freebsd-nq/sys/dev/xdma/xdma_sg.c
Ruslan Bukin 101869a8f0 o Implement a bounce buffer based on device reserved memory.
Grab device reserved physical memory regions from FDT using standard
  "memory-region" property and use vmem(9) to allocate buffers from it.
  The same vmem could be used by DMA engine drivers to allocate memory for
  DMA descriptors.
  This is required for platforms that provide uncached memory region
  reserved exclusively for DMA operations.
o Change sleepable sx(9) lock type to non-sleepable mutex(9) since
  network drivers usually hold mutex during DMA operations. So we don't
  take sleepable lock after non-sleepable.

Tested on U.S. Government Furnished Equipment (GFE) 64-bit RISC-V cores.

Sponsored by:	DARPA, AFRL
2019-05-08 15:22:27 +00:00

647 lines
14 KiB
C

/*-
* SPDX-License-Identifier: BSD-2-Clause
*
* Copyright (c) 2018-2019 Ruslan Bukin <br@bsdpad.com>
*
* This software was developed by SRI International and the University of
* Cambridge Computer Laboratory under DARPA/AFRL contract FA8750-10-C-0237
* ("CTSRD"), as part of the DARPA CRASH research programme.
*
* Redistribution and use in source and binary forms, with or without
* modification, are permitted provided that the following conditions
* are met:
* 1. Redistributions of source code must retain the above copyright
* notice, this list of conditions and the following disclaimer.
* 2. 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.
*
* THIS SOFTWARE IS PROVIDED BY THE AUTHOR AND CONTRIBUTORS ``AS IS'' AND
* ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
* IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
* ARE DISCLAIMED. IN NO EVENT SHALL THE AUTHOR OR CONTRIBUTORS BE LIABLE
* FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
* DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
* OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
* HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
* LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
* OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
* SUCH DAMAGE.
*/
#include <sys/cdefs.h>
__FBSDID("$FreeBSD$");
#include "opt_platform.h"
#include <sys/param.h>
#include <sys/conf.h>
#include <sys/bus.h>
#include <sys/kernel.h>
#include <sys/malloc.h>
#include <sys/mbuf.h>
#include <sys/rwlock.h>
#include <machine/bus.h>
#include <vm/vm.h>
#include <vm/vm_extern.h>
#include <vm/vm_page.h>
#ifdef FDT
#include <dev/fdt/fdt_common.h>
#include <dev/ofw/ofw_bus.h>
#include <dev/ofw/ofw_bus_subr.h>
#endif
#include <dev/xdma/xdma.h>
#include <xdma_if.h>
struct seg_load_request {
struct bus_dma_segment *seg;
uint32_t nsegs;
uint32_t error;
};
static void
xchan_bufs_free_reserved(xdma_channel_t *xchan)
{
struct xdma_request *xr;
vm_size_t size;
int i;
for (i = 0; i < xchan->xr_num; i++) {
xr = &xchan->xr_mem[i];
size = xr->buf.size;
if (xr->buf.vaddr) {
pmap_kremove_device(xr->buf.vaddr, size);
kva_free(xr->buf.vaddr, size);
xr->buf.vaddr = 0;
}
if (xr->buf.paddr) {
