ath9k: Add Rx EDMA support
Signed-off-by: Felix Fietkau <nbd@openwrt.org>
Signed-off-by: Vasanthakumar Thiagarajan <vasanth@atheros.com>
Signed-off-by: John W. Linville <linville@tuxdriver.com>
diff --git a/drivers/net/wireless/ath/ath9k/recv.c b/drivers/net/wireless/ath/ath9k/recv.c
index 94560e2..ffb599c 100644
--- a/drivers/net/wireless/ath/ath9k/recv.c
+++ b/drivers/net/wireless/ath/ath9k/recv.c
@@ -16,6 +16,8 @@
#include "ath9k.h"
+#define SKB_CB_ATHBUF(__skb) (*((struct ath_buf **)__skb->cb))
+
static struct ieee80211_hw * ath_get_virt_hw(struct ath_softc *sc,
struct ieee80211_hdr *hdr)
{
@@ -115,6 +117,190 @@
ath9k_hw_setmcastfilter(ah, mfilt[0], mfilt[1]);
}
+static bool ath_rx_edma_buf_link(struct ath_softc *sc,
+ enum ath9k_rx_qtype qtype)
+{
+ struct ath_hw *ah = sc->sc_ah;
+ struct ath_rx_edma *rx_edma;
+ struct sk_buff *skb;
+ struct ath_buf *bf;
+
+ rx_edma = &sc->rx.rx_edma[qtype];
+ if (skb_queue_len(&rx_edma->rx_fifo) >= rx_edma->rx_fifo_hwsize)
+ return false;
+
+ bf = list_first_entry(&sc->rx.rxbuf, struct ath_buf, list);
+ list_del_init(&bf->list);
+
+ skb = bf->bf_mpdu;
+
+ ATH_RXBUF_RESET(bf);
+ memset(skb->data, 0, ah->caps.rx_status_len);
+ dma_sync_single_for_device(sc->dev, bf->bf_buf_addr,
+ ah->caps.rx_status_len, DMA_TO_DEVICE);
+
+ SKB_CB_ATHBUF(skb) = bf;
+ ath9k_hw_addrxbuf_edma(ah, bf->bf_buf_addr, qtype);
+ skb_queue_tail(&rx_edma->rx_fifo, skb);
+
+ return true;
+}
+
+static void ath_rx_addbuffer_edma(struct ath_softc *sc,
+ enum ath9k_rx_qtype qtype, int size)
+{
+ struct ath_rx_edma *rx_edma;
+ struct ath_common *common = ath9k_hw_common(sc->sc_ah);
+ u32 nbuf = 0;
+
+ rx_edma = &sc->rx.rx_edma[qtype];
+ if (list_empty(&sc->rx.rxbuf)) {
+ ath_print(common, ATH_DBG_QUEUE, "No free rx buf available\n");
+ return;
+ }
+
+ while (!list_empty(&sc->rx.rxbuf)) {
+ nbuf++;
+
+ if (!ath_rx_edma_buf_link(sc, qtype))
+ break;
+
+ if (nbuf >= size)
+ break;
+ }
+}
+
+static void ath_rx_remove_buffer(struct ath_softc *sc,
+ enum ath9k_rx_qtype qtype)
+{
+ struct ath_buf *bf;
+ struct ath_rx_edma *rx_edma;
+ struct sk_buff *skb;
+
+ rx_edma = &sc->rx.rx_edma[qtype];
+
+ while ((skb = skb_dequeue(&rx_edma->rx_fifo)) != NULL) {
+ bf = SKB_CB_ATHBUF(skb);
+ BUG_ON(!bf);
+ list_add_tail(&bf->list, &sc->rx.rxbuf);
+ }
+}
+
+static void ath_rx_edma_cleanup(struct ath_softc *sc)
+{
+ struct ath_buf *bf;
+
+ ath_rx_remove_buffer(sc, ATH9K_RX_QUEUE_LP);
+ ath_rx_remove_buffer(sc, ATH9K_RX_QUEUE_HP);
+
+ list_for_each_entry(bf, &sc->rx.rxbuf, list) {
+ if (bf->bf_mpdu)
+ dev_kfree_skb_any(bf->bf_mpdu);
+ }
+
+ INIT_LIST_HEAD(&sc->rx.rxbuf);
+
+ kfree(sc->rx.rx_bufptr);
+ sc->rx.rx_bufptr = NULL;
+}
+
+static void ath_rx_edma_init_queue(struct ath_rx_edma *rx_edma, int size)
