mf.c

底层驱动开发

/*
  * mf.c
  * Copyright (C) 2001 Troy D. Armstrong  IBM Corporation
  * Copyright (C) 2004-2005 Stephen Rothwell  IBM Corporation
  *
  * This modules exists as an interface between a Linux secondary partition
  * running on an iSeries and the primary partition's Virtual Service
  * Processor (VSP) object.  The VSP has final authority over powering on/off
  * all partitions in the iSeries.  It also provides miscellaneous low-level
  * machine facility type operations.
  *
  *
  * 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 Free Software Foundation; either version 2 of the License, or
  * (at your option) any later version.
  *
  * This program is distributed in the hope that it will be useful,
  * but WITHOUT ANY WARRANTY; without even the implied warranty of
  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
  * GNU General Public License for more details.
  *
  * You should have received a copy of the GNU General Public License
  * along with this program; if not, write to the Free Software
  * Foundation, Inc., 59 Temple Place, Suite 330, Boston, MA  02111-1307 USA
  */

#include <linux/types.h>
#include <linux/errno.h>
#include <linux/kernel.h>
#include <linux/init.h>
#include <linux/completion.h>
#include <linux/delay.h>
#include <linux/dma-mapping.h>
#include <linux/bcd.h>

#include <asm/time.h>
#include <asm/uaccess.h>
#include <asm/paca.h>
#include <asm/iSeries/vio.h>
#include <asm/iSeries/mf.h>
#include <asm/iSeries/HvLpConfig.h>
#include <asm/iSeries/ItLpQueue.h>

/*
 * This is the structure layout for the Machine Facilites LPAR event
 * flows.
 */
struct vsp_cmd_data {
	u64 token;
	u16 cmd;
	HvLpIndex lp_index;
	u8 result_code;
	u32 reserved;
	union {
		u64 state;	/* GetStateOut */
		u64 ipl_type;	/* GetIplTypeOut, Function02SelectIplTypeIn */
		u64 ipl_mode;	/* GetIplModeOut, Function02SelectIplModeIn */
		u64 page[4];	/* GetSrcHistoryIn */
		u64 flag;	/* GetAutoIplWhenPrimaryIplsOut,
				   SetAutoIplWhenPrimaryIplsIn,
				   WhiteButtonPowerOffIn,
				   Function08FastPowerOffIn,
				   IsSpcnRackPowerIncompleteOut */
		struct {
			u64 token;
			u64 address_type;
			u64 side;
			u32 length;
			u32 offset;
		} kern;		/* SetKernelImageIn, GetKernelImageIn,
				   SetKernelCmdLineIn, GetKernelCmdLineIn */
		u32 length_out;	/* GetKernelImageOut, GetKernelCmdLineOut */
		u8 reserved[80];
	} sub_data;
};

struct vsp_rsp_data {
	struct completion com;
	struct vsp_cmd_data *response;
};

struct alloc_data {
	u16 size;
	u16 type;
	u32 count;
	u16 reserved1;
	u8 reserved2;
	HvLpIndex target_lp;
};

struct ce_msg_data;

typedef void (*ce_msg_comp_hdlr)(void *token, struct ce_msg_data *vsp_cmd_rsp);

struct ce_msg_comp_data {
	ce_msg_comp_hdlr handler;
	void *token;
};

struct ce_msg_data {
	u8 ce_msg[12];
	char reserved[4];
	struct ce_msg_comp_data *completion;
};

struct io_mf_lp_event {
	struct HvLpEvent hp_lp_event;
	u16 subtype_result_code;
	u16 reserved1;
	u32 reserved2;
	union {
		struct alloc_data alloc;
		struct ce_msg_data ce_msg;
		struct vsp_cmd_data vsp_cmd;
	} data;
};

#define subtype_data(a, b, c, d)	\
		(((a) << 24) + ((b) << 16) + ((c) << 8) + (d))

/*
 * All outgoing event traffic is kept on a FIFO queue.  The first
 * pointer points to the one that is outstanding, and all new
 * requests get stuck on the end.  Also, we keep a certain number of
 * preallocated pending events so that we can operate very early in
 * the boot up sequence (before kmalloc is ready).
 */
struct pending_event {
	struct pending_event *next;
	struct io_mf_lp_event event;
	MFCompleteHandler hdlr;
	char dma_data[72];
	unsigned dma_data_length;
	unsigned remote_address;
};
static spinlock_t pending_event_spinlock;
static struct pending_event *pending_event_head;
static struct pending_event *pending_event_tail;
static struct pending_event *pending_event_avail;
static struct pending_event pending_event_prealloc[16];

/*
 * Put a pending event onto the available queue, so it can get reused.
 * Attention! You must have the pending_event_spinlock before calling!
 */
static void free_pending_event(struct pending_event *ev)
{
	if (ev != NULL) {
		ev->next = pending_event_avail;
		pending_event_avail = ev;
	}
}

