mcctrl: support remote CPU MSR read/write operations

arch/x86/kernel/cpu.c

@@ -1834,4 +1834,22 @@ mod_nmi_ctx(void *nmi_ctx, void (*func)())
 	l[i++] = (unsigned long)(l + 27);	// ols rsp
 	l[i++] = 0x28;				// KERNEL DS
 }
+
+int arch_cpu_read_write_register(
+		struct mcctrl_os_cpu_register *desc,
+		enum mcctrl_os_cpu_operation op)
+{
+	if (op == MCCTRL_OS_CPU_READ_REGISTER) {
+		desc->val = rdmsr(desc->addr);
+	}
+	else if (op == MCCTRL_OS_CPU_WRITE_REGISTER) {
+		wrmsr(desc->addr, desc->val);
+	}
+	else {
+		return -1;
+	}
+
+	return 0;
+}
+
 /*** end of file ***/

executer/kernel/mcctrl/control.c

@@ -40,6 +40,7 @@
 #include <asm/io.h>
 #include "../../config.h"
 #include "mcctrl.h"
+#include "mcctrl_public.h"
 #include <ihk/ihk_host_user.h>
 
 //#define DEBUG
@@ -1937,3 +1938,149 @@ void mcexec_prepare_ack(ihk_os_t os, unsigned long arg, int err)
 	mcctrl_put_per_proc_data(ppd);
 }
 
+/* Per-CPU register manipulation functions */
+struct mcctrl_os_cpu_response {
+	int done;
+	unsigned long val;
+	wait_queue_head_t wq;
+};
+
+int mcctrl_get_request_os_cpu(ihk_os_t *ret_os, int *ret_cpu)
+{
+	ihk_os_t os;
+	struct mcctrl_usrdata *usrdata;
+	struct mcctrl_per_proc_data *ppd;
+	struct ikc_scd_packet *packet;
+	struct ihk_ikc_channel_desc *ch;
+	int ret = 0;
+
+	/* Look up IHK OS structure
+	 * TODO: iterate all possible indeces, currently only for OS 0
+	 */
+	os = ihk_host_find_os(0, NULL);
+	if (!os) {
+		printk("%s: ERROR: no OS found for index 0\n", __FUNCTION__);
+		return -EINVAL;
+	}
+
+	/* Look up per-OS mcctrl structure */
+	usrdata = ihk_host_os_get_usrdata(os);
+	if (!usrdata) {
+		printk("%s: ERROR: no usrdata found for OS %p\n", __FUNCTION__, os);
+		return -EINVAL;
+	}
+
+	/* Look up per-process structure */
+	ppd = mcctrl_get_per_proc_data(usrdata, task_tgid_vnr(current));
+	if (!ppd) {
+		kprintf("%s: ERROR: no per-process structure for PID %d??\n",
+				__FUNCTION__, task_tgid_vnr(current));
+		return -EINVAL;
+	}
+
+	/* Look up per-thread structure */
+	packet = (struct ikc_scd_packet *)mcctrl_get_per_thread_data(ppd, current);
+	if (!packet) {
+		ret = -EINVAL;
+		printk("%s: ERROR: no packet registered for TID %d\n",
+				__FUNCTION__, task_pid_vnr(current));
+		goto out_put_ppd;
+	}
+
+	*ret_os = os;
+	/* TODO: define a new IHK query function instead of
+	 * accessing internals directly */
+	ch = (usrdata->channels + packet->ref)->c;
+	*ret_cpu = ch->send.queue->read_cpu;
+	ret = 0;
+
+	printk("%s: OS: %p, CPU: %d\n", __FUNCTION__, os, *ret_cpu);
+
+out_put_ppd:
+	mcctrl_put_per_proc_data(ppd);
+
+	return ret;
+}
+
+EXPORT_SYMBOL(mcctrl_get_request_os_cpu);
+
+void mcctrl_os_read_write_cpu_response(ihk_os_t os,
+		struct ikc_scd_packet *pisp)
+{
+	struct mcctrl_os_cpu_response *resp;
+
+	/* XXX: What if caller thread is unblocked by a signal
+	 * before this message arrives? */
+	resp = pisp->resp;
+	if (!resp) {
+		return;
+	}
+
+	resp->val = pisp->desc.val;
+	resp->done = 1;
+	wake_up_interruptible(&resp->wq);
+}
+
+int __mcctrl_os_read_write_cpu_register(ihk_os_t os, int cpu,
+		struct mcctrl_os_cpu_register *desc,
+		enum mcctrl_os_cpu_operation op)
+{
+	struct ikc_scd_packet isp;
+	struct mcctrl_os_cpu_response resp;
+	int ret = -EINVAL;
+
+	memset(&isp, '\0', sizeof(struct ikc_scd_packet));
+	isp.msg = SCD_MSG_CPU_RW_REG;
+	isp.op = op;
+	isp.desc = *desc;
+	isp.resp = &resp;
+
+	resp.done = 0;
+	init_waitqueue_head(&resp.wq);
+
+	mb();
+	ret = mcctrl_ikc_send(os, cpu, &isp);
+	if (ret < 0) {
+		printk("%s: ERROR sending IKC msg: %d\n", __FUNCTION__, ret);
+		goto out;
+	}
+
+	/* Wait for response */
+	ret = wait_event_interruptible(resp.wq, resp.done);
+	if (ret < 0) {
+		printk("%s: ERROR after wait: %d\n", __FUNCTION__, ret);
+		goto out;
+	}
+
+	/* Update if read */
+	if (ret == 0 && op == MCCTRL_OS_CPU_READ_REGISTER) {
+		desc->val = resp.val;
+	}
+
+	printk("%s: MCCTRL_OS_CPU_%s_REGISTER: reg: 0x%lx, val: 0x%lx\n",
+		__FUNCTION__,
+		(op == MCCTRL_OS_CPU_READ_REGISTER ? "READ" : "WRITE"),
+		desc->addr, desc->val);
+
+out:
+	return ret;
+}
+
+int mcctrl_os_read_cpu_register(ihk_os_t os, int cpu,
+		struct mcctrl_os_cpu_register *desc)
+{
+	return __mcctrl_os_read_write_cpu_register(os, cpu,
+			desc, MCCTRL_OS_CPU_READ_REGISTER);
+}
+
+EXPORT_SYMBOL(mcctrl_os_read_cpu_register);
+
+int mcctrl_os_write_cpu_register(ihk_os_t os, int cpu,
+		struct mcctrl_os_cpu_register *desc)
+{
+	return __mcctrl_os_read_write_cpu_register(os, cpu,
+			desc, MCCTRL_OS_CPU_WRITE_REGISTER);
+}
+
+EXPORT_SYMBOL(mcctrl_os_write_cpu_register);
+

