1 files changed, 180 insertions, 0 deletions

diff --git a/patches/boot_time_opt_guest/0152-x86-kvm-Notify-host-to-release-pages.patch b/patches/boot_time_opt_guest/0152-x86-kvm-Notify-host-to-release-pages.patch
new file mode 100644
index 0000000..ff9d8c0
--- /dev/null
+++ b/patches/boot_time_opt_guest/0152-x86-kvm-Notify-host-to-release-pages.patch
@@ -0,0 +1,180 @@
+From d28921b5f797829e4e676f7968ae688ef96b7992 Mon Sep 17 00:00:00 2001
+From: Sebastien Boeuf <sebastien.boeuf@intel.com>
+Date: Mon, 23 Jan 2017 15:08:55 -0800
+Subject: [PATCH 152/154] x86: kvm: Notify host to release pages
+
+In context of hypervisors managing several virtual machines, we
+want those virtual machines to give the memory they used back to
+the host when they don't need it anymore.
+
+This patch introduces a new hypercall KVM_HC_RETURN_MEM, allowing
+the guest kernel to notify the host kernel when such event occurs.
+And relying on do_madvise() function that we have previously exported,
+it issues a call to this function when it receives the new hypercall.
+
+Use of do_madvise() with MADV_DONTNEED flag will allow the guest to
+ask for a new page without going through a new hypercall. Instead,
+it will be able to start using that memory again as it will get
+faulted back in as a fresh new page. That's why do_madvise() is more
+efficient than doing vm_unmap() to return some memory to the host.
+
+This patch introduces also a new sysctl kvm_madv_instant_free,
+allowing user to set MADV_FREE advice instead of MADV_DONTNEED.
+Indeed, MADV_FREE saves more performances than using MADV_DONTNEED
+because it does not zero the pages in case the memory has not been
+freed by the kernel. This can happen when there was no need for the
+kernel to get this memory back, meaning it was keeping those pages
+in the right state to be re-used by the same application.
+MADV_FREE being a very recent advice introduced in kernel 4.5, we
+only want to enable it through a sysctl in case the user want to
+use it.
+
+Suggested-by: Arjan van de Ven <arjan.van.de.ven@intel.com>
+Signed-off-by: Sebastien Boeuf <sebastien.boeuf@intel.com>
+---
+ arch/x86/kvm/x86.c            | 17 +++++++++++++++++
+ include/linux/mm.h            |  5 +++++
+ include/uapi/linux/kvm_para.h |  3 +++
+ kernel/sysctl.c               |  7 +++++++
+ mm/Makefile                   |  2 +-
+ mm/kvm.c                      | 25 +++++++++++++++++++++++++
+ 6 files changed, 58 insertions(+), 1 deletion(-)
+ create mode 100644 mm/kvm.c
+
+diff --git a/arch/x86/kvm/x86.c b/arch/x86/kvm/x86.c
+index 582c75311f95..683a94dd5f03 100644
+--- a/arch/x86/kvm/x86.c
++++ b/arch/x86/kvm/x86.c
+@@ -46,6 +46,7 @@
+ #include <linux/user-return-notifier.h>
+ #include <linux/srcu.h>
+ #include <linux/slab.h>
++#include <linux/mm.h>
+ #include <linux/perf_event.h>
+ #include <linux/uaccess.h>
+ #include <linux/hash.h>
+@@ -6019,6 +6020,19 @@ static void kvm_pv_kick_cpu_op(struct kvm *kvm, unsigned long flags, int apicid)
+        kvm_irq_delivery_to_apic(kvm, NULL, &lapic_irq, NULL);
+ }
+ 
++static int kvm_pv_return_mem_op(struct kvm *kvm, gpa_t gpa, size_t len)
++{
++       unsigned long start = gfn_to_hva(kvm, gpa_to_gfn(gpa));
++
++       if (len > KVM_MAX_RET_MEM_SIZE)
++               return KVM_EPERM;
++
++       if (kvm_is_error_hva(start + len))
++               return KVM_EFAULT;
++
++       return do_madvise(start, len, kvm_ret_mem_advice);
++}
++
+ void kvm_vcpu_deactivate_apicv(struct kvm_vcpu *vcpu)
+ {
+        vcpu->arch.apicv_active = false;
+@@ -6065,6 +6079,9 @@ int kvm_emulate_hypercall(struct kvm_vcpu *vcpu)
