android_kernel_xiaomi_sm8350/drivers/infiniband/hw/ipath/ipath_cq.c
Steve Wise 00f7ec36c9 RDMA/core: Add memory management extensions support
This patch adds support for the IB "base memory management extension"
(BMME) and the equivalent iWARP operations (which the iWARP verbs
mandates all devices must implement).  The new operations are:

 - Allocate an ib_mr for use in fast register work requests.

 - Allocate/free a physical buffer lists for use in fast register work
   requests.  This allows device drivers to allocate this memory as
   needed for use in posting send requests (eg via dma_alloc_coherent).

 - New send queue work requests:
   * send with remote invalidate
   * fast register memory region
   * local invalidate memory region
   * RDMA read with invalidate local memory region (iWARP only)

Consumer interface details:

 - A new device capability flag IB_DEVICE_MEM_MGT_EXTENSIONS is added
   to indicate device support for these features.

 - New send work request opcodes IB_WR_FAST_REG_MR, IB_WR_LOCAL_INV,
   IB_WR_RDMA_READ_WITH_INV are added.

 - A new consumer API function, ib_alloc_mr() is added to allocate
   fast register memory regions.

 - New consumer API functions, ib_alloc_fast_reg_page_list() and
   ib_free_fast_reg_page_list() are added to allocate and free
   device-specific memory for fast registration page lists.

 - A new consumer API function, ib_update_fast_reg_key(), is added to
   allow the key portion of the R_Key and L_Key of a fast registration
   MR to be updated.  Consumers call this if desired before posting
   a IB_WR_FAST_REG_MR work request.

Consumers can use this as follows:

 - MR is allocated with ib_alloc_mr().

 - Page list memory is allocated with ib_alloc_fast_reg_page_list().

 - MR R_Key/L_Key "key" field is updated with ib_update_fast_reg_key().

 - MR made VALID and bound to a specific page list via
   ib_post_send(IB_WR_FAST_REG_MR)

 - MR made INVALID via ib_post_send(IB_WR_LOCAL_INV),
   ib_post_send(IB_WR_RDMA_READ_WITH_INV) or an incoming send with
   invalidate operation.

 - MR is deallocated with ib_dereg_mr()

 - page lists dealloced via ib_free_fast_reg_page_list().

Applications can allocate a fast register MR once, and then can
repeatedly bind the MR to different physical block lists (PBLs) via
posting work requests to a send queue (SQ).  For each outstanding
MR-to-PBL binding in the SQ pipe, a fast_reg_page_list needs to be
allocated (the fast_reg_page_list is owned by the low-level driver
from the consumer posting a work request until the request completes).
Thus pipelining can be achieved while still allowing device-specific
page_list processing.

The 32-bit fast register memory key/STag is composed of a 24-bit index
and an 8-bit key.  The application can change the key each time it
fast registers thus allowing more control over the peer's use of the
key/STag (ie it can effectively be changed each time the rkey is
rebound to a page list).

Signed-off-by: Steve Wise <swise@opengridcomputing.com>
Signed-off-by: Roland Dreier <rolandd@cisco.com>
2008-07-14 23:48:45 -07:00

