ipath_file_ops.c

来自「linux 内核源代码」· C语言 代码 · 共 2,267 行 · 第 1/5 页

/*
 * Copyright (c) 2006, 2007 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
 * licenses.  You may choose to be licensed under the terms of the GNU
 * General Public License (GPL) Version 2, available from the file
 * COPYING in the main directory of this source tree, or the
 * OpenIB.org BSD license below:
 *
 *     Redistribution and use in source and binary forms, with or
 *     without modification, are permitted provided that the following
 *     conditions are met:
 *
 *      - Redistributions of source code must retain the above
 *        copyright notice, this list of conditions and the following
 *        disclaimer.
 *
 *      - 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.
 *
 * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND,
 * EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF
 * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND
 * NONINFRINGEMENT. IN NO EVENT SHALL THE AUTHORS OR COPYRIGHT HOLDERS
 * BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN
 * ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN
 * CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
 * SOFTWARE.
 */

#include <linux/pci.h>
#include <linux/poll.h>
#include <linux/cdev.h>
#include <linux/swap.h>
#include <linux/vmalloc.h>
#include <asm/pgtable.h>

#include "ipath_kernel.h"
#include "ipath_common.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 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,
	.open = ipath_open,
	.release = ipath_close,
	.poll = ipath_poll,
	.mmap = ipath_mmap
};

/*
 * Convert kernel virtual addresses to physical addresses so they don't
 * potentially conflict with the chip addresses used as mmap offsets.
 * It doesn't really matter what mmap offset we use as long as we can
 * interpret it correctly.
 */
static u64 cvt_kvaddr(void *p)
{
	struct page *page;
	u64 paddr = 0;

	page = vmalloc_to_page(p);
	if (page)
		paddr = page_to_pfn(page) << PAGE_SHIFT;

	return paddr;
}

static int ipath_get_base_info(struct file *fp,
			       void __user *ubase, size_t ubase_size)
{
	struct ipath_portdata *pd = port_fp(fp);
	int ret = 0;
	struct ipath_base_info *kinfo = NULL;
	struct ipath_devdata *dd = pd->port_dd;
	unsigned subport_cnt;
	int shared, master;
	size_t sz;

	subport_cnt = pd->port_subport_cnt;
	if (!subport_cnt) {
		shared = 0;
		master = 0;
		subport_cnt = 1;
	} else {
		shared = 1;
		master = !subport_fp(fp);
	}

	sz = sizeof(*kinfo);
	/* If port sharing is not requested, allow the old size structure */
	if (!shared)
		sz -= 7 * sizeof(u64);
	if (ubase_size < sz) {
		ipath_cdbg(PROC,
			   "Base size %zu, need %zu (version mismatch?)\n",
			   ubase_size, sz);
		ret = -EINVAL;
		goto bail;
	}

	kinfo = kzalloc(sizeof(*kinfo), GFP_KERNEL);
	if (kinfo == NULL) {
		ret = -ENOMEM;
		goto bail;
	}

	ret = dd->ipath_f_get_base_info(pd, kinfo);
	if (ret < 0)
		goto bail;

	kinfo->spi_rcvhdr_cnt = dd->ipath_rcvhdrcnt;
	kinfo->spi_rcvhdrent_size = dd->ipath_rcvhdrentsize;
	kinfo->spi_tidegrcnt = dd->ipath_rcvegrcnt;
	kinfo->spi_rcv_egrbufsize = dd->ipath_rcvegrbufsize;
	/*
	 * have to mmap whole thing
	 */
	kinfo->spi_rcv_egrbuftotlen =
		pd->port_rcvegrbuf_chunks * pd->port_rcvegrbuf_size;
	kinfo->spi_rcv_egrperchunk = pd->port_rcvegrbufs_perchunk;
	kinfo->spi_rcv_egrchunksize = kinfo->spi_rcv_egrbuftotlen /
		pd->port_rcvegrbuf_chunks;
	kinfo->spi_tidcnt = dd->ipath_rcvtidcnt / subport_cnt;
	if (master)
		kinfo->spi_tidcnt += dd->ipath_rcvtidcnt % subport_cnt;
	/*
	 * for this use, may be ipath_cfgports summed over all chips that
	 * are are configured and present
	 */
	kinfo->spi_nports = dd->ipath_cfgports;
	/* unit (chip/board) our port is on */
	kinfo->spi_unit = dd->ipath_unit;
	/* for now, only a single page */
	kinfo->spi_tid_maxsize = PAGE_SIZE;

