ipath_file_ops.c

来自「LINUX 2.6.17.4的源码」· C语言 代码 · 共 1,915 行 · 第 1/4 页

	return ret;
}

static int mmap_piobufs(struct vm_area_struct *vma,
			struct ipath_devdata *dd,
			struct ipath_portdata *pd)
{
	unsigned long phys;
	int ret;

	/*
	 * When we map the PIO buffers, we want to map them as writeonly, no
	 * read possible.
	 */

	if ((vma->vm_end - vma->vm_start) >
	    (dd->ipath_pbufsport * dd->ipath_palign)) {
		dev_info(&dd->pcidev->dev, "FAIL mmap piobufs: "
			 "reqlen %lx > PAGE\n",
			 vma->vm_end - vma->vm_start);
		ret = -EFAULT;
		goto bail;
	}

	phys = dd->ipath_physaddr + pd->port_piobufs;
	/*
	 * Do *NOT* mark this as non-cached (PWT bit), or we don't get the
	 * write combining behavior we want on the PIO buffers!
	 * vma->vm_page_prot =
	 *        pgprot_noncached(vma->vm_page_prot);
	 */

	if (vma->vm_flags & VM_READ) {
		dev_info(&dd->pcidev->dev,
			 "Can't map piobufs as readable (flags=%lx)\n",
			 vma->vm_flags);
		ret = -EPERM;
		goto bail;
	}

	/* don't allow them to later change to readable with mprotect */

	vma->vm_flags &= ~VM_MAYWRITE;
	vma->vm_flags |= VM_DONTCOPY | VM_DONTEXPAND;

	ret = io_remap_pfn_range(vma, vma->vm_start, phys >> PAGE_SHIFT,
				 vma->vm_end - vma->vm_start,
				 vma->vm_page_prot);
bail:
	return ret;
}

static int mmap_rcvegrbufs(struct vm_area_struct *vma,
			   struct ipath_portdata *pd)
{
	struct ipath_devdata *dd = pd->port_dd;
	unsigned long start, size;
	size_t total_size, i;
	dma_addr_t *phys;
	int ret;

	if (!pd->port_rcvegrbuf) {
		ret = -EFAULT;
		goto bail;
	}

	size = pd->port_rcvegrbuf_size;
	total_size = pd->port_rcvegrbuf_chunks * size;
	if ((vma->vm_end - vma->vm_start) > total_size) {
		dev_info(&dd->pcidev->dev, "FAIL on egr bufs: "
			 "reqlen %lx > actual %lx\n",
			 vma->vm_end - vma->vm_start,
			 (unsigned long) total_size);
		ret = -EFAULT;
		goto bail;
	}

	if (vma->vm_flags & VM_WRITE) {
		dev_info(&dd->pcidev->dev, "Can't map eager buffers as "
			 "writable (flags=%lx)\n", vma->vm_flags);
		ret = -EPERM;
		goto bail;
	}

	start = vma->vm_start;
	phys = pd->port_rcvegrbuf_phys;

	/* don't allow them to later change to writeable with mprotect */
	vma->vm_flags &= ~VM_MAYWRITE;

	for (i = 0; i < pd->port_rcvegrbuf_chunks; i++, start += size) {
		ret = remap_pfn_range(vma, start, phys[i] >> PAGE_SHIFT,
				      size, vma->vm_page_prot);
		if (ret < 0)
			goto bail;
	}
	ret = 0;

bail:
	return ret;
}

static int mmap_rcvhdrq(struct vm_area_struct *vma,
			struct ipath_portdata *pd)
{
	struct ipath_devdata *dd = pd->port_dd;
	size_t total_size;
	int ret;