vmem_free(xchan->vmem, xr->buf.paddr, size);
xr->buf.paddr = 0;
}
xr->buf.size = 0;
}
}
static int
xchan_bufs_alloc_reserved(xdma_channel_t *xchan)
{
xdma_controller_t *xdma;
struct xdma_request *xr;
vmem_addr_t addr;
vm_size_t size;
int i;
xdma = xchan->xdma;
if (xchan->vmem == NULL)
return (ENOBUFS);
for (i = 0; i < xchan->xr_num; i++) {
xr = &xchan->xr_mem[i];
size = round_page(xchan->maxsegsize);
if (vmem_alloc(xchan->vmem, size,
M_BESTFIT | M_NOWAIT, &addr)) {
device_printf(xdma->dev,
"%s: Can't allocate memory\n", __func__);
xchan_bufs_free_reserved(xchan);
return (ENOMEM);
}
xr->buf.size = size;
xr->buf.paddr = addr;
xr->buf.vaddr = kva_alloc(size);
if (xr->buf.vaddr == 0) {
device_printf(xdma->dev,
"%s: Can't allocate KVA\n", __func__);
xchan_bufs_free_reserved(xchan);
return (ENOMEM);
}
pmap_kenter_device(xr->buf.vaddr, size, addr);
}
return (0);
}
static int
xchan_bufs_alloc_busdma(xdma_channel_t *xchan)
{
xdma_controller_t *xdma;
struct xdma_request *xr;
int err;
int i;
xdma = xchan->xdma;
/* Create bus_dma tag */
err = bus_dma_tag_create(
bus_get_dma_tag(xdma->dev), /* Parent tag. */
xchan->alignment, /* alignment */
xchan->boundary, /* boundary */
xchan->lowaddr, /* lowaddr */
xchan->highaddr, /* highaddr */
NULL, NULL, /* filter, filterarg */
xchan->maxsegsize * xchan->maxnsegs, /* maxsize */
xchan->maxnsegs, /* nsegments */
xchan->maxsegsize, /* maxsegsize */
0, /* flags */
NULL, NULL, /* lockfunc, lockarg */
&xchan->dma_tag_bufs);
if (err != 0) {
device_printf(xdma->dev,
"%s: Can't create bus_dma tag.\n", __func__);
return (-1);
}
for (i = 0; i < xchan->xr_num; i++) {
xr = &xchan->xr_mem[i];
err = bus_dmamap_create(xchan->dma_tag_bufs, 0,
&xr->buf.map);
if (err != 0) {
device_printf(xdma->dev,
"%s: Can't create buf DMA map.\n", __func__);
/* Cleanup. */
bus_dma_tag_destroy(xchan->dma_tag_bufs);
return (-1);
}
}
return (0);
}
static int
xchan_bufs_alloc(xdma_channel_t *xchan)
{
xdma_controller_t *xdma;
int ret;
xdma = xchan->xdma;
if (xdma == NULL) {
device_printf(xdma->dev,
"%s: Channel was not allocated properly.\n", __func__);
return (-1);
}
if (xchan->caps & XCHAN_CAP_BUSDMA)
ret = xchan_bufs_alloc_busdma(xchan);
else {
ret = xchan_bufs_alloc_reserved(xchan);
}
if (ret != 0) {
device_printf(xdma->dev,
"%s: Can't allocate bufs.\n", __func__);
return (-1);
}
xchan->flags |= XCHAN_BUFS_ALLOCATED;
return (0);
}
static int
xchan_bufs_free(xdma_channel_t *xchan)
{
struct xdma_request *xr;
struct xchan_buf *b;
int i;
if ((xchan->flags & XCHAN_BUFS_ALLOCATED) == 0)
return (-1);
if (xchan->caps & XCHAN_CAP_BUSDMA) {
for (i = 0; i < xchan->xr_num; i++) {
xr = &xchan->xr_mem[i];
b = &xr->buf;
bus_dmamap_destroy(xchan->dma_tag_bufs, b->map);
}
bus_dma_tag_destroy(xchan->dma_tag_bufs);
} else
xchan_bufs_free_reserved(xchan);
xchan->flags &= ~XCHAN_BUFS_ALLOCATED;
return (0);
}
void
xdma_channel_free_sg(xdma_channel_t *xchan)
{
xchan_bufs_free(xchan);
xchan_sglist_free(xchan);
xchan_bank_free(xchan);
}
/*
* Prepare xchan for a scatter-gather transfer.
