endmenu
source "palacios/src/interfaces/Kconfig"
-
+source "palacios/src/extensions/Kconfig"
config TELEMETRY
bool "Enable VMM telemetry support"
Enable the Vnet in Palacios
-config INSPECTOR
- bool "Enable VM inspection"
- depends on EXPERIMENTAL
- default n
- help
- Enable inspection framework for vm internal state
-
endmenu
libs-y := palacios/lib/$(ARCH)/
devices-y := palacios/src/devices/
interfaces-y := palacios/src/interfaces/
+extensions-y := palacios/src/extensions/
modules-y := modules/
palacios-dirs := $(patsubst %/,%,$(filter %/, \
- $(core-y) $(devices-y) $(interfaces-y) $(libs-y)) $(modules-y))
+ $(core-y) $(devices-y) $(interfaces-y) $(extensions-y) $(libs-y)) $(modules-y))
palacios-cleandirs := $(sort $(palacios-dirs) $(patsubst %/,%,$(filter %/, \
$(core-n) $(core-) $(devices-n) $(devices-) \
- $(interfaces-n) $(interfaces-) $(modules-n) $(modules-))))
+ $(interfaces-n) $(interfaces-) $(extensions-n) $(extensions-) $(modules-n) $(modules-))))
core-y := $(patsubst %/, %/built-in.o, $(core-y))
devices-y := $(patsubst %/, %/built-in.o, $(devices-y))
interfaces-y := $(patsubst %/, %/built-in.o, $(interfaces-y))
+extensions-y := $(patsubst %/, %/built-in.o, $(extensions-y))
libs-y := $(patsubst %/, %/built-in.o, $(libs-y))
modules-y := $(patsubst %/, %/built-in.o, $(modules-y))
#lnxmod-y := $(patsubst %/, %/built-in.o, $(lnxmod-y))
-palacios := $(core-y) $(devices-y) $(interfaces-y) $(libs-y) $(modules-y)
+palacios := $(core-y) $(devices-y) $(interfaces-y) $(extensions-y) $(libs-y) $(modules-y)
# Rule to link palacios - also used during CONFIG_CONFIGKALLSYMS
#include <linux/debugfs.h>
#include <linux/uaccess.h>
-#include <palacios/vmm_inspector.h>
+#include <interfaces/inspector.h>
#include "palacios.h"
#define READ_ONLY 2
#define HOOKED 4
-struct v3_inspector_state {
- struct v3_mtree state_tree;
-
-};
-
-
int v3_init_inspector(struct v3_vm_info * vm);
int v3_init_inspector_core(struct guest_info * core);
struct v3_sym_core_state;
#endif
-#ifdef CONFIG_INSPECTOR
-#include <palacios/vmm_inspector.h>
-#endif
#include <palacios/vmm_config.h>
struct v3_telemetry_state telemetry;
#endif
-#ifdef CONFIG_INSPECTOR
- struct v3_inspector_state inspector;
-#endif
uint64_t yield_cycle_period;
#ifdef __V3VEE__
#include <palacios/vmm.h>
-#include <palacios/vmm_list.h>
#include <palacios/vmm_config.h>
+#include <palacios/vmm_list.h>
struct v3_vm_info;
char * name;
int (*init)(struct v3_vm_info * vm, v3_cfg_tree_t * cfg, void ** priv_data);
int (*deinit)(struct v3_vm_info * vm, void * priv_data);
- int (*core_init)(struct guest_info * core);
- int (*core_deinit)(struct guest_info * core);
- int (*on_entry)(struct guest_info * core);
- int (*on_exit)(struct guest_info * core);
+ int (*core_init)(struct guest_info * core, void * priv_data);
+ int (*core_deinit)(struct guest_info * core, void * priv_data);
+ int (*on_entry)(struct guest_info * core, void * priv_data);
+ int (*on_exit)(struct guest_info * core, void * priv_data);
};
struct v3_extension {
int v3_init_ext_manager(struct v3_vm_info * vm);
int v3_add_extension(struct v3_vm_info * vm, const char * name, v3_cfg_tree_t * cfg);
+int v3_init_core_extensions(struct guest_info * core);
+
+void * v3_get_extension_state(struct v3_vm_info * vm, const char * name);
#define register_extension(ext) \
--- /dev/null
+menu "Extensions"
+
+config EXT_VTIME
+ bool "Enable Time virtualization"
+ default n
+ help
+ Enables the timer virtualization extensions. These hide the cost of
+ running inside the VMM context. This can aid the consistency of
+ time between multiple timers, but can cause the guest to run
+ a good bit slower than the host in VM-intensive parts of the code.