executer/kernel/mcctrl/ikc.c

@@ -54,6 +54,8 @@ static void mcctrl_ikc_init(ihk_os_t os, int cpu, unsigned long rphys, struct ih
 int mcexec_syscall(struct mcctrl_usrdata *ud, struct ikc_scd_packet *packet);
 void sig_done(unsigned long arg, int err);
 void mcctrl_perf_ack(ihk_os_t os, struct ikc_scd_packet *packet);
+void mcctrl_os_read_write_cpu_response(ihk_os_t os,
+		struct ikc_scd_packet *pisp);
 
 /* XXX: this runs in atomic context! */
 static int syscall_packet_handler(struct ihk_ikc_channel_desc *c,
@@ -113,6 +115,11 @@ static int syscall_packet_handler(struct ihk_ikc_channel_desc *c,
 	case SCD_MSG_PERF_ACK:
 		mcctrl_perf_ack(__os, pisp);
 		break;
+
+	case SCD_MSG_CPU_RW_REG_RESP:
+		mcctrl_os_read_write_cpu_response(__os, pisp);
+		break;
+
 	default:
 		printk(KERN_ERR "mcctrl:syscall_packet_handler:"
 				"unknown message (%d.%d.%d.%d.%d.%#lx)\n",

executer/kernel/mcctrl/mcctrl.h

@@ -44,6 +44,7 @@
 #include <linux/rwlock.h>
 #include <linux/threads.h>
 #include "sysfs.h"
+#include "mcctrl_public.h"
 
 #define SCD_MSG_PREPARE_PROCESS         0x1
 #define SCD_MSG_PREPARE_PROCESS_ACKED   0x2
@@ -95,6 +96,9 @@
 #define SCD_MSG_PERF_CTRL               0x50
 #define SCD_MSG_PERF_ACK                0x51
 
+#define SCD_MSG_CPU_RW_REG              0x52
+#define SCD_MSG_CPU_RW_REG_RESP         0x53
+
 #define DMA_PIN_SHIFT                   21
 
