ocf-linux: version bump to 20110720
[15.05/openwrt.git] / target / linux / generic / files / crypto / ocf / ubsec_ssb / ubsec_ssb.c
diff --git a/target/linux/generic/files/crypto/ocf/ubsec_ssb/ubsec_ssb.c b/target/linux/generic/files/crypto/ocf/ubsec_ssb/ubsec_ssb.c
new file mode 100644 (file)
index 0000000..aa0733f
--- /dev/null
@@ -0,0 +1,2220 @@
+
+/*
+ * Copyright (c) 2008 Daniel Mueller (daniel@danm.de)
+ * Copyright (c) 2007 David McCullough (david_mccullough@securecomputing.com)
+ * Copyright (c) 2000 Jason L. Wright (jason@thought.net)
+ * Copyright (c) 2000 Theo de Raadt (deraadt@openbsd.org)
+ * Copyright (c) 2001 Patrik Lindergren (patrik@ipunplugged.com)
+ * 
+ * Redistribution and use in source and binary forms, with or without
+ * modification, are permitted provided that the following conditions
+ * are met:
+ * 1. Redistributions of source code must retain the above copyright
+ *    notice, this list of conditions and the following disclaimer.
+ * 2. Redistributions in binary form must reproduce the above copyright
+ *    notice, this list of conditions and the following disclaimer in the
+ *    documentation and/or other materials provided with the distribution.
+ *
+ * THIS SOFTWARE IS PROVIDED BY THE AUTHOR ``AS IS'' AND ANY EXPRESS OR
+ * IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE IMPLIED
+ * WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE ARE
+ * DISCLAIMED.  IN NO EVENT SHALL THE AUTHOR BE LIABLE FOR ANY DIRECT,
+ * INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES
+ * (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR
+ * SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
+ * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT,
+ * STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN
+ * ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE
+ * POSSIBILITY OF SUCH DAMAGE.
+ *
+ * Effort sponsored in part by the Defense Advanced Research Projects
+ * Agency (DARPA) and Air Force Research Laboratory, Air Force
+ * Materiel Command, USAF, under agreement number F30602-01-2-0537.
+ *
+ */
+#undef UBSEC_DEBUG
+#undef UBSEC_VERBOSE_DEBUG
+
+#ifdef UBSEC_VERBOSE_DEBUG
+#define UBSEC_DEBUG
+#endif
+
+/*
+ * uBsec BCM5365 hardware crypto accelerator
+ */
+
+#include <linux/kernel.h>
+#include <linux/module.h>
+#include <linux/moduleparam.h>
+#include <linux/proc_fs.h>
+#include <linux/types.h>
+#include <linux/init.h>
+#include <linux/delay.h>
+#include <linux/interrupt.h>
+#include <linux/fs.h>
+#include <linux/random.h>
+#include <linux/skbuff.h>
+#include <linux/stat.h>
+#include <asm/io.h>
+
+#include <linux/ssb/ssb.h>
+
+/*
+ * BSD queue
+ */
+//#include "bsdqueue.h"
+
+/* 
+ * OCF
+ */
+#include <cryptodev.h>
+#include <uio.h>
+
+#define HMAC_HACK 1
+
+#define HMAC_HACK 1
+#ifdef HMAC_HACK
+#include <safe/hmachack.h>
+#include <safe/md5.h>
+#include <safe/md5.c>
+#include <safe/sha1.h>
+#include <safe/sha1.c>
+#endif
+
+#include "bsdqueue.h"
+#include "ubsecreg.h"
+#include "ubsecvar.h"
+
+#define DRV_MODULE_NAME     "ubsec_ssb"
+#define PFX DRV_MODULE_NAME ": "
+#define DRV_MODULE_VERSION  "0.02"
+#define DRV_MODULE_RELDATE  "Feb 21, 2009"
+
+#if 1
+#define DPRINTF(a...) \
+    if (debug) \
+    { \
+        printk(DRV_MODULE_NAME ": " a); \
+    }
+#else
+#define DPRINTF(a...)
+#endif
+
+/*
+ * Prototypes 
+ */
+static irqreturn_t ubsec_ssb_isr(int, void *, struct pt_regs *);
+static int __devinit ubsec_ssb_probe(struct ssb_device *sdev,
+    const struct ssb_device_id *ent);
+static void __devexit ubsec_ssb_remove(struct ssb_device *sdev);
+int ubsec_attach(struct ssb_device *sdev, const struct ssb_device_id *ent, 
+    struct device *self);
+static void ubsec_setup_mackey(struct ubsec_session *ses, int algo, 
+    caddr_t key, int klen);
+static int dma_map_skb(struct ubsec_softc *sc, 
+    struct ubsec_dma_alloc* q_map, struct sk_buff *skb, int *mlen);
+static int dma_map_uio(struct ubsec_softc *sc, 
+    struct ubsec_dma_alloc *q_map, struct uio *uio, int *mlen);
+static void dma_unmap(struct ubsec_softc *sc, 
+    struct ubsec_dma_alloc *q_map, int mlen);
+static int ubsec_dmamap_aligned(struct ubsec_softc *sc, 
+    const struct ubsec_dma_alloc *q_map, int mlen);
+
+#ifdef UBSEC_DEBUG
+static int proc_read(char *buf, char **start, off_t offset,
+    int size, int *peof, void *data);
+#endif
+
+void ubsec_reset_board(struct ubsec_softc *);
+void ubsec_init_board(struct ubsec_softc *);
+void ubsec_cleanchip(struct ubsec_softc *);
+void ubsec_totalreset(struct ubsec_softc *);
+int  ubsec_free_q(struct ubsec_softc*, struct ubsec_q *);
+
+static int ubsec_newsession(device_t, u_int32_t *, struct cryptoini *);
+static int ubsec_freesession(device_t, u_int64_t);
+static int ubsec_process(device_t, struct cryptop *, int);
+
+void    ubsec_callback(struct ubsec_softc *, struct ubsec_q *);
+void    ubsec_feed(struct ubsec_softc *);
+void    ubsec_mcopy(struct sk_buff *, struct sk_buff *, int, int);
+void    ubsec_dma_free(struct ubsec_softc *, struct ubsec_dma_alloc *);
+int     ubsec_dma_malloc(struct ubsec_softc *, struct ubsec_dma_alloc *,
+        size_t, int);
+
+/* DEBUG crap... */
+void ubsec_dump_pb(struct ubsec_pktbuf *);
+void ubsec_dump_mcr(struct ubsec_mcr *);
+
+#define READ_REG(sc,r) \
+    ssb_read32((sc)->sdev, (r));
+#define WRITE_REG(sc,r,val) \
+    ssb_write32((sc)->sdev, (r), (val));
+#define READ_REG_SDEV(sdev,r) \
+    ssb_read32((sdev), (r));
+#define WRITE_REG_SDEV(sdev,r,val) \
+    ssb_write32((sdev), (r), (val));
+
+#define SWAP32(x) (x) = htole32(ntohl((x)))
+#define HTOLE32(x) (x) = htole32(x)
+
+#ifdef __LITTLE_ENDIAN
+#define letoh16(x) (x)
+#define letoh32(x) (x)
+#endif
+
+static int debug;
+module_param(debug, int, 0644);
+MODULE_PARM_DESC(debug, "Enable debug output");
+
+#define UBSEC_SSB_MAX_CHIPS 1
+static struct ubsec_softc *ubsec_chip_idx[UBSEC_SSB_MAX_CHIPS];
+static struct ubsec_stats ubsecstats;
+
+#ifdef UBSEC_DEBUG
+static struct proc_dir_entry *procdebug;
+#endif
+
+static struct ssb_device_id ubsec_ssb_tbl[] = {
+    /* Broadcom BCM5365P IPSec Core */
+    SSB_DEVICE(SSB_VENDOR_BROADCOM, SSB_DEV_IPSEC, SSB_ANY_REV),
+    SSB_DEVTABLE_END
+};
+
+static struct ssb_driver ubsec_ssb_driver = {
+    .name       = DRV_MODULE_NAME,
+    .id_table   = ubsec_ssb_tbl,
+    .probe      = ubsec_ssb_probe,
+    .remove     = __devexit_p(ubsec_ssb_remove),
+     /*
+    .suspend    = ubsec_ssb_suspend,
+    .resume     = ubsec_ssb_resume
+    */
+};
+
+static device_method_t ubsec_ssb_methods = {
+    /* crypto device methods */
+    DEVMETHOD(cryptodev_newsession, ubsec_newsession),
+    DEVMETHOD(cryptodev_freesession,ubsec_freesession),
+    DEVMETHOD(cryptodev_process,    ubsec_process),
+};
+
+#ifdef UBSEC_DEBUG
+static int 
+proc_read(char *buf, char **start, off_t offset,
+    int size, int *peof, void *data)
+{
+    int i = 0, byteswritten = 0, ret;
+    unsigned int stat, ctrl;
+#ifdef UBSEC_VERBOSE_DEBUG
+    struct ubsec_q *q;
+    struct ubsec_dma *dmap;
+#endif
+   
+    while ((i < UBSEC_SSB_MAX_CHIPS) && (ubsec_chip_idx[i] != NULL))
+    {
+        struct ubsec_softc *sc = ubsec_chip_idx[i];
+        
+        stat = READ_REG(sc, BS_STAT);
+        ctrl = READ_REG(sc, BS_CTRL);
+        ret = snprintf((buf + byteswritten), 
+            (size - byteswritten) , 
+            "DEV %d, DMASTAT %08x, DMACTRL %08x\n", i, stat, ctrl);
+
+        byteswritten += ret;
+
+#ifdef UBSEC_VERBOSE_DEBUG
+        printf("DEV %d, DMASTAT %08x, DMACTRL %08x\n", i, stat, ctrl);
+
+        /* Dump all queues MCRs */
+        if (!BSD_SIMPLEQ_EMPTY(&sc->sc_qchip)) {
+            BSD_SIMPLEQ_FOREACH(q, &sc->sc_qchip, q_next)
+            {
+                dmap = q->q_dma;
+                ubsec_dump_mcr(&dmap->d_dma->d_mcr);
+            }
+        }
+#endif
+
+        i++;
+    }
+
+    *peof = 1;
+
+    return byteswritten;
+}
+#endif
+
+/*
+ * map in a given sk_buff
+ */
+static int
+dma_map_skb(struct ubsec_softc *sc, struct ubsec_dma_alloc* q_map, struct sk_buff *skb, int *mlen)
+{
+    int i = 0;
+    dma_addr_t tmp;
+
+#ifdef UBSEC_DEBUG
+    DPRINTF("%s()\n", __FUNCTION__);
+#endif
+
+    /*
+     * We support only a limited number of fragments.
