[permalink] [raw]

Subject: [PATCHv3 2/2] add mt7601u kbuild and others

From: Jakub Kicinski <[email protected]>

Signed-off-by: Jakub Kicinski <[email protected]>
---
MAINTAINERS | 6 ++++++
drivers/net/wireless/Kconfig | 1 +
drivers/net/wireless/Makefile | 2 ++
drivers/net/wireless/mediatek/Kconfig | 10 ++++++++++
drivers/net/wireless/mediatek/Makefile | 1 +
drivers/net/wireless/mediatek/mt7601u/Kconfig | 6 ++++++
drivers/net/wireless/mediatek/mt7601u/Makefile | 9 +++++++++
7 files changed, 35 insertions(+)
create mode 100644 drivers/net/wireless/mediatek/Kconfig
create mode 100644 drivers/net/wireless/mediatek/Makefile
create mode 100644 drivers/net/wireless/mediatek/mt7601u/Kconfig
create mode 100644 drivers/net/wireless/mediatek/mt7601u/Makefile

diff --git a/MAINTAINERS b/MAINTAINERS
index df106f87a3ba..c6f4576efd56 100644
--- a/MAINTAINERS
+++ b/MAINTAINERS
@@ -6365,6 +6365,12 @@ F: include/uapi/linux/meye.h
F: include/uapi/linux/ivtv*
F: include/uapi/linux/uvcvideo.h

+MEDIATEK MT7601U WIRELESS LAN DRIVER
+M: Jakub Kicinski <[email protected]>
+L: [email protected]
+S: Maintained
+F: drivers/net/wireless/mediatek/mt7601u/
+
MEGARAID SCSI/SAS DRIVERS
M: Kashyap Desai <[email protected]>
M: Sumit Saxena <[email protected]>
diff --git a/drivers/net/wireless/Kconfig b/drivers/net/wireless/Kconfig
index 16604bdf5197..a63ab2e83105 100644
--- a/drivers/net/wireless/Kconfig
+++ b/drivers/net/wireless/Kconfig
@@ -277,6 +277,7 @@ source "drivers/net/wireless/libertas/Kconfig"
source "drivers/net/wireless/orinoco/Kconfig"
source "drivers/net/wireless/p54/Kconfig"
source "drivers/net/wireless/rt2x00/Kconfig"
+source "drivers/net/wireless/mediatek/Kconfig"
source "drivers/net/wireless/rtlwifi/Kconfig"
source "drivers/net/wireless/ti/Kconfig"
source "drivers/net/wireless/zd1211rw/Kconfig"
diff --git a/drivers/net/wireless/Makefile b/drivers/net/wireless/Makefile
index 0c8891686718..6b9e729dd8ac 100644
--- a/drivers/net/wireless/Makefile
+++ b/drivers/net/wireless/Makefile
@@ -45,6 +45,8 @@ obj-$(CONFIG_IWLWIFI) += iwlwifi/
obj-$(CONFIG_IWLEGACY) += iwlegacy/
obj-$(CONFIG_RT2X00) += rt2x00/

+obj-$(CONFIG_WL_MEDIATEK) += mediatek/
+
obj-$(CONFIG_P54_COMMON) += p54/

obj-$(CONFIG_ATH_CARDS) += ath/
diff --git a/drivers/net/wireless/mediatek/Kconfig b/drivers/net/wireless/mediatek/Kconfig
new file mode 100644
index 000000000000..cba300c6b5da
--- /dev/null
+++ b/drivers/net/wireless/mediatek/Kconfig
@@ -0,0 +1,10 @@
+menuconfig WL_MEDIATEK
+ bool "Mediatek Wireless LAN support"
+ ---help---
+ Enable community drivers for MediaTek WiFi devices.
+ Those drivers make use of the Linux mac80211 stack.
+
+
+if WL_MEDIATEK
+source "drivers/net/wireless/mediatek/mt7601u/Kconfig"
+endif # WL_MEDIATEK
diff --git a/drivers/net/wireless/mediatek/Makefile b/drivers/net/wireless/mediatek/Makefile
new file mode 100644
index 000000000000..9d5f182fd7fd
--- /dev/null
+++ b/drivers/net/wireless/mediatek/Makefile
@@ -0,0 +1 @@
+obj-$(CONFIG_MT7601U) += mt7601u/
diff --git a/drivers/net/wireless/mediatek/mt7601u/Kconfig b/drivers/net/wireless/mediatek/mt7601u/Kconfig
new file mode 100644
index 000000000000..f46bed92796b
--- /dev/null
+++ b/drivers/net/wireless/mediatek/mt7601u/Kconfig
@@ -0,0 +1,6 @@
+config MT7601U
+ tristate "MediaTek MT7601U (USB) support"
+ depends on MAC80211
+ depends on USB
+ ---help---
+ This adds support for MT7601U-based wireless USB dongles.
diff --git a/drivers/net/wireless/mediatek/mt7601u/Makefile b/drivers/net/wireless/mediatek/mt7601u/Makefile
new file mode 100644
index 000000000000..ea9ed8a5db4d
--- /dev/null
+++ b/drivers/net/wireless/mediatek/mt7601u/Makefile
@@ -0,0 +1,9 @@
+ccflags-y += -D__CHECK_ENDIAN__
+
+obj-$(CONFIG_MT7601U) += mt7601u.o
+
+mt7601u-objs = \
+ usb.o init.o main.o mcu.o trace.o dma.o core.o eeprom.o phy.o \
+ mac.o util.o debugfs.o tx.o
+
+CFLAGS_trace.o := -I$(src)
--
2.1.0

