2 * Intel MIC Platform Software Stack (MPSS)
4 * Copyright(c) 2016 Intel Corporation.
6 * This program is free software; you can redistribute it and/or modify
7 * it under the terms of the GNU General Public License, version 2, as
8 * published by the Free Software Foundation.
10 * This program is distributed in the hope that it will be useful, but
11 * WITHOUT ANY WARRANTY; without even the implied warranty of
12 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
13 * General Public License for more details.
15 * The full GNU General Public License is included in this distribution in
16 * the file called "COPYING".
18 * Intel Virtio Over PCIe (VOP) driver.
21 #include <linux/sched.h>
22 #include <linux/poll.h>
23 #include <linux/dma-mapping.h>
25 #include <linux/mic_common.h>
26 #include "../common/mic_dev.h"
28 #include <linux/mic_ioctl.h>
31 /* Helper API to obtain the VOP PCIe device */
32 static inline struct device *vop_dev(struct vop_vdev *vdev)
34 return vdev->vpdev->dev.parent;
37 /* Helper API to check if a virtio device is initialized */
38 static inline int vop_vdev_inited(struct vop_vdev *vdev)
42 /* Device has not been created yet */
43 if (!vdev->dd || !vdev->dd->type) {
44 dev_err(vop_dev(vdev), "%s %d err %d\n",
45 __func__, __LINE__, -EINVAL);
48 /* Device has been removed/deleted */
49 if (vdev->dd->type == -1) {
50 dev_dbg(vop_dev(vdev), "%s %d err %d\n",
51 __func__, __LINE__, -ENODEV);
57 static void _vop_notify(struct vringh *vrh)
59 struct vop_vringh *vvrh = container_of(vrh, struct vop_vringh, vrh);
60 struct vop_vdev *vdev = vvrh->vdev;
61 struct vop_device *vpdev = vdev->vpdev;
62 s8 db = vdev->dc->h2c_vdev_db;
65 vpdev->hw_ops->send_intr(vpdev, db);
68 static void vop_virtio_init_post(struct vop_vdev *vdev)
70 struct mic_vqconfig *vqconfig = mic_vq_config(vdev->dd);
71 struct vop_device *vpdev = vdev->vpdev;
74 for (i = 0; i < vdev->dd->num_vq; i++) {
75 used_size = PAGE_ALIGN(sizeof(u16) * 3 +
76 sizeof(struct vring_used_elem) *
77 le16_to_cpu(vqconfig->num));
78 if (!le64_to_cpu(vqconfig[i].used_address)) {
79 dev_warn(vop_dev(vdev), "used_address zero??\n");
82 vdev->vvr[i].vrh.vring.used =
83 (void __force *)vpdev->hw_ops->ioremap(
85 le64_to_cpu(vqconfig[i].used_address),
89 vdev->dc->used_address_updated = 0;
91 dev_info(vop_dev(vdev), "%s: device type %d LINKUP\n",
92 __func__, vdev->virtio_id);
95 static inline void vop_virtio_device_reset(struct vop_vdev *vdev)
99 dev_dbg(vop_dev(vdev), "%s: status %d device type %d RESET\n",
100 __func__, vdev->dd->status, vdev->virtio_id);
102 for (i = 0; i < vdev->dd->num_vq; i++)
104 * Avoid lockdep false positive. The + 1 is for the vop
105 * mutex which is held in the reset devices code path.