+
+
+config EXT_VTSC
+ bool "Fully virtualize guest TSC"
+ default n
+ depends on EXT_VTIME
+ help
+ Virtualize the processor time stamp counter in the guest,
+ generally increasing consistency between various time sources
+ but also potentially making guest time run slower than real time.
+
+config EXT_MTRRS
+ bool "Support virtualized MTTRs"
+ default n
+ help
+ Provides a virtualized set of MTTR registers
+
+config EXT_MACH_CHECK
+ bool "Support Machine Check functionality"
+ default n
+ help
+ Provides a virtualized machine check architecture
+
+
+config EXT_INSPECTOR
+ bool "VM Inspector"
+ default n
+ help
+ Provides the inspection extension
+
+endmenu
--- /dev/null
+obj-$(CONFIG_EXT_MTRRS) += ext_mtrr.o
+obj-$(CONFIG_EXT_VTSC) += ext_vtsc.o
+obj-$(CONFIG_EXT_VTIME) += ext_vtime.o
+obj-$(CONFIG_EXT_INSPECTOR) += ext_inspector.o
*/
-#include <palacios/vmm_inspector.h>
+//#include <palacios/vmm_inspector.h>
#include <palacios/vmm.h>
#include <palacios/vm_guest.h>
#include <palacios/vmm_sprintf.h>
+#include <palacios/vmm_extensions.h>
+
+#include <palacios/vmm_multitree.h>
+#include <interfaces/inspector.h>
// Note that v3_inspect_node_t is actuall a struct v3_mtree
// Its set as void for opaque portability
+struct v3_inspector_state {
+ struct v3_mtree state_tree;
+
+};
-int v3_init_inspector(struct v3_vm_info * vm) {
- struct v3_inspector_state * state = (struct v3_inspector_state *)&(vm->inspector);
+static int init_inspector(struct v3_vm_info * vm, v3_cfg_tree_t * cfg, void ** priv_data) {
+ struct v3_inspector_state * state = V3_Malloc(sizeof(struct v3_inspector_state));
memset(state, 0, sizeof(struct v3_inspector_state));
strncpy(state->state_tree.name, "vm->name", 50);
state->state_tree.subtree = 1;
+ *priv_data = state;
+
return 0;
}
-int v3_init_inspector_core(struct guest_info * core) {
- struct v3_inspector_state * vm_state = &(core->vm_info->inspector);
+static int init_inspector_core(struct guest_info * core, void * priv_data) {
+ struct v3_inspector_state * vm_state = priv_data;
char core_name[50];
snprintf(core_name, 50, "core.%d", core->cpu_id);
v3_inspect_64(cr_node, "EFER", (uint64_t *)&(core->ctrl_regs.efer));
- // struct v3_mtree * seg_node = v3_mtree_create_subtree(core_node, "SEGMENTS");
+ //struct v3_mtree * seg_node = v3_mtree_create_subtree(core_node, "SEGMENTS");
}
+
+
+
+static struct v3_extension_impl inspector_impl = {
+ .name = "inspector",
+ .init = init_inspector,
+ .deinit = NULL,
+ .core_init = init_inspector_core,
+ .core_deinit = NULL,
+ .on_entry = NULL,
+ .on_exit = NULL
+};
+
+
+register_extension(&inspector_impl);
+
+
v3_inspect_node_t * v3_inspect_add_subtree(v3_inspect_node_t * root, char * name) {
return v3_mtree_create_subtree(root, name);
}
-
-
int v3_find_inspection_value(v3_inspect_node_t * node, char * name,
struct v3_inspection_value * value) {
struct v3_mtree * mt_node = v3_mtree_find_node(node, name);
v3_inspect_node_t * v3_get_inspection_root(struct v3_vm_info * vm) {
- return &(vm->inspector.state_tree);
+ struct v3_inspector_state * inspector = v3_get_extension_state(vm, inspector_impl.name);
+
+ if (inspector == NULL) {
+ return NULL;
+ }
+
+ return &(inspector->state_tree);
}
v3_inspect_node_t * v3_get_inspection_subtree(v3_inspect_node_t * root, char * name) {
v3_inspect_node_t * v3_inspection_first_child(v3_inspect_node_t * root) {
return v3_mtree_first_child(root);
}
+
+
+
+
--- /dev/null
+/*
+ * This file is part of the Palacios Virtual Machine Monitor developed
+ * by the V3VEE Project with funding from the United States National
+ * Science Foundation and the Department of Energy.