	/*
	 * kmalloc'ed memory, physically contiguous; this is from
	 * spi_rcvhdr_base; we allow user to map read-write so they can
	 * write hdrq entries to allow protocol code to directly poll
	 * whether a hdrq entry has been written.
	 */
	total_size = ALIGN(dd->ipath_rcvhdrcnt * dd->ipath_rcvhdrentsize *
			   sizeof(u32), PAGE_SIZE);
	if ((vma->vm_end - vma->vm_start) > total_size) {
		dev_info(&dd->pcidev->dev,
			 "FAIL on rcvhdrq: reqlen %lx > actual %lx\n",
			 vma->vm_end - vma->vm_start,
			 (unsigned long) total_size);
		ret = -EFAULT;
		goto bail;
	}

	ret = remap_pfn_range(vma, vma->vm_start,
			      pd->port_rcvhdrq_phys >> PAGE_SHIFT,
			      vma->vm_end - vma->vm_start,
			      vma->vm_page_prot);
bail:
	return ret;
}

static int mmap_pioavailregs(struct vm_area_struct *vma,
			     struct ipath_portdata *pd)
{
	struct ipath_devdata *dd = pd->port_dd;
	int ret;

	/*
	 * when we map the PIO bufferavail registers, we want to map them as
	 * readonly, no write possible.
	 *
	 * kmalloc'ed memory, physically contiguous, one page only, readonly
	 */

	if ((vma->vm_end - vma->vm_start) > PAGE_SIZE) {
		dev_info(&dd->pcidev->dev, "FAIL on pioavailregs_dma: "
			 "reqlen %lx > actual %lx\n",
			 vma->vm_end - vma->vm_start,
			 (unsigned long) PAGE_SIZE);
		ret = -EFAULT;
		goto bail;
	}

	if (vma->vm_flags & VM_WRITE) {
		dev_info(&dd->pcidev->dev,
			 "Can't map pioavailregs as writable (flags=%lx)\n",
			 vma->vm_flags);
		ret = -EPERM;
		goto bail;
	}

	/* don't allow them to later change with mprotect */
	vma->vm_flags &= ~VM_MAYWRITE;

	ret = remap_pfn_range(vma, vma->vm_start,
			      dd->ipath_pioavailregs_phys >> PAGE_SHIFT,
			      PAGE_SIZE, vma->vm_page_prot);
bail:
	return ret;
}

/**
 * ipath_mmap - mmap various structures into user space
 * @fp: the file pointer
 * @vma: the VM area
 *
 * We use this to have a shared buffer between the kernel and the user code
 * for the rcvhdr queue, egr buffers, and the per-port user regs and pio
 * buffers in the chip.  We have the open and close entries so we can bump
 * the ref count and keep the driver from being unloaded while still mapped.
 */
static int ipath_mmap(struct file *fp, struct vm_area_struct *vma)
{
	struct ipath_portdata *pd;
	struct ipath_devdata *dd;
	u64 pgaddr, ureg;
	int ret;

	pd = port_fp(fp);
	dd = pd->port_dd;
	/*
	 * This is the ipath_do_user_init() code, mapping the shared buffers
	 * into the user process. The address referred to by vm_pgoff is the
	 * virtual, not physical, address; we only do one mmap for each
	 * space mapped.
	 */
	pgaddr = vma->vm_pgoff << PAGE_SHIFT;

	/*
	 * note that ureg does *NOT* have the kregvirt as part of it, to be
	 * sure that for 32 bit programs, we don't end up trying to map a >
	 * 44 address.  Has to match ipath_get_base_info() code that sets
	 * __spi_uregbase
	 */

	ureg = dd->ipath_uregbase + dd->ipath_palign * pd->port_port;

	ipath_cdbg(MM, "ushare: pgaddr %llx vm_start=%lx, vmlen %lx\n",
		   (unsigned long long) pgaddr, vma->vm_start,
		   vma->vm_end - vma->vm_start);

	if (pgaddr == ureg)
		ret = mmap_ureg(vma, dd, ureg);
	else if (pgaddr == pd->port_piobufs)
		ret = mmap_piobufs(vma, dd, pd);
	else if (pgaddr == (u64) pd->port_rcvegr_phys)
		ret = mmap_rcvegrbufs(vma, pd);
	else if (pgaddr == (u64) pd->port_rcvhdrq_phys)
		ret = mmap_rcvhdrq(vma, pd);
	else if (pgaddr == dd->ipath_pioavailregs_phys)
		ret = mmap_pioavailregs(vma, pd);
	else
		ret = -EINVAL;