2015-05-05 00:49:54

by Jakub Kicinski

[permalink] [raw]

Subject: Re: [PATCH 1/2] add mt7601u driver

On Mon, 04 May 2015 12:15:11 +0200, Johannes Berg wrote:
> On Mon, 2015-05-04 at 12:04 +0200, Jakub Kiciński wrote:
>
> > > Don't know how your buffers are set up, but if the DMA engine consumes
> > > pages you could consider using paged RX instead of the memcpy().
> >
> > DMA engine can concatenate multiple frames into a single USB bulk
> > transfer to a large continuous buffer. There is no way to request
> > any alignment of the frames within that large buffer so I think paged
> > RX is not an option.
>
> You could probably still do it because the networking stack only
> requires the *headers* to be aligned. But if they headers aren't in the
> skb->data (skb->head) then they'll be pulled into that by the network
> stack, where they'll be aligned.

I *think* I got it. Would you care to take a look? I basically
allocate a compound page with dev_alloc_pages() and run get_page() for
every fragment.

------>8-----------------------------------

diff --git a/dma.c b/dma.c
index e865cf8384a5..a92993a64aeb 100644
--- a/dma.c
+++ b/dma.c
@@ -16,6 +16,9 @@
#include "usb.h"
#include "trace.h"

+static int mt7601u_submit_rx_buf(struct mt7601u_dev *dev,
+ struct mt7601u_dma_buf_rx *e, gfp_t gfp);
+
static unsigned int ieee80211_get_hdrlen_from_buf(const u8 *data, unsigned len)
{
const struct ieee80211_hdr *hdr = (const struct ieee80211_hdr *)data;
@@ -60,7 +63,7 @@ mt7601u_rx_process_seg(struct mt7601u_dev *dev, u8 *data, u32 seg_len)
{
struct sk_buff *skb;
struct mt7601u_rxwi *rxwi;
- u32 fce_info;
+ u32 true_len, fce_info;

/* DMA_INFO field at the beginning of the segment contains only some of
* the information, we need to read the FCE descriptor from the end.
@@ -86,11 +89,81 @@ mt7601u_rx_process_seg(struct mt7601u_dev *dev, u8 *data, u32 seg_len)
if (!skb)
return;

- if (mt76_mac_process_rx(dev, skb, rxwi)) {
- dev_kfree_skb(skb);
- return;
+ true_len = mt76_mac_process_rx(dev, skb, skb->data, rxwi);
+ skb_trim(skb, true_len);
+
+ ieee80211_rx_ni(dev->hw, skb);
+}
+
+static void mt7601u_rx_add_frags(struct mt7601u_dev *dev, struct sk_buff *skb,
+ void *data, u32 true_len, u32 truesize,
+ struct page *p)
+{
+ unsigned long addr = (unsigned long) data;
+ u32 overpage = 0;
+ u32 p_off = (data - page_address(p)) >> PAGE_SHIFT;
+
+ if ((addr & ~PAGE_MASK) + true_len > PAGE_SIZE)
+ overpage = (addr & ~PAGE_MASK) + true_len - PAGE_SIZE;
+
+ skb_add_rx_frag(skb, skb_shinfo(skb)->nr_frags, p + p_off,
+ addr & ~PAGE_MASK, true_len - overpage, truesize);
+ get_page(p + p_off);
+
+ /* Do we really need to split the buffer if it crosses a page boundary?
+ * Does networking code care?
+ */
+ if (overpage) {
+ skb_add_rx_frag(skb, skb_shinfo(skb)->nr_frags, p + p_off + 1,
+ 0, overpage, 0);
+ get_page(p + p_off + 1);
+ }
+}
+
+static void mt7601u_rx_process_seg_paged(struct mt7601u_dev *dev, u8 *data,
+ u32 seg_len, struct page *p)
+{
+ struct sk_buff *skb;
+ struct mt7601u_rxwi *rxwi;
+ u32 true_len, fce_info, truesize = seg_len;
+
+ /* DMA_INFO field at the beginning of the segment contains only some of
+ * the information, we need to read the FCE descriptor from the end.
+ */
+ fce_info = get_unaligned_le32(data + seg_len - MT_FCE_INFO_LEN);
+ seg_len -= MT_FCE_INFO_LEN;
+
+ data += MT_DMA_HDR_LEN;
+ seg_len -= MT_DMA_HDR_LEN;
+
+ rxwi = (struct mt7601u_rxwi *) data;
+ data += sizeof(struct mt7601u_rxwi);
+ seg_len -= sizeof(struct mt7601u_rxwi);
+
+ if (unlikely(rxwi->zero[0] || rxwi->zero[1] || rxwi->zero[2]))
+ dev_err_once(dev->dev, "Error: RXWI zero fields are set\n");
+ if (unlikely(MT76_GET(MT_RXD_INFO_TYPE, fce_info)))
+ dev_err_once(dev->dev, "Error: RX path seen a non-pkt urb\n");
+
+ trace_mt_rx(dev, rxwi, fce_info);
+
+ if (rxwi->rxinfo & cpu_to_le32(MT_RXINFO_L2PAD)) {
+ int hdr_len = ieee80211_get_hdrlen_from_buf(data, seg_len);
+
+ memmove(data + 2, data, hdr_len);
+ seg_len -= 2;
+ data += 2;
}