+{
+ skb_queue_head_init(&rx_edma->rx_fifo);
+ skb_queue_head_init(&rx_edma->rx_buffers);
+ rx_edma->rx_fifo_hwsize = size;
+}
+
+static int ath_rx_edma_init(struct ath_softc *sc, int nbufs)
+{
+ struct ath_common *common = ath9k_hw_common(sc->sc_ah);
+ struct ath_hw *ah = sc->sc_ah;
+ struct sk_buff *skb;
+ struct ath_buf *bf;
+ int error = 0, i;
+ u32 size;
+
+
+ common->rx_bufsize = roundup(IEEE80211_MAX_MPDU_LEN +
+ ah->caps.rx_status_len,
+ min(common->cachelsz, (u16)64));
+
+ ath9k_hw_set_rx_bufsize(ah, common->rx_bufsize -
+ ah->caps.rx_status_len);
+
+ ath_rx_edma_init_queue(&sc->rx.rx_edma[ATH9K_RX_QUEUE_LP],
+ ah->caps.rx_lp_qdepth);
+ ath_rx_edma_init_queue(&sc->rx.rx_edma[ATH9K_RX_QUEUE_HP],
+ ah->caps.rx_hp_qdepth);
+
+ size = sizeof(struct ath_buf) * nbufs;
+ bf = kzalloc(size, GFP_KERNEL);
+ if (!bf)
+ return -ENOMEM;
+
+ INIT_LIST_HEAD(&sc->rx.rxbuf);
+ sc->rx.rx_bufptr = bf;
+
+ for (i = 0; i < nbufs; i++, bf++) {
+ skb = ath_rxbuf_alloc(common, common->rx_bufsize, GFP_KERNEL);
+ if (!skb) {
+ error = -ENOMEM;
+ goto rx_init_fail;
+ }
+
+ memset(skb->data, 0, common->rx_bufsize);
+ bf->bf_mpdu = skb;
+
+ bf->bf_buf_addr = dma_map_single(sc->dev, skb->data,
+ common->rx_bufsize,
+ DMA_BIDIRECTIONAL);
+ if (unlikely(dma_mapping_error(sc->dev,
+ bf->bf_buf_addr))) {
+ dev_kfree_skb_any(skb);
+ bf->bf_mpdu = NULL;
+ ath_print(common, ATH_DBG_FATAL,
+ "dma_mapping_error() on RX init\n");
+ error = -ENOMEM;
+ goto rx_init_fail;
+ }
+
+ list_add_tail(&bf->list, &sc->rx.rxbuf);
+ }
+
+ return 0;
+
+rx_init_fail:
+ ath_rx_edma_cleanup(sc);
+ return error;
+}
+
+static void ath_edma_start_recv(struct ath_softc *sc)
+{
+ spin_lock_bh(&sc->rx.rxbuflock);
+
+ ath9k_hw_rxena(sc->sc_ah);
+
+ ath_rx_addbuffer_edma(sc, ATH9K_RX_QUEUE_HP,
+ sc->rx.rx_edma[ATH9K_RX_QUEUE_HP].rx_fifo_hwsize);
+
+ ath_rx_addbuffer_edma(sc, ATH9K_RX_QUEUE_LP,
+ sc->rx.rx_edma[ATH9K_RX_QUEUE_LP].rx_fifo_hwsize);
+
+ spin_unlock_bh(&sc->rx.rxbuflock);
+
+ ath_opmode_init(sc);
+
+ ath9k_hw_startpcureceive(sc->sc_ah);
+}
+
+static void ath_edma_stop_recv(struct ath_softc *sc)
+{
+ spin_lock_bh(&sc->rx.rxbuflock);
+ ath_rx_remove_buffer(sc, ATH9K_RX_QUEUE_HP);
+ ath_rx_remove_buffer(sc, ATH9K_RX_QUEUE_LP);
+ spin_unlock_bh(&sc->rx.rxbuflock);
+}
+
int ath_rx_init(struct ath_softc *sc, int nbufs)
{
struct ath_common *common = ath9k_hw_common(sc->sc_ah);
@@ -126,45 +312,51 @@
sc->sc_flags &= ~SC_OP_RXFLUSH;
spin_lock_init(&sc->rx.rxbuflock);
- common->rx_bufsize = roundup(IEEE80211_MAX_MPDU_LEN,
- min(common->cachelsz, (u16)64));
+ if (sc->sc_ah->caps.hw_caps & ATH9K_HW_CAP_EDMA) {
+ return ath_rx_edma_init(sc, nbufs);
+ } else {
+ common->rx_bufsize = roundup(IEEE80211_MAX_MPDU_LEN,