 #define DO_USER_MODE
@@ -106,6 +110,12 @@ struct coretable {
 	unsigned long addr;
 };
 
+enum mcctrl_os_cpu_operation {
+	MCCTRL_OS_CPU_READ_REGISTER,
+	MCCTRL_OS_CPU_WRITE_REGISTER,
+	MCCTRL_OS_CPU_MAX_OP
+};
+
 struct ikc_scd_packet {
 	int msg;
 	int err;
@@ -131,6 +141,13 @@ struct ikc_scd_packet {
 		struct {
 			int ttid;
 		};
+
+		/* SCD_MSG_CPU_RW_REG */
+		struct {
+			struct mcctrl_os_cpu_register desc;
+			enum mcctrl_os_cpu_operation op;
+			void *resp;
+		};
 	};
 	char padding[12];
 };

executer/kernel/mcctrl/mcctrl_public.h

@@ -0,0 +1,20 @@
+#ifndef __MCCTRL_PUBLIC_H
+#define __MCCTRL_PUBLIC_H
+
+#include <ihk/ihk_host_user.h>
+#include <ikc/queue.h>
+
+struct mcctrl_os_cpu_register {
+	unsigned long addr;
+	unsigned long val;
+	unsigned long addr_ext;
+};
+
+int mcctrl_os_read_cpu_register(ihk_os_t os, int cpu,
+		struct mcctrl_os_cpu_register *desc);
+int mcctrl_os_write_cpu_register(ihk_os_t os, int cpu,
+		struct mcctrl_os_cpu_register *desc);
+int mcctrl_get_request_os_cpu(ihk_os_t *os, int *cpu);
+
+
+#endif // __MCCTRL_PUBLIC_H

kernel/host.c

@@ -644,6 +644,17 @@ static int syscall_packet_handler(struct ihk_ikc_channel_desc *c,
 		ret = 0;
 		break;
 
+	case SCD_MSG_CPU_RW_REG:
+
+		pckt.msg = SCD_MSG_CPU_RW_REG_RESP;
+		memcpy(&pckt.desc, &packet->desc,
+				sizeof(struct mcctrl_os_cpu_register));
+		pckt.resp = packet->resp;
+		pckt.err = arch_cpu_read_write_register(&pckt.desc, pckt.op);
+
+		ihk_ikc_send(resp_channel, &pckt, 0);
+		break;
+
 	default:
 		kprintf("syscall_pakcet_handler:unknown message "
 				"(%d.%d.%d.%d.%d.%#lx)\n",

kernel/include/syscall.h

@@ -78,6 +78,9 @@
 #define SCD_MSG_PERF_CTRL               0x50
 #define SCD_MSG_PERF_ACK                0x51
 
+#define SCD_MSG_CPU_RW_REG              0x52
+#define SCD_MSG_CPU_RW_REG_RESP         0x53
+
 /* Cloning flags.  */
 # define CSIGNAL       0x000000ff /* Signal mask to be sent at exit.  */
 # define CLONE_VM      0x00000100 /* Set if VM shared between processes.  */
@@ -210,6 +213,18 @@ struct syscall_request {
 	unsigned long args[6];
 };
 
+struct mcctrl_os_cpu_register {
+	unsigned long addr;
+	unsigned long val;
+	unsigned long addr_ext;
+};
+
+enum mcctrl_os_cpu_operation {
+	MCCTRL_OS_CPU_READ_REGISTER,
+	MCCTRL_OS_CPU_WRITE_REGISTER,
+	MCCTRL_OS_CPU_MAX_OP
+};
+
 struct ikc_scd_packet {
 	int msg;
 	int err;
@@ -235,6 +250,13 @@ struct ikc_scd_packet {
 		struct {
 			int ttid;
 		};
+
+		/* SCD_MSG_CPU_RW_REG */
+		struct {
+			struct mcctrl_os_cpu_register desc;
+			enum mcctrl_os_cpu_operation op;
+			void *resp;
+		};
 	};
 	char padding[12];
 };
@@ -412,6 +434,8 @@ int do_shmget(key_t key, size_t size, int shmflg);
 struct process_vm;
 int arch_map_vdso(struct process_vm *vm);	/* arch dependent */
 int arch_setup_vdso(void);
+int arch_cpu_read_write_register(struct mcctrl_os_cpu_register *desc,
+		enum mcctrl_os_cpu_operation op);
 
 #define VDSO_MAXPAGES 2
 struct vdso {