+ skb = __netdev_alloc_skb(NULL, 256, GFP_ATOMIC);
+ if (!skb)
+ return;
+
+ true_len = mt76_mac_process_rx(dev, skb, data, rxwi);
+ WARN_ON(true_len > seg_len || true_len < seg_len - 3);
+
+ mt7601u_rx_add_frags(dev, skb, data, true_len, truesize, p);
+
ieee80211_rx_ni(dev->hw, skb);
}

@@ -110,17 +183,31 @@ static u16 mt7601u_rx_next_seg_len(u8 *data, u32 data_len)
}

static void
-mt7601u_rx_process_entry(struct mt7601u_dev *dev, struct mt7601u_dma_buf *e)
+mt7601u_rx_process_entry(struct mt7601u_dev *dev, struct mt7601u_dma_buf_rx *e)
{
u32 seg_len, data_len = e->urb->actual_length;
- u8 *data = e->buf;
+ u8 *data = page_address(e->p);
+ struct page *new_p = NULL;
+ bool paged = true;
int cnt = 0;

if (!test_bit(MT7601U_STATE_INITIALIZED, &dev->state))
return;

+ /* Copy if there is very little data in the buffer. */
+ if (data_len < 512) {
+ paged = false;
+ } else {
+ new_p = dev_alloc_pages(MT_RX_ORDER);
+ if (!new_p)
+ paged = false;
+ }
+
while ((seg_len = mt7601u_rx_next_seg_len(data, data_len))) {
- mt7601u_rx_process_seg(dev, data, seg_len);
+ if (paged)
+ mt7601u_rx_process_seg_paged(dev, data, seg_len, e->p);
+ else
+ mt7601u_rx_process_seg(dev, data, seg_len);

data_len -= seg_len;
data += seg_len;
@@ -128,14 +215,21 @@ mt7601u_rx_process_entry(struct mt7601u_dev *dev, struct mt7601u_dma_buf *e)
}

if (cnt > 1)
- trace_mt_rx_dma_aggr(dev, cnt);
+ trace_mt_rx_dma_aggr(dev, cnt, paged);
+
+ if (paged) {
+ /* we have one extra ref from the allcator */
+ put_page(e->p);
+
+ e->p = new_p;
+ }
}

-static struct mt7601u_dma_buf *
+static struct mt7601u_dma_buf_rx *
mt7601u_rx_get_pending_entry(struct mt7601u_dev *dev)
{
struct mt7601u_rx_queue *q = &dev->rx_q;
- struct mt7601u_dma_buf *buf = NULL;
+ struct mt7601u_dma_buf_rx *buf = NULL;
unsigned long flags;

spin_lock_irqsave(&dev->rx_lock, flags);
@@ -175,15 +269,14 @@ out:
static void mt7601u_rx_tasklet(unsigned long data)
{
struct mt7601u_dev *dev = (struct mt7601u_dev *) data;
- struct mt7601u_dma_buf *e;
+ struct mt7601u_dma_buf_rx *e;

while ((e = mt7601u_rx_get_pending_entry(dev))) {
if (e->urb->status)
continue;

mt7601u_rx_process_entry(dev, e);
- mt7601u_usb_submit_buf(dev, USB_DIR_IN, MT_EP_IN_PKT_RX, e,
- GFP_ATOMIC, mt7601u_complete_rx, dev);
+ mt7601u_submit_rx_buf(dev, e, GFP_ATOMIC);
}
}

@@ -325,14 +418,33 @@ static void mt7601u_kill_rx(struct mt7601u_dev *dev)
spin_unlock_irqrestore(&dev->rx_lock, flags);
}

