KQEMU on FreeBSD partially works. (Was: [Qemu-devel] Accelerator for FreeBSD?)

[Antony T Curtis <antony.t.curtis@ntlworld.com>]

[-- Attachment #1: Type: text/plain, Size: 1371 bytes --]

On Sun, 2005-03-20 at 19:42 -0600, Ryan Rempel wrote:
> There is a reference here [1] to the possibility of porting the
> Accelerator to *BSD by porting a small "C stub".
> 
> Is anyone working on this at the moment? If not, would it be possible
> to make the "C stub" available so that someone could give it a try?
> I'd be interested in taking a crack at it myself, but there may be
> others better situated (I've not done FreeBSD kernel development
> before, but I have done some Darwin kernel development).
> 
> That's assuming that the reference is at all accurate :-)
> 
> [1] http://www.dad-answers.com/qemu-forum/viewtopic.php?t=333

I have made some headway... The attached stub appears to work but in my
limited testing so far, it may be a bit unstable. The driver stub is for
FreeBSD 4 (tested on 4.9) and I may work on a 5.3+ one soon.

YMMV. Use with caution...

Simply copy the attached files into the same directory that the kqemu
distribution is untarred and build with

# make -f Makefile.bsd

Install  with

# make -f Makefile.bsd
# kldload kqemu

You will need to create the /dev nodes

# mknod /dev/kqemu c 250 0 root:wheel
# chmod ug+rw /dev/kqemu

-- 
Antony T Curtis, BSc.                   UNIX, Linux, *BSD, Networking
antony.t.curtis@ntlworld.com            C++, J2EE, Perl, MySQL, Apache
                                        IT Consultancy.

[-- Attachment #2: Makefile.bsd --]
[-- Type: text/plain, Size: 89 bytes --]

# 

KMOD=	kqemu
SRCS=	kqemu.h kmod_bsd.c kqemu-mod-i386.o
NOMAN=

.include <bsd.kmod.mk>

[-- Attachment #3: kmod_bsd.c --]
[-- Type: text/x-csrc, Size: 8450 bytes --]

/*
 * FreeBSD kernel wrapper for KQEMU
 * Copyright (c) 2005 Antony T Curtis
 *
 * Based upon the Linux wrapper by Fabrice Bellard
 */

#include <sys/param.h>
#include <sys/systm.h>
#include <sys/kernel.h>
#include <sys/buf.h>
#include <sys/uio.h>
#include <sys/conf.h>
#include <sys/ctype.h>
#include <sys/fcntl.h>
#include <sys/malloc.h>
#include <sys/proc.h>
#include <sys/ioccom.h>
#include <sys/signalvar.h>
#include <sys/resourcevar.h>

#include <vm/vm.h>
#include <vm/vm_param.h>
#include <vm/pmap.h>
#include <vm/vm_map.h>
#include <vm/vm_page.h>
#include <vm/vm_kern.h>
#include <vm/vm_extern.h>

#include <machine/stdarg.h>

#define __KERNEL__

#include "kqemu.h"

static MALLOC_DEFINE(M_KQEMU, "KQEMU", "KQEMU Resources");

/* lock the page at virtual address 'user_addr' and return its
   page index. Return -1 if error */
unsigned long CDECL kqemu_lock_user_page(unsigned long user_addr)
{
    int rc;
    caddr_t addr = (caddr_t) user_addr;
    vm_page_t m;
    vm_offset_t paddr;
    
    /*kqemu_log("kqemu_lock_user_page(0x%08x)\n", addr);*/
    
    rc = vm_fault_quick(addr, VM_PROT_READ|VM_PROT_WRITE);
    if (rc < 0) {    
	/*kqemu_log("vm_fault_quick failed rc=%d\n",rc);*/
	return -1;	
    }
    paddr = vtophys(addr);
    m = PHYS_TO_VM_PAGE(paddr);
    vm_page_wire(m);
    
    return (long)(paddr >> PAGE_SHIFT);
}

void CDECL kqemu_unlock_user_page(unsigned long page_index)
{
    vm_page_t m;
    vm_offset_t paddr;
    /*kqemu_log("kqemu_unlock_user_page(0x%08x)\n",page_index);*/

    paddr = (vm_offset_t)(page_index << PAGE_SHIFT);
    m = PHYS_TO_VM_PAGE(paddr);
    vm_page_unwire(m, 1);
}