	/*
	 * Doing this per port, and based on the skip value, etc.  This has
	 * to be the actual buffer size, since the protocol code treats it
	 * as an array.
	 *
	 * These have to be set to user addresses in the user code via mmap.
	 * These values are used on return to user code for the mmap target
	 * addresses only.  For 32 bit, same 44 bit address problem, so use
	 * the physical address, not virtual.  Before 2.6.11, using the
	 * page_address() macro worked, but in 2.6.11, even that returns the
	 * full 64 bit address (upper bits all 1's).  So far, using the
	 * physical addresses (or chip offsets, for chip mapping) works, but
	 * no doubt some future kernel release will change that, and we'll be
	 * on to yet another method of dealing with this.
	 */
	kinfo->spi_rcvhdr_base = (u64) pd->port_rcvhdrq_phys;
	kinfo->spi_rcvhdr_tailaddr = (u64) pd->port_rcvhdrqtailaddr_phys;
	kinfo->spi_rcv_egrbufs = (u64) pd->port_rcvegr_phys;
	kinfo->spi_pioavailaddr = (u64) dd->ipath_pioavailregs_phys;
	kinfo->spi_status = (u64) kinfo->spi_pioavailaddr +
		(void *) dd->ipath_statusp -
		(void *) dd->ipath_pioavailregs_dma;
	if (!shared) {
		kinfo->spi_piocnt = dd->ipath_pbufsport;
		kinfo->spi_piobufbase = (u64) pd->port_piobufs;
		kinfo->__spi_uregbase = (u64) dd->ipath_uregbase +
			dd->ipath_palign * pd->port_port;
	} else if (master) {
		kinfo->spi_piocnt = (dd->ipath_pbufsport / subport_cnt) +
				    (dd->ipath_pbufsport % 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);
	} else {
		unsigned slave = subport_fp(fp) - 1;

		kinfo->spi_piocnt = dd->ipath_pbufsport / 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_palign * pd->port_port;
		kinfo->spi_port_rcvegrbuf = kinfo->spi_rcv_egrbufs;
		kinfo->spi_port_rcvhdr_base = kinfo->spi_rcvhdr_base;
		kinfo->spi_port_rcvhdr_tailaddr = kinfo->spi_rcvhdr_tailaddr;

		kinfo->__spi_uregbase = cvt_kvaddr(pd->subport_uregbase +
			PAGE_SIZE * subport_fp(fp));

		kinfo->spi_rcvhdr_base = cvt_kvaddr(pd->subport_rcvhdr_base +
			pd->port_rcvhdrq_size * subport_fp(fp));
		kinfo->spi_rcvhdr_tailaddr = 0;
		kinfo->spi_rcv_egrbufs = cvt_kvaddr(pd->subport_rcvegrbuf +
			pd->port_rcvegrbuf_chunks * pd->port_rcvegrbuf_size *
			subport_fp(fp));

		kinfo->spi_subport_uregbase =
			cvt_kvaddr(pd->subport_uregbase);
		kinfo->spi_subport_rcvegrbuf =
			cvt_kvaddr(pd->subport_rcvegrbuf);
		kinfo->spi_subport_rcvhdr_base =
			cvt_kvaddr(pd->subport_rcvhdr_base);
		ipath_cdbg(PROC, "port %u flags %x %llx %llx %llx\n",
			kinfo->spi_port, kinfo->spi_runtime_flags,
			(unsigned long long) kinfo->spi_subport_uregbase,
			(unsigned long long) kinfo->spi_subport_rcvegrbuf,
			(unsigned long long) kinfo->spi_subport_rcvhdr_base);
	}

	kinfo->spi_pioindex = (kinfo->spi_piobufbase - dd->ipath_piobufbase) /
		dd->ipath_palign;
	kinfo->spi_pioalign = dd->ipath_palign;

	kinfo->spi_qpair = IPATH_KD_QP;
	kinfo->spi_piosize = dd->ipath_ibmaxlen;
	kinfo->spi_mtu = dd->ipath_ibmaxlen;	/* maxlen, not ibmtu */
	kinfo->spi_port = pd->port_port;
	kinfo->spi_subport = subport_fp(fp);
	kinfo->spi_sw_version = IPATH_KERN_SWVERSION;
	kinfo->spi_hw_version = dd->ipath_revision;