+static int mt7601u_submit_rx_buf(struct mt7601u_dev *dev,
+ struct mt7601u_dma_buf_rx *e, gfp_t gfp)
+{
+ struct usb_device *usb_dev = mt7601u_to_usb_dev(dev);
+ u8 *buf = page_address(e->p);
+ unsigned pipe;
+ int ret;
+
+ pipe = usb_rcvbulkpipe(usb_dev, dev->in_eps[MT_EP_IN_PKT_RX]);
+
+ usb_fill_bulk_urb(e->urb, usb_dev, pipe, buf, MT_RX_URB_SIZE,
+ mt7601u_complete_rx, dev);
+
+ trace_mt_submit_urb(dev, e->urb);
+ ret = usb_submit_urb(e->urb, gfp);
+ if (ret)
+ dev_err(dev->dev, "Error: submit RX URB failed:%d\n", ret);
+
+ return ret;
+}
+
static int mt7601u_submit_rx(struct mt7601u_dev *dev)
{
int i, ret;

for (i = 0; i < dev->rx_q.entries; i++) {
- ret = mt7601u_usb_submit_buf(dev, USB_DIR_IN, MT_EP_IN_PKT_RX,
- &dev->rx_q.e[i], GFP_KERNEL,
- mt7601u_complete_rx, dev);
+ ret = mt7601u_submit_rx_buf(dev, &dev->rx_q.e[i], GFP_KERNEL);
if (ret)
return ret;
}
@@ -344,8 +456,10 @@ static void mt7601u_free_rx(struct mt7601u_dev *dev)
{
int i;

- for (i = 0; i < dev->rx_q.entries; i++)
- mt7601u_usb_free_buf(dev, &dev->rx_q.e[i]);
+ for (i = 0; i < dev->rx_q.entries; i++) {
+ __free_pages(dev->rx_q.e[i].p, MT_RX_ORDER);
+ usb_free_urb(dev->rx_q.e[i].urb);
+ }
}

static int mt7601u_alloc_rx(struct mt7601u_dev *dev)
@@ -356,9 +470,13 @@ static int mt7601u_alloc_rx(struct mt7601u_dev *dev)
dev->rx_q.dev = dev;
dev->rx_q.entries = N_RX_ENTRIES;

- for (i = 0; i < N_RX_ENTRIES; i++)
- if (mt7601u_usb_alloc_buf(dev, MT_RX_URB_SIZE, &dev->rx_q.e[i]))
+ for (i = 0; i < N_RX_ENTRIES; i++) {
+ dev->rx_q.e[i].urb = usb_alloc_urb(0, GFP_KERNEL);
+ dev->rx_q.e[i].p = dev_alloc_pages(MT_RX_ORDER);
+
+ if (!dev->rx_q.e[i].urb || !dev->rx_q.e[i].p)
return -ENOMEM;
+ }

return 0;
}
diff --git a/mac.c b/mac.c
index 539751362b41..8802366f7181 100644
--- a/mac.c
+++ b/mac.c
@@ -441,30 +441,28 @@ mt7601u_rx_monitor_beacon(struct mt7601u_dev *dev, struct mt7601u_rxwi *rxwi,
}

static int
-mt7601u_rx_is_our_beacon(struct mt7601u_dev *dev, struct sk_buff *skb)
+mt7601u_rx_is_our_beacon(struct mt7601u_dev *dev, u8 *data)
{
- struct ieee80211_hdr *hdr = (struct ieee80211_hdr *)skb->data;
+ struct ieee80211_hdr *hdr = (struct ieee80211_hdr *)data;

return ieee80211_is_beacon(hdr->frame_control) &&
ether_addr_equal(hdr->addr2, dev->ap_bssid);
}

-int mt76_mac_process_rx(struct mt7601u_dev *dev, struct sk_buff *skb, void *rxi)
+u32 mt76_mac_process_rx(struct mt7601u_dev *dev, struct sk_buff *skb,
+ u8 *data, void *rxi)
{
struct ieee80211_rx_status *status = IEEE80211_SKB_RXCB(skb);
struct mt7601u_rxwi *rxwi = rxi;
u32 ctl = le32_to_cpu(rxwi->ctl);
u16 rate = le16_to_cpu(rxwi->rate);
- int len, rssi;
+ int rssi;

if (rxwi->rxinfo & cpu_to_le32(MT_RXINFO_DECRYPT)) {
status->flag |= RX_FLAG_DECRYPTED;
status->flag |= RX_FLAG_IV_STRIPPED | RX_FLAG_MMIC_STRIPPED;
}

- len = MT76_GET(MT_RXWI_CTL_MPDU_LEN, ctl);
- skb_trim(skb, len);
-
status->chains = BIT(0);
rssi = mt7601u_phy_get_rssi(dev, rxwi, rate);
status->chain_signal[0] = status->signal = rssi;
@@ -474,13 +472,13 @@ int mt76_mac_process_rx(struct mt7601u_dev *dev, struct sk_buff *skb, void *rxi)
mt76_mac_process_rate(status, rate);

spin_lock_bh(&dev->con_mon_lock);
- if (mt7601u_rx_is_our_beacon(dev, skb))
+ if (mt7601u_rx_is_our_beacon(dev, data))
mt7601u_rx_monitor_beacon(dev, rxwi, rate, rssi);
else if (rxwi->rxinfo & cpu_to_le32(MT_RXINFO_U2M))
dev->avg_rssi = (dev->avg_rssi * 15) / 16 + (rssi << 8);
spin_unlock_bh(&dev->con_mon_lock);