/*
 * Enqueue the outbound event onto the stack.  If the queue was
 * empty to begin with, we must also issue it via the Hypervisor
 * interface.  There is a section of code below that will touch
 * the first stack pointer without the protection of the pending_event_spinlock.
 * This is OK, because we know that nobody else will be modifying
 * the first pointer when we do this.
 */
static int signal_event(struct pending_event *ev)
{
	int rc = 0;
	unsigned long flags;
	int go = 1;
	struct pending_event *ev1;
	HvLpEvent_Rc hv_rc;

	/* enqueue the event */
	if (ev != NULL) {
		ev->next = NULL;
		spin_lock_irqsave(&pending_event_spinlock, flags);
		if (pending_event_head == NULL)
			pending_event_head = ev;
		else {
			go = 0;
			pending_event_tail->next = ev;
		}
		pending_event_tail = ev;
		spin_unlock_irqrestore(&pending_event_spinlock, flags);
	}

	/* send the event */
	while (go) {
		go = 0;

		/* any DMA data to send beforehand? */
		if (pending_event_head->dma_data_length > 0)
			HvCallEvent_dmaToSp(pending_event_head->dma_data,
					pending_event_head->remote_address,
					pending_event_head->dma_data_length,
					HvLpDma_Direction_LocalToRemote);

		hv_rc = HvCallEvent_signalLpEvent(
				&pending_event_head->event.hp_lp_event);
		if (hv_rc != HvLpEvent_Rc_Good) {
			printk(KERN_ERR "mf.c: HvCallEvent_signalLpEvent() "
					"failed with %d\n", (int)hv_rc);

			spin_lock_irqsave(&pending_event_spinlock, flags);
			ev1 = pending_event_head;
			pending_event_head = pending_event_head->next;
			if (pending_event_head != NULL)
				go = 1;
			spin_unlock_irqrestore(&pending_event_spinlock, flags);

			if (ev1 == ev)
				rc = -EIO;
			else if (ev1->hdlr != NULL)
				(*ev1->hdlr)((void *)ev1->event.hp_lp_event.xCorrelationToken, -EIO);

			spin_lock_irqsave(&pending_event_spinlock, flags);
			free_pending_event(ev1);
			spin_unlock_irqrestore(&pending_event_spinlock, flags);
		}
	}

	return rc;
}

/*
 * Allocate a new pending_event structure, and initialize it.
 */
static struct pending_event *new_pending_event(void)
{
	struct pending_event *ev = NULL;
	HvLpIndex primary_lp = HvLpConfig_getPrimaryLpIndex();
	unsigned long flags;
	struct HvLpEvent *hev;

	spin_lock_irqsave(&pending_event_spinlock, flags);
	if (pending_event_avail != NULL) {
		ev = pending_event_avail;
		pending_event_avail = pending_event_avail->next;
	}
	spin_unlock_irqrestore(&pending_event_spinlock, flags);
	if (ev == NULL) {
		ev = kmalloc(sizeof(struct pending_event), GFP_ATOMIC);
		if (ev == NULL) {
			printk(KERN_ERR "mf.c: unable to kmalloc %ld bytes\n",
					sizeof(struct pending_event));
			return NULL;
		}
	}
	memset(ev, 0, sizeof(struct pending_event));
	hev = &ev->event.hp_lp_event;
	hev->xFlags.xValid = 1;
	hev->xFlags.xAckType = HvLpEvent_AckType_ImmediateAck;
	hev->xFlags.xAckInd = HvLpEvent_AckInd_DoAck;
	hev->xFlags.xFunction = HvLpEvent_Function_Int;
	hev->xType = HvLpEvent_Type_MachineFac;
	hev->xSourceLp = HvLpConfig_getLpIndex();
	hev->xTargetLp = primary_lp;
	hev->xSizeMinus1 = sizeof(ev->event) - 1;
	hev->xRc = HvLpEvent_Rc_Good;
	hev->xSourceInstanceId = HvCallEvent_getSourceLpInstanceId(primary_lp,
			HvLpEvent_Type_MachineFac);
	hev->xTargetInstanceId = HvCallEvent_getTargetLpInstanceId(primary_lp,
			HvLpEvent_Type_MachineFac);

	return ev;
}

static int signal_vsp_instruction(struct vsp_cmd_data *vsp_cmd)
{
	struct pending_event *ev = new_pending_event();
	int rc;
	struct vsp_rsp_data response;

	if (ev == NULL)
		return -ENOMEM;

	init_completion(&response.com);
	response.response = vsp_cmd;
	ev->event.hp_lp_event.xSubtype = 6;
	ev->event.hp_lp_event.x.xSubtypeData =
		subtype_data('M', 'F',  'V',  'I');
	ev->event.data.vsp_cmd.token = (u64)&response;
	ev->event.data.vsp_cmd.cmd = vsp_cmd->cmd;
	ev->event.data.vsp_cmd.lp_index = HvLpConfig_getLpIndex();
	ev->event.data.vsp_cmd.result_code = 0xFF;
	ev->event.data.vsp_cmd.reserved = 0;
	memcpy(&(ev->event.data.vsp_cmd.sub_data),
			&(vsp_cmd->sub_data), sizeof(vsp_cmd->sub_data));
	mb();

	rc = signal_event(ev);
	if (rc == 0)
		wait_for_completion(&response.com);
	return rc;
}