107 mutex_lock_nested(&vdev->vvr[i].vr_mutex, i + 1);
109 /* 0 status means "reset" */
110 vdev->dd->status = 0;
111 vdev->dc->vdev_reset = 0;
112 vdev->dc->host_ack = 1;
114 for (i = 0; i < vdev->dd->num_vq; i++) {
115 struct vringh *vrh = &vdev->vvr[i].vrh;
117 vdev->vvr[i].vring.info->avail_idx = 0;
119 vrh->last_avail_idx = 0;
120 vrh->last_used_idx = 0;
123 for (i = 0; i < vdev->dd->num_vq; i++)
124 mutex_unlock(&vdev->vvr[i].vr_mutex);
127 static void vop_virtio_reset_devices(struct vop_info *vi)
129 struct list_head *pos, *tmp;
130 struct vop_vdev *vdev;
132 list_for_each_safe(pos, tmp, &vi->vdev_list) {
133 vdev = list_entry(pos, struct vop_vdev, list);
134 vop_virtio_device_reset(vdev);
136 wake_up(&vdev->waitq);
140 static void vop_bh_handler(struct work_struct *work)
142 struct vop_vdev *vdev = container_of(work, struct vop_vdev,
145 if (vdev->dc->used_address_updated)
146 vop_virtio_init_post(vdev);
148 if (vdev->dc->vdev_reset)
149 vop_virtio_device_reset(vdev);
152 wake_up(&vdev->waitq);
155 static irqreturn_t _vop_virtio_intr_handler(int irq, void *data)
157 struct vop_vdev *vdev = data;
158 struct vop_device *vpdev = vdev->vpdev;
160 vpdev->hw_ops->ack_interrupt(vpdev, vdev->virtio_db);
161 schedule_work(&vdev->virtio_bh_work);
165 static int vop_virtio_config_change(struct vop_vdev *vdev, void *argp)
167 DECLARE_WAIT_QUEUE_HEAD_ONSTACK(wake);
168 int ret = 0, retry, i;
169 struct vop_device *vpdev = vdev->vpdev;
170 struct vop_info *vi = dev_get_drvdata(&vpdev->dev);
171 struct mic_bootparam *bootparam = vpdev->hw_ops->get_dp(vpdev);
172 s8 db = bootparam->h2c_config_db;
174 mutex_lock(&vi->vop_mutex);
175 for (i = 0; i < vdev->dd->num_vq; i++)
176 mutex_lock_nested(&vdev->vvr[i].vr_mutex, i + 1);
178 if (db == -1 || vdev->dd->type == -1) {
183 memcpy(mic_vq_configspace(vdev->dd), argp, vdev->dd->config_len);
184 vdev->dc->config_change = MIC_VIRTIO_PARAM_CONFIG_CHANGED;
185 vpdev->hw_ops->send_intr(vpdev, db);
187 for (retry = 100; retry--;) {
188 ret = wait_event_timeout(wake, vdev->dc->guest_ack,
189 msecs_to_jiffies(100));
194 dev_dbg(vop_dev(vdev),
195 "%s %d retry: %d\n", __func__, __LINE__, retry);
196 vdev->dc->config_change = 0;
197 vdev->dc->guest_ack = 0;
199 for (i = 0; i < vdev->dd->num_vq; i++)
200 mutex_unlock(&vdev->vvr[i].vr_mutex);
201 mutex_unlock(&vi->vop_mutex);
205 static int vop_copy_dp_entry(struct vop_vdev *vdev,
206 struct mic_device_desc *argp, __u8 *type,
207 struct mic_device_desc **devpage)
209 struct vop_device *vpdev = vdev->vpdev;
210 struct mic_device_desc *devp;
211 struct mic_vqconfig *vqconfig;
213 bool slot_found = false;
215 vqconfig = mic_vq_config(argp);
216 for (i = 0; i < argp->num_vq; i++) {
217 if (le16_to_cpu(vqconfig[i].num) > MIC_MAX_VRING_ENTRIES) {
219 dev_err(vop_dev(vdev), "%s %d err %d\n",
220 __func__, __LINE__, ret);
225 /* Find the first free device page entry */
226 for (i = sizeof(struct mic_bootparam);
227 i < MIC_DP_SIZE - mic_total_desc_size(argp);
228 i += mic_total_desc_size(devp)) {
229 devp = vpdev->hw_ops->get_dp(vpdev) + i;
230 if (devp->type == 0 || devp->type == -1) {
237 dev_err(vop_dev(vdev), "%s %d err %d\n",
238 __func__, __LINE__, ret);
242 * Save off the type before doing the memcpy. Type will be set in the
243 * end after completing all initialization for the new device.