- return 0;
+ return MT76_GET(MT_RXWI_CTL_MPDU_LEN, ctl);
}

static enum mt76_cipher_type
diff --git a/mac.h b/mac.h
index 90cf147f37e7..2c22d63c63a2 100644
--- a/mac.h
+++ b/mac.h
@@ -160,8 +160,8 @@ struct mt76_txwi {
#define MT_TXWI_CTL_CHAN_CHECK_PKT BIT(4)
#define MT_TXWI_CTL_PIFS_REV BIT(6)

-int
-mt76_mac_process_rx(struct mt7601u_dev *dev, struct sk_buff *skb, void *rxwi);
+u32 mt76_mac_process_rx(struct mt7601u_dev *dev, struct sk_buff *skb,
+ u8 *data, void *rxi);
int mt76_mac_wcid_set_key(struct mt7601u_dev *dev, u8 idx,
struct ieee80211_key_conf *key);
void mt76_mac_wcid_set_rate(struct mt7601u_dev *dev, struct mt76_wcid *wcid,
diff --git a/mt7601u.h b/mt7601u.h
index f9957350a187..d6f83e03cb58 100644
--- a/mt7601u.h
+++ b/mt7601u.h
@@ -37,6 +37,7 @@
#define MT_USB_AGGR_SIZE_LIMIT 21 /* * 1024B */
#define MT_USB_AGGR_TIMEOUT 0x80 /* * 33ns */
#define MT_RX_URB_SIZE (24 * 1024)
+#define MT_RX_ORDER 3

struct mt7601u_dma_buf {
struct urb *urb;
@@ -73,7 +74,10 @@ struct mac_stats {
struct mt7601u_rx_queue {
struct mt7601u_dev *dev;

- struct mt7601u_dma_buf e[N_RX_ENTRIES];
+ struct mt7601u_dma_buf_rx {
+ struct urb *urb;
+ struct page *p;
+ } e[N_RX_ENTRIES];

unsigned int start;
unsigned int end;
diff --git a/trace.h b/trace.h
index db86272401ff..289897300ef0 100644
--- a/trace.h
+++ b/trace.h
@@ -352,17 +352,20 @@ TRACE_EVENT(mt_tx_status,
);

TRACE_EVENT(mt_rx_dma_aggr,
- TP_PROTO(struct mt7601u_dev *dev, int cnt),
- TP_ARGS(dev, cnt),
+ TP_PROTO(struct mt7601u_dev *dev, int cnt, bool paged),
+ TP_ARGS(dev, cnt, paged),
TP_STRUCT__entry(
DEV_ENTRY
__field(u8, cnt)
+ __field(bool, paged)
),
TP_fast_assign(
DEV_ASSIGN;
__entry->cnt = cnt;
+ __entry->paged = paged;
),
- TP_printk(DEV_PR_FMT "%d", DEV_PR_ARG, __entry->cnt)
+ TP_printk(DEV_PR_FMT "cnt:%d paged:%d",
+ DEV_PR_ARG, __entry->cnt, __entry->paged)
);

DEFINE_EVENT(dev_simple_evt, set_key,

2015-05-05 10:44:22

by Jakub Kicinski

[permalink] [raw]

Subject: Re: [PATCH 1/2] add mt7601u driver

On Tue, 05 May 2015 08:45:42 +0200, Johannes Berg wrote:
> On Tue, 2015-05-05 at 02:49 +0200, Jakub Kiciński wrote:
> > On Mon, 04 May 2015 12:15:11 +0200, Johannes Berg wrote:
> > > On Mon, 2015-05-04 at 12:04 +0200, Jakub Kiciński wrote:
> > >
> > > > > Don't know how your buffers are set up, but if the DMA engine consumes
> > > > > pages you could consider using paged RX instead of the memcpy().
> > > >
> > > > DMA engine can concatenate multiple frames into a single USB bulk
> > > > transfer to a large continuous buffer. There is no way to request
> > > > any alignment of the frames within that large buffer so I think paged
> > > > RX is not an option.
> > >
> > > You could probably still do it because the networking stack only
> > > requires the *headers* to be aligned. But if they headers aren't in the
> > > skb->data (skb->head) then they'll be pulled into that by the network
> > > stack, where they'll be aligned.
> >
> > I *think* I got it. Would you care to take a look? I basically
> > allocate a compound page with dev_alloc_pages() and run get_page() for
> > every fragment.
>
> I think it looks fine - except the compound part. I'm pretty sure you
> need to do references always with the pointer to the first page (and
> consequently don't need to split on the page boundary, and use the same
> page pointer just with a bigger offset)

I wasn't sure about the splitting and when I split and took the
reference on head page -> mapcount got corrupted. But you are right,
no splitting and reference on head works fine.

> We have a comment on this code (from Eric) saying
>
> /* Dont use dev_alloc_skb(), we'll have enough headroom once
> * ieee80211_hdr pulled.
>
> which probably applies here as well. We also just use 128 instead of 256
> and haven't seen a need for more.
>
> We also pull the entire frame only if it fits into those 128 bytes, and
> we preload the 802.11 header + 8 bytes for snap and possibly crypto
> headroom so we can do it all at once instead of later in multiple
> places. You can check out iwl_mvm_pass_packet_to_mac80211() to see the
> details.

Done.