	if (master) {
		kinfo->spi_runtime_flags |= IPATH_RUNTIME_MASTER;
	}

	sz = (ubase_size < sizeof(*kinfo)) ? ubase_size : sizeof(*kinfo);
	if (copy_to_user(ubase, kinfo, sz))
		ret = -EFAULT;

bail:
	kfree(kinfo);
	return ret;
}

/**
 * ipath_tid_update - update a port TID
 * @pd: the port
 * @fp: the ipath device file
 * @ti: the TID information
 *
 * The new implementation as of Oct 2004 is that the driver assigns
 * the tid and returns it to the caller.   To make it easier to
 * catch bugs, and to reduce search time, we keep a cursor for
 * each port, walking the shadow tid array to find one that's not
 * in use.
 *
 * For now, if we can't allocate the full list, we fail, although
 * in the long run, we'll allocate as many as we can, and the
 * caller will deal with that by trying the remaining pages later.
 * That means that when we fail, we have to mark the tids as not in
 * use again, in our shadow copy.
 *
 * It's up to the caller to free the tids when they are done.
 * We'll unlock the pages as they free them.
 *
 * Also, right now we are locking one page at a time, but since
 * the intended use of this routine is for a single group of
 * virtually contiguous pages, that should change to improve
 * performance.
 */
static int ipath_tid_update(struct ipath_portdata *pd, struct file *fp,
			    const struct ipath_tid_info *ti)
{
	int ret = 0, ntids;
	u32 tid, porttid, cnt, i, tidcnt, tidoff;
	u16 *tidlist;
	struct ipath_devdata *dd = pd->port_dd;
	u64 physaddr;
	unsigned long vaddr;
	u64 __iomem *tidbase;
	unsigned long tidmap[8];
	struct page **pagep = NULL;
	unsigned subport = subport_fp(fp);

	if (!dd->ipath_pageshadow) {
		ret = -ENOMEM;
		goto done;
	}

	cnt = ti->tidcnt;
	if (!cnt) {
		ipath_dbg("After copyin, tidcnt 0, tidlist %llx\n",
			  (unsigned long long) ti->tidlist);
		/*
		 * Should we treat as success?  likely a bug
		 */
		ret = -EFAULT;
		goto done;
	}
	porttid = pd->port_port * dd->ipath_rcvtidcnt;
	if (!pd->port_subport_cnt) {
		tidcnt = dd->ipath_rcvtidcnt;
		tid = pd->port_tidcursor;
		tidoff = 0;
	} else if (!subport) {
		tidcnt = (dd->ipath_rcvtidcnt / pd->port_subport_cnt) +
			 (dd->ipath_rcvtidcnt % pd->port_subport_cnt);
		tidoff = dd->ipath_rcvtidcnt - tidcnt;
		porttid += tidoff;
		tid = tidcursor_fp(fp);
	} else {
		tidcnt = dd->ipath_rcvtidcnt / pd->port_subport_cnt;
		tidoff = tidcnt * (subport - 1);
		porttid += tidoff;
		tid = tidcursor_fp(fp);
	}
	if (cnt > tidcnt) {
		/* make sure it all fits in port_tid_pg_list */
		dev_info(&dd->pcidev->dev, "Process tried to allocate %u "
			 "TIDs, only trying max (%u)\n", cnt, tidcnt);
		cnt = tidcnt;
	}
	pagep = &((struct page **) pd->port_tid_pg_list)[tidoff];
	tidlist = &((u16 *) &pagep[dd->ipath_rcvtidcnt])[tidoff];

	memset(tidmap, 0, sizeof(tidmap));
	/* before decrement; chip actual # */
	ntids = tidcnt;
	tidbase = (u64 __iomem *) (((char __iomem *) dd->ipath_kregbase) +
				   dd->ipath_rcvtidbase +
				   porttid * sizeof(*tidbase));

	ipath_cdbg(VERBOSE, "Port%u %u tids, cursor %u, tidbase %p\n",
		   pd->port_port, cnt, tid, tidbase);