unsigned long CDECL kqemu_alloc_zeroed_page(void)
{
    void *addr;
    /*kqemu_log("kqemu_alloc_zeroed_page()\n");*/
    addr = contigmalloc(PAGE_SIZE, M_KQEMU, M_WAITOK, 0, ~0ul, PAGE_SIZE, 0);
    if (!addr) {
	/*kqemu_log("contigmalloc failed\n");*/
        return -1;
    }
    memset(addr, 0, PAGE_SIZE);
    return ((long) addr) >> PAGE_SHIFT;
}

void CDECL kqemu_free_page(unsigned long page_index)
{
    /*kqemu_log("kqemu_free_page(0x%08x)\n", page_index);*/
    contigfree((void *)(page_index << PAGE_SHIFT), PAGE_SIZE, M_KQEMU);
}

void * CDECL kqemu_page_kaddr(unsigned long page_index)
{
    caddr_t addr;
    /*kqemu_log("kqemu_page_kaddr(0x%08x)\n", page_index);*/
    addr = (caddr_t)(page_index << PAGE_SHIFT);
    return (void *)addr;
}

/* contraint: each page of the vmalloced area must be in the first 4
   GB of physical memory */
void * CDECL kqemu_vmalloc(unsigned int size)
{
    /*kqemu_log("kqemu_vmalloc(0x%08x)\n", size);*/
    return malloc(size, M_KQEMU, M_WAITOK);
}

void CDECL kqemu_vfree(void *ptr)
{
    /*kqemu_log("kqemu_vfree(0x%08x)\n", ptr);*/
    return free(ptr, M_KQEMU);
}

unsigned long CDECL kqemu_vmalloc_to_phys(const void *vaddr)
{
    vm_offset_t paddr = (vm_offset_t)vaddr;
    return (long)(paddr >> PAGE_SHIFT);
}

static int
curpriority_cmp(struct proc *p)
{
        int c_class, p_class;

        c_class = RTP_PRIO_BASE(curproc->p_rtprio.type);
        p_class = RTP_PRIO_BASE(p->p_rtprio.type);
        if (p_class != c_class)
                return (p_class - c_class);
        if (p_class == RTP_PRIO_NORMAL)
                return (((int)p->p_priority - (int)curpriority) / PPQ);
        return ((int)p->p_rtprio.prio - (int)curproc->p_rtprio.prio);
}

/* return TRUE if a signal is pending (i.e. the guest must stop
   execution) */
int CDECL kqemu_schedule(void)
{
    struct proc *p = curproc;
    if (curpriority_cmp(p) > 0) {
        int s = splhigh();
	p->p_priority = MAXPRI;
        setrunqueue(p);
        p->p_stats->p_ru.ru_nvcsw++;
        mi_switch();
        splx(s);
    }
    return issignal(curproc) != 0;
}


static char log_buf[4096];

void CDECL kqemu_log(const char *fmt, ...)
{
    va_list ap;
    va_start(ap, fmt);
    vsnprintf(log_buf, sizeof(log_buf), fmt, ap);
    printf("kqemu: %s", log_buf);
    va_end(ap);
}

/*********************************************************/


#define KQEMU_MAX_INSTANCES 4

struct kqemu_instance {
    struct kqemu_state *state;
};

static int kqemu_ref_count = 0;
static int max_locked_pages;
static dev_t kqemu_dev;


static	d_open_t	kqemu_open;
static	d_close_t	kqemu_close;
static	d_ioctl_t	kqemu_ioctl;

static struct cdevsw kqemu_cdevsw = {
	/* open */	kqemu_open,
	/* close */	kqemu_close,
	/* read */	noread,
	/* write */	nowrite,
	/* ioctl */	kqemu_ioctl,
	/* poll */	nopoll,
	/* mmap */	nommap,
	/* strategy */	nostrategy,
	/* name */	"kqemu",
	/* maj */	KQEMU_MAJOR,
	/* dump */	nodump,
	/* psize */	nopsize,
	/* flags */	0,
	/* bmaj */	-1
};

int
kqemu_open(dev, flags, fmt, p)
	dev_t		dev;
	int		flags, fmt;
	struct proc	*p;
{
    struct kqemu_instance *ks;
    
    if (dev->si_drv1 || kqemu_ref_count >= KQEMU_MAX_INSTANCES)
    	return(EBUSY);

    if ((flags & (FREAD|FWRITE)) == FREAD)
    	return(EPERM);

    ks = (struct kqemu_instance *) malloc(sizeof(*ks), M_KQEMU, M_WAITOK);
    if (ks == NULL)
        return(ENOMEM);
    memset(ks, 0, sizeof *ks);
    
    dev->si_drv1 = ks;
    kqemu_ref_count++;
    kqemu_log("opened by pid=%d\n", p->p_pid);
    return(0);
}