	vma->vm_private_data = NULL;

	if (ret < 0)
		dev_info(&dd->pcidev->dev,
			 "Failure %d on addr %lx, off %lx\n",
			 -ret, vma->vm_start, vma->vm_pgoff);

	return ret;
}

static unsigned int ipath_poll(struct file *fp,
			       struct poll_table_struct *pt)
{
	struct ipath_portdata *pd;
	u32 head, tail;
	int bit;
	struct ipath_devdata *dd;

	pd = port_fp(fp);
	dd = pd->port_dd;

	bit = pd->port_port + INFINIPATH_R_INTRAVAIL_SHIFT;
	set_bit(bit, &dd->ipath_rcvctrl);

	/*
	 * Before blocking, make sure that head is still == tail,
	 * reading from the chip, so we can be sure the interrupt
	 * enable has made it to the chip.  If not equal, disable
	 * interrupt again and return immediately.  This avoids races,
	 * and the overhead of the chip read doesn't matter much at
	 * this point, since we are waiting for something anyway.
	 */

	ipath_write_kreg(dd, dd->ipath_kregs->kr_rcvctrl,
			 dd->ipath_rcvctrl);

	head = ipath_read_ureg32(dd, ur_rcvhdrhead, pd->port_port);
	tail = ipath_read_ureg32(dd, ur_rcvhdrtail, pd->port_port);

	if (tail == head) {
		set_bit(IPATH_PORT_WAITING_RCV, &pd->port_flag);
		if(dd->ipath_rhdrhead_intr_off) /* arm rcv interrupt */
			(void)ipath_write_ureg(dd, ur_rcvhdrhead,
					       dd->ipath_rhdrhead_intr_off
					       | head, pd->port_port);
		poll_wait(fp, &pd->port_wait, pt);

		if (test_bit(IPATH_PORT_WAITING_RCV, &pd->port_flag)) {
			/* timed out, no packets received */
			clear_bit(IPATH_PORT_WAITING_RCV, &pd->port_flag);
			pd->port_rcvwait_to++;
		}
	}
	else {
		/* it's already happened; don't do wait_event overhead */
		pd->port_rcvnowait++;
	}

	clear_bit(bit, &dd->ipath_rcvctrl);
	ipath_write_kreg(dd, dd->ipath_kregs->kr_rcvctrl,
			 dd->ipath_rcvctrl);

	return 0;
}

static int try_alloc_port(struct ipath_devdata *dd, int port,
			  struct file *fp)
{
	int ret;

	if (!dd->ipath_pd[port]) {
		void *p, *ptmp;

		p = kzalloc(sizeof(struct ipath_portdata), GFP_KERNEL);

		/*
		 * Allocate memory for use in ipath_tid_update() just once
		 * at open, not per call.  Reduces cost of expected send
		 * setup.
		 */
		ptmp = kmalloc(dd->ipath_rcvtidcnt * sizeof(u16) +
			       dd->ipath_rcvtidcnt * sizeof(struct page **),
			       GFP_KERNEL);
		if (!p || !ptmp) {
			ipath_dev_err(dd, "Unable to allocate portdata "
				      "memory, failing open\n");
			ret = -ENOMEM;
			kfree(p);
			kfree(ptmp);
			goto bail;
		}
		dd->ipath_pd[port] = p;
		dd->ipath_pd[port]->port_port = port;
		dd->ipath_pd[port]->port_dd = dd;
		dd->ipath_pd[port]->port_tid_pg_list = ptmp;
		init_waitqueue_head(&dd->ipath_pd[port]->port_wait);
	}
	if (!dd->ipath_pd[port]->port_cnt) {
		dd->ipath_pd[port]->port_cnt = 1;
		fp->private_data = (void *) dd->ipath_pd[port];
		ipath_cdbg(PROC, "%s[%u] opened unit:port %u:%u\n",
			   current->comm, current->pid, dd->ipath_unit,
			   port);
		dd->ipath_pd[port]->port_pid = current->pid;
		strncpy(dd->ipath_pd[port]->port_comm, current->comm,
			sizeof(dd->ipath_pd[port]->port_comm));
		ipath_stats.sps_ports++;
		ret = 0;
		goto bail;
	}
	ret = -EBUSY;