+     */
+    if (unlikely((skb_shinfo(skb)->nr_frags + 1) >= UBS_MAX_SCATTER))
+    {
+        printk(KERN_ERR "Only %d scatter fragments are supported.\n", UBS_MAX_SCATTER);
+        return (-ENOMEM);
+    }
+
+#ifdef UBSEC_VERBOSE_DEBUG
+    DPRINTF("%s - map %d 0x%x %d\n", __FUNCTION__, 0, (unsigned int)skb->data, skb_headlen(skb));
+#endif
+
+    /* first data package */
+    tmp = dma_map_single(sc->sc_dv,
+                         skb->data,
+                         skb_headlen(skb),
+                         DMA_BIDIRECTIONAL);
+    
+    q_map[i].dma_paddr = tmp;
+    q_map[i].dma_vaddr = skb->data;
+    q_map[i].dma_size = skb_headlen(skb);
+
+    if (unlikely(tmp == 0))
+    {
+        printk(KERN_ERR "Could not map memory region for dma.\n");
+        return (-EINVAL);
+    }
+
+#ifdef UBSEC_VERBOSE_DEBUG
+    DPRINTF("%s - map %d done physical addr 0x%x\n", __FUNCTION__, 0, (unsigned int)tmp);
+#endif
+
+
+    /* all other data packages */    
+    for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) {
+
+#ifdef UBSEC_VERBOSE_DEBUG
+        DPRINTF("%s - map %d 0x%x %d\n", __FUNCTION__, i + 1, 
+            (unsigned int)page_address(skb_shinfo(skb)->frags[i].page) +
+            skb_shinfo(skb)->frags[i].page_offset, skb_shinfo(skb)->frags[i].size);
+#endif
+
+        tmp = dma_map_single(sc->sc_dv,
+                             page_address(skb_shinfo(skb)->frags[i].page) +
+                                 skb_shinfo(skb)->frags[i].page_offset, 
+                             skb_shinfo(skb)->frags[i].size,
+                             DMA_BIDIRECTIONAL);
+
+        q_map[i + 1].dma_paddr = tmp;
+        q_map[i + 1].dma_vaddr = (void*)(page_address(skb_shinfo(skb)->frags[i].page) +
+                                  skb_shinfo(skb)->frags[i].page_offset);
+        q_map[i + 1].dma_size = skb_shinfo(skb)->frags[i].size;
+
+        if (unlikely(tmp == 0))
+        {
+            printk(KERN_ERR "Could not map memory region for dma.\n");
+            return (-EINVAL);
+        }
+
+#ifdef UBSEC_VERBOSE_DEBUG
+        DPRINTF("%s - map %d done physical addr 0x%x\n", __FUNCTION__, i + 1, (unsigned int)tmp);
+#endif
+
+    }
+    *mlen = i + 1;
+
+    return(0);
+}
+
+/*
+ * map in a given uio buffer
+ */
+
+static int
+dma_map_uio(struct ubsec_softc *sc, struct ubsec_dma_alloc *q_map, struct uio *uio, int *mlen)
+{
+    struct iovec *iov = uio->uio_iov;
+    int n;
+    dma_addr_t tmp;
+
+#ifdef UBSEC_DEBUG
+    DPRINTF("%s()\n", __FUNCTION__);
+#endif
+
+    /*
+     * We support only a limited number of fragments.
+     */
+    if (unlikely(uio->uio_iovcnt >= UBS_MAX_SCATTER))
+    {
+        printk(KERN_ERR "Only %d scatter fragments are supported.\n", UBS_MAX_SCATTER);
+        return (-ENOMEM);
+    }
+
+    for (n = 0; n < uio->uio_iovcnt; n++) {
+#ifdef UBSEC_VERBOSE_DEBUG
+        DPRINTF("%s - map %d 0x%x %d\n", __FUNCTION__, n, (unsigned int)iov->iov_base, iov->iov_len);
+#endif
+        tmp = dma_map_single(sc->sc_dv,
+                             iov->iov_base,
+                             iov->iov_len,
+                             DMA_BIDIRECTIONAL);
+
+        q_map[n].dma_paddr = tmp;
+        q_map[n].dma_vaddr = iov->iov_base;
+        q_map[n].dma_size = iov->iov_len;
+
+        if (unlikely(tmp == 0))
+                       {
+            printk(KERN_ERR "Could not map memory region for dma.\n");
+            return (-EINVAL);
+        }
+
+#ifdef UBSEC_VERBOSE_DEBUG
+        DPRINTF("%s - map %d done physical addr 0x%x\n", __FUNCTION__, n, (unsigned int)tmp);
+#endif
+
+        iov++;
+    }
+    *mlen = n;
+
+    return(0);
+}
+
+static void
+dma_unmap(struct ubsec_softc *sc, struct ubsec_dma_alloc *q_map, int mlen)
+{
+    int i;
+
+#ifdef UBSEC_DEBUG
+    DPRINTF("%s()\n", __FUNCTION__);
+#endif
+
+    for(i = 0; i < mlen; i++)
+    {
+#ifdef UBSEC_VERBOSE_DEBUG
+        DPRINTF("%s - unmap %d 0x%x %d\n", __FUNCTION__, i, (unsigned int)q_map[i].dma_paddr, q_map[i].dma_size);
+#endif
+        dma_unmap_single(sc->sc_dv,
+                         q_map[i].dma_paddr,
+                         q_map[i].dma_size,
+                         DMA_BIDIRECTIONAL);
+    }
+    return;
+}
+
+/*
+ * Is the operand suitable aligned for direct DMA.  Each
+ * segment must be aligned on a 32-bit boundary and all
+ * but the last segment must be a multiple of 4 bytes.
+ */
+static int
+ubsec_dmamap_aligned(struct ubsec_softc *sc, const struct ubsec_dma_alloc *q_map, int mlen)
+{
+    int i;
+
+#ifdef UBSEC_DEBUG
+    DPRINTF("%s()\n", __FUNCTION__);
+#endif
+
+    for (i = 0; i < mlen; i++) {
+        if (q_map[i].dma_paddr & 3)
+            return (0);
+        if (i != (mlen - 1) && (q_map[i].dma_size & 3))
+            return (0);
+    }
+    return (1);
+}
+
+
+#define N(a)    (sizeof(a) / sizeof (a[0]))
+static void
+ubsec_setup_mackey(struct ubsec_session *ses, int algo, caddr_t key, int klen)
+{
+#ifdef HMAC_HACK
+    MD5_CTX md5ctx;
+    SHA1_CTX sha1ctx;
+    int i;
+
+#ifdef UBSEC_DEBUG
+    DPRINTF("%s()\n", __FUNCTION__);
+#endif
+
+    for (i = 0; i < klen; i++)
+        key[i] ^= HMAC_IPAD_VAL;
+
+    if (algo == CRYPTO_MD5_HMAC) {
+        MD5Init(&md5ctx);
+        MD5Update(&md5ctx, key, klen);
+        MD5Update(&md5ctx, hmac_ipad_buffer, MD5_HMAC_BLOCK_LEN - klen);
+        bcopy(md5ctx.md5_st8, ses->ses_hminner, sizeof(md5ctx.md5_st8));
+    } else {
+        SHA1Init(&sha1ctx);
+        SHA1Update(&sha1ctx, key, klen);
+        SHA1Update(&sha1ctx, hmac_ipad_buffer,
+            SHA1_HMAC_BLOCK_LEN - klen);
+        bcopy(sha1ctx.h.b32, ses->ses_hminner, sizeof(sha1ctx.h.b32));
+    }
+
+    for (i = 0; i < klen; i++)
+        key[i] ^= (HMAC_IPAD_VAL ^ HMAC_OPAD_VAL);
+
+    if (algo == CRYPTO_MD5_HMAC) {
+        MD5Init(&md5ctx);
+        MD5Update(&md5ctx, key, klen);
+        MD5Update(&md5ctx, hmac_opad_buffer, MD5_HMAC_BLOCK_LEN - klen);
+        bcopy(md5ctx.md5_st8, ses->ses_hmouter, sizeof(md5ctx.md5_st8));
+    } else {
+        SHA1Init(&sha1ctx);
+        SHA1Update(&sha1ctx, key, klen);
+        SHA1Update(&sha1ctx, hmac_opad_buffer,
+            SHA1_HMAC_BLOCK_LEN - klen);
+        bcopy(sha1ctx.h.b32, ses->ses_hmouter, sizeof(sha1ctx.h.b32));
+    }
+
+    for (i = 0; i < klen; i++)
+        key[i] ^= HMAC_OPAD_VAL;
+
+#else /* HMAC_HACK */
+    DPRINTF("md5/sha not implemented\n");
+#endif /* HMAC_HACK */
+}
+#undef N
+
+static int 
+__devinit ubsec_ssb_probe(struct ssb_device *sdev, 
+    const struct ssb_device_id *ent) 
+{
+    int err;
+
+#ifdef UBSEC_DEBUG
+    DPRINTF("%s()\n", __FUNCTION__);
+#endif
+
+    err = ssb_bus_powerup(sdev->bus, 0);
+    if (err) {
+        dev_err(sdev->dev, "Failed to powerup the bus\n");
+       goto err_out;
+    }
+
+    err = request_irq(sdev->irq, (irq_handler_t)ubsec_ssb_isr, 
+        IRQF_DISABLED | IRQF_SHARED, DRV_MODULE_NAME, sdev);
+    if (err) {
+        dev_err(sdev->dev, "Could not request irq\n");
+        goto err_out_powerdown;
+    }
+
+#if (LINUX_VERSION_CODE >= KERNEL_VERSION(2,6,36))
+    err = dma_set_mask(sdev->dma_dev, DMA_BIT_MASK(32)) || 
+         dma_set_coherent_mask(sdev->dma_dev, DMA_BIT_MASK(32));
+#else
+    err = ssb_dma_set_mask(sdev, DMA_32BIT_MASK);
+#endif
+    if (err) {
+        dev_err(sdev->dev,
+        "Required 32BIT DMA mask unsupported by the system.\n");
+        goto err_out_free_irq;
+    }
+
+    printk(KERN_INFO "Sentry5(tm) ROBOGateway(tm) IPSec Core at IRQ %u\n",
+        sdev->irq);
+
+    DPRINTF("Vendor: %x, core id: %x, revision: %x\n",
+        sdev->id.vendor, sdev->id.coreid, sdev->id.revision);
+
+    ssb_device_enable(sdev, 0);
+
+    if (ubsec_attach(sdev, ent, sdev->dev) != 0)
+        goto err_out_disable;
+
+#ifdef UBSEC_DEBUG
+    procdebug = create_proc_entry(DRV_MODULE_NAME, S_IRUSR, NULL);
+    if (procdebug)
+    {
+        procdebug->read_proc = proc_read;