247 memcpy(devp, argp, mic_desc_size(argp));
254 static void vop_init_device_ctrl(struct vop_vdev *vdev,
255 struct mic_device_desc *devpage)
257 struct mic_device_ctrl *dc;
259 dc = (void *)devpage + mic_aligned_desc_size(devpage);
261 dc->config_change = 0;
265 dc->used_address_updated = 0;
266 dc->c2h_vdev_db = -1;
267 dc->h2c_vdev_db = -1;
271 static int vop_virtio_add_device(struct vop_vdev *vdev,
272 struct mic_device_desc *argp)
274 struct vop_info *vi = vdev->vi;
275 struct vop_device *vpdev = vi->vpdev;
276 struct mic_device_desc *dd = NULL;
277 struct mic_vqconfig *vqconfig;
278 int vr_size, i, j, ret;
282 struct mic_bootparam *bootparam;
286 bootparam = vpdev->hw_ops->get_dp(vpdev);
287 init_waitqueue_head(&vdev->waitq);
288 INIT_LIST_HEAD(&vdev->list);
291 ret = vop_copy_dp_entry(vdev, argp, &type, &dd);
293 dev_err(vop_dev(vdev), "%s %d err %d\n",
294 __func__, __LINE__, ret);
298 vop_init_device_ctrl(vdev, dd);
301 vdev->virtio_id = type;
302 vqconfig = mic_vq_config(dd);
303 INIT_WORK(&vdev->virtio_bh_work, vop_bh_handler);
305 for (i = 0; i < dd->num_vq; i++) {
306 struct vop_vringh *vvr = &vdev->vvr[i];
307 struct mic_vring *vr = &vdev->vvr[i].vring;
309 num = le16_to_cpu(vqconfig[i].num);
310 mutex_init(&vvr->vr_mutex);
311 vr_size = PAGE_ALIGN(vring_size(num, MIC_VIRTIO_RING_ALIGN) +
312 sizeof(struct _mic_vring_info));
314 __get_free_pages(GFP_KERNEL | __GFP_ZERO,
318 dev_err(vop_dev(vdev), "%s %d err %d\n",
319 __func__, __LINE__, ret);
323 vr->info = vr->va + vring_size(num, MIC_VIRTIO_RING_ALIGN);
324 vr->info->magic = cpu_to_le32(MIC_MAGIC + vdev->virtio_id + i);
325 vr_addr = dma_map_single(&vpdev->dev, vr->va, vr_size,
327 if (dma_mapping_error(&vpdev->dev, vr_addr)) {
328 free_pages((unsigned long)vr->va, get_order(vr_size));
330 dev_err(vop_dev(vdev), "%s %d err %d\n",
331 __func__, __LINE__, ret);
334 vqconfig[i].address = cpu_to_le64(vr_addr);
336 vring_init(&vr->vr, num, vr->va, MIC_VIRTIO_RING_ALIGN);
337 ret = vringh_init_kern(&vvr->vrh,
338 *(u32 *)mic_vq_features(vdev->dd),
339 num, false, vr->vr.desc, vr->vr.avail,
342 dev_err(vop_dev(vdev), "%s %d err %d\n",
343 __func__, __LINE__, ret);
346 vringh_kiov_init(&vvr->riov, NULL, 0);
347 vringh_kiov_init(&vvr->wiov, NULL, 0);
348 vvr->head = USHRT_MAX;
350 vvr->vrh.notify = _vop_notify;
352 "%s %d index %d va %p info %p vr_size 0x%x\n",
353 __func__, __LINE__, i, vr->va, vr->info, vr_size);
354 vvr->buf = (void *)__get_free_pages(GFP_KERNEL,
355 get_order(VOP_INT_DMA_BUF_SIZE));
356 vvr->buf_da = dma_map_single(&vpdev->dev,
357 vvr->buf, VOP_INT_DMA_BUF_SIZE,
361 snprintf(irqname, sizeof(irqname), "vop%dvirtio%d", vpdev->index,
363 vdev->virtio_db = vpdev->hw_ops->next_db(vpdev);
364 vdev->virtio_cookie = vpdev->hw_ops->request_irq(vpdev,
365 _vop_virtio_intr_handler, irqname, vdev,
367 if (IS_ERR(vdev->virtio_cookie)) {
368 ret = PTR_ERR(vdev->virtio_cookie);
369 dev_dbg(&vpdev->dev, "request irq failed\n");
373 vdev->dc->c2h_vdev_db = vdev->virtio_db;
376 * Order the type update with previous stores. This write barrier
377 * is paired with the corresponding read barrier before the uncached
378 * system memory read of the type, on the card while scanning the
386 db = bootparam->h2c_config_db;
