/*
- * Copyright (c) 2006, 2007 QLogic Corporation. All rights reserved.
+ * Copyright (c) 2006, 2007, 2008 QLogic Corporation. All rights reserved.
* Copyright (c) 2003, 2004, 2005, 2006 PathScale, Inc. All rights reserved.
*
* This software is available to you under a choice of one of two
#include <linux/cdev.h>
#include <linux/swap.h>
#include <linux/vmalloc.h>
+#include <linux/highmem.h>
+#include <linux/io.h>
+#include <linux/jiffies.h>
+#include <linux/smp_lock.h>
#include <asm/pgtable.h>
#include "ipath_kernel.h"
#include "ipath_common.h"
+#include "ipath_user_sdma.h"
static int ipath_open(struct inode *, struct file *);
static int ipath_close(struct inode *, struct file *);
static ssize_t ipath_write(struct file *, const char __user *, size_t,
loff_t *);
+static ssize_t ipath_writev(struct kiocb *, const struct iovec *,
+ unsigned long , loff_t);
static unsigned int ipath_poll(struct file *, struct poll_table_struct *);
static int ipath_mmap(struct file *, struct vm_area_struct *);
static const struct file_operations ipath_file_ops = {
.owner = THIS_MODULE,
.write = ipath_write,
+ .aio_write = ipath_writev,
.open = ipath_open,
.release = ipath_close,
.poll = ipath_poll,
(void *) dd->ipath_statusp -
(void *) dd->ipath_pioavailregs_dma;
if (!shared) {
- kinfo->spi_piocnt = dd->ipath_pbufsport;
+ kinfo->spi_piocnt = pd->port_piocnt;
kinfo->spi_piobufbase = (u64) pd->port_piobufs;
kinfo->__spi_uregbase = (u64) dd->ipath_uregbase +
dd->ipath_ureg_align * pd->port_port;
} else if (master) {
- kinfo->spi_piocnt = (dd->ipath_pbufsport / subport_cnt) +
- (dd->ipath_pbufsport % subport_cnt);
+ kinfo->spi_piocnt = (pd->port_piocnt / subport_cnt) +
+ (pd->port_piocnt % subport_cnt);
/* Master's PIO buffers are after all the slave's */
kinfo->spi_piobufbase = (u64) pd->port_piobufs +
dd->ipath_palign *
- (dd->ipath_pbufsport - kinfo->spi_piocnt);
+ (pd->port_piocnt - kinfo->spi_piocnt);
} else {
unsigned slave = subport_fp(fp) - 1;
- kinfo->spi_piocnt = dd->ipath_pbufsport / subport_cnt;
+ kinfo->spi_piocnt = pd->port_piocnt / subport_cnt;
kinfo->spi_piobufbase = (u64) pd->port_piobufs +
dd->ipath_palign * kinfo->spi_piocnt * slave;
}
+
if (shared) {
kinfo->spi_port_uregbase = (u64) dd->ipath_uregbase +
dd->ipath_ureg_align * pd->port_port;
(unsigned long long) kinfo->spi_subport_rcvhdr_base);
}
- kinfo->spi_pioindex = (kinfo->spi_piobufbase - dd->ipath_piobufbase) /
- dd->ipath_palign;
+ /*
+ * All user buffers are 2KB buffers. If we ever support
+ * giving 4KB buffers to user processes, this will need some
+ * work.
+ */
+ kinfo->spi_pioindex = (kinfo->spi_piobufbase -
+ (dd->ipath_piobufbase & 0xffffffff)) / dd->ipath_palign;
kinfo->spi_pioalign = dd->ipath_palign;
kinfo->spi_qpair = IPATH_KD_QP;
- kinfo->spi_piosize = dd->ipath_ibmaxlen;
+ /*
+ * user mode PIO buffers are always 2KB, even when 4KB can
+ * be received, and sent via the kernel; this is ibmaxlen
+ * for 2K MTU.