+        procdebug->data = NULL;
+    } else 
+        DPRINTF("Unable to create proc file.\n");
+#endif
+
+    return 0;
+
+err_out_disable:
+    ssb_device_disable(sdev, 0);
+
+err_out_free_irq:
+    free_irq(sdev->irq, sdev);
+
+err_out_powerdown:
+    ssb_bus_may_powerdown(sdev->bus);
+
+err_out:
+    return err;
+}
+
+static void __devexit ubsec_ssb_remove(struct ssb_device *sdev) {
+
+    struct ubsec_softc *sc;
+    unsigned int ctrlflgs;
+    struct ubsec_dma *dmap;
+    u_int32_t i;
+
+#ifdef UBSEC_DEBUG
+    DPRINTF("%s()\n", __FUNCTION__);
+#endif
+
+    ctrlflgs = READ_REG_SDEV(sdev, BS_CTRL);
+    /* disable all IPSec Core interrupts globally */
+    ctrlflgs ^= (BS_CTRL_MCR1INT | BS_CTRL_MCR2INT |
+        BS_CTRL_DMAERR);
+    WRITE_REG_SDEV(sdev, BS_CTRL, ctrlflgs);
+
+    free_irq(sdev->irq, sdev);
+
+    sc = (struct ubsec_softc *)ssb_get_drvdata(sdev);
+
+    /* unregister all crypto algorithms */
+    crypto_unregister_all(sc->sc_cid);
+
+    /* Free queue / dma memory */
+    for (i = 0; i < UBS_MAX_NQUEUE; i++) {
+        struct ubsec_q *q;
+
+        q = sc->sc_queuea[i];
+        if (q != NULL)
+        {
+            dmap = q->q_dma;
+            if (dmap != NULL)
+            {
+                ubsec_dma_free(sc, &dmap->d_alloc);
+                q->q_dma = NULL;
+            }
+            kfree(q);
+        }
+        sc->sc_queuea[i] = NULL;
+    }
+
+    ssb_device_disable(sdev, 0);
+    ssb_bus_may_powerdown(sdev->bus);
+    ssb_set_drvdata(sdev, NULL);
+
+#ifdef UBSEC_DEBUG
+    if (procdebug)
+        remove_proc_entry(DRV_MODULE_NAME, NULL);
+#endif
+
+}
+
+
+int
+ubsec_attach(struct ssb_device *sdev, const struct ssb_device_id *ent, 
+    struct device *self)
+{
+    struct ubsec_softc *sc = NULL;
+    struct ubsec_dma *dmap;
+    u_int32_t i;
+    static int num_chips = 0;
+
+#ifdef UBSEC_DEBUG
+    DPRINTF("%s()\n", __FUNCTION__);
+#endif
+
+    sc = (struct ubsec_softc *) kmalloc(sizeof(*sc), GFP_KERNEL);
+    if (!sc)
+        return(-ENOMEM);
+    memset(sc, 0, sizeof(*sc));
+
+    sc->sc_dv = sdev->dev;
+    sc->sdev = sdev;
+
+    spin_lock_init(&sc->sc_ringmtx);
+
+    softc_device_init(sc, "ubsec_ssb", num_chips, ubsec_ssb_methods);
+
+    /* Maybe someday there are boards with more than one chip available */
+    if (num_chips < UBSEC_SSB_MAX_CHIPS) {
+        ubsec_chip_idx[device_get_unit(sc->sc_dev)] = sc;
+        num_chips++;
+    }
+
+    ssb_set_drvdata(sdev, sc);
+
+    BSD_SIMPLEQ_INIT(&sc->sc_queue);
+    BSD_SIMPLEQ_INIT(&sc->sc_qchip);
+    BSD_SIMPLEQ_INIT(&sc->sc_queue2);
+    BSD_SIMPLEQ_INIT(&sc->sc_qchip2);
+    BSD_SIMPLEQ_INIT(&sc->sc_q2free);
+
+    sc->sc_statmask = BS_STAT_MCR1_DONE | BS_STAT_DMAERR;
+
+    sc->sc_cid = crypto_get_driverid(softc_get_device(sc), CRYPTOCAP_F_HARDWARE);
+    if (sc->sc_cid < 0) {
+        device_printf(sc->sc_dev, "could not get crypto driver id\n");
+        return -1;
+    }
+
+    BSD_SIMPLEQ_INIT(&sc->sc_freequeue);
+    dmap = sc->sc_dmaa;
+    for (i = 0; i < UBS_MAX_NQUEUE; i++, dmap++) {
+        struct ubsec_q *q;
+
+        q = (struct ubsec_q *)kmalloc(sizeof(struct ubsec_q), GFP_KERNEL);
+        if (q == NULL) {
+            printf(": can't allocate queue buffers\n");
+            break;
+        }
+
+        if (ubsec_dma_malloc(sc, &dmap->d_alloc, sizeof(struct ubsec_dmachunk),0)) {
+            printf(": can't allocate dma buffers\n");
+            kfree(q);
+            break;
+        }
+        dmap->d_dma = (struct ubsec_dmachunk *)dmap->d_alloc.dma_vaddr;
+
+        q->q_dma = dmap;
+        sc->sc_queuea[i] = q;
+
+        BSD_SIMPLEQ_INSERT_TAIL(&sc->sc_freequeue, q, q_next);
+    }
+
+    /*
+     * Reset Broadcom chip
+     */
+    ubsec_reset_board(sc);
+
+    /*
+     * Init Broadcom chip
+     */
+    ubsec_init_board(sc);
+
+    /* supported crypto algorithms */
+    crypto_register(sc->sc_cid, CRYPTO_3DES_CBC, 0, 0);
+    crypto_register(sc->sc_cid, CRYPTO_DES_CBC, 0, 0);
+
+    if (sc->sc_flags & UBS_FLAGS_AES) {
+        crypto_register(sc->sc_cid, CRYPTO_AES_CBC, 0, 0);
+        printf(KERN_INFO DRV_MODULE_NAME ": DES 3DES AES128 AES192 AES256 MD5_HMAC SHA1_HMAC\n");
+    }
+    else
+        printf(KERN_INFO DRV_MODULE_NAME ": DES 3DES MD5_HMAC SHA1_HMAC\n");
+
+    crypto_register(sc->sc_cid, CRYPTO_MD5_HMAC, 0, 0);
+    crypto_register(sc->sc_cid, CRYPTO_SHA1_HMAC, 0, 0);
+
+    return 0;
+}
+
+/*
+ * UBSEC Interrupt routine
+ */
+static irqreturn_t 
+ubsec_ssb_isr(int irq, void *arg, struct pt_regs *regs) 
+{
+    struct ubsec_softc *sc = NULL;
+    volatile u_int32_t stat;
+    struct ubsec_q *q;
+    struct ubsec_dma *dmap;
+    int npkts = 0, i;
+
+#ifdef UBSEC_VERBOSE_DEBUG
+    DPRINTF("%s()\n", __FUNCTION__);
+#endif
+
+    sc = (struct ubsec_softc *)ssb_get_drvdata(arg);
+
+    stat = READ_REG(sc, BS_STAT);
+
+    stat &= sc->sc_statmask;
+    if (stat == 0)
+        return IRQ_NONE;
+
+    WRITE_REG(sc, BS_STAT, stat);       /* IACK */
+
+    /*
+     * Check to see if we have any packets waiting for us
+     */
+    if ((stat & BS_STAT_MCR1_DONE)) {
+        while (!BSD_SIMPLEQ_EMPTY(&sc->sc_qchip)) {
+            q = BSD_SIMPLEQ_FIRST(&sc->sc_qchip);
+            dmap = q->q_dma;
+
+            if ((dmap->d_dma->d_mcr.mcr_flags & htole16(UBS_MCR_DONE)) == 0)
+            {
+                DPRINTF("error while processing MCR. Flags = %x\n", dmap->d_dma->d_mcr.mcr_flags);
+                break;
+            }
+
+            BSD_SIMPLEQ_REMOVE_HEAD(&sc->sc_qchip, q_next);
+
+            npkts = q->q_nstacked_mcrs;
+            /*
+             * search for further sc_qchip ubsec_q's that share
+             * the same MCR, and complete them too, they must be
+             * at the top.
+             */
+            for (i = 0; i < npkts; i++) {
+                if(q->q_stacked_mcr[i])
+                    ubsec_callback(sc, q->q_stacked_mcr[i]);
+                else
+                    break;
+            }
+            ubsec_callback(sc, q);
+        }
+
+        /*
+         * Don't send any more packet to chip if there has been
+         * a DMAERR.
+         */
+        if (likely(!(stat & BS_STAT_DMAERR)))
+            ubsec_feed(sc);
+        else
+            DPRINTF("DMA error occurred. Stop feeding crypto chip.\n");
+    }
+
+    /*
+     * Check to see if we got any DMA Error
+     */
+    if (stat & BS_STAT_DMAERR) {
+        volatile u_int32_t a = READ_REG(sc, BS_ERR);
+
+        printf(KERN_ERR "%s: dmaerr %s@%08x\n", DRV_MODULE_NAME,
+            (a & BS_ERR_READ) ? "read" : "write", a & BS_ERR_ADDR);
+
+        ubsecstats.hst_dmaerr++;
+        ubsec_totalreset(sc);
+        ubsec_feed(sc);
+    }
+
+    return IRQ_HANDLED;
+}
+
+/*
+ * ubsec_feed() - aggregate and post requests to chip
+ *        It is assumed that the caller set splnet()
+ */
+void
+ubsec_feed(struct ubsec_softc *sc)
+{
+#ifdef UBSEC_VERBOSE_DEBUG
+    static int max;
+#endif 
+    struct ubsec_q *q, *q2;
+    int npkts, i;
+    void *v;
+    u_int32_t stat;
+
+    npkts = sc->sc_nqueue;
+    if (npkts > UBS_MAX_AGGR)
+        npkts = UBS_MAX_AGGR;
+    if (npkts < 2)
+        goto feed1;
+
+    stat = READ_REG(sc, BS_STAT);
+
+    if (stat & (BS_STAT_MCR1_FULL | BS_STAT_DMAERR)) {
+        if(stat & BS_STAT_DMAERR) {
+            ubsec_totalreset(sc);
+            ubsecstats.hst_dmaerr++;
+        }
+        return;
+    }
+
+#ifdef UBSEC_VERBOSE_DEBUG
+    DPRINTF("merging %d records\n", npkts);
+
+    /* XXX temporary aggregation statistics reporting code */
+    if (max < npkts) {
+        max = npkts;
+        DPRINTF("%s: new max aggregate %d\n", DRV_MODULE_NAME, max);
+    }
+#endif /* UBSEC_VERBOSE_DEBUG */
+
+    q = BSD_SIMPLEQ_FIRST(&sc->sc_queue);
+    BSD_SIMPLEQ_REMOVE_HEAD(&sc->sc_queue, q_next);
+    --sc->sc_nqueue;
+
+#if 0
+    /* 
+     * XXX 
+     * We use dma_map_single() - no sync required!
