linux/drivers/net/ethernet/synopsys/dwc-xlgmac-desc.c

645 lines
16 KiB
C
Raw Normal View History

/* Synopsys DesignWare Core Enterprise Ethernet (XLGMAC) Driver
*
* Copyright (c) 2017 Synopsys, Inc. (www.synopsys.com)
*
* This program is dual-licensed; you may select either version 2 of
* the GNU General Public License ("GPL") or BSD license ("BSD").
*
* This Synopsys DWC XLGMAC software driver and associated documentation
* (hereinafter the "Software") is an unsupported proprietary work of
* Synopsys, Inc. unless otherwise expressly agreed to in writing between
* Synopsys and you. The Software IS NOT an item of Licensed Software or a
* Licensed Product under any End User Software License Agreement or
* Agreement for Licensed Products with Synopsys or any supplement thereto.
* Synopsys is a registered trademark of Synopsys, Inc. Other names included
* in the SOFTWARE may be the trademarks of their respective owners.
*/
#include "dwc-xlgmac.h"
#include "dwc-xlgmac-reg.h"
static void xlgmac_unmap_desc_data(struct xlgmac_pdata *pdata,
struct xlgmac_desc_data *desc_data)
{
if (desc_data->skb_dma) {
if (desc_data->mapped_as_page) {
dma_unmap_page(pdata->dev, desc_data->skb_dma,
desc_data->skb_dma_len, DMA_TO_DEVICE);
} else {
dma_unmap_single(pdata->dev, desc_data->skb_dma,
desc_data->skb_dma_len, DMA_TO_DEVICE);
}
desc_data->skb_dma = 0;
desc_data->skb_dma_len = 0;
}
if (desc_data->skb) {
dev_kfree_skb_any(desc_data->skb);
desc_data->skb = NULL;
}
if (desc_data->rx.hdr.pa.pages)
put_page(desc_data->rx.hdr.pa.pages);
if (desc_data->rx.hdr.pa_unmap.pages) {
dma_unmap_page(pdata->dev, desc_data->rx.hdr.pa_unmap.pages_dma,
desc_data->rx.hdr.pa_unmap.pages_len,
DMA_FROM_DEVICE);
put_page(desc_data->rx.hdr.pa_unmap.pages);
}
if (desc_data->rx.buf.pa.pages)
put_page(desc_data->rx.buf.pa.pages);
if (desc_data->rx.buf.pa_unmap.pages) {
dma_unmap_page(pdata->dev, desc_data->rx.buf.pa_unmap.pages_dma,
desc_data->rx.buf.pa_unmap.pages_len,
DMA_FROM_DEVICE);
put_page(desc_data->rx.buf.pa_unmap.pages);
}
memset(&desc_data->tx, 0, sizeof(desc_data->tx));
memset(&desc_data->rx, 0, sizeof(desc_data->rx));
desc_data->mapped_as_page = 0;
if (desc_data->state_saved) {
desc_data->state_saved = 0;
desc_data->state.skb = NULL;
desc_data->state.len = 0;
desc_data->state.error = 0;
}
}
static void xlgmac_free_ring(struct xlgmac_pdata *pdata,
struct xlgmac_ring *ring)
{
struct xlgmac_desc_data *desc_data;
unsigned int i;
if (!ring)
return;
if (ring->desc_data_head) {
for (i = 0; i < ring->dma_desc_count; i++) {
desc_data = XLGMAC_GET_DESC_DATA(ring, i);
xlgmac_unmap_desc_data(pdata, desc_data);
}
kfree(ring->desc_data_head);
ring->desc_data_head = NULL;
}
if (ring->rx_hdr_pa.pages) {
dma_unmap_page(pdata->dev, ring->rx_hdr_pa.pages_dma,
ring->rx_hdr_pa.pages_len, DMA_FROM_DEVICE);
put_page(ring->rx_hdr_pa.pages);
ring->rx_hdr_pa.pages = NULL;
ring->rx_hdr_pa.pages_len = 0;
ring->rx_hdr_pa.pages_offset = 0;
ring->rx_hdr_pa.pages_dma = 0;
}
if (ring->rx_buf_pa.pages) {
dma_unmap_page(pdata->dev, ring->rx_buf_pa.pages_dma,
ring->rx_buf_pa.pages_len, DMA_FROM_DEVICE);
put_page(ring->rx_buf_pa.pages);
ring->rx_buf_pa.pages = NULL;
ring->rx_buf_pa.pages_len = 0;
ring->rx_buf_pa.pages_offset = 0;
ring->rx_buf_pa.pages_dma = 0;
}
if (ring->dma_desc_head) {
dma_free_coherent(pdata->dev,
(sizeof(struct xlgmac_dma_desc) *