+ min(common->cachelsz, (u16)64));
- ath_print(common, ATH_DBG_CONFIG, "cachelsz %u rxbufsize %u\n",
- common->cachelsz, common->rx_bufsize);
+ ath_print(common, ATH_DBG_CONFIG, "cachelsz %u rxbufsize %u\n",
+ common->cachelsz, common->rx_bufsize);
- /* Initialize rx descriptors */
+ /* Initialize rx descriptors */
- error = ath_descdma_setup(sc, &sc->rx.rxdma, &sc->rx.rxbuf,
- "rx", nbufs, 1);
- if (error != 0) {
- ath_print(common, ATH_DBG_FATAL,
- "failed to allocate rx descriptors: %d\n", error);
- goto err;
- }
-
- list_for_each_entry(bf, &sc->rx.rxbuf, list) {
- skb = ath_rxbuf_alloc(common, common->rx_bufsize, GFP_KERNEL);
- if (skb == NULL) {
- error = -ENOMEM;
- goto err;
- }
-
- bf->bf_mpdu = skb;
- bf->bf_buf_addr = dma_map_single(sc->dev, skb->data,
- common->rx_bufsize,
- DMA_FROM_DEVICE);
- if (unlikely(dma_mapping_error(sc->dev,
- bf->bf_buf_addr))) {
- dev_kfree_skb_any(skb);
- bf->bf_mpdu = NULL;
+ error = ath_descdma_setup(sc, &sc->rx.rxdma, &sc->rx.rxbuf,
+ "rx", nbufs, 1);
+ if (error != 0) {
ath_print(common, ATH_DBG_FATAL,
- "dma_mapping_error() on RX init\n");
- error = -ENOMEM;
+ "failed to allocate rx descriptors: %d\n",
+ error);
goto err;
}
- bf->bf_dmacontext = bf->bf_buf_addr;
+
+ list_for_each_entry(bf, &sc->rx.rxbuf, list) {
+ skb = ath_rxbuf_alloc(common, common->rx_bufsize,
+ GFP_KERNEL);
+ if (skb == NULL) {
+ error = -ENOMEM;
+ goto err;
+ }
+
+ bf->bf_mpdu = skb;
+ bf->bf_buf_addr = dma_map_single(sc->dev, skb->data,
+ common->rx_bufsize,
+ DMA_FROM_DEVICE);
+ if (unlikely(dma_mapping_error(sc->dev,
+ bf->bf_buf_addr))) {
+ dev_kfree_skb_any(skb);
+ bf->bf_mpdu = NULL;
+ ath_print(common, ATH_DBG_FATAL,
+ "dma_mapping_error() on RX init\n");
+ error = -ENOMEM;
+ goto err;
+ }
+ bf->bf_dmacontext = bf->bf_buf_addr;
+ }
+ sc->rx.rxlink = NULL;
}
- sc->rx.rxlink = NULL;
err:
if (error)
@@ -180,17 +372,23 @@
struct sk_buff *skb;
struct ath_buf *bf;
- list_for_each_entry(bf, &sc->rx.rxbuf, list) {
- skb = bf->bf_mpdu;
- if (skb) {
- dma_unmap_single(sc->dev, bf->bf_buf_addr,
- common->rx_bufsize, DMA_FROM_DEVICE);
- dev_kfree_skb(skb);
+ if (sc->sc_ah->caps.hw_caps & ATH9K_HW_CAP_EDMA) {
+ ath_rx_edma_cleanup(sc);
+ return;
+ } else {
+ list_for_each_entry(bf, &sc->rx.rxbuf, list) {
+ skb = bf->bf_mpdu;
+ if (skb) {
+ dma_unmap_single(sc->dev, bf->bf_buf_addr,
+ common->rx_bufsize,
+ DMA_FROM_DEVICE);
+ dev_kfree_skb(skb);
+ }
}
- }
- if (sc->rx.rxdma.dd_desc_len != 0)
- ath_descdma_cleanup(sc, &sc->rx.rxdma, &sc->rx.rxbuf);
+ if (sc->rx.rxdma.dd_desc_len != 0)
+ ath_descdma_cleanup(sc, &sc->rx.rxdma, &sc->rx.rxbuf);
+ }
}
/*
@@ -273,6 +471,11 @@
struct ath_hw *ah = sc->sc_ah;
struct ath_buf *bf, *tbf;