+     */
+
+    bus_dmamap_sync(sc->sc_dmat, q->q_src_map,
+        0, q->q_src_map->dm_mapsize, BUS_DMASYNC_PREWRITE);
+    if (q->q_dst_map != NULL)
+        bus_dmamap_sync(sc->sc_dmat, q->q_dst_map,
+            0, q->q_dst_map->dm_mapsize, BUS_DMASYNC_PREREAD);
+#endif
+
+    q->q_nstacked_mcrs = npkts - 1;     /* Number of packets stacked */
+
+    for (i = 0; i < q->q_nstacked_mcrs; i++) {
+        q2 = BSD_SIMPLEQ_FIRST(&sc->sc_queue);
+
+#if 0
+        bus_dmamap_sync(sc->sc_dmat, q2->q_src_map,
+            0, q2->q_src_map->dm_mapsize, BUS_DMASYNC_PREWRITE);
+        if (q2->q_dst_map != NULL)
+            bus_dmamap_sync(sc->sc_dmat, q2->q_dst_map,
+                0, q2->q_dst_map->dm_mapsize, BUS_DMASYNC_PREREAD);
+#endif
+        BSD_SIMPLEQ_REMOVE_HEAD(&sc->sc_queue, q_next);
+        --sc->sc_nqueue;
+
+        v = ((char *)&q2->q_dma->d_dma->d_mcr) + sizeof(struct ubsec_mcr) -
+            sizeof(struct ubsec_mcr_add);
+        bcopy(v, &q->q_dma->d_dma->d_mcradd[i], sizeof(struct ubsec_mcr_add));
+        q->q_stacked_mcr[i] = q2;
+    }
+    q->q_dma->d_dma->d_mcr.mcr_pkts = htole16(npkts);
+    BSD_SIMPLEQ_INSERT_TAIL(&sc->sc_qchip, q, q_next);
+#if 0
+    bus_dmamap_sync(sc->sc_dmat, q->q_dma->d_alloc.dma_map,
+        0, q->q_dma->d_alloc.dma_map->dm_mapsize,
+        BUS_DMASYNC_PREREAD | BUS_DMASYNC_PREWRITE);
+#endif
+    WRITE_REG(sc, BS_MCR1, q->q_dma->d_alloc.dma_paddr +
+        offsetof(struct ubsec_dmachunk, d_mcr));
+#ifdef UBSEC_VERBOSE_DEBUG
+    DPRINTF("feed (1): q->chip %p %08x %08x\n", q,
+        (u_int32_t)q->q_dma->d_alloc.dma_paddr,
+        (u_int32_t)(q->q_dma->d_alloc.dma_paddr +
+        offsetof(struct ubsec_dmachunk, d_mcr)));
+#endif /* UBSEC_DEBUG */
+    return;
+
+feed1:
+    while (!BSD_SIMPLEQ_EMPTY(&sc->sc_queue)) {
+        stat = READ_REG(sc, BS_STAT);
+
+        if (stat & (BS_STAT_MCR1_FULL | BS_STAT_DMAERR)) {
+            if(stat & BS_STAT_DMAERR) {
+                ubsec_totalreset(sc);
+                ubsecstats.hst_dmaerr++;
+            }
+            break;
+        }
+
+        q = BSD_SIMPLEQ_FIRST(&sc->sc_queue);
+
+#if 0
+        bus_dmamap_sync(sc->sc_dmat, q->q_src_map,
+            0, q->q_src_map->dm_mapsize, BUS_DMASYNC_PREWRITE);
+        if (q->q_dst_map != NULL)
+            bus_dmamap_sync(sc->sc_dmat, q->q_dst_map,
+                0, q->q_dst_map->dm_mapsize, BUS_DMASYNC_PREREAD);
+        bus_dmamap_sync(sc->sc_dmat, q->q_dma->d_alloc.dma_map,
+            0, q->q_dma->d_alloc.dma_map->dm_mapsize,
+            BUS_DMASYNC_PREREAD | BUS_DMASYNC_PREWRITE);
+#endif
+
+        WRITE_REG(sc, BS_MCR1, q->q_dma->d_alloc.dma_paddr +
+            offsetof(struct ubsec_dmachunk, d_mcr));
+#ifdef UBSEC_VERBOSE_DEBUG
+        DPRINTF("feed (2): q->chip %p %08x %08x\n", q, 
+            (u_int32_t)q->q_dma->d_alloc.dma_paddr,
+            (u_int32_t)(q->q_dma->d_alloc.dma_paddr +
+            offsetof(struct ubsec_dmachunk, d_mcr)));
+#endif /* UBSEC_DEBUG */
+        BSD_SIMPLEQ_REMOVE_HEAD(&sc->sc_queue, q_next);
+        --sc->sc_nqueue;
+        BSD_SIMPLEQ_INSERT_TAIL(&sc->sc_qchip, q, q_next);
+    }
+}
+
+/*
+ * Allocate a new 'session' and return an encoded session id.  'sidp'
+ * contains our registration id, and should contain an encoded session
+ * id on successful allocation.
+ */
+static int
+ubsec_newsession(device_t dev, u_int32_t *sidp, struct cryptoini *cri)
+{
+    struct cryptoini *c, *encini = NULL, *macini = NULL;
+    struct ubsec_softc *sc = NULL;
+    struct ubsec_session *ses = NULL;
+    int sesn, i;
+
+#ifdef UBSEC_DEBUG
+    DPRINTF("%s()\n", __FUNCTION__);
+#endif
+
+    if (sidp == NULL || cri == NULL)
+        return (EINVAL);
+
+    sc = device_get_softc(dev);
+
+    if (sc == NULL)
+        return (EINVAL);
+
+    for (c = cri; c != NULL; c = c->cri_next) {
+        if (c->cri_alg == CRYPTO_MD5_HMAC ||
+            c->cri_alg == CRYPTO_SHA1_HMAC) {
+            if (macini)
+                return (EINVAL);
+            macini = c;
+        } else if (c->cri_alg == CRYPTO_DES_CBC ||
+            c->cri_alg == CRYPTO_3DES_CBC ||
+            c->cri_alg == CRYPTO_AES_CBC) {
+            if (encini)
+                return (EINVAL);
+            encini = c;
+        } else
+            return (EINVAL);
+    }
+    if (encini == NULL && macini == NULL)
+        return (EINVAL);
+
+    if (sc->sc_sessions == NULL) {
+        ses = sc->sc_sessions = (struct ubsec_session *)kmalloc(
+            sizeof(struct ubsec_session), SLAB_ATOMIC);
+        if (ses == NULL)
+            return (ENOMEM);
+        memset(ses, 0, sizeof(struct ubsec_session));
+        sesn = 0;
+        sc->sc_nsessions = 1;
+    } else {
+        for (sesn = 0; sesn < sc->sc_nsessions; sesn++) {
+            if (sc->sc_sessions[sesn].ses_used == 0) {
+                ses = &sc->sc_sessions[sesn];
+                break;
+            }
+        }
+
+        if (ses == NULL) {
+            sesn = sc->sc_nsessions;
+            ses = (struct ubsec_session *)kmalloc((sesn + 1) *
+                sizeof(struct ubsec_session), SLAB_ATOMIC);
+            if (ses == NULL)
+                return (ENOMEM);
+            memset(ses, 0, (sesn + 1) * sizeof(struct ubsec_session));
+            bcopy(sc->sc_sessions, ses, sesn *
+                sizeof(struct ubsec_session));
+            bzero(sc->sc_sessions, sesn *
+                sizeof(struct ubsec_session));
+            kfree(sc->sc_sessions);
+            sc->sc_sessions = ses;
+            ses = &sc->sc_sessions[sesn];
+            sc->sc_nsessions++;
+        }
+    }
+
+    bzero(ses, sizeof(struct ubsec_session));
+    ses->ses_used = 1;
+    if (encini) {
+        /* get an IV */
+        /* XXX may read fewer than requested */
+        read_random(ses->ses_iv, sizeof(ses->ses_iv));
+
+        /* Go ahead and compute key in ubsec's byte order */
+        if (encini->cri_alg == CRYPTO_DES_CBC) {
+            /* DES uses the same key three times:
+             * 1st encrypt -> 2nd decrypt -> 3nd encrypt */
+            bcopy(encini->cri_key, &ses->ses_key[0], 8);
+            bcopy(encini->cri_key, &ses->ses_key[2], 8);
+            bcopy(encini->cri_key, &ses->ses_key[4], 8);
+            ses->ses_keysize = 192; /* Fake! Actually its only 64bits .. 
+                                       oh no it is even less: 54bits. */
+        } else if(encini->cri_alg == CRYPTO_3DES_CBC) {
+            bcopy(encini->cri_key, ses->ses_key, 24);
+            ses->ses_keysize = 192;
+        } else if(encini->cri_alg == CRYPTO_AES_CBC) {
+            ses->ses_keysize = encini->cri_klen;
+
+            if (ses->ses_keysize != 128 &&
+                ses->ses_keysize != 192 &&
+                ses->ses_keysize != 256)
+            {
+                DPRINTF("unsupported AES key size: %d\n", ses->ses_keysize);
+                return (EINVAL);
+            }
+            bcopy(encini->cri_key, ses->ses_key, (ses->ses_keysize / 8));
+        }
+
+        /* Hardware requires the keys in little endian byte order */
+        for (i=0; i < (ses->ses_keysize / 32); i++)
+            SWAP32(ses->ses_key[i]);
+    }
+
+    if (macini) {
+        ses->ses_mlen = macini->cri_mlen;
+
+        if (ses->ses_mlen == 0 ||
+            ses->ses_mlen > SHA1_HASH_LEN) {
+
+            if (macini->cri_alg == CRYPTO_MD5_HMAC ||
+                macini->cri_alg == CRYPTO_SHA1_HMAC)
+            {
+                ses->ses_mlen = DEFAULT_HMAC_LEN;
+            } else
+            {
+                /*
+                 * Reserved for future usage. MD5/SHA1 calculations have
+                 * different hash sizes.
+                 */
+                printk(KERN_ERR DRV_MODULE_NAME ": unsupported hash operation with mac/hash len: %d\n", ses->ses_mlen);
+                return (EINVAL);
+            }
+            
+        }
+
+        if (macini->cri_key != NULL) {
+            ubsec_setup_mackey(ses, macini->cri_alg, macini->cri_key,
+                macini->cri_klen / 8);
+        }
+    }
+
+    *sidp = UBSEC_SID(device_get_unit(sc->sc_dev), sesn);
+    return (0);
+}
+
+/*
+ * Deallocate a session.