ring->dma_desc_count),
ring->dma_desc_head,
ring->dma_desc_head_addr);
ring->dma_desc_head = NULL;
}
}
static int xlgmac_init_ring(struct xlgmac_pdata *pdata,
struct xlgmac_ring *ring,
unsigned int dma_desc_count)
{
if (!ring)
return 0;
/* Descriptors */
ring->dma_desc_count = dma_desc_count;
ring->dma_desc_head = dma_alloc_coherent(pdata->dev,
(sizeof(struct xlgmac_dma_desc) *
dma_desc_count),
&ring->dma_desc_head_addr,
GFP_KERNEL);
if (!ring->dma_desc_head)
return -ENOMEM;
/* Array of descriptor data */
ring->desc_data_head = kcalloc(dma_desc_count,
sizeof(struct xlgmac_desc_data),
GFP_KERNEL);
if (!ring->desc_data_head)
return -ENOMEM;
netif_dbg(pdata, drv, pdata->netdev,
"dma_desc_head=%p, dma_desc_head_addr=%pad, desc_data_head=%p\n",
ring->dma_desc_head,
&ring->dma_desc_head_addr,
ring->desc_data_head);
return 0;
}
static void xlgmac_free_rings(struct xlgmac_pdata *pdata)
{
struct xlgmac_channel *channel;
unsigned int i;
if (!pdata->channel_head)
return;
channel = pdata->channel_head;
for (i = 0; i < pdata->channel_count; i++, channel++) {
xlgmac_free_ring(pdata, channel->tx_ring);
xlgmac_free_ring(pdata, channel->rx_ring);
}
}
static int xlgmac_alloc_rings(struct xlgmac_pdata *pdata)
{
struct xlgmac_channel *channel;
unsigned int i;
int ret;
channel = pdata->channel_head;
for (i = 0; i < pdata->channel_count; i++, channel++) {
netif_dbg(pdata, drv, pdata->netdev, "%s - Tx ring:\n",
channel->name);
ret = xlgmac_init_ring(pdata, channel->tx_ring,
pdata->tx_desc_count);
if (ret) {
netdev_alert(pdata->netdev,
"error initializing Tx ring");
goto err_init_ring;
}
netif_dbg(pdata, drv, pdata->netdev, "%s - Rx ring:\n",
channel->name);
ret = xlgmac_init_ring(pdata, channel->rx_ring,
pdata->rx_desc_count);
if (ret) {
netdev_alert(pdata->netdev,
"error initializing Rx ring\n");
goto err_init_ring;
}
}
return 0;
err_init_ring:
xlgmac_free_rings(pdata);
return ret;
}
static void xlgmac_free_channels(struct xlgmac_pdata *pdata)
{
if (!pdata->channel_head)
return;
kfree(pdata->channel_head->tx_ring);
pdata->channel_head->tx_ring = NULL;
kfree(pdata->channel_head->rx_ring);
pdata->channel_head->rx_ring = NULL;
kfree(pdata->channel_head);
pdata->channel_head = NULL;
pdata->channel_count = 0;
}
static int xlgmac_alloc_channels(struct xlgmac_pdata *pdata)
{
struct xlgmac_channel *channel_head, *channel;
struct xlgmac_ring *tx_ring, *rx_ring;
int ret = -ENOMEM;
unsigned int i;
channel_head = kcalloc(pdata->channel_count,
sizeof(struct xlgmac_channel), GFP_KERNEL);
if (!channel_head)
return ret;
netif_dbg(pdata, drv, pdata->netdev,
"channel_head=%p\n", channel_head);
tx_ring = kcalloc(pdata->tx_ring_count, sizeof(struct xlgmac_ring),
GFP_KERNEL);
if (!tx_ring)
goto err_tx_ring;
rx_ring = kcalloc(pdata->rx_ring_count, sizeof(struct xlgmac_ring),
GFP_KERNEL);
if (!rx_ring)
goto err_rx_ring;
for (i = 0, channel = channel_head; i < pdata->channel_count;
i++, channel++) {
snprintf(channel->name, sizeof(channel->name), "channel-%u", i);
channel->pdata = pdata;
channel->queue_index = i;
channel->dma_regs = pdata->mac_regs + DMA_CH_BASE +
(DMA_CH_INC * i);
if (pdata->per_channel_irq) {
/* Get the per DMA interrupt */
ret = pdata->channel_irq[i];
if (ret < 0) {
netdev_err(pdata->netdev,
"get_irq %u failed\n",
i + 1);
goto err_irq;
}
channel->dma_irq = ret;
}
if (i < pdata->tx_ring_count)
channel->tx_ring = tx_ring++;