Thanks a lot for your comments, here is the improved version. I
will give it some more testing and send v2 of the entire driver.

------------>8--------------------------------

diff --git a/dma.c b/dma.c
index e865cf8384a5..4510b3a1357b 100644
--- a/dma.c
+++ b/dma.c
@@ -16,6 +16,9 @@
#include "usb.h"
#include "trace.h"

+static int mt7601u_submit_rx_buf(struct mt7601u_dev *dev,
+ struct mt7601u_dma_buf_rx *e, gfp_t gfp);
+
static unsigned int ieee80211_get_hdrlen_from_buf(const u8 *data, unsigned len)
{
const struct ieee80211_hdr *hdr = (const struct ieee80211_hdr *)data;
@@ -34,6 +37,7 @@ mt7601u_rx_skb_from_seg(struct mt7601u_dev *dev, struct mt7601u_rxwi *rxwi,
u8 *data, u32 seg_len)
{
struct sk_buff *skb;
+ u32 true_len;

if (rxwi->rxinfo & cpu_to_le32(MT_RXINFO_L2PAD))
seg_len -= 2;
@@ -52,15 +56,55 @@ mt7601u_rx_skb_from_seg(struct mt7601u_dev *dev, struct mt7601u_rxwi *rxwi,

memcpy(skb_put(skb, seg_len), data, seg_len);

+ true_len = mt76_mac_process_rx(dev, skb, skb->data, rxwi);
+ skb_trim(skb, true_len);
+
return skb;
}

-static void
-mt7601u_rx_process_seg(struct mt7601u_dev *dev, u8 *data, u32 seg_len)
+static struct sk_buff *
+mt7601u_rx_skb_from_seg_paged(struct mt7601u_dev *dev,
+ struct mt7601u_rxwi *rxwi, void *data,
+ u32 seg_len, u32 truesize, struct page *p)
+{
+ unsigned int hdr_len = ieee80211_get_hdrlen_from_buf(data, seg_len);
+ unsigned int true_len, copy, frag;
+ struct sk_buff *skb;
+
+ skb = alloc_skb(128, GFP_ATOMIC);
+ if (!skb)
+ return NULL;
+
+ true_len = mt76_mac_process_rx(dev, skb, data, rxwi);
+
+ if (rxwi->rxinfo & cpu_to_le32(MT_RXINFO_L2PAD)) {
+ memcpy(skb_put(skb, hdr_len), data, hdr_len);
+ data += hdr_len + 2;
+ true_len -= hdr_len;
+ hdr_len = 0;
+ }
+
+ copy = (true_len <= skb_tailroom(skb)) ? true_len : hdr_len + 8;
+ frag = true_len - copy;
+
+ memcpy(skb_put(skb, copy), data, copy);
+ data += copy;
+
+ if (frag) {
+ skb_add_rx_frag(skb, 0, p, data - page_address(p),
+ frag, truesize);
+ get_page(p);
+ }
+
+ return skb;
+}
+
+static void mt7601u_rx_process_seg(struct mt7601u_dev *dev, u8 *data,
+ u32 seg_len, struct page *p, bool paged)
{
struct sk_buff *skb;
struct mt7601u_rxwi *rxwi;
- u32 fce_info;
+ u32 fce_info, truesize = seg_len;

/* DMA_INFO field at the beginning of the segment contains only some of
* the information, we need to read the FCE descriptor from the end.
@@ -82,15 +126,14 @@ mt7601u_rx_process_seg(struct mt7601u_dev *dev, u8 *data, u32 seg_len)

trace_mt_rx(dev, rxwi, fce_info);

- skb = mt7601u_rx_skb_from_seg(dev, rxwi, data, seg_len);
+ if (paged)
+ skb = mt7601u_rx_skb_from_seg_paged(dev, rxwi, data, seg_len,
+ truesize, p);
+ else
+ skb = mt7601u_rx_skb_from_seg(dev, rxwi, data, seg_len);
if (!skb)
return;

- if (mt76_mac_process_rx(dev, skb, rxwi)) {
- dev_kfree_skb(skb);
- return;
- }
-
ieee80211_rx_ni(dev->hw, skb);
}

@@ -110,17 +153,28 @@ static u16 mt7601u_rx_next_seg_len(u8 *data, u32 data_len)
}

static void
-mt7601u_rx_process_entry(struct mt7601u_dev *dev, struct mt7601u_dma_buf *e)
+mt7601u_rx_process_entry(struct mt7601u_dev *dev, struct mt7601u_dma_buf_rx *e)
{
u32 seg_len, data_len = e->urb->actual_length;
- u8 *data = e->buf;
+ u8 *data = page_address(e->p);
+ struct page *new_p = NULL;
+ bool paged = true;
int cnt = 0;

if (!test_bit(MT7601U_STATE_INITIALIZED, &dev->state))
return;