+ *
+ * The V3VEE Project is a joint project between Northwestern University
+ * and the University of New Mexico. You can find out more at
+ * http://www.v3vee.org
+ *
+ * Copyright (c) 2008, Jack Lange <jarusl@cs.northwestern.edu>
+ * Copyright (c) 2008, The V3VEE Project <http://www.v3vee.org>
+ * All rights reserved.
+ *
+ * Author: Jack Lange <jarusl@cs.northwestern.edu>
+ * Patrick G. Bridges <bridges@cs.unm.edu>
+ *
+ * This is free software. You are permitted to use,
+ * redistribute, and modify it as specified in the file "V3VEE_LICENSE".
+ */
+#include <palacios/vmm.h>
+#include <palacios/vmm_time.h>
+#include <palacios/vm_guest.h>
+
+
+
+
+/* Overview
+ *
+ * Time handling in VMMs is challenging, and Palacios uses the highest
+ * resolution, lowest overhead timer on modern CPUs that it can - the
+ * processor timestamp counter (TSC). Note that on somewhat old processors
+ * this can be problematic; in particular, older AMD processors did not
+ * have a constant rate timestamp counter in the face of power management
+ * events. However, the latest Intel and AMD CPUs all do (should...) have a
+ * constant rate TSC, and Palacios relies on this fact.
+ *
+ * Basically, Palacios keeps track of three quantities as it runs to manage
+ * the passage of time:
+ * (1) The host timestamp counter - read directly from HW and never written
+ * (2) A monotonic guest timestamp counter used to measure the progression of
+ * time in the guest. This is computed using an offsets from (1) above.
+ * (3) The actual guest timestamp counter (which can be written by
+ * writing to the guest TSC MSR - MSR 0x10) from the monotonic guest TSC.
+ * This is also computed as an offset from (2) above when the TSC and
+ * this offset is updated when the TSC MSR is written.
+ *
+ * The value used to offset the guest TSC from the host TSC is the *sum* of all
+ * of these offsets (2 and 3) above
+ *
+ * Because all other devices are slaved off of the passage of time in the guest,
+ * it is (2) above that drives the firing of other timers in the guest,
+ * including timer devices such as the Programmable Interrupt Timer (PIT).
+ *
+ * Future additions:
+ * (1) Add support for temporarily skewing guest time off of where it should
+ * be to support slack simulation of guests. The idea is that simulators
+ * set this skew to be the difference between how much time passed for a
+ * simulated feature and a real implementation of that feature, making
+ * pass at a different rate from real time on this core. The VMM will then
+ * attempt to move this skew back towards 0 subject to resolution/accuracy
+ * constraints from various system timers.
+ *
+ * The main effort in doing this will be to get accuracy/resolution
+ * information from each local timer and to use this to bound how much skew
+ * is removed on each exit.
+ */
+
+
+
+struct vtime_state {
+ uint32_t guest_cpu_freq; // can be lower than host CPU freq!
+ uint64_t initial_time; // Time when VMM started.