+ */
+ kinfo->spi_piosize = dd->ipath_piosize2k - 2 * sizeof(u32);
kinfo->spi_mtu = dd->ipath_ibmaxlen; /* maxlen, not ibmtu */
kinfo->spi_port = pd->port_port;
kinfo->spi_subport = subport_fp(fp);
p = dd->ipath_pageshadow[porttid + tid];
dd->ipath_pageshadow[porttid + tid] = NULL;
ipath_cdbg(VERBOSE, "PID %u freeing TID %u\n",
- pd->port_pid, tid);
+ pid_nr(pd->port_pid), tid);
dd->ipath_f_put_tid(dd, &tidbase[tid],
RCVHQ_RCV_TYPE_EXPECTED,
dd->ipath_tidinvalid);
ureg = dd->ipath_uregbase + dd->ipath_ureg_align * pd->port_port;
if (!pd->port_subport_cnt) {
/* port is not shared */
- piocnt = dd->ipath_pbufsport;
+ piocnt = pd->port_piocnt;
piobufs = pd->port_piobufs;
} else if (!subport_fp(fp)) {
/* caller is the master */
- piocnt = (dd->ipath_pbufsport / pd->port_subport_cnt) +
- (dd->ipath_pbufsport % pd->port_subport_cnt);
+ piocnt = (pd->port_piocnt / pd->port_subport_cnt) +
+ (pd->port_piocnt % pd->port_subport_cnt);
piobufs = pd->port_piobufs +
- dd->ipath_palign * (dd->ipath_pbufsport - piocnt);
+ dd->ipath_palign * (pd->port_piocnt - piocnt);
} else {
unsigned slave = subport_fp(fp) - 1;
/* caller is a slave */
- piocnt = dd->ipath_pbufsport / pd->port_subport_cnt;
+ piocnt = pd->port_piocnt / pd->port_subport_cnt;
piobufs = pd->port_piobufs + dd->ipath_palign * piocnt * slave;
}
port);
pd->port_cnt = 1;
port_fp(fp) = pd;
- pd->port_pid = current->pid;
+ pd->port_pid = get_pid(task_pid(current));
strncpy(pd->port_comm, current->comm, sizeof(pd->port_comm));
ipath_stats.sps_ports++;
ret = 0;
* InfiniPath chip to that processor (we assume reasonable connectivity,
* for now). This code assumes that if affinity has been set
* before this point, that at most one cpu is set; for now this
- * is reasonable. I check for both cpus_empty() and cpus_full(),
+ * is reasonable. I check for both cpumask_empty() and cpumask_full(),
* in case some kernel variant sets none of the bits when no
* affinity is set. 2.6.11 and 12 kernels have all present
* cpus set. Some day we'll have to fix it up further to handle
* information. There may be some issues with dual core numbering
* as well. This needs more work prior to release.
*/
- if (!cpus_empty(current->cpus_allowed) &&
- !cpus_full(current->cpus_allowed)) {
+ if (!cpumask_empty(¤t->cpus_allowed) &&
+ !cpumask_full(¤t->cpus_allowed)) {
int ncpus = num_online_cpus(), curcpu = -1, nset = 0;
for (i = 0; i < ncpus; i++)
- if (cpu_isset(i, current->cpus_allowed)) {
+ if (cpumask_test_cpu(i, ¤t->cpus_allowed)) {
ipath_cdbg(PROC, "%s[%u] affinity set for "
"cpu %d/%d\n", current->comm,
current->pid, i, ncpus);
for (ndev = 0; ndev < devmax; ndev++) {
struct ipath_devdata *dd = ipath_lookup(ndev);
- if (!dd)
+ if (!usable(dd))
continue;
for (i = 1; i < dd->ipath_cfgports; i++) {
struct ipath_portdata *pd = dd->ipath_pd[i];
}
port_fp(fp) = pd;
subport_fp(fp) = pd->port_cnt++;
- pd->port_subpid[subport_fp(fp)] = current->pid;
+ pd->port_subpid[subport_fp(fp)] =
+ get_pid(task_pid(current));
tidcursor_fp(fp) = 0;
pd->active_slaves |= 1 << subport_fp(fp);
ipath_cdbg(PROC,
"%s[%u] %u sharing %s[%u] unit:port %u:%u\n",
current->comm, current->pid,
subport_fp(fp),
- pd->port_comm, pd->port_pid,
