               page = alloc_page(flags);
                if (!page) {
                        if (flags == VMW_PAGE_ALLOC_CANSLEEP) {
                                /*
@@ -580,9 +787,13 @@ static void vmballoon_inflate(struct vmballoon *b)
                        continue;
                }
 
-               error = vmballoon_lock_page(b, page);
-               if (error)
-                       break;
+               b->ops->add_page(b, num_pages++, page);
+               if (num_pages == b->batch_max_pages) {
+                       error = b->ops->lock(b, num_pages);
+                       num_pages = 0;
+                       if (error)
+                               break;
+               }
 
                if (++allocations > VMW_BALLOON_YIELD_THRESHOLD) {
                        cond_resched();
@@ -595,6 +806,9 @@ static void vmballoon_inflate(struct vmballoon *b)
                }
        }
 
+       if (num_pages > 0)
+               b->ops->lock(b, num_pages);
+
        /*
         * We reached our goal without failures so try increasing
         * allocation rate.
@@ -618,6 +832,7 @@ static void vmballoon_deflate(struct vmballoon *b)
        struct page *page, *next;
        unsigned int i = 0;
        unsigned int goal;
+       unsigned int num_pages = 0;
        int error;
 
        pr_debug("%s - size: %d, target %d\n", __func__, b->size, b->target);
@@ -629,21 +844,94 @@ static void vmballoon_deflate(struct vmballoon *b)
 
        /* free pages to reach target */
        list_for_each_entry_safe(page, next, &b->pages, lru) {
-               error = vmballoon_release_page(b, page);
-               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) {
+                       error = b->ops->unlock(b, num_pages);
+                       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;
+                       }
                }
 
                if (++i >= goal)
                        break;
        }
 
+       if (num_pages > 0)
+               b->ops->unlock(b, num_pages);
+
        /* slowly increase rate if there were no errors */
-       b->rate_free = min(b->rate_free + VMW_BALLOON_RATE_FREE_INC,
-                          VMW_BALLOON_RATE_FREE_MAX);
+       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 = {
+       .add_page = vmballoon_add_page,
+       .lock = vmballoon_lock_page,
+       .unlock = vmballoon_unlock_page
+};
+
+static const struct vmballoon_ops vmballoon_batched_ops = {
+       .add_page = vmballoon_add_batched_page,
+       .lock = vmballoon_lock_batched_page,
+       .unlock = vmballoon_unlock_batched_page
+};
+
+static bool vmballoon_init_batching(struct vmballoon *b)
+{
+       b->page = alloc_page(VMW_PAGE_ALLOC_NOSLEEP);
+       if (!b->page)
+               return false;
+
+       b->batch_page = vmap(&b->page, 1, VM_MAP, PAGE_KERNEL);
+       if (!b->batch_page) {
+               __free_page(b->page);
+               return false;
+       }
+
+       return true;
+}
+
+/*
+ * Perform standard reset sequence by popping the balloon (in case it
+ * is not  empty) and then restarting protocol. This operation normally
+ * happens when host responds with VMW_BALLOON_ERROR_RESET to a command.
+ */
+static void vmballoon_reset(struct vmballoon *b)
+{
+       /* free all pages, skipping monitor unlock */
+       vmballoon_pop(b);
+
+       if (!vmballoon_send_start(b, VMW_BALLOON_CAPABILITIES))
+               return;
+
+       if ((b->capabilities & VMW_BALLOON_BATCHED_CMDS) != 0) {
+               b->ops = &vmballoon_batched_ops;
+               b->batch_max_pages = VMW_BALLOON_BATCH_MAX_PAGES;
+               if (!vmballoon_init_batching(b)) {
+                       /*
+                        * We failed to initialize batching, inform the monitor
+                        * about it by sending a null capability.
+                        *
+                        * The guest will retry in one second.
+                        */
+                       vmballoon_send_start(b, 0);
+                       return;
+               }
+       } else if ((b->capabilities & VMW_BALLOON_BASIC_CMDS) != 0) {
+               b->ops = &vmballoon_basic_ops;
+               b->batch_max_pages = 1;
+       }
+
+       b->reset_required = false;
+       if (!vmballoon_send_guest_id(b))
+               pr_err("failed to send guest ID to the host\n");
 }
 
 /*
@@ -802,11 +1090,23 @@ static int __init vmballoon_init(void)
        /*
         * Start balloon.
         */
-       if (!vmballoon_send_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;
@@ -833,7 +1133,7 @@ static void __exit vmballoon_exit(void)
         * Reset connection before deallocating memory to avoid potential for
         * additional spurious resets from guest touching deallocated pages.
         */
-       vmballoon_send_start(&balloon);
+       vmballoon_send_start(&balloon, VMW_BALLOON_CAPABILITIES);
        vmballoon_pop(&balloon);
 }
 module_exit(vmballoon_exit);
This page took 0.033289 seconds and 5 git commands to generate.