388 vpdev->hw_ops->send_intr(vpdev, db);
390 dev_dbg(&vpdev->dev, "Added virtio id %d db %d\n", dd->type, db);
393 vqconfig = mic_vq_config(dd);
394 for (j = 0; j < i; j++) {
395 struct vop_vringh *vvr = &vdev->vvr[j];
397 dma_unmap_single(&vpdev->dev, le64_to_cpu(vqconfig[j].address),
398 vvr->vring.len, DMA_BIDIRECTIONAL);
399 free_pages((unsigned long)vvr->vring.va,
400 get_order(vvr->vring.len));
405 static void vop_dev_remove(struct vop_info *pvi, struct mic_device_ctrl *devp,
406 struct vop_device *vpdev)
408 struct mic_bootparam *bootparam = vpdev->hw_ops->get_dp(vpdev);
411 DECLARE_WAIT_QUEUE_HEAD_ONSTACK(wake);
413 devp->config_change = MIC_VIRTIO_PARAM_DEV_REMOVE;
414 db = bootparam->h2c_config_db;
416 vpdev->hw_ops->send_intr(vpdev, db);
419 for (retry = 15; retry--;) {
420 ret = wait_event_timeout(wake, devp->guest_ack,
421 msecs_to_jiffies(1000));
426 devp->config_change = 0;
430 static void vop_virtio_del_device(struct vop_vdev *vdev)
432 struct vop_info *vi = vdev->vi;
433 struct vop_device *vpdev = vdev->vpdev;
435 struct mic_vqconfig *vqconfig;
436 struct mic_bootparam *bootparam = vpdev->hw_ops->get_dp(vpdev);
439 goto skip_hot_remove;
440 vop_dev_remove(vi, vdev->dc, vpdev);
442 vpdev->hw_ops->free_irq(vpdev, vdev->virtio_cookie, vdev);
443 flush_work(&vdev->virtio_bh_work);
444 vqconfig = mic_vq_config(vdev->dd);
445 for (i = 0; i < vdev->dd->num_vq; i++) {
446 struct vop_vringh *vvr = &vdev->vvr[i];
448 dma_unmap_single(&vpdev->dev,
449 vvr->buf_da, VOP_INT_DMA_BUF_SIZE,
451 free_pages((unsigned long)vvr->buf,
452 get_order(VOP_INT_DMA_BUF_SIZE));
453 vringh_kiov_cleanup(&vvr->riov);
454 vringh_kiov_cleanup(&vvr->wiov);
455 dma_unmap_single(&vpdev->dev, le64_to_cpu(vqconfig[i].address),
456 vvr->vring.len, DMA_BIDIRECTIONAL);
457 free_pages((unsigned long)vvr->vring.va,
458 get_order(vvr->vring.len));
461 * Order the type update with previous stores. This write barrier
462 * is paired with the corresponding read barrier before the uncached
463 * system memory read of the type, on the card while scanning the
471 * vop_sync_dma - Wrapper for synchronous DMAs.
473 * @dev - The address of the pointer to the device instance used
474 * for DMA registration.
475 * @dst - destination DMA address.
476 * @src - source DMA address.
477 * @len - size of the transfer.
479 * Return DMA_SUCCESS on success
481 static int vop_sync_dma(struct vop_vdev *vdev, dma_addr_t dst, dma_addr_t src,
485 struct dma_device *ddev;
486 struct dma_async_tx_descriptor *tx;
487 struct vop_info *vi = dev_get_drvdata(&vdev->vpdev->dev);
488 struct dma_chan *vop_ch = vi->dma_ch;
494 ddev = vop_ch->device;
495 tx = ddev->device_prep_dma_memcpy(vop_ch, dst, src, len,
503 cookie = tx->tx_submit(tx);
504 if (dma_submit_error(cookie)) {
508 dma_async_issue_pending(vop_ch);
509 err = dma_sync_wait(vop_ch, cookie);
513 dev_err(&vi->vpdev->dev, "%s %d err %d\n",
514 __func__, __LINE__, err);
518 #define VOP_USE_DMA true
521 * Initiates the copies across the PCIe bus from card memory to a user
522 * space buffer. When transfers are done using DMA, source/destination
523 * addresses and transfer length must follow the alignment requirements of
524 * the MIC DMA engine.