+ if (ah->caps.hw_caps & ATH9K_HW_CAP_EDMA) {
+ ath_edma_start_recv(sc);
+ return 0;
+ }
+
spin_lock_bh(&sc->rx.rxbuflock);
if (list_empty(&sc->rx.rxbuf))
goto start_recv;
@@ -306,7 +509,11 @@
ath9k_hw_stoppcurecv(ah);
ath9k_hw_setrxfilter(ah, 0);
stopped = ath9k_hw_stopdmarecv(ah);
- sc->rx.rxlink = NULL;
+
+ if (sc->sc_ah->caps.hw_caps & ATH9K_HW_CAP_EDMA)
+ ath_edma_stop_recv(sc);
+ else
+ sc->rx.rxlink = NULL;
return stopped;
}
@@ -315,7 +522,9 @@
{
spin_lock_bh(&sc->rx.rxflushlock);
sc->sc_flags |= SC_OP_RXFLUSH;
- ath_rx_tasklet(sc, 1);
+ if (sc->sc_ah->caps.hw_caps & ATH9K_HW_CAP_EDMA)
+ ath_rx_tasklet(sc, 1, true);
+ ath_rx_tasklet(sc, 1, false);
sc->sc_flags &= ~SC_OP_RXFLUSH;
spin_unlock_bh(&sc->rx.rxflushlock);
}
@@ -469,14 +678,147 @@
ieee80211_rx(hw, skb);
}
-int ath_rx_tasklet(struct ath_softc *sc, int flush)
+static bool ath_edma_get_buffers(struct ath_softc *sc,
+ enum ath9k_rx_qtype qtype)
{
-#define PA2DESC(_sc, _pa) \
- ((struct ath_desc *)((caddr_t)(_sc)->rx.rxdma.dd_desc + \
- ((_pa) - (_sc)->rx.rxdma.dd_desc_paddr)))
-
+ struct ath_rx_edma *rx_edma = &sc->rx.rx_edma[qtype];
+ struct ath_hw *ah = sc->sc_ah;
+ struct ath_common *common = ath9k_hw_common(ah);
+ struct sk_buff *skb;
struct ath_buf *bf;
+ int ret;
+
+ skb = skb_peek(&rx_edma->rx_fifo);
+ if (!skb)
+ return false;
+
+ bf = SKB_CB_ATHBUF(skb);
+ BUG_ON(!bf);
+
+ dma_sync_single_for_device(sc->dev, bf->bf_buf_addr,
+ common->rx_bufsize, DMA_FROM_DEVICE);
+
+ ret = ath9k_hw_process_rxdesc_edma(ah, NULL, skb->data);
+ if (ret == -EINPROGRESS)
+ return false;
+
+ __skb_unlink(skb, &rx_edma->rx_fifo);
+ if (ret == -EINVAL) {
+ /* corrupt descriptor, skip this one and the following one */
+ list_add_tail(&bf->list, &sc->rx.rxbuf);
+ ath_rx_edma_buf_link(sc, qtype);
+ skb = skb_peek(&rx_edma->rx_fifo);
+ if (!skb)
+ return true;
+
+ bf = SKB_CB_ATHBUF(skb);
+ BUG_ON(!bf);
+
+ __skb_unlink(skb, &rx_edma->rx_fifo);
+ list_add_tail(&bf->list, &sc->rx.rxbuf);
+ ath_rx_edma_buf_link(sc, qtype);
+ }
+ skb_queue_tail(&rx_edma->rx_buffers, skb);
+
+ return true;
+}
+
+static struct ath_buf *ath_edma_get_next_rx_buf(struct ath_softc *sc,
+ struct ath_rx_status *rs,
+ enum ath9k_rx_qtype qtype)
+{
+ struct ath_rx_edma *rx_edma = &sc->rx.rx_edma[qtype];
+ struct sk_buff *skb;
+ struct ath_buf *bf;
+
+ while (ath_edma_get_buffers(sc, qtype));
+ skb = __skb_dequeue(&rx_edma->rx_buffers);
+ if (!skb)
+ return NULL;
+
+ bf = SKB_CB_ATHBUF(skb);
+ ath9k_hw_process_rxdesc_edma(sc->sc_ah, rs, skb->data);
+ return bf;
+}
+
+static struct ath_buf *ath_get_next_rx_buf(struct ath_softc *sc,
+ struct ath_rx_status *rs)
+{
+ struct ath_hw *ah = sc->sc_ah;
+ struct ath_common *common = ath9k_hw_common(ah);
struct ath_desc *ds;