/*
 * Send a 12-byte CE message to the primary partition VSP object
 */
static int signal_ce_msg(char *ce_msg, struct ce_msg_comp_data *completion)
{
	struct pending_event *ev = new_pending_event();

	if (ev == NULL)
		return -ENOMEM;

	ev->event.hp_lp_event.xSubtype = 0;
	ev->event.hp_lp_event.x.xSubtypeData =
		subtype_data('M',  'F',  'C',  'E');
	memcpy(ev->event.data.ce_msg.ce_msg, ce_msg, 12);
	ev->event.data.ce_msg.completion = completion;
	return signal_event(ev);
}

/*
 * Send a 12-byte CE message (with no data) to the primary partition VSP object
 */
static int signal_ce_msg_simple(u8 ce_op, struct ce_msg_comp_data *completion)
{
	u8 ce_msg[12];

	memset(ce_msg, 0, sizeof(ce_msg));
	ce_msg[3] = ce_op;
	return signal_ce_msg(ce_msg, completion);
}

/*
 * Send a 12-byte CE message and DMA data to the primary partition VSP object
 */
static int dma_and_signal_ce_msg(char *ce_msg,
		struct ce_msg_comp_data *completion, void *dma_data,
		unsigned dma_data_length, unsigned remote_address)
{
	struct pending_event *ev = new_pending_event();

	if (ev == NULL)
		return -ENOMEM;

	ev->event.hp_lp_event.xSubtype = 0;
	ev->event.hp_lp_event.x.xSubtypeData =
		subtype_data('M', 'F', 'C', 'E');
	memcpy(ev->event.data.ce_msg.ce_msg, ce_msg, 12);
	ev->event.data.ce_msg.completion = completion;
	memcpy(ev->dma_data, dma_data, dma_data_length);
	ev->dma_data_length = dma_data_length;
	ev->remote_address = remote_address;
	return signal_event(ev);
}

/*
 * Initiate a nice (hopefully) shutdown of Linux.  We simply are
 * going to try and send the init process a SIGINT signal.  If
 * this fails (why?), we'll simply force it off in a not-so-nice
 * manner.
 */
static int shutdown(void)
{
	int rc = kill_proc(1, SIGINT, 1);

	if (rc) {
		printk(KERN_ALERT "mf.c: SIGINT to init failed (%d), "
				"hard shutdown commencing\n", rc);
		mf_power_off();
	} else
		printk(KERN_INFO "mf.c: init has been successfully notified "
				"to proceed with shutdown\n");
	return rc;
}

/*
 * The primary partition VSP object is sending us a new
 * event flow.  Handle it...
 */
static void handle_int(struct io_mf_lp_event *event)
{
	struct ce_msg_data *ce_msg_data;
	struct ce_msg_data *pce_msg_data;
	unsigned long flags;
	struct pending_event *pev;

	/* ack the interrupt */
	event->hp_lp_event.xRc = HvLpEvent_Rc_Good;
	HvCallEvent_ackLpEvent(&event->hp_lp_event);

	/* process interrupt */
	switch (event->hp_lp_event.xSubtype) {
	case 0:	/* CE message */
		ce_msg_data = &event->data.ce_msg;
		switch (ce_msg_data->ce_msg[3]) {
		case 0x5B:	/* power control notification */
			if ((ce_msg_data->ce_msg[5] & 0x20) != 0) {
				printk(KERN_INFO "mf.c: Commencing partition shutdown\n");
				if (shutdown() == 0)
					signal_ce_msg_simple(0xDB, NULL);
			}
			break;
		case 0xC0:	/* get time */
			spin_lock_irqsave(&pending_event_spinlock, flags);
			pev = pending_event_head;
			if (pev != NULL)
				pending_event_head = pending_event_head->next;
			spin_unlock_irqrestore(&pending_event_spinlock, flags);
			if (pev == NULL)
				break;
			pce_msg_data = &pev->event.data.ce_msg;
			if (pce_msg_data->ce_msg[3] != 0x40)
				break;
			if (pce_msg_data->completion != NULL) {
				ce_msg_comp_hdlr handler =
					pce_msg_data->completion->handler;
				void *token = pce_msg_data->completion->token;

				if (handler != NULL)
					(*handler)(token, ce_msg_data);
			}
			spin_lock_irqsave(&pending_event_spinlock, flags);
			free_pending_event(pev);
			spin_unlock_irqrestore(&pending_event_spinlock, flags);
			/* send next waiting event */
			if (pending_event_head != NULL)
				signal_event(NULL);
			break;
		}
		break;
	case 1:	/* IT sys shutdown */
		printk(KERN_INFO "mf.c: Commencing system shutdown\n");
		shutdown();
		break;
	}