526 static int vop_virtio_copy_to_user(struct vop_vdev *vdev, void __user *ubuf,
527 size_t len, u64 daddr, size_t dlen,
530 struct vop_device *vpdev = vdev->vpdev;
531 void __iomem *dbuf = vpdev->hw_ops->ioremap(vpdev, daddr, len);
532 struct vop_vringh *vvr = &vdev->vvr[vr_idx];
533 struct vop_info *vi = dev_get_drvdata(&vpdev->dev);
534 size_t dma_alignment = 1 << vi->dma_ch->device->copy_align;
535 bool x200 = is_dma_copy_aligned(vi->dma_ch->device, 1, 1, 1);
536 size_t dma_offset, partlen;
540 if (copy_to_user(ubuf, (void __force *)dbuf, len)) {
542 dev_err(vop_dev(vdev), "%s %d err %d\n",
543 __func__, __LINE__, err);
546 vdev->in_bytes += len;
551 dma_offset = daddr - round_down(daddr, dma_alignment);
555 * X100 uses DMA addresses as seen by the card so adding
556 * the aperture base is not required for DMA. However x200
557 * requires DMA addresses to be an offset into the bar so
558 * add the aperture base for x200.
561 daddr += vpdev->aper->pa;
563 partlen = min_t(size_t, len, VOP_INT_DMA_BUF_SIZE);
564 err = vop_sync_dma(vdev, vvr->buf_da, daddr,
565 ALIGN(partlen, dma_alignment));
567 dev_err(vop_dev(vdev), "%s %d err %d\n",
568 __func__, __LINE__, err);
571 if (copy_to_user(ubuf, vvr->buf + dma_offset,
572 partlen - dma_offset)) {
574 dev_err(vop_dev(vdev), "%s %d err %d\n",
575 __func__, __LINE__, err);
581 vdev->in_bytes_dma += partlen;
582 vdev->in_bytes += partlen;
588 vpdev->hw_ops->iounmap(vpdev, dbuf);
589 dev_dbg(vop_dev(vdev),
590 "%s: ubuf %p dbuf %p len 0x%lx vr_idx 0x%x\n",
591 __func__, ubuf, dbuf, len, vr_idx);
596 * Initiates copies across the PCIe bus from a user space buffer to card
597 * memory. When transfers are done using DMA, source/destination addresses
598 * and transfer length must follow the alignment requirements of the MIC
601 static int vop_virtio_copy_from_user(struct vop_vdev *vdev, void __user *ubuf,
602 size_t len, u64 daddr, size_t dlen,
605 struct vop_device *vpdev = vdev->vpdev;
606 void __iomem *dbuf = vpdev->hw_ops->ioremap(vpdev, daddr, len);
607 struct vop_vringh *vvr = &vdev->vvr[vr_idx];
608 struct vop_info *vi = dev_get_drvdata(&vdev->vpdev->dev);
609 size_t dma_alignment = 1 << vi->dma_ch->device->copy_align;
610 bool x200 = is_dma_copy_aligned(vi->dma_ch->device, 1, 1, 1);
612 bool dma = VOP_USE_DMA;
615 if (daddr & (dma_alignment - 1)) {
616 vdev->tx_dst_unaligned += len;
618 } else if (ALIGN(len, dma_alignment) > dlen) {
619 vdev->tx_len_unaligned += len;
627 * X100 uses DMA addresses as seen by the card so adding
628 * the aperture base is not required for DMA. However x200
629 * requires DMA addresses to be an offset into the bar so
630 * add the aperture base for x200.
633 daddr += vpdev->aper->pa;
635 partlen = min_t(size_t, len, VOP_INT_DMA_BUF_SIZE);
637 if (copy_from_user(vvr->buf, ubuf, partlen)) {
639 dev_err(vop_dev(vdev), "%s %d err %d\n",
640 __func__, __LINE__, err);
643 err = vop_sync_dma(vdev, daddr, vvr->buf_da,
644 ALIGN(partlen, dma_alignment));
646 dev_err(vop_dev(vdev), "%s %d err %d\n",
647 __func__, __LINE__, err);
653 vdev->out_bytes_dma += partlen;
654 vdev->out_bytes += partlen;
659 * We are copying to IO below and should ideally use something
660 * like copy_from_user_toio(..) if it existed.
662 if (copy_from_user((void __force *)dbuf, ubuf, len)) {
664 dev_err(vop_dev(vdev), "%s %d err %d\n",
665 __func__, __LINE__, err);
668 vdev->out_bytes += len;
671 vpdev->hw_ops->iounmap(vpdev, dbuf);
672 dev_dbg(vop_dev(vdev),
673 "%s: ubuf %p dbuf %p len 0x%lx vr_idx 0x%x\n",
674 __func__, ubuf, dbuf, len, vr_idx);
678 #define MIC_VRINGH_READ true
680 /* Determine the total number of bytes consumed in a VRINGH KIOV */
681 static inline u32 vop_vringh_iov_consumed(struct vringh_kiov *iov)
684 u32 total = iov->consumed;
686 for (i = 0; i < iov->i; i++)
687 total += iov->iov[i].iov_len;
692 * Traverse the VRINGH KIOV and issue the APIs to trigger the copies.