+ struct ath_buf *bf;
+ int ret;
+
+ if (list_empty(&sc->rx.rxbuf)) {
+ sc->rx.rxlink = NULL;
+ return NULL;
+ }
+
+ bf = list_first_entry(&sc->rx.rxbuf, struct ath_buf, list);
+ ds = bf->bf_desc;
+
+ /*
+ * Must provide the virtual address of the current
+ * descriptor, the physical address, and the virtual
+ * address of the next descriptor in the h/w chain.
+ * This allows the HAL to look ahead to see if the
+ * hardware is done with a descriptor by checking the
+ * done bit in the following descriptor and the address
+ * of the current descriptor the DMA engine is working
+ * on. All this is necessary because of our use of
+ * a self-linked list to avoid rx overruns.
+ */
+ ret = ath9k_hw_rxprocdesc(ah, ds, rs, 0);
+ if (ret == -EINPROGRESS) {
+ struct ath_rx_status trs;
+ struct ath_buf *tbf;
+ struct ath_desc *tds;
+
+ memset(&trs, 0, sizeof(trs));
+ if (list_is_last(&bf->list, &sc->rx.rxbuf)) {
+ sc->rx.rxlink = NULL;
+ return NULL;
+ }
+
+ tbf = list_entry(bf->list.next, struct ath_buf, list);
+
+ /*
+ * On some hardware the descriptor status words could
+ * get corrupted, including the done bit. Because of
+ * this, check if the next descriptor's done bit is
+ * set or not.
+ *
+ * If the next descriptor's done bit is set, the current
+ * descriptor has been corrupted. Force s/w to discard
+ * this descriptor and continue...
+ */
+
+ tds = tbf->bf_desc;
+ ret = ath9k_hw_rxprocdesc(ah, tds, &trs, 0);
+ if (ret == -EINPROGRESS)
+ return NULL;
+ }
+
+ if (!bf->bf_mpdu)
+ return bf;
+
+ /*
+ * Synchronize the DMA transfer with CPU before
+ * 1. accessing the frame
+ * 2. requeueing the same buffer to h/w
+ */
+ dma_sync_single_for_device(sc->dev, bf->bf_buf_addr,
+ common->rx_bufsize,
+ DMA_FROM_DEVICE);
+
+ return bf;
+}
+
+
+int ath_rx_tasklet(struct ath_softc *sc, int flush, bool hp)
+{
+ struct ath_buf *bf;
struct sk_buff *skb = NULL, *requeue_skb;
struct ieee80211_rx_status *rxs;
struct ath_hw *ah = sc->sc_ah;
@@ -491,7 +833,16 @@
int retval;
bool decrypt_error = false;
struct ath_rx_status rs;
+ enum ath9k_rx_qtype qtype;
+ bool edma = !!(ah->caps.hw_caps & ATH9K_HW_CAP_EDMA);
+ int dma_type;
+ if (edma)
+ dma_type = DMA_FROM_DEVICE;
+ else
+ dma_type = DMA_BIDIRECTIONAL;
+
+ qtype = hp ? ATH9K_RX_QUEUE_HP : ATH9K_RX_QUEUE_LP;
spin_lock_bh(&sc->rx.rxbuflock);
do {
@@ -499,71 +850,19 @@
if ((sc->sc_flags & SC_OP_RXFLUSH) && (flush == 0))
break;
- if (list_empty(&sc->rx.rxbuf)) {
- sc->rx.rxlink = NULL;
- break;
- }
-
- bf = list_first_entry(&sc->rx.rxbuf, struct ath_buf, list);
- ds = bf->bf_desc;
-
- /*
- * Must provide the virtual address of the current
- * descriptor, the physical address, and the virtual
- * address of the next descriptor in the h/w chain.