+                kvm_pv_kick_cpu_op(vcpu->kvm, a0, a1);
+                ret = 0;
+                break;
++       case KVM_HC_RETURN_MEM:
++               ret = kvm_pv_return_mem_op(vcpu->kvm, a0, a1);
++               break;
+        default:
+                ret = -KVM_ENOSYS;
+                break;
+diff --git a/include/linux/mm.h b/include/linux/mm.h
+index 925ec25f99a8..833f23d98baa 100644
+--- a/include/linux/mm.h
++++ b/include/linux/mm.h
+@@ -2303,6 +2303,11 @@ extern bool process_shares_mm(struct task_struct *p, struct mm_struct *mm);
+ extern int sysctl_drop_caches;
+ int drop_caches_sysctl_handler(struct ctl_table *, int,
+                                        void __user *, size_t *, loff_t *);
++extern int sysctl_kvm_madv_instant_free;
++extern int kvm_ret_mem_advice;
++int kvm_madv_instant_free_sysctl_handler(struct ctl_table *table, int write,
++                                        void __user *buffer, size_t *length,
++                                        loff_t *ppos);
+ #endif
+ 
+ void drop_slab(void);
+diff --git a/include/uapi/linux/kvm_para.h b/include/uapi/linux/kvm_para.h
+index bf6cd7d5cac2..7d90f77d87d0 100644
+--- a/include/uapi/linux/kvm_para.h
++++ b/include/uapi/linux/kvm_para.h
+@@ -23,6 +23,9 @@
+ #define KVM_HC_MIPS_GET_CLOCK_FREQ     6
+ #define KVM_HC_MIPS_EXIT_VM            7
+ #define KVM_HC_MIPS_CONSOLE_OUTPUT     8
++#define KVM_HC_RETURN_MEM              10
++
++#define KVM_MAX_RET_MEM_SIZE           (1 << 22) // 4MiB
+ 
+ /*
+  * hypercalls use architecture specific
+diff --git a/kernel/sysctl.c b/kernel/sysctl.c
+index c1095cdc0fe2..d8ae774fa042 100644
+--- a/kernel/sysctl.c
++++ b/kernel/sysctl.c
+@@ -1398,6 +1398,13 @@ static struct ctl_table vm_table[] = {
+                .extra1         = &one,
+                .extra2         = &four,
+        },
++       {
++               .procname       = "kvm_madv_instant_free",
++               .data           = &sysctl_kvm_madv_instant_free,
++               .maxlen         = sizeof(int),
++               .mode           = 0644,
++               .proc_handler   = kvm_madv_instant_free_sysctl_handler,
++       },
+ #ifdef CONFIG_COMPACTION
+        {
+                .procname       = "compact_memory",
+diff --git a/mm/Makefile b/mm/Makefile
+index 295bd7a9f76b..651ce0aff140 100644
+--- a/mm/Makefile
++++ b/mm/Makefile
+@@ -37,7 +37,7 @@ obj-y                 := filemap.o mempool.o oom_kill.o \
+                           mm_init.o mmu_context.o percpu.o slab_common.o \
+                           compaction.o vmacache.o \
+                           interval_tree.o list_lru.o workingset.o \
+-                          prfile.o debug.o $(mmu-y)
++                          prfile.o debug.o kvm.o $(mmu-y)
+ 
+ obj-y += init-mm.o
+ 
+diff --git a/mm/kvm.c b/mm/kvm.c
+new file mode 100644
+index 000000000000..8945f6a311b9
+--- /dev/null
++++ b/mm/kvm.c
+@@ -0,0 +1,25 @@
++#include <linux/mman.h>
++
++int sysctl_kvm_madv_instant_free;
++
++int kvm_ret_mem_advice = MADV_DONTNEED;
++EXPORT_SYMBOL_GPL(kvm_ret_mem_advice);
++
++int kvm_madv_instant_free_sysctl_handler(struct ctl_table *table, int write,
++       void __user *buffer, size_t *length, loff_t *ppos)
++{
++       int ret;
++
++       ret = proc_dointvec(table, write, buffer, length, ppos);
++       if (ret)
++               return ret;
++
++#ifdef MADV_FREE
++       if (sysctl_kvm_madv_instant_free > 0)
++               kvm_ret_mem_advice = MADV_FREE;
++       else
++               kvm_ret_mem_advice = MADV_DONTNEED;
++#endif
++
++       return 0;
++}
+-- 
+2.12.1
+