693 * This API is heavily based on the vringh_iov_xfer(..) implementation
694 * in vringh.c. The reason we cannot reuse vringh_iov_pull_kern(..)
695 * and vringh_iov_push_kern(..) directly is because there is no
696 * way to override the VRINGH xfer(..) routines as of v3.10.
698 static int vop_vringh_copy(struct vop_vdev *vdev, struct vringh_kiov *iov,
699 void __user *ubuf, size_t len, bool read, int vr_idx,
703 size_t partlen, tot_len = 0;
705 while (len && iov->i < iov->used) {
706 struct kvec *kiov = &iov->iov[iov->i];
708 partlen = min(kiov->iov_len, len);
710 ret = vop_virtio_copy_to_user(vdev, ubuf, partlen,
715 ret = vop_virtio_copy_from_user(vdev, ubuf, partlen,
720 dev_err(vop_dev(vdev), "%s %d err %d\n",
721 __func__, __LINE__, ret);
727 iov->consumed += partlen;
728 kiov->iov_len -= partlen;
729 kiov->iov_base += partlen;
730 if (!kiov->iov_len) {
731 /* Fix up old iov element then increment. */
732 kiov->iov_len = iov->consumed;
733 kiov->iov_base -= iov->consumed;
744 * Use the standard VRINGH infrastructure in the kernel to fetch new
745 * descriptors, initiate the copies and update the used ring.
747 static int _vop_virtio_copy(struct vop_vdev *vdev, struct mic_copy_desc *copy)
750 u32 iovcnt = copy->iovcnt;
752 struct iovec __user *u_iov = copy->iov;
753 void __user *ubuf = NULL;
754 struct vop_vringh *vvr = &vdev->vvr[copy->vr_idx];
755 struct vringh_kiov *riov = &vvr->riov;
756 struct vringh_kiov *wiov = &vvr->wiov;
757 struct vringh *vrh = &vvr->vrh;
758 u16 *head = &vvr->head;
759 struct mic_vring *vr = &vvr->vring;
760 size_t len = 0, out_len;
763 /* Fetch a new IOVEC if all previous elements have been processed */
764 if (riov->i == riov->used && wiov->i == wiov->used) {
765 ret = vringh_getdesc_kern(vrh, riov, wiov,
767 /* Check if there are available descriptors */
773 /* Copy over a new iovec from user space. */
774 ret = copy_from_user(&iov, u_iov, sizeof(*u_iov));
777 dev_err(vop_dev(vdev), "%s %d err %d\n",
778 __func__, __LINE__, ret);
784 /* Issue all the read descriptors first */
785 ret = vop_vringh_copy(vdev, riov, ubuf, len,
786 MIC_VRINGH_READ, copy->vr_idx, &out_len);
788 dev_err(vop_dev(vdev), "%s %d err %d\n",
789 __func__, __LINE__, ret);
794 copy->out_len += out_len;
795 /* Issue the write descriptors next */
796 ret = vop_vringh_copy(vdev, wiov, ubuf, len,
797 !MIC_VRINGH_READ, copy->vr_idx, &out_len);
799 dev_err(vop_dev(vdev), "%s %d err %d\n",
800 __func__, __LINE__, ret);
805 copy->out_len += out_len;
807 /* One user space iovec is now completed */
811 /* Exit loop if all elements in KIOVs have been processed. */
812 if (riov->i == riov->used && wiov->i == wiov->used)
816 * Update the used ring if a descriptor was available and some data was
817 * copied in/out and the user asked for a used ring update.