+ pd->port_comm, pid_nr(pd->port_pid),
dd->ipath_unit, pd->port_port);
ret = 1;
goto done;
static int ipath_open(struct inode *in, struct file *fp)
{
/* The real work is performed later in ipath_assign_port() */
+ cycle_kernel_lock();
fp->private_data = kzalloc(sizeof(struct ipath_filedata), GFP_KERNEL);
return fp->private_data ? 0 : -ENOMEM;
}
if (ipath_compatible_subports(swmajor, swminor) &&
uinfo->spu_subport_cnt &&
(ret = find_shared_port(fp, uinfo))) {
- mutex_unlock(&ipath_mutex);
if (ret > 0)
ret = 0;
- goto done;
+ goto done_chk_sdma;
}
i_minor = iminor(fp->f_path.dentry->d_inode) - IPATH_USER_MINOR_BASE;
else
ret = find_best_unit(fp, uinfo);
+done_chk_sdma:
+ if (!ret) {
+ struct ipath_filedata *fd = fp->private_data;
+ const struct ipath_portdata *pd = fd->pd;
+ const struct ipath_devdata *dd = pd->port_dd;
+
+ fd->pq = ipath_user_sdma_queue_create(&dd->pcidev->dev,
+ dd->ipath_unit,
+ pd->port_port,
+ fd->subport);
+
+ if (!fd->pq)
+ ret = -ENOMEM;
+ }
+
mutex_unlock(&ipath_mutex);
done:
/* for now we do nothing with rcvhdrcnt: uinfo->spu_rcvhdrcnt */
+ /* some ports may get extra buffers, calculate that here */
+ if (pd->port_port <= dd->ipath_ports_extrabuf)
+ pd->port_piocnt = dd->ipath_pbufsport + 1;
+ else
+ pd->port_piocnt = dd->ipath_pbufsport;
+
/* for right now, kernel piobufs are at end, so port 1 is at 0 */
+ if (pd->port_port <= dd->ipath_ports_extrabuf)
+ pd->port_pio_base = (dd->ipath_pbufsport + 1)
+ * (pd->port_port - 1);
+ else
+ pd->port_pio_base = dd->ipath_ports_extrabuf +
+ dd->ipath_pbufsport * (pd->port_port - 1);
pd->port_piobufs = dd->ipath_piobufbase +
- dd->ipath_pbufsport * (pd->port_port - 1) * dd->ipath_palign;
- ipath_cdbg(VERBOSE, "Set base of piobufs for port %u to 0x%x\n",
- pd->port_port, pd->port_piobufs);
+ pd->port_pio_base * dd->ipath_palign;
+ ipath_cdbg(VERBOSE, "piobuf base for port %u is 0x%x, piocnt %u,"
+ " first pio %u\n", pd->port_port, pd->port_piobufs,
+ pd->port_piocnt, pd->port_pio_base);
+ ipath_chg_pioavailkernel(dd, pd->port_pio_base, pd->port_piocnt, 0);
/*
* Now allocate the rcvhdr Q and eager TIDs; skip the TID
pd->port_hdrqfull_poll = pd->port_hdrqfull;
/*
- * now enable the port; the tail registers will be written to memory
- * by the chip as soon as it sees the write to
- * dd->ipath_kregs->kr_rcvctrl. The update only happens on
- * transition from 0 to 1, so clear it first, then set it as part of
- * enabling the port. This will (very briefly) affect any other
- * open ports, but it shouldn't be long enough to be an issue.
- * We explictly set the in-memory copy to 0 beforehand, so we don't
- * have to wait to be sure the DMA update has happened.
+ * Now enable the port for receive.
+ * For chips that are set to DMA the tail register to memory
+ * when they change (and when the update bit transitions from
+ * 0 to 1. So for those chips, we turn it off and then back on.
+ * This will (very briefly) affect any other open ports, but the
+ * duration is very short, and therefore isn't an issue. We
+ * explictly set the in-memory tail copy to 0 beforehand, so we
+ * don't have to wait to be sure the DMA update has happened
+ * (chip resets head/tail to 0 on transition to enable).