if (i < pdata->rx_ring_count)
channel->rx_ring = rx_ring++;
netif_dbg(pdata, drv, pdata->netdev,
"%s: dma_regs=%p, tx_ring=%p, rx_ring=%p\n",
channel->name, channel->dma_regs,
channel->tx_ring, channel->rx_ring);
}
pdata->channel_head = channel_head;
return 0;
err_irq:
kfree(rx_ring);
err_rx_ring:
kfree(tx_ring);
err_tx_ring:
kfree(channel_head);
return ret;
}
static void xlgmac_free_channels_and_rings(struct xlgmac_pdata *pdata)
{
xlgmac_free_rings(pdata);
xlgmac_free_channels(pdata);
}
static int xlgmac_alloc_channels_and_rings(struct xlgmac_pdata *pdata)
{
int ret;
ret = xlgmac_alloc_channels(pdata);
if (ret)
goto err_alloc;
ret = xlgmac_alloc_rings(pdata);
if (ret)
goto err_alloc;
return 0;
err_alloc:
xlgmac_free_channels_and_rings(pdata);
return ret;
}
static int xlgmac_alloc_pages(struct xlgmac_pdata *pdata,
struct xlgmac_page_alloc *pa,
gfp_t gfp, int order)
{
struct page *pages = NULL;
dma_addr_t pages_dma;
/* Try to obtain pages, decreasing order if necessary */
mm: remove __GFP_COLD As the page free path makes no distinction between cache hot and cold pages, there is no real useful ordering of pages in the free list that allocation requests can take advantage of. Juding from the users of __GFP_COLD, it is likely that a number of them are the result of copying other sites instead of actually measuring the impact. Remove the __GFP_COLD parameter which simplifies a number of paths in the page allocator. This is potentially controversial but bear in mind that the size of the per-cpu pagelists versus modern cache sizes means that the whole per-cpu list can often fit in the L3 cache. Hence, there is only a potential benefit for microbenchmarks that alloc/free pages in a tight loop. It's even worse when THP is taken into account which has little or no chance of getting a cache-hot page as the per-cpu list is bypassed and the zeroing of multiple pages will thrash the cache anyway. The truncate microbenchmarks are not shown as this patch affects the allocation path and not the free path. A page fault microbenchmark was tested but it showed no sigificant difference which is not surprising given that the __GFP_COLD branches are a miniscule percentage of the fault path. Link: http://lkml.kernel.org/r/20171018075952.10627-9-mgorman@techsingularity.net Signed-off-by: Mel Gorman <mgorman@techsingularity.net> Acked-by: Vlastimil Babka <vbabka@suse.cz> Cc: Andi Kleen <ak@linux.intel.com> Cc: Dave Chinner <david@fromorbit.com> Cc: Dave Hansen <dave.hansen@intel.com> Cc: Jan Kara <jack@suse.cz> Cc: Johannes Weiner <hannes@cmpxchg.org> Signed-off-by: Andrew Morton <akpm@linux-foundation.org> Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2017-11-16 09:38:03 +08:00
gfp |= __GFP_COMP | __GFP_NOWARN;
while (order >= 0) {
pages = alloc_pages(gfp, order);
if (pages)
break;
order--;
}
if (!pages)
return -ENOMEM;
/* Map the pages */
pages_dma = dma_map_page(pdata->dev, pages, 0,
PAGE_SIZE << order, DMA_FROM_DEVICE);
if (dma_mapping_error(pdata->dev, pages_dma)) {
put_page(pages);
return -ENOMEM;
}
pa->pages = pages;
pa->pages_len = PAGE_SIZE << order;
pa->pages_offset = 0;
pa->pages_dma = pages_dma;
return 0;
}
static void xlgmac_set_buffer_data(struct xlgmac_buffer_data *bd,
struct xlgmac_page_alloc *pa,
unsigned int len)
{
get_page(pa->pages);
bd->pa = *pa;
bd->dma_base = pa->pages_dma;