int
kqemu_close(dev, flags, fmt, p)
	dev_t		dev;
	int		flags, fmt;
	struct proc	*p;
{
    struct kqemu_instance *ks = (struct kqemu_instance *) dev->si_drv1;

    if (ks->state) {
	kqemu_delete(ks->state);
	ks->state = NULL;
    }
    
    free(ks, M_KQEMU);
    dev->si_drv1 = NULL;
    kqemu_ref_count--;
    kqemu_log("closed by pid=%d\n", p->p_pid);
    return(0);
}

int
kqemu_ioctl(dev, cmd, cmdarg, flags, p)
	dev_t		dev;
	unsigned long	cmd;
	caddr_t		cmdarg;
	int		flags;
	struct proc	*p;
{
    struct kqemu_instance *ks = (struct kqemu_instance *) dev->si_drv1;
    struct kqemu_state *s = ks->state;
    long ret;
    /*int error;*/

    switch (cmd) {
    case KQEMU_INIT:
    	/*kqemu_log("KQEMU_INIT data=0x%08x\n",cmdarg);*/
    {
        if (s)
	    return(EIO);

        if (!(s = kqemu_init((struct kqemu_init *)cmdarg, max_locked_pages)))
	    return(ENOMEM);

        ks->state = s;
	ret = 0;
	break;
    }
    case KQEMU_EXEC:
    	/*kqemu_log("KQEMU_EXEC data=0x%08x\n",cmdarg);*/
    {
        struct kqemu_cpu_state *ctx;
	
        if (!s)
	    return(EIO);
            
        ctx = kqemu_get_cpu_state(s);
	memcpy((void *)ctx, (void *)cmdarg, sizeof(struct kqemu_cpu_state));

        ret = kqemu_exec(s);

	memcpy((void *)cmdarg, (void *)ctx, sizeof(struct kqemu_cpu_state));

        break;
    }
    case KQEMU_GET_VERSION:
    	/*kqemu_log("KQEMU_GET_VERSION data=0x%08x\n",cmdarg);*/
    {
    	*(int *)cmdarg = KQEMU_VERSION;
	ret = 0;
	break;
    }
    default:
    	/*kqemu_log("ioctl unknown 0x%08x\n",cmd);*/
        return(ENXIO);
    }
    return(ret);
}

static int
init_module(void)
{
    int rc;
    
    printf("QEMU Accelerator Module version %d.%d.%d, Copyright (c) 2005 Fabrice Bellard\n"
	   "FreeBSD wrapper port, Copyright (c) 2005  Antony T Curtis\n"
           "This is a proprietary product. Read the LICENSE file for more information\n"
           "Redistribution of this module is prohibited without authorization\n",
           (KQEMU_VERSION >> 16),
           (KQEMU_VERSION >> 8) & 0xff,
           (KQEMU_VERSION) & 0xff);

    max_locked_pages = physmem / (2 * KQEMU_MAX_INSTANCES);
    if (max_locked_pages > 32768)
        max_locked_pages = 32768;
	
    if ((rc = cdevsw_add(&kqemu_cdevsw))) {
    	kqemu_log("error registering cdevsw, rc=%d\n", rc);
	return(ENOENT);
    }
	
    kqemu_dev = make_dev(&kqemu_cdevsw, 0, UID_ROOT, GID_WHEEL, 0600, "kqemu");

    kqemu_log("KQEMU installed, max_instances=%d max_locked_mem=%dkB.\n",
              KQEMU_MAX_INSTANCES, max_locked_pages * 4);

    kqemu_ref_count = 0;
    return 0;
}

static void
cleanup_module(void)
{
    int rc;
    
    destroy_dev(kqemu_dev);
    if ((rc = cdevsw_remove(&kqemu_cdevsw)))
	kqemu_log("error unregistering, rc=%d\n", rc);
}

static int
kqemu_modevent(module_t mod, int type, void *data)
{
    int err = 0;
    switch (type) {
    case MOD_LOAD:
    	err = init_module();
	break;
    case MOD_UNLOAD:
    	if (kqemu_ref_count > 0) {
	    err = EBUSY;
	    break;
	}
	/* fall through */
    case MOD_SHUTDOWN:
        cleanup_module();
    	break;
    default:
        err = EINVAL;
    	break;
    }
    return(err);
}

static moduledata_t kqemu_mod = {
	"kqemu_driver",
	kqemu_modevent,
	NULL
};

DECLARE_MODULE(kqemu, kqemu_mod, SI_SUB_DRIVERS, SI_ORDER_MIDDLE);