*/
- if (pd->port_rcvhdrtail_kvaddr)
- ipath_clear_rcvhdrtail(pd);
set_bit(dd->ipath_r_portenable_shift + pd->port_port,
&dd->ipath_rcvctrl);
- ipath_write_kreg(dd, dd->ipath_kregs->kr_rcvctrl,
+ if (!(dd->ipath_flags & IPATH_NODMA_RTAIL)) {
+ if (pd->port_rcvhdrtail_kvaddr)
+ ipath_clear_rcvhdrtail(pd);
+ ipath_write_kreg(dd, dd->ipath_kregs->kr_rcvctrl,
dd->ipath_rcvctrl &
~(1ULL << dd->ipath_r_tailupd_shift));
+ }
ipath_write_kreg(dd, dd->ipath_kregs->kr_rcvctrl,
dd->ipath_rcvctrl);
/* Notify any waiting slaves */
ipath_cdbg(VERBOSE, "Port %u unlocking any locked expTID pages\n",
pd->port_port);
for (i = port_tidbase; i < maxtid; i++) {
- if (!dd->ipath_pageshadow[i])
+ struct page *ps = dd->ipath_pageshadow[i];
+
+ if (!ps)
continue;
+ dd->ipath_pageshadow[i] = NULL;
pci_unmap_page(dd->pcidev, dd->ipath_physshadow[i],
PAGE_SIZE, PCI_DMA_FROMDEVICE);
- ipath_release_user_pages_on_close(&dd->ipath_pageshadow[i],
- 1);
- dd->ipath_pageshadow[i] = NULL;
+ ipath_release_user_pages_on_close(&ps, 1);
cnt++;
ipath_stats.sps_pageunlocks++;
}
struct ipath_filedata *fd;
struct ipath_portdata *pd;
struct ipath_devdata *dd;
+ unsigned long flags;
unsigned port;
+ struct pid *pid;
ipath_cdbg(VERBOSE, "close on dev %lx, private data %p\n",
(long)in->i_rdev, fp->private_data);
mutex_unlock(&ipath_mutex);
goto bail;
}
+
+ dd = pd->port_dd;
+
+ /* drain user sdma queue */
+ ipath_user_sdma_queue_drain(dd, fd->pq);
+ ipath_user_sdma_queue_destroy(fd->pq);
+
if (--pd->port_cnt) {
/*
* XXX If the master closes the port before the slave(s),
* the slave(s) don't wait for receive data forever.
*/
pd->active_slaves &= ~(1 << fd->subport);
- pd->port_subpid[fd->subport] = 0;
+ put_pid(pd->port_subpid[fd->subport]);
+ pd->port_subpid[fd->subport] = NULL;
mutex_unlock(&ipath_mutex);
goto bail;
}
+ /* early; no interrupt users after this */
+ spin_lock_irqsave(&dd->ipath_uctxt_lock, flags);
port = pd->port_port;
- dd = pd->port_dd;
-
- if (pd->port_hdrqfull) {
- ipath_cdbg(PROC, "%s[%u] had %u rcvhdrqfull errors "
- "during run\n", pd->port_comm, pd->port_pid,
- pd->port_hdrqfull);
- pd->port_hdrqfull = 0;
- }
+ dd->ipath_pd[port] = NULL;
+ pid = pd->port_pid;
+ pd->port_pid = NULL;
+ spin_unlock_irqrestore(&dd->ipath_uctxt_lock, flags);
if (pd->port_rcvwait_to || pd->port_piowait_to
|| pd->port_rcvnowait || pd->port_pionowait) {
pd->port_rcvnowait = pd->port_pionowait = 0;
}
if (pd->port_flag) {
- ipath_dbg("port %u port_flag still set to 0x%lx\n",
+ ipath_cdbg(PROC, "port %u port_flag set: 0x%lx\n",
pd->port_port, pd->port_flag);
pd->port_flag = 0;
}
if (dd->ipath_kregbase) {
- int i;
/* atomically clear receive enable port and intr avail. */
clear_bit(dd->ipath_r_portenable_shift + port,
&dd->ipath_rcvctrl);
ipath_write_kreg_port(dd, dd->ipath_kregs->kr_rcvhdraddr,
pd->port_port, dd->ipath_dummy_hdrq_phys);
- i = dd->ipath_pbufsport * (port - 1);
- ipath_disarm_piobufs(dd, i, dd->ipath_pbufsport);
+ ipath_disarm_piobufs(dd, pd->port_pio_base, pd->port_piocnt);
+ ipath_chg_pioavailkernel(dd, pd->port_pio_base,