* xr_num - xdma requests queue size,
* maxsegsize - maximum allowed scatter-gather list element size in bytes
*/
int
xdma_prep_sg(xdma_channel_t *xchan, uint32_t xr_num,
bus_size_t maxsegsize, bus_size_t maxnsegs,
bus_size_t alignment, bus_addr_t boundary,
bus_addr_t lowaddr, bus_addr_t highaddr)
{
xdma_controller_t *xdma;
int ret;
xdma = xchan->xdma;
KASSERT(xdma != NULL, ("xdma is NULL"));
if (xchan->flags & XCHAN_CONFIGURED) {
device_printf(xdma->dev,
"%s: Channel is already configured.\n", __func__);
return (-1);
}
xchan->xr_num = xr_num;
xchan->maxsegsize = maxsegsize;
xchan->maxnsegs = maxnsegs;
xchan->alignment = alignment;
xchan->boundary = boundary;
xchan->lowaddr = lowaddr;
xchan->highaddr = highaddr;
if (xchan->maxnsegs > XDMA_MAX_SEG) {
device_printf(xdma->dev, "%s: maxnsegs is too big\n",
__func__);
return (-1);
}
xchan_bank_init(xchan);
/* Allocate sglist. */
ret = xchan_sglist_alloc(xchan);
if (ret != 0) {
device_printf(xdma->dev,
"%s: Can't allocate sglist.\n", __func__);
return (-1);
}
/* Allocate buffers if required. */
if ((xchan->caps & XCHAN_CAP_NOBUFS) == 0) {
ret = xchan_bufs_alloc(xchan);
if (ret != 0) {
device_printf(xdma->dev,
"%s: Can't allocate bufs.\n", __func__);
/* Cleanup */
xchan_sglist_free(xchan);
xchan_bank_free(xchan);
return (-1);
}
}
xchan->flags |= (XCHAN_CONFIGURED | XCHAN_TYPE_SG);
XCHAN_LOCK(xchan);
ret = XDMA_CHANNEL_PREP_SG(xdma->dma_dev, xchan);
if (ret != 0) {
device_printf(xdma->dev,
"%s: Can't prepare SG transfer.\n", __func__);
XCHAN_UNLOCK(xchan);
return (-1);
}
XCHAN_UNLOCK(xchan);
return (0);
}
void
xchan_seg_done(xdma_channel_t *xchan,
struct xdma_transfer_status *st)
{
struct xdma_request *xr;
xdma_controller_t *xdma;
struct xchan_buf *b;
xdma = xchan->xdma;
xr = TAILQ_FIRST(&xchan->processing);
if (xr == NULL)
panic("request not found\n");
b = &xr->buf;
atomic_subtract_int(&b->nsegs_left, 1);
if (b->nsegs_left == 0) {
if (xchan->caps & XCHAN_CAP_BUSDMA) {
if (xr->direction == XDMA_MEM_TO_DEV)
bus_dmamap_sync(xchan->dma_tag_bufs, b->map,
BUS_DMASYNC_POSTWRITE);
else
bus_dmamap_sync(xchan->dma_tag_bufs, b->map,
BUS_DMASYNC_POSTREAD);
bus_dmamap_unload(xchan->dma_tag_bufs, b->map);
} else {
if (xr->req_type == XR_TYPE_MBUF &&
xr->direction == XDMA_DEV_TO_MEM)
m_copyback(xr->m, 0, st->transferred,
(void *)xr->buf.vaddr);
}
xr->status.error = st->error;
xr->status.transferred = st->transferred;
QUEUE_PROC_LOCK(xchan);
TAILQ_REMOVE(&xchan->processing, xr, xr_next);
QUEUE_PROC_UNLOCK(xchan);
QUEUE_OUT_LOCK(xchan);
TAILQ_INSERT_TAIL(&xchan->queue_out, xr, xr_next);
QUEUE_OUT_UNLOCK(xchan);