+ /* Copy if there is very little data in the buffer. */
+ if (data_len < 512) {
+ paged = false;
+ } else {
+ new_p = dev_alloc_pages(MT_RX_ORDER);
+ if (!new_p)
+ paged = false;
+ }
+
while ((seg_len = mt7601u_rx_next_seg_len(data, data_len))) {
- mt7601u_rx_process_seg(dev, data, seg_len);
+ mt7601u_rx_process_seg(dev, data, seg_len, e->p, paged);

data_len -= seg_len;
data += seg_len;
@@ -128,14 +182,21 @@ mt7601u_rx_process_entry(struct mt7601u_dev *dev, struct mt7601u_dma_buf *e)
}

if (cnt > 1)
- trace_mt_rx_dma_aggr(dev, cnt);
+ trace_mt_rx_dma_aggr(dev, cnt, paged);
+
+ if (paged) {
+ /* we have one extra ref from the allocator */
+ __free_pages(e->p, MT_RX_ORDER);
+
+ e->p = new_p;
+ }
}

-static struct mt7601u_dma_buf *
+static struct mt7601u_dma_buf_rx *
mt7601u_rx_get_pending_entry(struct mt7601u_dev *dev)
{
struct mt7601u_rx_queue *q = &dev->rx_q;
- struct mt7601u_dma_buf *buf = NULL;
+ struct mt7601u_dma_buf_rx *buf = NULL;
unsigned long flags;

spin_lock_irqsave(&dev->rx_lock, flags);
@@ -175,15 +236,14 @@ out:
static void mt7601u_rx_tasklet(unsigned long data)
{
struct mt7601u_dev *dev = (struct mt7601u_dev *) data;
- struct mt7601u_dma_buf *e;
+ struct mt7601u_dma_buf_rx *e;

while ((e = mt7601u_rx_get_pending_entry(dev))) {
if (e->urb->status)
continue;

mt7601u_rx_process_entry(dev, e);
- mt7601u_usb_submit_buf(dev, USB_DIR_IN, MT_EP_IN_PKT_RX, e,
- GFP_ATOMIC, mt7601u_complete_rx, dev);
+ mt7601u_submit_rx_buf(dev, e, GFP_ATOMIC);
}
}

@@ -325,14 +385,33 @@ static void mt7601u_kill_rx(struct mt7601u_dev *dev)
spin_unlock_irqrestore(&dev->rx_lock, flags);
}

+static int mt7601u_submit_rx_buf(struct mt7601u_dev *dev,
+ struct mt7601u_dma_buf_rx *e, gfp_t gfp)
+{
+ struct usb_device *usb_dev = mt7601u_to_usb_dev(dev);
+ u8 *buf = page_address(e->p);
+ unsigned pipe;
+ int ret;
+
+ pipe = usb_rcvbulkpipe(usb_dev, dev->in_eps[MT_EP_IN_PKT_RX]);
+
+ usb_fill_bulk_urb(e->urb, usb_dev, pipe, buf, MT_RX_URB_SIZE,
+ mt7601u_complete_rx, dev);
+
+ trace_mt_submit_urb(dev, e->urb);
+ ret = usb_submit_urb(e->urb, gfp);
+ if (ret)
+ dev_err(dev->dev, "Error: submit RX URB failed:%d\n", ret);
+
+ return ret;
+}
+
static int mt7601u_submit_rx(struct mt7601u_dev *dev)
{
int i, ret;

for (i = 0; i < dev->rx_q.entries; i++) {
- ret = mt7601u_usb_submit_buf(dev, USB_DIR_IN, MT_EP_IN_PKT_RX,
- &dev->rx_q.e[i], GFP_KERNEL,
- mt7601u_complete_rx, dev);
+ ret = mt7601u_submit_rx_buf(dev, &dev->rx_q.e[i], GFP_KERNEL);
if (ret)
return ret;
}
@@ -344,8 +423,10 @@ static void mt7601u_free_rx(struct mt7601u_dev *dev)
{
int i;

- for (i = 0; i < dev->rx_q.entries; i++)
- mt7601u_usb_free_buf(dev, &dev->rx_q.e[i]);
+ for (i = 0; i < dev->rx_q.entries; i++) {
+ __free_pages(dev->rx_q.e[i].p, MT_RX_ORDER);
+ usb_free_urb(dev->rx_q.e[i].urb);
+ }
}

static int mt7601u_alloc_rx(struct mt7601u_dev *dev)
@@ -356,9 +437,13 @@ static int mt7601u_alloc_rx(struct mt7601u_dev *dev)
dev->rx_q.dev = dev;
dev->rx_q.entries = N_RX_ENTRIES;

- for (i = 0; i < N_RX_ENTRIES; i++)
- if (mt7601u_usb_alloc_buf(dev, MT_RX_URB_SIZE, &dev->rx_q.e[i]))
+ for (i = 0; i < N_RX_ENTRIES; i++) {
+ dev->rx_q.e[i].urb = usb_alloc_urb(0, GFP_KERNEL);
+ dev->rx_q.e[i].p = dev_alloc_pages(MT_RX_ORDER);
+
+ if (!dev->rx_q.e[i].urb || !dev->rx_q.e[i].p)
return -ENOMEM;
+ }

return 0;
}
diff --git a/mac.c b/mac.c
index 539751362b41..8802366f7181 100644
--- a/mac.c
+++ b/mac.c
@@ -441,30 +441,28 @@ mt7601u_rx_monitor_beacon(struct mt7601u_dev *dev, struct mt7601u_rxwi *rxwi,
}

static int
-mt7601u_rx_is_our_beacon(struct mt7601u_dev *dev, struct sk_buff *skb)
+mt7601u_rx_is_our_beacon(struct mt7601u_dev *dev, u8 *data)
{
- struct ieee80211_hdr *hdr = (struct ieee80211_hdr *)skb->data;
+ struct ieee80211_hdr *hdr = (struct ieee80211_hdr *)data;

return ieee80211_is_beacon(hdr->frame_control) &&
ether_addr_equal(hdr->addr2, dev->ap_bssid);
}