+ */
+static int
+ubsec_freesession(device_t dev, u_int64_t tid)
+{
+    struct ubsec_softc *sc = device_get_softc(dev);
+    int session;
+    u_int32_t sid = ((u_int32_t)tid) & 0xffffffff;
+
+#ifdef UBSEC_DEBUG
+    DPRINTF("%s()\n", __FUNCTION__);
+#endif
+
+    if (sc == NULL)
+        return (EINVAL);
+
+    session = UBSEC_SESSION(sid);
+    if (session < sc->sc_nsessions) {
+        bzero(&sc->sc_sessions[session], sizeof(sc->sc_sessions[session]));
+        return (0);
+    } else
+        return (EINVAL);
+}
+
+static int
+ubsec_process(device_t dev, struct cryptop *crp, int hint)
+{
+    struct ubsec_q *q = NULL;
+    int err = 0, i, j, nicealign;
+    struct ubsec_softc *sc = device_get_softc(dev);
+    struct cryptodesc *crd1, *crd2, *maccrd, *enccrd;
+    int encoffset = 0, macoffset = 0, cpskip, cpoffset;
+    int sskip, dskip, stheend, dtheend, ivsize = 8;
+    int16_t coffset;
+    struct ubsec_session *ses;
+    struct ubsec_generic_ctx ctx;
+    struct ubsec_dma *dmap = NULL;
+    unsigned long flags;
+
+#ifdef UBSEC_DEBUG
+    DPRINTF("%s()\n", __FUNCTION__);
+#endif
+
+    if (unlikely(crp == NULL || crp->crp_callback == NULL)) {
+        ubsecstats.hst_invalid++;
+        return (EINVAL);
+    }
+
+    if (unlikely(sc == NULL))
+        return (EINVAL);
+
+#ifdef UBSEC_VERBOSE_DEBUG
+    DPRINTF("spin_lock_irqsave\n");
+#endif
+    spin_lock_irqsave(&sc->sc_ringmtx, flags);
+    //spin_lock_irq(&sc->sc_ringmtx);
+
+    if (BSD_SIMPLEQ_EMPTY(&sc->sc_freequeue)) {
+        ubsecstats.hst_queuefull++;
+#ifdef UBSEC_VERBOSE_DEBUG
+        DPRINTF("spin_unlock_irqrestore\n");
+#endif
+        spin_unlock_irqrestore(&sc->sc_ringmtx, flags);
+        //spin_unlock_irq(&sc->sc_ringmtx);
+        err = ENOMEM;
+        goto errout2;
+    }
+
+    q = BSD_SIMPLEQ_FIRST(&sc->sc_freequeue);
+    BSD_SIMPLEQ_REMOVE_HEAD(&sc->sc_freequeue, q_next);
+#ifdef UBSEC_VERBOSE_DEBUG
+    DPRINTF("spin_unlock_irqrestore\n");
+#endif
+    spin_unlock_irqrestore(&sc->sc_ringmtx, flags);
+    //spin_unlock_irq(&sc->sc_ringmtx);
+
+    dmap = q->q_dma; /* Save dma pointer */
+    bzero(q, sizeof(struct ubsec_q));
+    bzero(&ctx, sizeof(ctx));
+
+    q->q_sesn = UBSEC_SESSION(crp->crp_sid);
+    q->q_dma = dmap;
+    ses = &sc->sc_sessions[q->q_sesn];
+
+    if (crp->crp_flags & CRYPTO_F_SKBUF) {
+        q->q_src_m = (struct sk_buff *)crp->crp_buf;
+        q->q_dst_m = (struct sk_buff *)crp->crp_buf;
+    } else if (crp->crp_flags & CRYPTO_F_IOV) {
+        q->q_src_io = (struct uio *)crp->crp_buf;
+        q->q_dst_io = (struct uio *)crp->crp_buf;
+    } else {
+        err = EINVAL;
+        goto errout;    /* XXX we don't handle contiguous blocks! */
+    }
+
+    bzero(&dmap->d_dma->d_mcr, sizeof(struct ubsec_mcr));
+
+    dmap->d_dma->d_mcr.mcr_pkts = htole16(1);
+    dmap->d_dma->d_mcr.mcr_flags = 0;
+    q->q_crp = crp;
+
+    crd1 = crp->crp_desc;
+    if (crd1 == NULL) {
+        err = EINVAL;
+        goto errout;
+    }
+    crd2 = crd1->crd_next;
+
+    if (crd2 == NULL) {
+        if (crd1->crd_alg == CRYPTO_MD5_HMAC ||
+            crd1->crd_alg == CRYPTO_SHA1_HMAC) {
+            maccrd = crd1;
+            enccrd = NULL;
+        } else if (crd1->crd_alg == CRYPTO_DES_CBC ||
+            crd1->crd_alg == CRYPTO_3DES_CBC || 
+            crd1->crd_alg == CRYPTO_AES_CBC) {
+            maccrd = NULL;
+            enccrd = crd1;
+        } else {
+            err = EINVAL;
+            goto errout;
+        }
+    } else {
+        if ((crd1->crd_alg == CRYPTO_MD5_HMAC ||
+            crd1->crd_alg == CRYPTO_SHA1_HMAC) &&
+            (crd2->crd_alg == CRYPTO_DES_CBC ||
+            crd2->crd_alg == CRYPTO_3DES_CBC ||
+            crd2->crd_alg == CRYPTO_AES_CBC) &&
+            ((crd2->crd_flags & CRD_F_ENCRYPT) == 0)) {
+            maccrd = crd1;
+            enccrd = crd2;
+        } else if ((crd1->crd_alg == CRYPTO_DES_CBC ||
+            crd1->crd_alg == CRYPTO_3DES_CBC ||
+            crd1->crd_alg == CRYPTO_AES_CBC) &&
+            (crd2->crd_alg == CRYPTO_MD5_HMAC ||
+            crd2->crd_alg == CRYPTO_SHA1_HMAC) &&
+            (crd1->crd_flags & CRD_F_ENCRYPT)) {
+            enccrd = crd1;
+            maccrd = crd2;
+        } else {
+            /*
+             * We cannot order the ubsec as requested
+             */
+            printk(KERN_ERR DRV_MODULE_NAME ": got wrong algorithm/signature order.\n");
+            err = EINVAL;
+            goto errout;
+        }
+    }
+
+    /* Encryption/Decryption requested */
+    if (enccrd) {
+        encoffset = enccrd->crd_skip;
+
+        if (enccrd->crd_alg == CRYPTO_DES_CBC ||
+            enccrd->crd_alg == CRYPTO_3DES_CBC)
+        {
+            ctx.pc_flags |= htole16(UBS_PKTCTX_ENC_3DES);
+            ctx.pc_type = htole16(UBS_PKTCTX_TYPE_IPSEC_DES);
+            ivsize = 8;     /* [3]DES uses 64bit IVs */
+        } else {
+            ctx.pc_flags |= htole16(UBS_PKTCTX_ENC_AES);
+            ctx.pc_type = htole16(UBS_PKTCTX_TYPE_IPSEC_AES);
+            ivsize = 16;    /* AES uses 128bit IVs / [3]DES 64bit IVs */
+
+            switch(ses->ses_keysize)
+            {
+                case 128:
+                    ctx.pc_flags |= htole16(UBS_PKTCTX_AES128);
+                    break;
+                case 192:
+                    ctx.pc_flags |= htole16(UBS_PKTCTX_AES192);
+                    break;
+                case 256:
+                    ctx.pc_flags |= htole16(UBS_PKTCTX_AES256);
+                    break;
+                default:
+                    DPRINTF("invalid AES key size: %d\n", ses->ses_keysize);
+                    err = EINVAL;
+                    goto errout;
+            }
+        }
+
+        if (enccrd->crd_flags & CRD_F_ENCRYPT) {
+            /* Direction: Outbound */
+
+            q->q_flags |= UBSEC_QFLAGS_COPYOUTIV;
+
+            if (enccrd->crd_flags & CRD_F_IV_EXPLICIT) {
+                bcopy(enccrd->crd_iv, ctx.pc_iv, ivsize);
+            } else {
+                for(i=0; i < (ivsize / 4); i++)
+                    ctx.pc_iv[i] = ses->ses_iv[i];
+            }
+
+            /* If there is no IV in the buffer -> copy it here */
+            if ((enccrd->crd_flags & CRD_F_IV_PRESENT) == 0) {
+                if (crp->crp_flags & CRYPTO_F_SKBUF)
+                    /*
+                    m_copyback(q->q_src_m,
+                        enccrd->crd_inject,
+                        8, ctx.pc_iv);
+                    */
+                    crypto_copyback(crp->crp_flags, (caddr_t)q->q_src_m,
+                        enccrd->crd_inject, ivsize, (caddr_t)ctx.pc_iv);
+                else if (crp->crp_flags & CRYPTO_F_IOV)
+                    /*
+                    cuio_copyback(q->q_src_io,
+                        enccrd->crd_inject,
+                        8, ctx.pc_iv);
+                    */
+                    crypto_copyback(crp->crp_flags, (caddr_t)q->q_src_io,
+                        enccrd->crd_inject, ivsize, (caddr_t)ctx.pc_iv);
+            }
+        } else {
+            /* Direction: Inbound */
+
+            ctx.pc_flags |= htole16(UBS_PKTCTX_INBOUND);
+
+            if (enccrd->crd_flags & CRD_F_IV_EXPLICIT)
+                bcopy(enccrd->crd_iv, ctx.pc_iv, ivsize);
+            else if (crp->crp_flags & CRYPTO_F_SKBUF)
+                /*
+                m_copydata(q->q_src_m, enccrd->crd_inject,
+                    8, (caddr_t)ctx.pc_iv);
+                */
+                crypto_copydata(crp->crp_flags, (caddr_t)q->q_src_m,
+                    enccrd->crd_inject, ivsize,
+                    (caddr_t)ctx.pc_iv);
+            else if (crp->crp_flags & CRYPTO_F_IOV)
+                /*
+                cuio_copydata(q->q_src_io,
+                    enccrd->crd_inject, 8,
+                    (caddr_t)ctx.pc_iv);
+                */
+                crypto_copydata(crp->crp_flags, (caddr_t)q->q_src_io,
+                    enccrd->crd_inject, ivsize,
+                    (caddr_t)ctx.pc_iv);
+
+        }
+
+        /* Even though key & IV sizes differ from cipher to cipher
+         * copy / swap the full array lengths. Let the compiler unroll
+         * the loop to increase the cpu pipeline performance... */
+        for(i=0; i < 8; i++)
+            ctx.pc_key[i] = ses->ses_key[i];
+        for(i=0; i < 4; i++)
+            SWAP32(ctx.pc_iv[i]);
+    }
+
+    /* Authentication requested */
+    if (maccrd) {
+        macoffset = maccrd->crd_skip;
+
+        if (maccrd->crd_alg == CRYPTO_MD5_HMAC)
+            ctx.pc_flags |= htole16(UBS_PKTCTX_AUTH_MD5);
+        else
+            ctx.pc_flags |= htole16(UBS_PKTCTX_AUTH_SHA1);
+
+        for (i = 0; i < 5; i++) {
+            ctx.pc_hminner[i] = ses->ses_hminner[i];
+            ctx.pc_hmouter[i] = ses->ses_hmouter[i];
+
+            HTOLE32(ctx.pc_hminner[i]);
+            HTOLE32(ctx.pc_hmouter[i]);
+        }
+    }
+
+    if (enccrd && maccrd) {
+        /*
+         * ubsec cannot handle packets where the end of encryption
+         * and authentication are not the same, or where the
+         * encrypted part begins before the authenticated part.