}
}
static void
xdma_dmamap_cb(void *arg, bus_dma_segment_t *segs, int nsegs, int error)
{
struct seg_load_request *slr;
struct bus_dma_segment *seg;
int i;
slr = arg;
seg = slr->seg;
if (error != 0) {
slr->error = error;
return;
}
slr->nsegs = nsegs;
for (i = 0; i < nsegs; i++) {
seg[i].ds_addr = segs[i].ds_addr;
seg[i].ds_len = segs[i].ds_len;
}
}
static int
_xdma_load_data_busdma(xdma_channel_t *xchan, struct xdma_request *xr,
struct bus_dma_segment *seg)
{
xdma_controller_t *xdma;
struct seg_load_request slr;
uint32_t nsegs;
void *addr;
int error;
xdma = xchan->xdma;
error = 0;
nsegs = 0;
switch (xr->req_type) {
case XR_TYPE_MBUF:
error = bus_dmamap_load_mbuf_sg(xchan->dma_tag_bufs,
xr->buf.map, xr->m, seg, &nsegs, BUS_DMA_NOWAIT);
break;
case XR_TYPE_BIO:
slr.nsegs = 0;
slr.error = 0;
slr.seg = seg;
error = bus_dmamap_load_bio(xchan->dma_tag_bufs,
xr->buf.map, xr->bp, xdma_dmamap_cb, &slr, BUS_DMA_NOWAIT);
if (slr.error != 0) {
device_printf(xdma->dma_dev,
"%s: bus_dmamap_load failed, err %d\n",
__func__, slr.error);
return (0);
}
nsegs = slr.nsegs;
break;
case XR_TYPE_VIRT:
switch (xr->direction) {
case XDMA_MEM_TO_DEV:
addr = (void *)xr->src_addr;
break;
case XDMA_DEV_TO_MEM:
addr = (void *)xr->dst_addr;
break;
default:
device_printf(xdma->dma_dev,
"%s: Direction is not supported\n", __func__);
return (0);
}
slr.nsegs = 0;
slr.error = 0;
slr.seg = seg;
error = bus_dmamap_load(xchan->dma_tag_bufs, xr->buf.map,
addr, (xr->block_len * xr->block_num),
xdma_dmamap_cb, &slr, BUS_DMA_NOWAIT);
if (slr.error != 0) {
device_printf(xdma->dma_dev,
"%s: bus_dmamap_load failed, err %d\n",
__func__, slr.error);
return (0);
}
nsegs = slr.nsegs;
break;
default:
break;
}
if (error != 0) {
if (error == ENOMEM) {
/*
* Out of memory. Try again later.
* TODO: count errors.
*/
} else
device_printf(xdma->dma_dev,
"%s: bus_dmamap_load failed with err %d\n",
__func__, error);
return (0);
}
if (xr->direction == XDMA_MEM_TO_DEV)
bus_dmamap_sync(xchan->dma_tag_bufs, xr->buf.map,
BUS_DMASYNC_PREWRITE);
else
bus_dmamap_sync(xchan->dma_tag_bufs, xr->buf.map,
BUS_DMASYNC_PREREAD);
return (nsegs);
}
static int
_xdma_load_data(xdma_channel_t *xchan, struct xdma_request *xr,
struct bus_dma_segment *seg)
{
xdma_controller_t *xdma;
struct mbuf *m;
uint32_t nsegs;
xdma = xchan->xdma;
m = xr->m;
nsegs = 1;
switch (xr->req_type) {
case XR_TYPE_MBUF:
if ((xchan->caps & XCHAN_CAP_NOBUFS) == 0) {
if (xr->direction == XDMA_MEM_TO_DEV)
m_copydata(m, 0, m->m_pkthdr.len,
(void *)xr->buf.vaddr);
seg[0].ds_addr = (bus_addr_t)xr->buf.paddr;
} else
seg[0].ds_addr = mtod(m, bus_addr_t);