819 if (*head != USHRT_MAX && copy->out_len && copy->update_used) {
822 /* Determine the total data consumed */
823 total += vop_vringh_iov_consumed(riov);
824 total += vop_vringh_iov_consumed(wiov);
825 vringh_complete_kern(vrh, *head, total);
827 if (vringh_need_notify_kern(vrh) > 0)
829 vringh_kiov_cleanup(riov);
830 vringh_kiov_cleanup(wiov);
831 /* Update avail idx for user space */
832 vr->info->avail_idx = vrh->last_avail_idx;
837 static inline int vop_verify_copy_args(struct vop_vdev *vdev,
838 struct mic_copy_desc *copy)
840 if (!vdev || copy->vr_idx >= vdev->dd->num_vq)
845 /* Copy a specified number of virtio descriptors in a chain */
846 static int vop_virtio_copy_desc(struct vop_vdev *vdev,
847 struct mic_copy_desc *copy)
850 struct vop_vringh *vvr;
852 err = vop_verify_copy_args(vdev, copy);
856 vvr = &vdev->vvr[copy->vr_idx];
857 mutex_lock(&vvr->vr_mutex);
858 if (!vop_vdevup(vdev)) {
860 dev_err(vop_dev(vdev), "%s %d err %d\n",
861 __func__, __LINE__, err);
864 err = _vop_virtio_copy(vdev, copy);
866 dev_err(vop_dev(vdev), "%s %d err %d\n",
867 __func__, __LINE__, err);
870 mutex_unlock(&vvr->vr_mutex);
874 static int vop_open(struct inode *inode, struct file *f)
876 struct vop_vdev *vdev;
877 struct vop_info *vi = container_of(f->private_data,
878 struct vop_info, miscdev);
880 vdev = kzalloc(sizeof(*vdev), GFP_KERNEL);
884 mutex_init(&vdev->vdev_mutex);
885 f->private_data = vdev;
886 init_completion(&vdev->destroy);
887 complete(&vdev->destroy);
891 static int vop_release(struct inode *inode, struct file *f)
893 struct vop_vdev *vdev = f->private_data, *vdev_tmp;
894 struct vop_info *vi = vdev->vi;
895 struct list_head *pos, *tmp;
898 mutex_lock(&vdev->vdev_mutex);
901 mutex_lock(&vi->vop_mutex);
902 list_for_each_safe(pos, tmp, &vi->vdev_list) {
903 vdev_tmp = list_entry(pos, struct vop_vdev, list);
904 if (vdev == vdev_tmp) {
905 vop_virtio_del_device(vdev);
911 mutex_unlock(&vi->vop_mutex);
913 mutex_unlock(&vdev->vdev_mutex);
915 wait_for_completion(&vdev->destroy);
916 f->private_data = NULL;
921 static long vop_ioctl(struct file *f, unsigned int cmd, unsigned long arg)
923 struct vop_vdev *vdev = f->private_data;
924 struct vop_info *vi = vdev->vi;
925 void __user *argp = (void __user *)arg;
929 case MIC_VIRTIO_ADD_DEVICE:
931 struct mic_device_desc dd, *dd_config;
933 if (copy_from_user(&dd, argp, sizeof(dd)))
936 if (mic_aligned_desc_size(&dd) > MIC_MAX_DESC_BLK_SIZE ||
937 dd.num_vq > MIC_MAX_VRINGS)
940 dd_config = kzalloc(mic_desc_size(&dd), GFP_KERNEL);
943 if (copy_from_user(dd_config, argp, mic_desc_size(&dd))) {
947 /* Ensure desc has not changed between the two reads */
948 if (memcmp(&dd, dd_config, sizeof(dd))) {
952 mutex_lock(&vdev->vdev_mutex);
953 mutex_lock(&vi->vop_mutex);
954 ret = vop_virtio_add_device(vdev, dd_config);
957 list_add_tail(&vdev->list, &vi->vdev_list);
959 mutex_unlock(&vi->vop_mutex);
960 mutex_unlock(&vdev->vdev_mutex);
965 case MIC_VIRTIO_COPY_DESC:
967 struct mic_copy_desc copy;
969 mutex_lock(&vdev->vdev_mutex);
970 ret = vop_vdev_inited(vdev);
974 if (copy_from_user(©, argp, sizeof(copy))) {
979 ret = vop_virtio_copy_desc(vdev, ©);
983 &((struct mic_copy_desc __user *)argp)->out_len,
984 ©.out_len, sizeof(copy.out_len)))
987 mutex_unlock(&vdev->vdev_mutex);
990 case MIC_VIRTIO_CONFIG_CHANGE:
994 mutex_lock(&vdev->vdev_mutex);
995 ret = vop_vdev_inited(vdev);
998 buf = kzalloc(vdev->dd->config_len, GFP_KERNEL);
1003 if (copy_from_user(buf, argp, vdev->dd->config_len)) {
1007 ret = vop_virtio_config_change(vdev, buf);
1011 mutex_unlock(&vdev->vdev_mutex);
1015 return -ENOIOCTLCMD;
1021 * We return POLLIN | POLLOUT from poll when new buffers are enqueued, and
1022 * not when previously enqueued buffers may be available. This means that
1023 * in the card->host (TX) path, when userspace is unblocked by poll it
1024 * must drain all available descriptors or it can stall.