+         */
+        if (((encoffset + enccrd->crd_len) !=
+            (macoffset + maccrd->crd_len)) ||
+            (enccrd->crd_skip < maccrd->crd_skip)) {
+            err = EINVAL;
+            goto errout;
+        }
+        sskip = maccrd->crd_skip;
+        cpskip = dskip = enccrd->crd_skip;
+        stheend = maccrd->crd_len;
+        dtheend = enccrd->crd_len;
+        coffset = enccrd->crd_skip - maccrd->crd_skip;
+        cpoffset = cpskip + dtheend;
+#ifdef UBSEC_DEBUG
+        DPRINTF("mac: skip %d, len %d, inject %d\n",
+            maccrd->crd_skip, maccrd->crd_len, maccrd->crd_inject);
+        DPRINTF("enc: skip %d, len %d, inject %d\n",
+            enccrd->crd_skip, enccrd->crd_len, enccrd->crd_inject);
+        DPRINTF("src: skip %d, len %d\n", sskip, stheend);
+        DPRINTF("dst: skip %d, len %d\n", dskip, dtheend);
+        DPRINTF("ubs: coffset %d, pktlen %d, cpskip %d, cpoffset %d\n",
+            coffset, stheend, cpskip, cpoffset);
+#endif
+    } else {
+        cpskip = dskip = sskip = macoffset + encoffset;
+        dtheend = stheend = (enccrd)?enccrd->crd_len:maccrd->crd_len;
+        cpoffset = cpskip + dtheend;
+        coffset = 0;
+    }
+    ctx.pc_offset = htole16(coffset >> 2);
+
+#if 0
+    if (bus_dmamap_create(sc->sc_dmat, 0xfff0, UBS_MAX_SCATTER,
+        0xfff0, 0, BUS_DMA_NOWAIT, &q->q_src_map) != 0) {
+        err = ENOMEM;
+        goto errout;
+    }
+#endif
+
+    if (crp->crp_flags & CRYPTO_F_SKBUF) {
+#if 0
+        if (bus_dmamap_load_mbuf(sc->sc_dmat, q->q_src_map,
+            q->q_src_m, BUS_DMA_NOWAIT) != 0) {
+            bus_dmamap_destroy(sc->sc_dmat, q->q_src_map);
+            q->q_src_map = NULL;
+            err = ENOMEM;
+            goto errout;
+        }
+#endif
+        err = dma_map_skb(sc, q->q_src_map, q->q_src_m, &q->q_src_len);
+        if (unlikely(err != 0))
+            goto errout;
+
+    } else if (crp->crp_flags & CRYPTO_F_IOV) {
+#if 0
+        if (bus_dmamap_load_uio(sc->sc_dmat, q->q_src_map,
+            q->q_src_io, BUS_DMA_NOWAIT) != 0) {
+            bus_dmamap_destroy(sc->sc_dmat, q->q_src_map);
+            q->q_src_map = NULL;
+            err = ENOMEM;
+            goto errout;
+        }
+#endif
+        err = dma_map_uio(sc, q->q_src_map, q->q_src_io, &q->q_src_len);
+        if (unlikely(err != 0))
+           goto errout;
+    }
+
+    /* 
+     * Check alignment 
+     */
+    nicealign = ubsec_dmamap_aligned(sc, q->q_src_map, q->q_src_len);
+
+    dmap->d_dma->d_mcr.mcr_pktlen = htole16(stheend);
+
+#ifdef UBSEC_DEBUG
+    DPRINTF("src skip: %d\n", sskip);
+#endif
+    for (i = j = 0; i < q->q_src_len; i++) {
+        struct ubsec_pktbuf *pb;
+        size_t packl = q->q_src_map[i].dma_size;
+        dma_addr_t packp = q->q_src_map[i].dma_paddr;
+
+        if (sskip >= packl) {
+            sskip -= packl;
+            continue;
+        }
+
+        packl -= sskip;
+        packp += sskip;
+        sskip = 0;
+
+        /* maximum fragment size is 0xfffc */
+        if (packl > 0xfffc) {
+            DPRINTF("Error: fragment size is bigger than 0xfffc.\n");
+            err = EIO;
+            goto errout;
+        }
+
+        if (j == 0)
+            pb = &dmap->d_dma->d_mcr.mcr_ipktbuf;
+        else
+            pb = &dmap->d_dma->d_sbuf[j - 1];
+
+        pb->pb_addr = htole32(packp);
+
+        if (stheend) {
+            if (packl > stheend) {
+                pb->pb_len = htole32(stheend);
+                stheend = 0;
+            } else {
+                pb->pb_len = htole32(packl);
+                stheend -= packl;
+            }
+        } else
+            pb->pb_len = htole32(packl);
+
+        if ((i + 1) == q->q_src_len)
+            pb->pb_next = 0;
+        else
+            pb->pb_next = htole32(dmap->d_alloc.dma_paddr +
+                offsetof(struct ubsec_dmachunk, d_sbuf[j]));
+        j++;
+    }
+
+    if (enccrd == NULL && maccrd != NULL) {
+        /* Authentication only */
+        dmap->d_dma->d_mcr.mcr_opktbuf.pb_addr = 0;
+        dmap->d_dma->d_mcr.mcr_opktbuf.pb_len = 0;
+        dmap->d_dma->d_mcr.mcr_opktbuf.pb_next =
+            htole32(dmap->d_alloc.dma_paddr +
+            offsetof(struct ubsec_dmachunk, d_macbuf[0]));
+#ifdef UBSEC_DEBUG
+        DPRINTF("opkt: %x %x %x\n",
+            dmap->d_dma->d_mcr.mcr_opktbuf.pb_addr,
+            dmap->d_dma->d_mcr.mcr_opktbuf.pb_len,
+            dmap->d_dma->d_mcr.mcr_opktbuf.pb_next);
+#endif
+    } else {
+        if (crp->crp_flags & CRYPTO_F_IOV) {
+            if (!nicealign) {
+                err = EINVAL;
+                goto errout;
+            }
+#if 0
+            if (bus_dmamap_create(sc->sc_dmat, 0xfff0,
+                UBS_MAX_SCATTER, 0xfff0, 0, BUS_DMA_NOWAIT,
+                &q->q_dst_map) != 0) {
+                err = ENOMEM;
+                goto errout;
+            }
+            if (bus_dmamap_load_uio(sc->sc_dmat, q->q_dst_map,
+                q->q_dst_io, BUS_DMA_NOWAIT) != 0) {
+                bus_dmamap_destroy(sc->sc_dmat, q->q_dst_map);
+                q->q_dst_map = NULL;
+                goto errout;
+            }
+#endif
+
+            /* HW shall copy the result into the source memory */
+            for(i = 0; i < q->q_src_len; i++)
+                q->q_dst_map[i] = q->q_src_map[i];
+
+            q->q_dst_len = q->q_src_len;
+            q->q_has_dst = 0;
+
+        } else if (crp->crp_flags & CRYPTO_F_SKBUF) {
+            if (nicealign) {
+
+                /* HW shall copy the result into the source memory */
+                q->q_dst_m = q->q_src_m;
+                for(i = 0; i < q->q_src_len; i++)
+                    q->q_dst_map[i] = q->q_src_map[i];
+
+                q->q_dst_len = q->q_src_len;
+                q->q_has_dst = 0;
+
+            } else {
+#ifdef NOTYET
+                int totlen, len;
+                struct sk_buff *m, *top, **mp;
+
+                totlen = q->q_src_map->dm_mapsize;
+                if (q->q_src_m->m_flags & M_PKTHDR) {
+                    len = MHLEN;
+                    MGETHDR(m, M_DONTWAIT, MT_DATA);
+                } else {
+                    len = MLEN;
+                    MGET(m, M_DONTWAIT, MT_DATA);
+                }
+                if (m == NULL) {
+                    err = ENOMEM;
+                    goto errout;
+                }
+                if (len == MHLEN)
+                    M_DUP_PKTHDR(m, q->q_src_m);
+                if (totlen >= MINCLSIZE) {
+                    MCLGET(m, M_DONTWAIT);
+                    if (m->m_flags & M_EXT)
+                        len = MCLBYTES;
+                }
+                m->m_len = len;
+                top = NULL;
+                mp = &top;
+
+                while (totlen > 0) {
+                    if (top) {
+                        MGET(m, M_DONTWAIT, MT_DATA);
+                        if (m == NULL) {
+                            m_freem(top);
+                            err = ENOMEM;
+                            goto errout;
+                        }
+                        len = MLEN;
+                    }
+                    if (top && totlen >= MINCLSIZE) {
+                        MCLGET(m, M_DONTWAIT);
+                        if (m->m_flags & M_EXT)
+                            len = MCLBYTES;
+                    }
+                    m->m_len = len = min(totlen, len);
+                    totlen -= len;
+                    *mp = m;
+                    mp = &m->m_next;
+                }
+                q->q_dst_m = top;
+                ubsec_mcopy(q->q_src_m, q->q_dst_m,
+                    cpskip, cpoffset);
+                if (bus_dmamap_create(sc->sc_dmat, 0xfff0,
+                    UBS_MAX_SCATTER, 0xfff0, 0, BUS_DMA_NOWAIT,
+                    &q->q_dst_map) != 0) {
+                    err = ENOMEM;
+                    goto errout;
+                }
+                if (bus_dmamap_load_mbuf(sc->sc_dmat,
+                    q->q_dst_map, q->q_dst_m,
+                    BUS_DMA_NOWAIT) != 0) {
+                    bus_dmamap_destroy(sc->sc_dmat,
+                    q->q_dst_map);
+                    q->q_dst_map = NULL;
+                    err = ENOMEM;
+                    goto errout;
+                }
+#else
+                device_printf(sc->sc_dev,
+                    "%s,%d: CRYPTO_F_SKBUF unaligned not implemented\n",
+                    __FILE__, __LINE__);
+                err = EINVAL;
+                goto errout;
+#endif
+            }
+        } else {
+            err = EINVAL;
+            goto errout;
+        }
+
+#ifdef UBSEC_DEBUG
+        DPRINTF("dst skip: %d\n", dskip);
+#endif
+        for (i = j = 0; i < q->q_dst_len; i++) {
+            struct ubsec_pktbuf *pb;
+            size_t packl = q->q_dst_map[i].dma_size;
+            dma_addr_t packp = q->q_dst_map[i].dma_paddr;
+
+            if (dskip >= packl) {
+                dskip -= packl;
+                continue;
+            }
+
+            packl -= dskip;
+            packp += dskip;
+            dskip = 0;
+
+            if (packl > 0xfffc) {