+ sint64_t guest_host_offset;// Offset of monotonic guest time from host time
+};
+
+
+
+
+static int offset_time( struct guest_info * info, sint64_t offset )
+{
+ struct vm_time * time_state = &(info->time_state);
+// PrintDebug("Adding additional offset of %lld to guest time.\n", offset);
+ time_state->guest_host_offset += offset;
+ return 0;
+}
+
+
+// Control guest time in relation to host time so that the two stay
+// appropriately synchronized to the extent possible.
+int v3_adjust_time(struct guest_info * info) {
+ struct vm_time * time_state = &(info->time_state);
+ uint64_t host_time, target_host_time;
+ uint64_t guest_time, target_guest_time, old_guest_time;
+ uint64_t guest_elapsed, host_elapsed, desired_elapsed;
+
+ /* Compute the target host time given how much time has *already*
+ * passed in the guest */
+ guest_time = v3_get_guest_time(time_state);
+ guest_elapsed = (guest_time - time_state->initial_time);
+ desired_elapsed = (guest_elapsed * time_state->host_cpu_freq) / time_state->guest_cpu_freq;
+ target_host_time = time_state->initial_time + desired_elapsed;
+
+ /* Now, let the host run while the guest is stopped to make the two
+ * sync up. */
+ host_time = v3_get_host_time(time_state);
+ old_guest_time = v3_get_guest_time(time_state);
+
+ while (target_host_time > host_time) {
+ v3_yield(info);
+ host_time = v3_get_host_time(time_state);
+ }
+
+ guest_time = v3_get_guest_time(time_state);
+
+ // We do *not* assume the guest timer was paused in the VM. If it was
+ // this offseting is 0. If it wasn't we need this.
+ offset_time(info, (sint64_t)old_guest_time - (sint64_t)guest_time);
+
+ /* Now the host may have gotten ahead of the guest because
+ * yielding is a coarse grained thing. Figure out what guest time
+ * we want to be at, and use the use the offsetting mechanism in
+ * the VMM to make the guest run forward. We limit *how* much we skew
+ * it forward to prevent the guest time making large jumps,
+ * however. */
+ host_elapsed = host_time - time_state->initial_time;
+ desired_elapsed = (host_elapsed * time_state->guest_cpu_freq) / time_state->host_cpu_freq;
+ target_guest_time = time_state->initial_time + desired_elapsed;
+
+ if (guest_time < target_guest_time) {
+ uint64_t max_skew, desired_skew, skew;
+
+ if (time_state->enter_time) {
+ max_skew = (time_state->exit_time - time_state->enter_time) / 10;
+ } else {
+ max_skew = 0;
+ }
+
+ desired_skew = target_guest_time - guest_time;
+ skew = desired_skew > max_skew ? max_skew : desired_skew;
+/* PrintDebug("Guest %llu cycles behind where it should be.\n",
+ desired_skew);
+ PrintDebug("Limit on forward skew is %llu. Skewing forward %llu.\n",
+ max_skew, skew); */
+
+ offset_time(info, skew);
+ }
+
+ return 0;
+}
+
+
+static int init() {
+ khz = v3_cfg_val(cfg_tree, "khz");
+
+ if (khz) {
+ time_state->guest_cpu_freq = atoi(khz);
+ PrintDebug("Core %d CPU frequency requested at %d khz.\n",
+ info->cpu_id, time_state->guest_cpu_freq);
+ }
+
+ if ( (khz == NULL) ||
+ (time_state->guest_cpu_freq <= 0) ||
+ (time_state->guest_cpu_freq > time_state->host_cpu_freq) ) {
+
+ time_state->guest_cpu_freq = time_state->host_cpu_freq;
+ }
+
+
+}
--- /dev/null
+/*
+ * This file is part of the Palacios Virtual Machine Monitor developed
+ * by the V3VEE Project with funding from the United States National
+ * Science Foundation and the Department of Energy.
+ *
+ * The V3VEE Project is a joint project between Northwestern University
+ * and the University of New Mexico. You can find out more at
+ * http://www.v3vee.org
+ *
+ * Copyright (c) 2008, Jack Lange <jarusl@cs.northwestern.edu>
+ * Copyright (c) 2008, The V3VEE Project <http://www.v3vee.org>
+ * All rights reserved.