	/* virtual address of first page in transfer */
	vaddr = ti->tidvaddr;
	if (!access_ok(VERIFY_WRITE, (void __user *) vaddr,
		       cnt * PAGE_SIZE)) {
		ipath_dbg("Fail vaddr %p, %u pages, !access_ok\n",
			  (void *)vaddr, cnt);
		ret = -EFAULT;
		goto done;
	}
	ret = ipath_get_user_pages(vaddr, cnt, pagep);
	if (ret) {
		if (ret == -EBUSY) {
			ipath_dbg("Failed to lock addr %p, %u pages "
				  "(already locked)\n",
				  (void *) vaddr, cnt);
			/*
			 * for now, continue, and see what happens but with
			 * the new implementation, this should never happen,
			 * unless perhaps the user has mpin'ed the pages
			 * themselves (something we need to test)
			 */
			ret = 0;
		} else {
			dev_info(&dd->pcidev->dev,
				 "Failed to lock addr %p, %u pages: "
				 "errno %d\n", (void *) vaddr, cnt, -ret);
			goto done;
		}
	}
	for (i = 0; i < cnt; i++, vaddr += PAGE_SIZE) {
		for (; ntids--; tid++) {
			if (tid == tidcnt)
				tid = 0;
			if (!dd->ipath_pageshadow[porttid + tid])
				break;
		}
		if (ntids < 0) {
			/*
			 * oops, wrapped all the way through their TIDs,
			 * and didn't have enough free; see comments at
			 * start of routine
			 */
			ipath_dbg("Not enough free TIDs for %u pages "
				  "(index %d), failing\n", cnt, i);
			i--;	/* last tidlist[i] not filled in */
			ret = -ENOMEM;
			break;
		}
		tidlist[i] = tid + tidoff;
		ipath_cdbg(VERBOSE, "Updating idx %u to TID %u, "
			   "vaddr %lx\n", i, tid + tidoff, vaddr);
		/* we "know" system pages and TID pages are same size */
		dd->ipath_pageshadow[porttid + tid] = pagep[i];
		dd->ipath_physshadow[porttid + tid] = ipath_map_page(
			dd->pcidev, pagep[i], 0, PAGE_SIZE,
			PCI_DMA_FROMDEVICE);
		/*
		 * don't need atomic or it's overhead
		 */
		__set_bit(tid, tidmap);
		physaddr = dd->ipath_physshadow[porttid + tid];
		ipath_stats.sps_pagelocks++;
		ipath_cdbg(VERBOSE,
			   "TID %u, vaddr %lx, physaddr %llx pgp %p\n",
			   tid, vaddr, (unsigned long long) physaddr,
			   pagep[i]);
		dd->ipath_f_put_tid(dd, &tidbase[tid], RCVHQ_RCV_TYPE_EXPECTED,
				    physaddr);
		/*
		 * don't check this tid in ipath_portshadow, since we
		 * just filled it in; start with the next one.
		 */
		tid++;
	}

	if (ret) {
		u32 limit;
	cleanup:
		/* jump here if copy out of updated info failed... */
		ipath_dbg("After failure (ret=%d), undo %d of %d entries\n",
			  -ret, i, cnt);
		/* same code that's in ipath_free_tid() */
		limit = sizeof(tidmap) * BITS_PER_BYTE;
		if (limit > tidcnt)
			/* just in case size changes in future */
			limit = tidcnt;
		tid = find_first_bit((const unsigned long *)tidmap, limit);
		for (; tid < limit; tid++) {
			if (!test_bit(tid, tidmap))
				continue;
			if (dd->ipath_pageshadow[porttid + tid]) {
				ipath_cdbg(VERBOSE, "Freeing TID %u\n",
					   tid);
				dd->ipath_f_put_tid(dd, &tidbase[tid],
						    RCVHQ_RCV_TYPE_EXPECTED,
						    dd->ipath_tidinvalid);
				pci_unmap_page(dd->pcidev,
					dd->ipath_physshadow[porttid + tid],
					PAGE_SIZE, PCI_DMA_FROMDEVICE);
				dd->ipath_pageshadow[porttid + tid] = NULL;
				ipath_stats.sps_pageunlocks++;
			}
		}
		ipath_release_user_pages(pagep, cnt);
	} else {
		/*
		 * Copy the updated array, with ipath_tid's filled in, back
		 * to user.  Since we did the copy in already, this "should
		 * never fail" If it does, we have to clean up...
		 */
		if (copy_to_user((void __user *)
				 (unsigned long) ti->tidlist,
				 tidlist, cnt * sizeof(*tidlist))) {
			ret = -EFAULT;
			goto cleanup;
		}
		if (copy_to_user((void __user *) (unsigned long) ti->tidmap,
				 tidmap, sizeof tidmap)) {
			ret = -EFAULT;
			goto cleanup;
		}
		if (tid == tidcnt)