+                DPRINTF("Error: fragment size is bigger than 0xfffc.\n");
+                err = EIO;
+                goto errout;
+            }
+
+            if (j == 0)
+                pb = &dmap->d_dma->d_mcr.mcr_opktbuf;
+            else
+                pb = &dmap->d_dma->d_dbuf[j - 1];
+
+            pb->pb_addr = htole32(packp);
+
+            if (dtheend) {
+                if (packl > dtheend) {
+                    pb->pb_len = htole32(dtheend);
+                    dtheend = 0;
+                } else {
+                    pb->pb_len = htole32(packl);
+                    dtheend -= packl;
+                }
+            } else
+                pb->pb_len = htole32(packl);
+
+            if ((i + 1) == q->q_dst_len) {
+                if (maccrd)
+                    /* Authentication:
+                     * The last fragment of the output buffer 
+                     * contains the HMAC. */
+                    pb->pb_next = htole32(dmap->d_alloc.dma_paddr +
+                        offsetof(struct ubsec_dmachunk, d_macbuf[0]));
+                else
+                    pb->pb_next = 0;
+            } else
+                pb->pb_next = htole32(dmap->d_alloc.dma_paddr +
+                    offsetof(struct ubsec_dmachunk, d_dbuf[j]));
+            j++;
+        }
+    }
+
+    dmap->d_dma->d_mcr.mcr_cmdctxp = htole32(dmap->d_alloc.dma_paddr +
+        offsetof(struct ubsec_dmachunk, d_ctx));
+
+    if (sc->sc_flags & UBS_FLAGS_LONGCTX) {
+        /* new Broadcom cards with dynamic long command context structure */
+
+        if (enccrd != NULL &&
+            enccrd->crd_alg == CRYPTO_AES_CBC)
+        {
+            struct ubsec_pktctx_aes128 *ctxaes128;    
+            struct ubsec_pktctx_aes192 *ctxaes192;    
+            struct ubsec_pktctx_aes256 *ctxaes256;    
+
+            switch(ses->ses_keysize)
+            {
+                /* AES 128bit */
+                case 128:
+                ctxaes128 = (struct ubsec_pktctx_aes128 *)
+                    (dmap->d_alloc.dma_vaddr + 
+                    offsetof(struct ubsec_dmachunk, d_ctx));
+
+                ctxaes128->pc_len = htole16(sizeof(struct ubsec_pktctx_aes128));
+                ctxaes128->pc_type = ctx.pc_type;
+                ctxaes128->pc_flags = ctx.pc_flags;
+                ctxaes128->pc_offset = ctx.pc_offset;
+                for (i = 0; i < 4; i++)
+                    ctxaes128->pc_aeskey[i] = ctx.pc_key[i];
+                for (i = 0; i < 5; i++)
+                    ctxaes128->pc_hminner[i] = ctx.pc_hminner[i];
+                for (i = 0; i < 5; i++)
+                    ctxaes128->pc_hmouter[i] = ctx.pc_hmouter[i];
+                for (i = 0; i < 4; i++)
+                    ctxaes128->pc_iv[i] = ctx.pc_iv[i];
+                break;
+
+                /* AES 192bit */
+                case 192:
+                ctxaes192 = (struct ubsec_pktctx_aes192 *)
+                    (dmap->d_alloc.dma_vaddr + 
+                    offsetof(struct ubsec_dmachunk, d_ctx));
+
+                ctxaes192->pc_len = htole16(sizeof(struct ubsec_pktctx_aes192));
+                ctxaes192->pc_type = ctx.pc_type;
+                ctxaes192->pc_flags = ctx.pc_flags;
+                ctxaes192->pc_offset = ctx.pc_offset;
+                for (i = 0; i < 6; i++)
+                    ctxaes192->pc_aeskey[i] = ctx.pc_key[i];
+                for (i = 0; i < 5; i++)
+                    ctxaes192->pc_hminner[i] = ctx.pc_hminner[i];
+                for (i = 0; i < 5; i++)
+                    ctxaes192->pc_hmouter[i] = ctx.pc_hmouter[i];
+                for (i = 0; i < 4; i++)
+                    ctxaes192->pc_iv[i] = ctx.pc_iv[i];
+                break;
+
+                /* AES 256bit */
+                case 256:
+                ctxaes256 = (struct ubsec_pktctx_aes256 *)
+                    (dmap->d_alloc.dma_vaddr + 
+                    offsetof(struct ubsec_dmachunk, d_ctx));
+
+                ctxaes256->pc_len = htole16(sizeof(struct ubsec_pktctx_aes256));
+                ctxaes256->pc_type = ctx.pc_type;
+                ctxaes256->pc_flags = ctx.pc_flags;
+                ctxaes256->pc_offset = ctx.pc_offset;
+                for (i = 0; i < 8; i++)
+                    ctxaes256->pc_aeskey[i] = ctx.pc_key[i];
+                for (i = 0; i < 5; i++)
+                    ctxaes256->pc_hminner[i] = ctx.pc_hminner[i];
+                for (i = 0; i < 5; i++)
+                    ctxaes256->pc_hmouter[i] = ctx.pc_hmouter[i];
+                for (i = 0; i < 4; i++)
+                    ctxaes256->pc_iv[i] = ctx.pc_iv[i];
+                break;
+
+            }
+        } else {
+            /* 
+             * [3]DES / MD5_HMAC / SHA1_HMAC
+             *
+             * MD5_HMAC / SHA1_HMAC can use the IPSEC 3DES operation without
+             * encryption.
+             */
+            struct ubsec_pktctx_des *ctxdes;
+
+            ctxdes = (struct ubsec_pktctx_des *)(dmap->d_alloc.dma_vaddr +
+                offsetof(struct ubsec_dmachunk, d_ctx));
+            
+            ctxdes->pc_len = htole16(sizeof(struct ubsec_pktctx_des));
+            ctxdes->pc_type = ctx.pc_type;
+            ctxdes->pc_flags = ctx.pc_flags;
+            ctxdes->pc_offset = ctx.pc_offset;
+            for (i = 0; i < 6; i++)
+                ctxdes->pc_deskey[i] = ctx.pc_key[i];
+            for (i = 0; i < 5; i++)
+                ctxdes->pc_hminner[i] = ctx.pc_hminner[i];
+            for (i = 0; i < 5; i++)
+                ctxdes->pc_hmouter[i] = ctx.pc_hmouter[i];   
+            ctxdes->pc_iv[0] = ctx.pc_iv[0];
+            ctxdes->pc_iv[1] = ctx.pc_iv[1];
+        }
+    } else
+    {
+        /* old Broadcom card with fixed small command context structure */
+
+        /*
+         * [3]DES / MD5_HMAC / SHA1_HMAC
+         */
+        struct ubsec_pktctx *ctxs;
+
+        ctxs = (struct ubsec_pktctx *)(dmap->d_alloc.dma_vaddr +
+                    offsetof(struct ubsec_dmachunk, d_ctx));
+        /* transform generic context into small context */
+        for (i = 0; i < 6; i++)
+            ctxs->pc_deskey[i] = ctx.pc_key[i];
+        for (i = 0; i < 5; i++)
+            ctxs->pc_hminner[i] = ctx.pc_hminner[i];
+        for (i = 0; i < 5; i++)
+            ctxs->pc_hmouter[i] = ctx.pc_hmouter[i];
+        ctxs->pc_iv[0] = ctx.pc_iv[0];
+        ctxs->pc_iv[1] = ctx.pc_iv[1];
+        ctxs->pc_flags = ctx.pc_flags;
+        ctxs->pc_offset = ctx.pc_offset;
+    }
+
+#ifdef UBSEC_VERBOSE_DEBUG
+    DPRINTF("spin_lock_irqsave\n");
+#endif
+    spin_lock_irqsave(&sc->sc_ringmtx, flags);
+    //spin_lock_irq(&sc->sc_ringmtx);
+
+    BSD_SIMPLEQ_INSERT_TAIL(&sc->sc_queue, q, q_next);
+    sc->sc_nqueue++;
+    ubsecstats.hst_ipackets++;
+    ubsecstats.hst_ibytes += stheend;
+    ubsec_feed(sc);
+
+#ifdef UBSEC_VERBOSE_DEBUG
+    DPRINTF("spin_unlock_irqrestore\n");
+#endif
+    spin_unlock_irqrestore(&sc->sc_ringmtx, flags);
+    //spin_unlock_irq(&sc->sc_ringmtx);
+    
+    return (0);
+
+errout:
+    if (q != NULL) {
+#ifdef NOTYET
+        if ((q->q_dst_m != NULL) && (q->q_src_m != q->q_dst_m))
+            m_freem(q->q_dst_m);
+#endif
+
+        if ((q->q_has_dst == 1) && q->q_dst_len > 0) {
+#if 0
+            bus_dmamap_unload(sc->sc_dmat, q->q_dst_map);
+            bus_dmamap_destroy(sc->sc_dmat, q->q_dst_map);
+#endif
+            dma_unmap(sc, q->q_dst_map, q->q_dst_len);
+        }
+        if (q->q_src_len > 0) {
+#if 0
+            bus_dmamap_unload(sc->sc_dmat, q->q_src_map);
+            bus_dmamap_destroy(sc->sc_dmat, q->q_src_map);
+#endif
+            dma_unmap(sc, q->q_src_map, q->q_src_len);
+        }
+
+#ifdef UBSEC_VERBOSE_DEBUG
+        DPRINTF("spin_lock_irqsave\n");
+#endif
+        spin_lock_irqsave(&sc->sc_ringmtx, flags);
+        //spin_lock_irq(&sc->sc_ringmtx);
+
+        BSD_SIMPLEQ_INSERT_TAIL(&sc->sc_freequeue, q, q_next);
+
+#ifdef UBSEC_VERBOSE_DEBUG
+       DPRINTF("spin_unlock_irqrestore\n");
+#endif
+        spin_unlock_irqrestore(&sc->sc_ringmtx, flags);
+        //spin_unlock_irq(&sc->sc_ringmtx);
+
+    }
+    if (err == EINVAL)
+        ubsecstats.hst_invalid++;
+    else
+        ubsecstats.hst_nomem++;
+errout2:
+    crp->crp_etype = err;
+    crypto_done(crp);
+
+#ifdef UBSEC_DEBUG
+    DPRINTF("%s() err = %x\n", __FUNCTION__, err);
+#endif
+
+    return (0);
+}
+
+void
+ubsec_callback(struct ubsec_softc *sc, struct ubsec_q *q)
+{
+    struct cryptop *crp = (struct cryptop *)q->q_crp;
+    struct cryptodesc *crd;
+    struct ubsec_dma *dmap = q->q_dma;
+    int ivsize = 8;
+
+#ifdef UBSEC_DEBUG
+    DPRINTF("%s()\n", __FUNCTION__);
+#endif
+
+    ubsecstats.hst_opackets++;
+    ubsecstats.hst_obytes += dmap->d_alloc.dma_size;
+
+#if 0
+    bus_dmamap_sync(sc->sc_dmat, dmap->d_alloc.dma_map, 0,
+        dmap->d_alloc.dma_map->dm_mapsize,
+        BUS_DMASYNC_POSTREAD|BUS_DMASYNC_POSTWRITE);
+    if (q->q_dst_map != NULL && q->q_dst_map != q->q_src_map) {
+        bus_dmamap_sync(sc->sc_dmat, q->q_dst_map,
+            0, q->q_dst_map->dm_mapsize, BUS_DMASYNC_POSTREAD);
+        bus_dmamap_unload(sc->sc_dmat, q->q_dst_map);