+ *
+ * Author: Jack Lange <jarusl@cs.northwestern.edu>
+ * Patrick G. Bridges <bridges@cs.unm.edu>
+ *
+ * This is free software. You are permitted to use,
+ * redistribute, and modify it as specified in the file "V3VEE_LICENSE".
+ */
+
+#include <palacios/vmm.h>
+#include <palacios/vmm_time.h>
+#include <palacios/vm_guest.h>
+
+
+// Functions for handling exits on the TSC when fully virtualizing
+// the timestamp counter.
+#define TSC_MSR 0x10
+#define TSC_AUX_MSR 0xC0000103
+
+int v3_handle_rdtscp(struct guest_info *info);
+int v3_handle_rdtsc(struct guest_info *info);
+
+
+struct vtsc_state {
+
+ struct v3_msr tsc_aux; // Auxilliary MSR for RDTSCP
+
+};
+
+
+
+/*
+ * Handle full virtualization of the time stamp counter. As noted
+ * above, we don't store the actual value of the TSC, only the guest's
+ * offset from monotonic guest's time. If the guest writes to the TSC, we
+ * handle this by changing that offset.
+ *
+ * Possible TODO: Proper hooking of TSC read/writes?
+ */
+
+static int rdtsc(struct guest_info * info) {
+ uint64_t tscval = v3_get_guest_tsc(&info->time_state);
+
+ info->vm_regs.rdx = tscval >> 32;
+ info->vm_regs.rax = tscval & 0xffffffffLL;
+
+ return 0;
+}
+
+int v3_handle_rdtsc(struct guest_info * info) {
+ rdtsc(info);
+
+ info->vm_regs.rax &= 0x00000000ffffffffLL;
+ info->vm_regs.rdx &= 0x00000000ffffffffLL;
+
+ info->rip += 2;
+
+ return 0;
+}
+
+int v3_rdtscp(struct guest_info * info) {
+ int ret;
+ /* First get the MSR value that we need. It's safe to futz with
+ * ra/c/dx here since they're modified by this instruction anyway. */
+ info->vm_regs.rcx = TSC_AUX_MSR;
+ ret = v3_handle_msr_read(info);
+
+ if (ret != 0) {
+ return ret;
+ }
+
+ info->vm_regs.rcx = info->vm_regs.rax;
+
+ /* Now do the TSC half of the instruction */
+ ret = v3_rdtsc(info);
+
+ if (ret != 0) {
+ return ret;
+ }
+
+ return 0;
+}
+
+
+int v3_handle_rdtscp(struct guest_info * info) {
+ PrintDebug("Handling virtual RDTSCP call.\n");
+
+ v3_rdtscp(info);
+
+ info->vm_regs.rax &= 0x00000000ffffffffLL;
+ info->vm_regs.rcx &= 0x00000000ffffffffLL;
+ info->vm_regs.rdx &= 0x00000000ffffffffLL;
+
+ info->rip += 3;
+
+ return 0;
+}
+
+
+
+
+static int tsc_aux_msr_read_hook(struct guest_info *info, uint_t msr_num,
+ struct v3_msr *msr_val, void *priv) {
+ struct vm_time * time_state = &(info->time_state);
+
+ V3_ASSERT(msr_num == TSC_AUX_MSR);
+
+ msr_val->lo = time_state->tsc_aux.lo;
+ msr_val->hi = time_state->tsc_aux.hi;
+
+ return 0;
+}
+
+
+static int tsc_aux_msr_write_hook(struct guest_info *info, uint_t msr_num,
+ struct v3_msr msr_val, void *priv) {
+ struct vm_time * time_state = &(info->time_state);
+
+ V3_ASSERT(msr_num == TSC_AUX_MSR);
+
+ time_state->tsc_aux.lo = msr_val.lo;
+ time_state->tsc_aux.hi = msr_val.hi;