478 lines
12 KiB
C

/*
* Copyright (c) 2006, 2007 QLogic Corporation. All rights reserved.
* Copyright (c) 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/err.h>
#include <linux/vmalloc.h>
#include "ipath_verbs.h"
/**
* ipath_cq_enter - add a new entry to the completion queue
* @cq: completion queue
* @entry: work completion entry to add
* @sig: true if @entry is a solicitated entry
*
* This may be called with qp->s_lock held.
*/
void ipath_cq_enter(struct ipath_cq *cq, struct ib_wc *entry, int solicited)
{
struct ipath_cq_wc *wc;
unsigned long flags;
u32 head;
u32 next;
spin_lock_irqsave(&cq->lock, flags);
/*
* Note that the head pointer might be writable by user processes.
* Take care to verify it is a sane value.
*/
wc = cq->queue;
head = wc->head;
if (head >= (unsigned) cq->ibcq.cqe) {
head = cq->ibcq.cqe;
next = 0;
} else
next = head + 1;
if (unlikely(next == wc->tail)) {
spin_unlock_irqrestore(&cq->lock, flags);
if (cq->ibcq.event_handler) {
struct ib_event ev;
ev.device = cq->ibcq.device;
ev.element.cq = &cq->ibcq;
ev.event = IB_EVENT_CQ_ERR;
cq->ibcq.event_handler(&ev, cq->ibcq.cq_context);
}
return;
}
if (cq->ip) {
wc->uqueue[head].wr_id = entry->wr_id;
wc->uqueue[head].status = entry->status;
wc->uqueue[head].opcode = entry->opcode;
wc->uqueue[head].vendor_err = entry->vendor_err;
wc->uqueue[head].byte_len = entry->byte_len;
wc->uqueue[head].ex.imm_data = (__u32 __force) entry->ex.imm_data;
wc->uqueue[head].qp_num = entry->qp->qp_num;
wc->uqueue[head].src_qp = entry->src_qp;
wc->uqueue[head].wc_flags = entry->wc_flags;
wc->uqueue[head].pkey_index = entry->pkey_index;
wc->uqueue[head].slid = entry->slid;
wc->uqueue[head].sl = entry->sl;
wc->uqueue[head].dlid_path_bits = entry->dlid_path_bits;
wc->uqueue[head].port_num = entry->port_num;
/* Make sure entry is written before the head index. */
smp_wmb();
} else
wc->kqueue[head] = *entry;
wc->head = next;
if (cq->notify == IB_CQ_NEXT_COMP ||
(cq->notify == IB_CQ_SOLICITED && solicited)) {
cq->notify = IB_CQ_NONE;
cq->triggered++;
/*
* This will cause send_complete() to be called in
* another thread.
*/
tasklet_hi_schedule(&cq->comptask);
}
spin_unlock_irqrestore(&cq->lock, flags);
if (entry->status != IB_WC_SUCCESS)
to_idev(cq->ibcq.device)->n_wqe_errs++;
}
/**
* ipath_poll_cq - poll for work completion entries
* @ibcq: the completion queue to poll
* @num_entries: the maximum number of entries to return
* @entry: pointer to array where work completions are placed
*
* Returns the number of completion entries polled.
*
* This may be called from interrupt context. Also called by ib_poll_cq()
* in the generic verbs code.
*/
int ipath_poll_cq(struct ib_cq *ibcq, int num_entries, struct ib_wc *entry)
{
struct ipath_cq *cq = to_icq(ibcq);
struct ipath_cq_wc *wc;
unsigned long flags;
int npolled;
u32 tail;
/* The kernel can only poll a kernel completion queue */
if (cq->ip) {
npolled = -EINVAL;
goto bail;
}
spin_lock_irqsave(&cq->lock, flags);
wc = cq->queue;
tail = wc->tail;
if (tail > (u32) cq->ibcq.cqe)
tail = (u32) cq->ibcq.cqe;
for (npolled = 0; npolled < num_entries; ++npolled, ++entry) {
if (tail == wc->head)
break;
/* The kernel doesn't need a RMB since it has the lock. */
*entry = wc->kqueue[tail];
if (tail >= cq->ibcq.cqe)
tail = 0;
else
tail++;
}
wc->tail = tail;
spin_unlock_irqrestore(&cq->lock, flags);
bail:
return npolled;
}
static void send_complete(unsigned long data)
{
struct ipath_cq *cq = (struct ipath_cq *)data;
/*
* The completion handler will most likely rearm the notification
* and poll for all pending entries. If a new completion entry
* is added while we are in this routine, tasklet_hi_schedule()
* won't call us again until we return so we check triggered to
* see if we need to call the handler again.
*/
for (;;) {
u8 triggered = cq->triggered;
cq->ibcq.comp_handler(&cq->ibcq, cq->ibcq.cq_context);
if (cq->triggered == triggered)
return;
}
}
/**
* ipath_create_cq - create a completion queue
* @ibdev: the device this completion queue is attached to
* @entries: the minimum size of the completion queue
* @context: unused by the InfiniPath driver
* @udata: unused by the InfiniPath driver
*
* Returns a pointer to the completion queue or negative errno values
* for failure.
*
* Called by ib_create_cq() in the generic verbs code.
*/
struct ib_cq *ipath_create_cq(struct ib_device *ibdev, int entries, int comp_vector,
struct ib_ucontext *context,
struct ib_udata *udata)
{
struct ipath_ibdev *dev = to_idev(ibdev);
struct ipath_cq *cq;
struct ipath_cq_wc *wc;
struct ib_cq *ret;
u32 sz;
if (entries < 1 || entries > ib_ipath_max_cqes) {
ret = ERR_PTR(-EINVAL);
goto done;
}
/* Allocate the completion queue structure. */
cq = kmalloc(sizeof(*cq), GFP_KERNEL);
if (!cq) {
ret = ERR_PTR(-ENOMEM);
goto done;
}
/*
* Allocate the completion queue entries and head/tail pointers.
* This is allocated separately so that it can be resized and
* also mapped into user space.
* We need to use vmalloc() in order to support mmap and large
* numbers of entries.
*/
sz = sizeof(*wc);
if (udata && udata->outlen >= sizeof(__u64))
sz += sizeof(struct ib_uverbs_wc) * (entries + 1);
else
sz += sizeof(struct ib_wc) * (entries + 1);
wc = vmalloc_user(sz);
if (!wc) {
ret = ERR_PTR(-ENOMEM);
goto bail_cq;
}
/*
* Return the address of the WC as the offset to mmap.
* See ipath_mmap() for details.
*/
if (udata && udata->outlen >= sizeof(__u64)) {
int err;
cq->ip = ipath_create_mmap_info(dev, sz, context, wc);
if (!cq->ip) {
ret = ERR_PTR(-ENOMEM);
goto bail_wc;
}
err = ib_copy_to_udata(udata, &cq->ip->offset,
sizeof(cq->ip->offset));
if (err) {
ret = ERR_PTR(err);
goto bail_ip;
}
} else
cq->ip = NULL;
spin_lock(&dev->n_cqs_lock);
if (dev->n_cqs_allocated == ib_ipath_max_cqs) {
spin_unlock(&dev->n_cqs_lock);
ret = ERR_PTR(-ENOMEM);
goto bail_ip;
}
dev->n_cqs_allocated++;
spin_unlock(&dev->n_cqs_lock);
if (cq->ip) {
spin_lock_irq(&dev->pending_lock);
list_add(&cq->ip->pending_mmaps, &dev->pending_mmaps);
spin_unlock_irq(&dev->pending_lock);
}
/*
* ib_create_cq() will initialize cq->ibcq except for cq->ibcq.cqe.
* The number of entries should be >= the number requested or return
* an error.
*/
cq->ibcq.cqe = entries;
cq->notify = IB_CQ_NONE;
cq->triggered = 0;
spin_lock_init(&cq->lock);
tasklet_init(&cq->comptask, send_complete, (unsigned long)cq);
wc->head = 0;
wc->tail = 0;
cq->queue = wc;
ret = &cq->ibcq;
goto done;
bail_ip:
kfree(cq->ip);
bail_wc:
vfree(wc);
bail_cq:
kfree(cq);
done:
return ret;
}
/**
* ipath_destroy_cq - destroy a completion queue
* @ibcq: the completion queue to destroy.
*
* Returns 0 for success.
*
* Called by ib_destroy_cq() in the generic verbs code.
*/
int ipath_destroy_cq(struct ib_cq *ibcq)
{