seg[0].ds_len = m->m_pkthdr.len;
break;
case XR_TYPE_BIO:
case XR_TYPE_VIRT:
default:
panic("implement me\n");
}
return (nsegs);
}
static int
xdma_load_data(xdma_channel_t *xchan,
struct xdma_request *xr, struct bus_dma_segment *seg)
{
xdma_controller_t *xdma;
int error;
int nsegs;
xdma = xchan->xdma;
error = 0;
nsegs = 0;
if (xchan->caps & XCHAN_CAP_BUSDMA)
nsegs = _xdma_load_data_busdma(xchan, xr, seg);
else
nsegs = _xdma_load_data(xchan, xr, seg);
if (nsegs == 0)
return (0); /* Try again later. */
xr->buf.nsegs = nsegs;
xr->buf.nsegs_left = nsegs;
return (nsegs);
}
static int
xdma_process(xdma_channel_t *xchan,
struct xdma_sglist *sg)
{
struct bus_dma_segment seg[XDMA_MAX_SEG];
struct xdma_request *xr;
struct xdma_request *xr_tmp;
xdma_controller_t *xdma;
uint32_t capacity;
uint32_t n;
uint32_t c;
int nsegs;
int ret;
XCHAN_ASSERT_LOCKED(xchan);
xdma = xchan->xdma;
n = 0;
c = 0;
ret = XDMA_CHANNEL_CAPACITY(xdma->dma_dev, xchan, &capacity);
if (ret != 0) {
device_printf(xdma->dev,
"%s: Can't get DMA controller capacity.\n", __func__);
return (-1);
}
TAILQ_FOREACH_SAFE(xr, &xchan->queue_in, xr_next, xr_tmp) {
switch (xr->req_type) {
case XR_TYPE_MBUF:
if ((xchan->caps & XCHAN_CAP_NOSEG) ||
(c > xchan->maxnsegs))
c = xdma_mbuf_defrag(xchan, xr);
break;
case XR_TYPE_BIO:
case XR_TYPE_VIRT:
default:
c = 1;
}
if (capacity <= (c + n)) {
/*
* No space yet available for the entire
* request in the DMA engine.
*/
break;
}
if ((c + n + xchan->maxnsegs) >= XDMA_SGLIST_MAXLEN) {
/* Sglist is full. */
break;
}
nsegs = xdma_load_data(xchan, xr, seg);
if (nsegs == 0)
break;
xdma_sglist_add(&sg[n], seg, nsegs, xr);
n += nsegs;
QUEUE_IN_LOCK(xchan);
TAILQ_REMOVE(&xchan->queue_in, xr, xr_next);
QUEUE_IN_UNLOCK(xchan);
QUEUE_PROC_LOCK(xchan);
TAILQ_INSERT_TAIL(&xchan->processing, xr, xr_next);
QUEUE_PROC_UNLOCK(xchan);
}
return (n);
}
int
xdma_queue_submit_sg(xdma_channel_t *xchan)
{
struct xdma_sglist *sg;
xdma_controller_t *xdma;
uint32_t sg_n;
int ret;
xdma = xchan->xdma;
KASSERT(xdma != NULL, ("xdma is NULL"));
XCHAN_ASSERT_LOCKED(xchan);
sg = xchan->sg;
if ((xchan->caps & XCHAN_CAP_NOBUFS) == 0 &&
(xchan->flags & XCHAN_BUFS_ALLOCATED) == 0) {
device_printf(xdma->dev,
"%s: Can't submit a transfer: no bufs\n",
__func__);
return (-1);
}
sg_n = xdma_process(xchan, sg);
if (sg_n == 0)
return (0); /* Nothing to submit */
/* Now submit sglist to DMA engine driver. */
ret = XDMA_CHANNEL_SUBMIT_SG(xdma->dma_dev, xchan, sg, sg_n);
if (ret != 0) {
device_printf(xdma->dev,
"%s: Can't submit an sglist.\n", __func__);
return (-1);
}
return (0);
}