+
+ return 0;
+}
+
+
+static int tsc_msr_read_hook(struct guest_info *info, uint_t msr_num,
+ struct v3_msr *msr_val, void *priv) {
+ uint64_t time = v3_get_guest_tsc(&info->time_state);
+
+ V3_ASSERT(msr_num == TSC_MSR);
+
+ msr_val->hi = time >> 32;
+ msr_val->lo = time & 0xffffffffLL;
+
+ return 0;
+}
+
+
+static int tsc_msr_write_hook(struct guest_info *info, uint_t msr_num,
+ struct v3_msr msr_val, void *priv) {
+ struct vm_time * time_state = &(info->time_state);
+ uint64_t guest_time, new_tsc;
+
+ V3_ASSERT(msr_num == TSC_MSR);
+
+ new_tsc = (((uint64_t)msr_val.hi) << 32) | (uint64_t)msr_val.lo;
+ guest_time = v3_get_guest_time(time_state);
+ time_state->tsc_guest_offset = (sint64_t)new_tsc - (sint64_t)guest_time;
+
+ return 0;
+}
+
+
+static int deinit() {
+ v3_unhook_msr(vm, TSC_MSR);
+ v3_unhook_msr(vm, TSC_AUX_MSR);
+}
+
+
+static int init() {
+
+ time_state->tsc_aux.lo = 0;
+ time_state->tsc_aux.hi = 0;
+
+
+
+ PrintDebug("Installing TSC MSR hook.\n");
+ ret = v3_hook_msr(vm, TSC_MSR,
+ tsc_msr_read_hook, tsc_msr_write_hook, NULL);
+
+ if (ret != 0) {
+ return ret;
+ }
+
+ PrintDebug("Installing TSC_AUX MSR hook.\n");
+ ret = v3_hook_msr(vm, TSC_AUX_MSR, tsc_aux_msr_read_hook,
+ tsc_aux_msr_write_hook, NULL);
+
+ if (ret != 0) {
+ return ret;
+ }
+}
vmm_multitree.o \
-obj-$(CONFIG_INSPECTOR) += vmm_inspector.o
obj-$(CONFIG_XED) += vmm_xed.o
v3_set_foreground_vm(vm);
}
-#ifdef CONFIG_INSPECTOR
- v3_init_inspector(vm);
-#endif
+
#ifdef CONFIG_TELEMETRY
v3_init_telemetry(vm);
v3_cpu_arch_t cpu_type = v3_get_cpu_type(V3_Get_CPU());
struct v3_vm_info * vm = core->vm_info;
-#ifdef CONFIG_INSPECTOR
- v3_init_inspector_core(core);
-#endif
+
/*
* Initialize the subsystem data strutures
}
+
+
int V3_deinit_extensions() {
v3_free_htable(ext_table, 0, 0);
return 0;
return 0;
}
+
+int v3_deinit_ext_manager(struct v3_vm_info * vm);
+
int v3_add_extension(struct v3_vm_info * vm, const char * name, v3_cfg_tree_t * cfg) {
struct v3_extension_impl * impl = NULL;
struct v3_extension * ext = NULL;
return 0;
}
+
+int v3_init_core_extensions(struct guest_info * core) {
+ struct v3_extension * ext = NULL;
+
+ list_for_each_entry(ext, &(core->vm_info->extensions.extensions), node) {
+ if ((ext->impl) && (ext->impl->core_init)) {
+ if (ext->impl->core_init(core, ext->priv_data) == -1) {
+ PrintError("Error configuring per core extension %s on core %d\n",
+ ext->impl->name, core->cpu_id);
+ return -1;
+ }
+ }
+ }
+
+ return 0;
+}
+
+
+
+
+void * v3_get_extension_state(struct v3_vm_info * vm, const char * name) {
+ struct v3_extension * ext = NULL;
+
+ list_for_each_entry(ext, &(vm->extensions.extensions), node) {
+ if (strncmp(ext->impl->name, name, strlen(ext->impl->name)) == 0) {
+ return ext->priv_data;
+ }
+ }
+
+ return NULL;
+}