struct ipath_ibdev *dev = to_idev(ibcq->device);
struct ipath_cq *cq = to_icq(ibcq);
tasklet_kill(&cq->comptask);
spin_lock(&dev->n_cqs_lock);
dev->n_cqs_allocated--;
spin_unlock(&dev->n_cqs_lock);
if (cq->ip)
kref_put(&cq->ip->ref, ipath_release_mmap_info);
else
vfree(cq->queue);
kfree(cq);
return 0;
}
/**
* ipath_req_notify_cq - change the notification type for a completion queue
* @ibcq: the completion queue
* @notify_flags: the type of notification to request
*
* Returns 0 for success.
*
* This may be called from interrupt context. Also called by
* ib_req_notify_cq() in the generic verbs code.
*/
int ipath_req_notify_cq(struct ib_cq *ibcq, enum ib_cq_notify_flags notify_flags)
{
struct ipath_cq *cq = to_icq(ibcq);
unsigned long flags;
int ret = 0;
spin_lock_irqsave(&cq->lock, flags);
/*
* Don't change IB_CQ_NEXT_COMP to IB_CQ_SOLICITED but allow
* any other transitions (see C11-31 and C11-32 in ch. 11.4.2.2).
*/
if (cq->notify != IB_CQ_NEXT_COMP)
cq->notify = notify_flags & IB_CQ_SOLICITED_MASK;
if ((notify_flags & IB_CQ_REPORT_MISSED_EVENTS) &&
cq->queue->head != cq->queue->tail)
ret = 1;
spin_unlock_irqrestore(&cq->lock, flags);
return ret;
}
/**
* ipath_resize_cq - change the size of the CQ
* @ibcq: the completion queue
*
* Returns 0 for success.
*/
int ipath_resize_cq(struct ib_cq *ibcq, int cqe, struct ib_udata *udata)
{
struct ipath_cq *cq = to_icq(ibcq);
struct ipath_cq_wc *old_wc;
struct ipath_cq_wc *wc;
u32 head, tail, n;
int ret;
u32 sz;
if (cqe < 1 || cqe > ib_ipath_max_cqes) {
ret = -EINVAL;
goto bail;
}
/*
* Need to use vmalloc() if we want to support large #s of entries.
*/
sz = sizeof(*wc);
if (udata && udata->outlen >= sizeof(__u64))
sz += sizeof(struct ib_uverbs_wc) * (cqe + 1);
else
sz += sizeof(struct ib_wc) * (cqe + 1);
wc = vmalloc_user(sz);
if (!wc) {
ret = -ENOMEM;
goto bail;
}
/* Check that we can write the offset to mmap. */
if (udata && udata->outlen >= sizeof(__u64)) {
__u64 offset = 0;
ret = ib_copy_to_udata(udata, &offset, sizeof(offset));
if (ret)
goto bail_free;
}
spin_lock_irq(&cq->lock);
/*
* Make sure head and tail are sane since they
* might be user writable.
*/
old_wc = cq->queue;
head = old_wc->head;
if (head > (u32) cq->ibcq.cqe)
head = (u32) cq->ibcq.cqe;
tail = old_wc->tail;
if (tail > (u32) cq->ibcq.cqe)
tail = (u32) cq->ibcq.cqe;
if (head < tail)
n = cq->ibcq.cqe + 1 + head - tail;
else
n = head - tail;
if (unlikely((u32)cqe < n)) {
ret = -EINVAL;
goto bail_unlock;
}
for (n = 0; tail != head; n++) {
if (cq->ip)
wc->uqueue[n] = old_wc->uqueue[tail];
else
wc->kqueue[n] = old_wc->kqueue[tail];
if (tail == (u32) cq->ibcq.cqe)
tail = 0;
else
tail++;
}
cq->ibcq.cqe = cqe;
wc->head = n;
wc->tail = 0;
cq->queue = wc;
spin_unlock_irq(&cq->lock);
vfree(old_wc);
if (cq->ip) {
struct ipath_ibdev *dev = to_idev(ibcq->device);
struct ipath_mmap_info *ip = cq->ip;
ipath_update_mmap_info(dev, ip, sz, wc);
/*
* Return the offset to mmap.
* See ipath_mmap() for details.
*/
if (udata && udata->outlen >= sizeof(__u64)) {
ret = ib_copy_to_udata(udata, &ip->offset,
sizeof(ip->offset));
if (ret)
goto bail;
}
spin_lock_irq(&dev->pending_lock);
if (list_empty(&ip->pending_mmaps))
list_add(&ip->pending_mmaps, &dev->pending_mmaps);
spin_unlock_irq(&dev->pending_lock);
}
ret = 0;
goto bail;
bail_unlock:
spin_unlock_irq(&cq->lock);
bail_free:
vfree(wc);
bail:
return ret;
}