bail:
	return ret;
}

static inline int usable(struct ipath_devdata *dd)
{
	return dd &&
		(dd->ipath_flags & IPATH_PRESENT) &&
		dd->ipath_kregbase &&
		dd->ipath_lid &&
		!(dd->ipath_flags & (IPATH_LINKDOWN | IPATH_DISABLED
				     | IPATH_LINKUNK));
}

static int find_free_port(int unit, struct file *fp)
{
	struct ipath_devdata *dd = ipath_lookup(unit);
	int ret, i;

	if (!dd) {
		ret = -ENODEV;
		goto bail;
	}

	if (!usable(dd)) {
		ret = -ENETDOWN;
		goto bail;
	}

	for (i = 0; i < dd->ipath_cfgports; i++) {
		ret = try_alloc_port(dd, i, fp);
		if (ret != -EBUSY)
			goto bail;
	}
	ret = -EBUSY;

bail:
	return ret;
}

static int find_best_unit(struct file *fp)
{
	int ret = 0, i, prefunit = -1, devmax;
	int maxofallports, npresent, nup;
	int ndev;

	(void) ipath_count_units(&npresent, &nup, &maxofallports);

	/*
	 * This code is present to allow a knowledgeable person to
	 * specify the layout of processes to processors before opening
	 * this driver, and then we'll assign the process to the "closest"
	 * HT-400 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(),
	 * 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
	 * a cpu subset.  This algorithm fails for two HT-400's connected
	 * in tunnel fashion.  Eventually this needs real topology
	 * 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)) {
		int ncpus = num_online_cpus(), curcpu = -1;
		for (i = 0; i < ncpus; i++)
			if (cpu_isset(i, current->cpus_allowed)) {
				ipath_cdbg(PROC, "%s[%u] affinity set for "
					   "cpu %d\n", current->comm,
					   current->pid, i);
				curcpu = i;
			}
		if (curcpu != -1) {
			if (npresent) {
				prefunit = curcpu / (ncpus / npresent);
				ipath_dbg("%s[%u] %d chips, %d cpus, "
					  "%d cpus/chip, select unit %d\n",
					  current->comm, current->pid,
					  npresent, ncpus, ncpus / npresent,
					  prefunit);
			}
		}
	}

	/*
	 * user ports start at 1, kernel port is 0
	 * For now, we do round-robin access across all chips
	 */

	if (prefunit != -1)
		devmax = prefunit + 1;
	else
		devmax = ipath_count_units(NULL, NULL, NULL);
recheck:
	for (i = 1; i < maxofallports; i++) {
		for (ndev = prefunit != -1 ? prefunit : 0; ndev < devmax;
		     ndev++) {
			struct ipath_devdata *dd = ipath_lookup(ndev);

			if (!usable(dd))
				continue; /* can't use this unit */
			if (i >= dd->ipath_cfgports)
				/*
				 * Maxed out on users of this unit. Try
				 * next.
				 */
				continue;
			ret = try_alloc_port(dd, i, fp);
			if (!ret)
				goto done;
		}
	}

	if (npresent) {
		if (nup == 0) {
			ret = -ENETDOWN;
			ipath_dbg("No ports available (none initialized "
				  "and ready)\n");
		} else {
			if (prefunit > 0) {
				/* if started above 0, retry from 0 */
				ipath_cdbg(PROC,
					   "%s[%u] no ports on prefunit "
					   "%d, clear and re-check\n",
					   current->comm, current->pid,
					   prefunit);
				devmax = ipath_count_units(NULL, NULL,
							   NULL);
				prefunit = -1;
				goto recheck;
			}
			ret = -EBUSY;
			ipath_dbg("No ports available\n");
		}
	} else {
		ret = -ENXIO;
		ipath_dbg("No boards found\n");
	}