1026 static __poll_t vop_poll(struct file *f, poll_table *wait)
1028 struct vop_vdev *vdev = f->private_data;
1031 mutex_lock(&vdev->vdev_mutex);
1032 if (vop_vdev_inited(vdev)) {
1036 poll_wait(f, &vdev->waitq, wait);
1037 if (vop_vdev_inited(vdev)) {
1039 } else if (vdev->poll_wake) {
1040 vdev->poll_wake = 0;
1041 mask = POLLIN | POLLOUT;
1044 mutex_unlock(&vdev->vdev_mutex);
1049 vop_query_offset(struct vop_vdev *vdev, unsigned long offset,
1050 unsigned long *size, unsigned long *pa)
1052 struct vop_device *vpdev = vdev->vpdev;
1053 unsigned long start = MIC_DP_SIZE;
1057 * MMAP interface is as follows:
1059 * 0x0 virtio device_page
1060 * 0x1000 first vring
1061 * 0x1000 + size of 1st vring second vring
1065 *pa = virt_to_phys(vpdev->hw_ops->get_dp(vpdev));
1066 *size = MIC_DP_SIZE;
1070 for (i = 0; i < vdev->dd->num_vq; i++) {
1071 struct vop_vringh *vvr = &vdev->vvr[i];
1073 if (offset == start) {
1074 *pa = virt_to_phys(vvr->vring.va);
1075 *size = vvr->vring.len;
1078 start += vvr->vring.len;
1084 * Maps the device page and virtio rings to user space for readonly access.
1086 static int vop_mmap(struct file *f, struct vm_area_struct *vma)
1088 struct vop_vdev *vdev = f->private_data;
1089 unsigned long offset = vma->vm_pgoff << PAGE_SHIFT;
1090 unsigned long pa, size = vma->vm_end - vma->vm_start, size_rem = size;
1093 err = vop_vdev_inited(vdev);
1096 if (vma->vm_flags & VM_WRITE) {
1101 i = vop_query_offset(vdev, offset, &size, &pa);
1106 err = remap_pfn_range(vma, vma->vm_start + offset,
1107 pa >> PAGE_SHIFT, size,
1118 static const struct file_operations vop_fops = {
1120 .release = vop_release,
1121 .unlocked_ioctl = vop_ioctl,
1124 .owner = THIS_MODULE,
1127 int vop_host_init(struct vop_info *vi)
1130 struct miscdevice *mdev;
1131 struct vop_device *vpdev = vi->vpdev;
1133 INIT_LIST_HEAD(&vi->vdev_list);
1134 vi->dma_ch = vpdev->dma_ch;
1135 mdev = &vi->miscdev;
1136 mdev->minor = MISC_DYNAMIC_MINOR;
1137 snprintf(vi->name, sizeof(vi->name), "vop_virtio%d", vpdev->index);
1138 mdev->name = vi->name;
1139 mdev->fops = &vop_fops;
1140 mdev->parent = &vpdev->dev;
1142 rc = misc_register(mdev);
1144 dev_err(&vpdev->dev, "%s failed rc %d\n", __func__, rc);
1148 void vop_host_uninit(struct vop_info *vi)
1150 struct list_head *pos, *tmp;
1151 struct vop_vdev *vdev;
1153 mutex_lock(&vi->vop_mutex);
1154 vop_virtio_reset_devices(vi);
1155 list_for_each_safe(pos, tmp, &vi->vdev_list) {
1156 vdev = list_entry(pos, struct vop_vdev, list);
1158 reinit_completion(&vdev->destroy);
1159 mutex_unlock(&vi->vop_mutex);
1160 mutex_lock(&vdev->vdev_mutex);
1161 vop_virtio_del_device(vdev);
1162 vdev->deleted = true;
1163 mutex_unlock(&vdev->vdev_mutex);
1164 complete(&vdev->destroy);
1165 mutex_lock(&vi->vop_mutex);
1167 mutex_unlock(&vi->vop_mutex);
1168 misc_deregister(&vi->miscdev);