+ pd->port_piocnt, 1);
dd->ipath_f_clear_tids(dd, pd->port_port);
unlock_expected_tids(pd);
ipath_stats.sps_ports--;
ipath_cdbg(PROC, "%s[%u] closed port %u:%u\n",
- pd->port_comm, pd->port_pid,
+ pd->port_comm, pid_nr(pid),
dd->ipath_unit, port);
}
- pd->port_pid = 0;
- dd->ipath_pd[pd->port_port] = NULL; /* before releasing mutex */
+ put_pid(pid);
mutex_unlock(&ipath_mutex);
ipath_free_pddata(dd, pd); /* after releasing the mutex */
return ret;
}
-static int ipath_force_pio_avail_update(struct ipath_devdata *dd)
+static int ipath_sdma_get_inflight(struct ipath_user_sdma_queue *pq,
+ u32 __user *inflightp)
{
- unsigned long flags;
+ const u32 val = ipath_user_sdma_inflight_counter(pq);
- spin_lock_irqsave(&dd->ipath_sendctrl_lock, flags);
- ipath_write_kreg(dd, dd->ipath_kregs->kr_sendctrl,
- dd->ipath_sendctrl & ~INFINIPATH_S_PIOBUFAVAILUPD);
- ipath_read_kreg64(dd, dd->ipath_kregs->kr_scratch);
- ipath_write_kreg(dd, dd->ipath_kregs->kr_sendctrl, dd->ipath_sendctrl);
- ipath_read_kreg64(dd, dd->ipath_kregs->kr_scratch);
- spin_unlock_irqrestore(&dd->ipath_sendctrl_lock, flags);
+ if (put_user(val, inflightp))
+ return -EFAULT;
+
+ return 0;
+}
+
+static int ipath_sdma_get_complete(struct ipath_devdata *dd,
+ struct ipath_user_sdma_queue *pq,
+ u32 __user *completep)
+{
+ u32 val;
+ int err;
+
+ err = ipath_user_sdma_make_progress(dd, pq);
+ if (err < 0)
+ return err;
+
+ val = ipath_user_sdma_complete_counter(pq);
+ if (put_user(val, completep))
+ return -EFAULT;
return 0;
}
dest = &cmd.cmd.poll_type;
src = &ucmd->cmd.poll_type;
break;
+ case IPATH_CMD_ARMLAUNCH_CTRL:
+ copy = sizeof(cmd.cmd.armlaunch_ctrl);
+ dest = &cmd.cmd.armlaunch_ctrl;
+ src = &ucmd->cmd.armlaunch_ctrl;
+ break;
+ case IPATH_CMD_SDMA_INFLIGHT:
+ copy = sizeof(cmd.cmd.sdma_inflight);
+ dest = &cmd.cmd.sdma_inflight;
+ src = &ucmd->cmd.sdma_inflight;
+ break;
+ case IPATH_CMD_SDMA_COMPLETE:
+ copy = sizeof(cmd.cmd.sdma_complete);
+ dest = &cmd.cmd.sdma_complete;
+ src = &ucmd->cmd.sdma_complete;
+ break;
default:
ret = -EINVAL;
goto bail;
cmd.cmd.slave_mask_addr);
break;
case IPATH_CMD_PIOAVAILUPD:
- ret = ipath_force_pio_avail_update(pd->port_dd);
+ ipath_force_pio_avail_update(pd->port_dd);
break;
case IPATH_CMD_POLL_TYPE:
pd->poll_type = cmd.cmd.poll_type;
break;
+ case IPATH_CMD_ARMLAUNCH_CTRL:
+ if (cmd.cmd.armlaunch_ctrl)
+ ipath_enable_armlaunch(pd->port_dd);
+ else
+ ipath_disable_armlaunch(pd->port_dd);
+ break;
+ case IPATH_CMD_SDMA_INFLIGHT:
+ ret = ipath_sdma_get_inflight(user_sdma_queue_fp(fp),
+ (u32 __user *) (unsigned long)
+ cmd.cmd.sdma_inflight);
+ break;
+ case IPATH_CMD_SDMA_COMPLETE:
+ ret = ipath_sdma_get_complete(pd->port_dd,
+ user_sdma_queue_fp(fp),
+ (u32 __user *) (unsigned long)
+ cmd.cmd.sdma_complete);
+ break;
}
if (ret >= 0)
return ret;
}
+static ssize_t ipath_writev(struct kiocb *iocb, const struct iovec *iov,
+ unsigned long dim, loff_t off)
+{
+ struct file *filp = iocb->ki_filp;
+ struct ipath_filedata *fp = filp->private_data;