+        bus_dmamap_destroy(sc->sc_dmat, q->q_dst_map);
+    }
+    bus_dmamap_sync(sc->sc_dmat, q->q_src_map,
+        0, q->q_src_map->dm_mapsize, BUS_DMASYNC_POSTWRITE);
+    bus_dmamap_unload(sc->sc_dmat, q->q_src_map);
+    bus_dmamap_destroy(sc->sc_dmat, q->q_src_map);
+#endif
+
+    if ((q->q_has_dst == 1) && q->q_dst_len > 0)
+        dma_unmap(sc, q->q_dst_map, q->q_dst_len);
+
+    dma_unmap(sc, q->q_src_map, q->q_src_len);
+
+#ifdef NOTYET
+    if ((crp->crp_flags & CRYPTO_F_SKBUF) && (q->q_src_m != q->q_dst_m)) {
+        m_freem(q->q_src_m);
+        crp->crp_buf = (caddr_t)q->q_dst_m;
+    }
+#endif
+
+    /* copy out IV for future use */
+    if (q->q_flags & UBSEC_QFLAGS_COPYOUTIV) {
+        for (crd = crp->crp_desc; crd; crd = crd->crd_next) {
+            if (crd->crd_alg != CRYPTO_DES_CBC &&
+                crd->crd_alg != CRYPTO_3DES_CBC &&
+                crd->crd_alg != CRYPTO_AES_CBC)
+                continue;
+
+            if (crd->crd_alg == CRYPTO_AES_CBC)
+                ivsize = 16;
+            else
+                ivsize = 8;
+
+            if (crp->crp_flags & CRYPTO_F_SKBUF)
+#if 0
+                m_copydata((struct sk_buff *)crp->crp_buf,
+                    crd->crd_skip + crd->crd_len - 8, 8,
+                    (caddr_t)sc->sc_sessions[q->q_sesn].ses_iv);
+#endif
+                crypto_copydata(crp->crp_flags, (caddr_t)crp->crp_buf,
+                    crd->crd_skip + crd->crd_len - ivsize, ivsize,
+                    (caddr_t)sc->sc_sessions[q->q_sesn].ses_iv);
+
+            else if (crp->crp_flags & CRYPTO_F_IOV) {
+#if 0
+                cuio_copydata((struct uio *)crp->crp_buf,
+                    crd->crd_skip + crd->crd_len - 8, 8,
+                    (caddr_t)sc->sc_sessions[q->q_sesn].ses_iv);
+#endif
+                crypto_copydata(crp->crp_flags, (caddr_t)crp->crp_buf,
+                    crd->crd_skip + crd->crd_len - ivsize, ivsize,
+                    (caddr_t)sc->sc_sessions[q->q_sesn].ses_iv);
+                    
+            }
+            break;
+        }
+    }
+
+    for (crd = crp->crp_desc; crd; crd = crd->crd_next) {
+        if (crd->crd_alg != CRYPTO_MD5_HMAC &&
+            crd->crd_alg != CRYPTO_SHA1_HMAC)
+            continue;
+#if 0
+        if (crp->crp_flags & CRYPTO_F_SKBUF)
+            m_copyback((struct sk_buff *)crp->crp_buf,
+                crd->crd_inject, 12,
+                dmap->d_dma->d_macbuf);
+#endif
+#if 0
+            /* BUG? it does not honor the mac len.. */
+            crypto_copyback(crp->crp_flags, crp->crp_buf,
+                crd->crd_inject, 12,
+                (caddr_t)dmap->d_dma->d_macbuf);
+#endif
+            crypto_copyback(crp->crp_flags, crp->crp_buf,
+                crd->crd_inject, 
+                sc->sc_sessions[q->q_sesn].ses_mlen,
+                (caddr_t)dmap->d_dma->d_macbuf);
+#if 0
+        else if (crp->crp_flags & CRYPTO_F_IOV && crp->crp_mac)
+            bcopy((caddr_t)dmap->d_dma->d_macbuf,
+                crp->crp_mac, 12);
+#endif
+        break;
+    }
+    BSD_SIMPLEQ_INSERT_TAIL(&sc->sc_freequeue, q, q_next);
+    crypto_done(crp);
+}
+
+void
+ubsec_mcopy(struct sk_buff *srcm, struct sk_buff *dstm, int hoffset, int toffset)
+{
+    int i, j, dlen, slen;
+    caddr_t dptr, sptr;
+
+    j = 0;
+    sptr = srcm->data;
+    slen = srcm->len;
+    dptr = dstm->data;
+    dlen = dstm->len;
+
+    while (1) {
+        for (i = 0; i < min(slen, dlen); i++) {
+            if (j < hoffset || j >= toffset)
+                *dptr++ = *sptr++;
+            slen--;
+            dlen--;
+            j++;
+        }
+        if (slen == 0) {
+            srcm = srcm->next;
+            if (srcm == NULL)
+                return;
+            sptr = srcm->data;
+            slen = srcm->len;
+        }
+        if (dlen == 0) {
+            dstm = dstm->next;
+            if (dstm == NULL)
+                return;
+            dptr = dstm->data;
+            dlen = dstm->len;
+        }
+    }
+}
+
+int
+ubsec_dma_malloc(struct ubsec_softc *sc, struct ubsec_dma_alloc *dma, 
+    size_t size, int mapflags)
+{
+    dma->dma_vaddr = dma_alloc_coherent(sc->sc_dv, 
+        size, &dma->dma_paddr, GFP_KERNEL);
+
+    if (likely(dma->dma_vaddr))
+    {
+        dma->dma_size = size;
+        return (0);
+    }
+
+    DPRINTF("could not allocate %d bytes of coherent memory.\n", size);
+
+    return (1);
+}
+
+void
+ubsec_dma_free(struct ubsec_softc *sc, struct ubsec_dma_alloc *dma)
+{
+    dma_free_coherent(sc->sc_dv, dma->dma_size, dma->dma_vaddr, 
+        dma->dma_paddr);
+}
+
+/*
+ * Resets the board.  Values in the regesters are left as is
+ * from the reset (i.e. initial values are assigned elsewhere).
+ */
+void
+ubsec_reset_board(struct ubsec_softc *sc)
+{
+    volatile u_int32_t ctrl;
+
+#ifdef UBSEC_DEBUG
+    DPRINTF("%s()\n", __FUNCTION__);
+#endif
+    DPRINTF("Send reset signal to chip.\n");
+
+    ctrl = READ_REG(sc, BS_CTRL);
+    ctrl |= BS_CTRL_RESET;
+    WRITE_REG(sc, BS_CTRL, ctrl);
+
+    /*
+     * Wait aprox. 30 PCI clocks = 900 ns = 0.9 us
+     */
+    DELAY(10);
+}
+
+/*
+ * Init Broadcom registers
+ */
+void
+ubsec_init_board(struct ubsec_softc *sc)
+{
+    u_int32_t ctrl;
+
+#ifdef UBSEC_DEBUG
+    DPRINTF("%s()\n", __FUNCTION__);
+#endif
+    DPRINTF("Initialize chip.\n");
+
+    ctrl = READ_REG(sc, BS_CTRL);
+    ctrl &= ~(BS_CTRL_BE32 | BS_CTRL_BE64);
+    ctrl |= BS_CTRL_LITTLE_ENDIAN | BS_CTRL_MCR1INT | BS_CTRL_DMAERR;
+
+    WRITE_REG(sc, BS_CTRL, ctrl);
+
+    /* Set chip capabilities (BCM5365P) */
+    sc->sc_flags |= UBS_FLAGS_LONGCTX | UBS_FLAGS_AES;
+}
+
+/*
+ * Clean up after a chip crash.
+ * It is assumed that the caller has spin_lock_irq(sc_ringmtx).
+ */
+void
+ubsec_cleanchip(struct ubsec_softc *sc)
+{
+    struct ubsec_q *q;
+
+#ifdef UBSEC_DEBUG
+    DPRINTF("%s()\n", __FUNCTION__);
+#endif
+    DPRINTF("Clean up queues after chip crash.\n");
+
+    while (!BSD_SIMPLEQ_EMPTY(&sc->sc_qchip)) {
+        q = BSD_SIMPLEQ_FIRST(&sc->sc_qchip);
+        BSD_SIMPLEQ_REMOVE_HEAD(&sc->sc_qchip, q_next);
+        ubsec_free_q(sc, q);
+    }
+}
+
+/*
+ * free a ubsec_q
+ * It is assumed that the caller has spin_lock_irq(sc_ringmtx).
+ */
+int
+ubsec_free_q(struct ubsec_softc *sc, struct ubsec_q *q)
+{
+    struct ubsec_q *q2;
+    struct cryptop *crp;
+    int npkts;
+    int i;
+
+#ifdef UBSEC_DEBUG
+    DPRINTF("%s()\n", __FUNCTION__);
+#endif
+
+    npkts = q->q_nstacked_mcrs;
+
+    for (i = 0; i < npkts; i++) {
+        if(q->q_stacked_mcr[i]) {
+            q2 = q->q_stacked_mcr[i];
+
+            if ((q2->q_dst_m != NULL) && (q2->q_src_m != q2->q_dst_m)) 
+#ifdef NOTYET
+                m_freem(q2->q_dst_m);
+#else
+                printk(KERN_ERR "%s,%d: SKB not supported\n", __FILE__, __LINE__);
+#endif
+
+            crp = (struct cryptop *)q2->q_crp;
+            
+            BSD_SIMPLEQ_INSERT_TAIL(&sc->sc_freequeue, q2, q_next);
+            
+            crp->crp_etype = EFAULT;
+            crypto_done(crp);
+        } else {
+            break;
+        }
+    }
+
+    /*
+     * Free header MCR
+     */
+    if ((q->q_dst_m != NULL) && (q->q_src_m != q->q_dst_m))
+#ifdef NOTYET
+        m_freem(q->q_dst_m);
+#else
+        printk(KERN_ERR "%s,%d: SKB not supported\n", __FILE__, __LINE__);
+#endif
+
+    crp = (struct cryptop *)q->q_crp;
+    
+    BSD_SIMPLEQ_INSERT_TAIL(&sc->sc_freequeue, q, q_next);
+    
+    crp->crp_etype = EFAULT;
+    crypto_done(crp);
+    return(0);
+}
+
+/*
+ * Routine to reset the chip and clean up.
+ * It is assumed that the caller has spin_lock_irq(sc_ringmtx).
+ */
+void
+ubsec_totalreset(struct ubsec_softc *sc)
+{
+
+#ifdef UBSEC_DEBUG
+    DPRINTF("%s()\n", __FUNCTION__);
+#endif
+    DPRINTF("initiate total chip reset.. \n");
+    ubsec_reset_board(sc);
+    ubsec_init_board(sc);
+    ubsec_cleanchip(sc);
+}
+
+void
+ubsec_dump_pb(struct ubsec_pktbuf *pb)
+{
+    printf("addr 0x%x (0x%x) next 0x%x\n",
+        pb->pb_addr, pb->pb_len, pb->pb_next);
+}
+
+void
+ubsec_dump_mcr(struct ubsec_mcr *mcr)
+{
+    struct ubsec_mcr_add *ma;
+    int i;
+
+    printf("MCR:\n");
+    printf(" pkts: %u, flags 0x%x\n",
+        letoh16(mcr->mcr_pkts), letoh16(mcr->mcr_flags));
+    ma = (struct ubsec_mcr_add *)&mcr->mcr_cmdctxp;
+    for (i = 0; i < letoh16(mcr->mcr_pkts); i++) {
+        printf(" %d: ctx 0x%x len 0x%x rsvd 0x%x\n", i,
+            letoh32(ma->mcr_cmdctxp), letoh16(ma->mcr_pktlen),
+            letoh16(ma->mcr_reserved));
+        printf(" %d: ipkt ", i);
+        ubsec_dump_pb(&ma->mcr_ipktbuf);
+        printf(" %d: opkt ", i);
+        ubsec_dump_pb(&ma->mcr_opktbuf);
+        ma++;
+    }
+    printf("END MCR\n");
+}
+
+static int __init mod_init(void) {
+        return ssb_driver_register(&ubsec_ssb_driver);
+}
+
+static void __exit mod_exit(void) {
+        ssb_driver_unregister(&ubsec_ssb_driver);
+}
+
+module_init(mod_init);
+module_exit(mod_exit);
+
+// Meta information
+MODULE_AUTHOR("Daniel Mueller <daniel@danm.de>");
+MODULE_LICENSE("BSD");
+MODULE_DESCRIPTION("OCF driver for BCM5365P IPSec Core");
+MODULE_VERSION(DRV_MODULE_VERSION);
+