bd->dma_off = pa->pages_offset;
bd->dma_len = len;
pa->pages_offset += len;
if ((pa->pages_offset + len) > pa->pages_len) {
/* This data descriptor is responsible for unmapping page(s) */
bd->pa_unmap = *pa;
/* Get a new allocation next time */
pa->pages = NULL;
pa->pages_len = 0;
pa->pages_offset = 0;
pa->pages_dma = 0;
}
}
static int xlgmac_map_rx_buffer(struct xlgmac_pdata *pdata,
struct xlgmac_ring *ring,
struct xlgmac_desc_data *desc_data)
{
int order, ret;
if (!ring->rx_hdr_pa.pages) {
ret = xlgmac_alloc_pages(pdata, &ring->rx_hdr_pa,
GFP_ATOMIC, 0);
if (ret)
return ret;
}
if (!ring->rx_buf_pa.pages) {
order = max_t(int, PAGE_ALLOC_COSTLY_ORDER - 1, 0);
ret = xlgmac_alloc_pages(pdata, &ring->rx_buf_pa,
GFP_ATOMIC, order);
if (ret)
return ret;
}
/* Set up the header page info */
xlgmac_set_buffer_data(&desc_data->rx.hdr, &ring->rx_hdr_pa,
XLGMAC_SKB_ALLOC_SIZE);
/* Set up the buffer page info */
xlgmac_set_buffer_data(&desc_data->rx.buf, &ring->rx_buf_pa,
pdata->rx_buf_size);
return 0;
}
static void xlgmac_tx_desc_init(struct xlgmac_pdata *pdata)
{
struct xlgmac_hw_ops *hw_ops = &pdata->hw_ops;
struct xlgmac_desc_data *desc_data;
struct xlgmac_dma_desc *dma_desc;
struct xlgmac_channel *channel;
struct xlgmac_ring *ring;
dma_addr_t dma_desc_addr;
unsigned int i, j;
channel = pdata->channel_head;
for (i = 0; i < pdata->channel_count; i++, channel++) {
ring = channel->tx_ring;
if (!ring)
break;
dma_desc = ring->dma_desc_head;
dma_desc_addr = ring->dma_desc_head_addr;
for (j = 0; j < ring->dma_desc_count; j++) {
desc_data = XLGMAC_GET_DESC_DATA(ring, j);
desc_data->dma_desc = dma_desc;
desc_data->dma_desc_addr = dma_desc_addr;
dma_desc++;
dma_desc_addr += sizeof(struct xlgmac_dma_desc);
}
ring->cur = 0;
ring->dirty = 0;
memset(&ring->tx, 0, sizeof(ring->tx));
hw_ops->tx_desc_init(channel);
}
}
static void xlgmac_rx_desc_init(struct xlgmac_pdata *pdata)
{
struct xlgmac_hw_ops *hw_ops = &pdata->hw_ops;
struct xlgmac_desc_data *desc_data;
struct xlgmac_dma_desc *dma_desc;
struct xlgmac_channel *channel;
struct xlgmac_ring *ring;
dma_addr_t dma_desc_addr;
unsigned int i, j;
channel = pdata->channel_head;
for (i = 0; i < pdata->channel_count; i++, channel++) {
ring = channel->rx_ring;
if (!ring)
break;
dma_desc = ring->dma_desc_head;
dma_desc_addr = ring->dma_desc_head_addr;
for (j = 0; j < ring->dma_desc_count; j++) {
desc_data = XLGMAC_GET_DESC_DATA(ring, j);
desc_data->dma_desc = dma_desc;
desc_data->dma_desc_addr = dma_desc_addr;
if (xlgmac_map_rx_buffer(pdata, ring, desc_data))
break;
dma_desc++;
dma_desc_addr += sizeof(struct xlgmac_dma_desc);
}
ring->cur = 0;
ring->dirty = 0;
hw_ops->rx_desc_init(channel);
}
}
static int xlgmac_map_tx_skb(struct xlgmac_channel *channel,
struct sk_buff *skb)
{
struct xlgmac_pdata *pdata = channel->pdata;
struct xlgmac_ring *ring = channel->tx_ring;
unsigned int start_index, cur_index;
struct xlgmac_desc_data *desc_data;
unsigned int offset, datalen, len;
struct xlgmac_pkt_info *pkt_info;
skb_frag_t *frag;
unsigned int tso, vlan;
dma_addr_t skb_dma;
unsigned int i;
offset = 0;
start_index = ring->cur;
cur_index = ring->cur;
pkt_info = &ring->pkt_info;
pkt_info->desc_count = 0;
pkt_info->length = 0;
tso = XLGMAC_GET_REG_BITS(pkt_info->attributes,
TX_PACKET_ATTRIBUTES_TSO_ENABLE_POS,
TX_PACKET_ATTRIBUTES_TSO_ENABLE_LEN);
vlan = XLGMAC_GET_REG_BITS(pkt_info->attributes,