+ struct ipath_portdata *pd = port_fp(filp);
+ struct ipath_user_sdma_queue *pq = fp->pq;
+
+ if (!dim)
+ return -EINVAL;
+
+ return ipath_user_sdma_writev(pd->port_dd, pq, iov, dim);
+}
+
static struct class *ipath_class;
static int init_cdev(int minor, char *name, const struct file_operations *fops,
- struct cdev **cdevp, struct class_device **class_devp)
+ struct cdev **cdevp, struct device **devp)
{
const dev_t dev = MKDEV(IPATH_MAJOR, minor);
struct cdev *cdev = NULL;
- struct class_device *class_dev = NULL;
+ struct device *device = NULL;
int ret;
cdev = cdev_alloc();
goto err_cdev;
}
- class_dev = class_device_create(ipath_class, NULL, dev, NULL, name);
+ device = device_create(ipath_class, NULL, dev, NULL, name);
- if (IS_ERR(class_dev)) {
- ret = PTR_ERR(class_dev);
+ if (IS_ERR(device)) {
+ ret = PTR_ERR(device);
printk(KERN_ERR IPATH_DRV_NAME ": Could not create "
- "class_dev for minor %d, %s (err %d)\n",
+ "device for minor %d, %s (err %d)\n",
minor, name, -ret);
goto err_cdev;
}
done:
if (ret >= 0) {
*cdevp = cdev;
- *class_devp = class_dev;
+ *devp = device;
} else {
*cdevp = NULL;
- *class_devp = NULL;
+ *devp = NULL;
}
return ret;
}
int ipath_cdev_init(int minor, char *name, const struct file_operations *fops,
- struct cdev **cdevp, struct class_device **class_devp)
+ struct cdev **cdevp, struct device **devp)
{
- return init_cdev(minor, name, fops, cdevp, class_devp);
+ return init_cdev(minor, name, fops, cdevp, devp);
}
static void cleanup_cdev(struct cdev **cdevp,
- struct class_device **class_devp)
+ struct device **devp)
{
- struct class_device *class_dev = *class_devp;
+ struct device *dev = *devp;
- if (class_dev) {
- class_device_unregister(class_dev);
- *class_devp = NULL;
+ if (dev) {
+ device_unregister(dev);
+ *devp = NULL;
}
if (*cdevp) {
}
void ipath_cdev_cleanup(struct cdev **cdevp,
- struct class_device **class_devp)
+ struct device **devp)
{
- cleanup_cdev(cdevp, class_devp);
+ cleanup_cdev(cdevp, devp);
}
static struct cdev *wildcard_cdev;
-static struct class_device *wildcard_class_dev;
+static struct device *wildcard_dev;
static const dev_t dev = MKDEV(IPATH_MAJOR, 0);
goto bail;
}
ret = init_cdev(0, "ipath", &ipath_file_ops, &wildcard_cdev,
- &wildcard_class_dev);
+ &wildcard_dev);
if (ret < 0) {
ipath_dev_err(dd, "Could not create wildcard "
"minor: error %d\n", -ret);
snprintf(name, sizeof(name), "ipath%d", dd->ipath_unit);
ret = init_cdev(dd->ipath_unit + 1, name, &ipath_file_ops,
- &dd->user_cdev, &dd->user_class_dev);
+ &dd->user_cdev, &dd->user_dev);
if (ret < 0)
ipath_dev_err(dd, "Could not create user minor %d, %s\n",
dd->ipath_unit + 1, name);
void ipath_user_remove(struct ipath_devdata *dd)
{
- cleanup_cdev(&dd->user_cdev, &dd->user_class_dev);
+ cleanup_cdev(&dd->user_cdev, &dd->user_dev);
if (atomic_dec_return(&user_count) == 0) {
if (atomic_read(&user_setup) == 0)
goto bail;
- cleanup_cdev(&wildcard_cdev, &wildcard_class_dev);
+ cleanup_cdev(&wildcard_cdev, &wildcard_dev);
user_cleanup();
atomic_set(&user_setup, 0);