- * This allows the HAL to look ahead to see if the
- * hardware is done with a descriptor by checking the
- * done bit in the following descriptor and the address
- * of the current descriptor the DMA engine is working
- * on. All this is necessary because of our use of
- * a self-linked list to avoid rx overruns.
- */
memset(&rs, 0, sizeof(rs));
- retval = ath9k_hw_rxprocdesc(ah, ds, &rs, 0);
- if (retval == -EINPROGRESS) {
- struct ath_rx_status trs;
- struct ath_buf *tbf;
- struct ath_desc *tds;
+ if (edma)
+ bf = ath_edma_get_next_rx_buf(sc, &rs, qtype);
+ else
+ bf = ath_get_next_rx_buf(sc, &rs);
- memset(&trs, 0, sizeof(trs));
- if (list_is_last(&bf->list, &sc->rx.rxbuf)) {
- sc->rx.rxlink = NULL;
- break;
- }
-
- tbf = list_entry(bf->list.next, struct ath_buf, list);
-
- /*
- * On some hardware the descriptor status words could
- * get corrupted, including the done bit. Because of
- * this, check if the next descriptor's done bit is
- * set or not.
- *
- * If the next descriptor's done bit is set, the current
- * descriptor has been corrupted. Force s/w to discard
- * this descriptor and continue...
- */
-
- tds = tbf->bf_desc;
- retval = ath9k_hw_rxprocdesc(ah, tds, &trs, 0);
- if (retval == -EINPROGRESS) {
- break;
- }
- }
+ if (!bf)
+ break;
skb = bf->bf_mpdu;
if (!skb)
continue;
- /*
- * Synchronize the DMA transfer with CPU before
- * 1. accessing the frame
- * 2. requeueing the same buffer to h/w
- */
- dma_sync_single_for_cpu(sc->dev, bf->bf_buf_addr,
- common->rx_bufsize,
- DMA_FROM_DEVICE);
-
hdr = (struct ieee80211_hdr *) skb->data;
rxs = IEEE80211_SKB_RXCB(skb);
@@ -597,9 +896,11 @@
/* Unmap the frame */
dma_unmap_single(sc->dev, bf->bf_buf_addr,
common->rx_bufsize,
- DMA_FROM_DEVICE);
+ dma_type);
- skb_put(skb, rs.rs_datalen);
+ skb_put(skb, rs.rs_datalen + ah->caps.rx_status_len);
+ if (ah->caps.rx_status_len)
+ skb_pull(skb, ah->caps.rx_status_len);
ath9k_cmn_rx_skb_postprocess(common, skb, &rs,
rxs, decrypt_error);
@@ -608,7 +909,7 @@
bf->bf_mpdu = requeue_skb;
bf->bf_buf_addr = dma_map_single(sc->dev, requeue_skb->data,
common->rx_bufsize,
- DMA_FROM_DEVICE);
+ dma_type);
if (unlikely(dma_mapping_error(sc->dev,
bf->bf_buf_addr))) {
dev_kfree_skb_any(requeue_skb);
@@ -639,12 +940,16 @@
ath_rx_send_to_mac80211(hw, sc, skb, rxs);
requeue:
- list_move_tail(&bf->list, &sc->rx.rxbuf);
- ath_rx_buf_link(sc, bf);
+ if (edma) {
+ list_add_tail(&bf->list, &sc->rx.rxbuf);
+ ath_rx_edma_buf_link(sc, qtype);
+ } else {
+ list_move_tail(&bf->list, &sc->rx.rxbuf);
+ ath_rx_buf_link(sc, bf);
+ }
} while (1);
spin_unlock_bh(&sc->rx.rxbuflock);
return 0;
-#undef PA2DESC
}