TX_PACKET_ATTRIBUTES_VLAN_CTAG_POS,
TX_PACKET_ATTRIBUTES_VLAN_CTAG_LEN);
/* Save space for a context descriptor if needed */
if ((tso && (pkt_info->mss != ring->tx.cur_mss)) ||
(vlan && (pkt_info->vlan_ctag != ring->tx.cur_vlan_ctag)))
cur_index++;
desc_data = XLGMAC_GET_DESC_DATA(ring, cur_index);
if (tso) {
/* Map the TSO header */
skb_dma = dma_map_single(pdata->dev, skb->data,
pkt_info->header_len, DMA_TO_DEVICE);
if (dma_mapping_error(pdata->dev, skb_dma)) {
netdev_alert(pdata->netdev, "dma_map_single failed\n");
goto err_out;
}
desc_data->skb_dma = skb_dma;
desc_data->skb_dma_len = pkt_info->header_len;
netif_dbg(pdata, tx_queued, pdata->netdev,
"skb header: index=%u, dma=%pad, len=%u\n",
cur_index, &skb_dma, pkt_info->header_len);
offset = pkt_info->header_len;
pkt_info->length += pkt_info->header_len;
cur_index++;
desc_data = XLGMAC_GET_DESC_DATA(ring, cur_index);
}
/* Map the (remainder of the) packet */
for (datalen = skb_headlen(skb) - offset; datalen; ) {
len = min_t(unsigned int, datalen, XLGMAC_TX_MAX_BUF_SIZE);
skb_dma = dma_map_single(pdata->dev, skb->data + offset, len,
DMA_TO_DEVICE);
if (dma_mapping_error(pdata->dev, skb_dma)) {
netdev_alert(pdata->netdev, "dma_map_single failed\n");
goto err_out;
}
desc_data->skb_dma = skb_dma;
desc_data->skb_dma_len = len;
netif_dbg(pdata, tx_queued, pdata->netdev,
"skb data: index=%u, dma=%pad, len=%u\n",
cur_index, &skb_dma, len);
datalen -= len;
offset += len;
pkt_info->length += len;
cur_index++;
desc_data = XLGMAC_GET_DESC_DATA(ring, cur_index);
}
for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) {
netif_dbg(pdata, tx_queued, pdata->netdev,
"mapping frag %u\n", i);
frag = &skb_shinfo(skb)->frags[i];
offset = 0;
for (datalen = skb_frag_size(frag); datalen; ) {
len = min_t(unsigned int, datalen,
XLGMAC_TX_MAX_BUF_SIZE);
skb_dma = skb_frag_dma_map(pdata->dev, frag, offset,
len, DMA_TO_DEVICE);
if (dma_mapping_error(pdata->dev, skb_dma)) {
netdev_alert(pdata->netdev,
"skb_frag_dma_map failed\n");
goto err_out;
}
desc_data->skb_dma = skb_dma;
desc_data->skb_dma_len = len;
desc_data->mapped_as_page = 1;
netif_dbg(pdata, tx_queued, pdata->netdev,
"skb frag: index=%u, dma=%pad, len=%u\n",
cur_index, &skb_dma, len);
datalen -= len;
offset += len;
pkt_info->length += len;
cur_index++;
desc_data = XLGMAC_GET_DESC_DATA(ring, cur_index);
}
}
/* Save the skb address in the last entry. We always have some data
* that has been mapped so desc_data is always advanced past the last
* piece of mapped data - use the entry pointed to by cur_index - 1.
*/
desc_data = XLGMAC_GET_DESC_DATA(ring, cur_index - 1);
desc_data->skb = skb;
/* Save the number of descriptor entries used */
pkt_info->desc_count = cur_index - start_index;
return pkt_info->desc_count;
err_out:
while (start_index < cur_index) {
desc_data = XLGMAC_GET_DESC_DATA(ring, start_index++);
xlgmac_unmap_desc_data(pdata, desc_data);
}
return 0;
}
void xlgmac_init_desc_ops(struct xlgmac_desc_ops *desc_ops)
{
desc_ops->alloc_channels_and_rings = xlgmac_alloc_channels_and_rings;
desc_ops->free_channels_and_rings = xlgmac_free_channels_and_rings;
desc_ops->map_tx_skb = xlgmac_map_tx_skb;
desc_ops->map_rx_buffer = xlgmac_map_rx_buffer;
desc_ops->unmap_desc_data = xlgmac_unmap_desc_data;
desc_ops->tx_desc_init = xlgmac_tx_desc_init;
desc_ops->rx_desc_init = xlgmac_rx_desc_init;
}