-int mt76_mac_process_rx(struct mt7601u_dev *dev, struct sk_buff *skb, void *rxi)
+u32 mt76_mac_process_rx(struct mt7601u_dev *dev, struct sk_buff *skb,
+ u8 *data, void *rxi)
{
struct ieee80211_rx_status *status = IEEE80211_SKB_RXCB(skb);
struct mt7601u_rxwi *rxwi = rxi;
u32 ctl = le32_to_cpu(rxwi->ctl);
u16 rate = le16_to_cpu(rxwi->rate);
- int len, rssi;
+ int rssi;

if (rxwi->rxinfo & cpu_to_le32(MT_RXINFO_DECRYPT)) {
status->flag |= RX_FLAG_DECRYPTED;
status->flag |= RX_FLAG_IV_STRIPPED | RX_FLAG_MMIC_STRIPPED;
}

- len = MT76_GET(MT_RXWI_CTL_MPDU_LEN, ctl);
- skb_trim(skb, len);
-
status->chains = BIT(0);
rssi = mt7601u_phy_get_rssi(dev, rxwi, rate);
status->chain_signal[0] = status->signal = rssi;
@@ -474,13 +472,13 @@ int mt76_mac_process_rx(struct mt7601u_dev *dev, struct sk_buff *skb, void *rxi)
mt76_mac_process_rate(status, rate);

spin_lock_bh(&dev->con_mon_lock);
- if (mt7601u_rx_is_our_beacon(dev, skb))
+ if (mt7601u_rx_is_our_beacon(dev, data))
mt7601u_rx_monitor_beacon(dev, rxwi, rate, rssi);
else if (rxwi->rxinfo & cpu_to_le32(MT_RXINFO_U2M))
dev->avg_rssi = (dev->avg_rssi * 15) / 16 + (rssi << 8);
spin_unlock_bh(&dev->con_mon_lock);

- return 0;
+ return MT76_GET(MT_RXWI_CTL_MPDU_LEN, ctl);
}

static enum mt76_cipher_type
diff --git a/mac.h b/mac.h
index 90cf147f37e7..2c22d63c63a2 100644
--- a/mac.h
+++ b/mac.h
@@ -160,8 +160,8 @@ struct mt76_txwi {
#define MT_TXWI_CTL_CHAN_CHECK_PKT BIT(4)
#define MT_TXWI_CTL_PIFS_REV BIT(6)

-int
-mt76_mac_process_rx(struct mt7601u_dev *dev, struct sk_buff *skb, void *rxwi);
+u32 mt76_mac_process_rx(struct mt7601u_dev *dev, struct sk_buff *skb,
+ u8 *data, void *rxi);
int mt76_mac_wcid_set_key(struct mt7601u_dev *dev, u8 idx,
struct ieee80211_key_conf *key);
void mt76_mac_wcid_set_rate(struct mt7601u_dev *dev, struct mt76_wcid *wcid,
diff --git a/mt7601u.h b/mt7601u.h
index f9957350a187..d6f83e03cb58 100644
--- a/mt7601u.h
+++ b/mt7601u.h
@@ -37,6 +37,7 @@
#define MT_USB_AGGR_SIZE_LIMIT 21 /* * 1024B */
#define MT_USB_AGGR_TIMEOUT 0x80 /* * 33ns */
#define MT_RX_URB_SIZE (24 * 1024)
+#define MT_RX_ORDER 3

struct mt7601u_dma_buf {
struct urb *urb;
@@ -73,7 +74,10 @@ struct mac_stats {
struct mt7601u_rx_queue {
struct mt7601u_dev *dev;

- struct mt7601u_dma_buf e[N_RX_ENTRIES];
+ struct mt7601u_dma_buf_rx {
+ struct urb *urb;
+ struct page *p;
+ } e[N_RX_ENTRIES];

unsigned int start;
unsigned int end;
diff --git a/trace.h b/trace.h
index db86272401ff..289897300ef0 100644
--- a/trace.h
+++ b/trace.h
@@ -352,17 +352,20 @@ TRACE_EVENT(mt_tx_status,
);

TRACE_EVENT(mt_rx_dma_aggr,
- TP_PROTO(struct mt7601u_dev *dev, int cnt),
- TP_ARGS(dev, cnt),
+ TP_PROTO(struct mt7601u_dev *dev, int cnt, bool paged),
+ TP_ARGS(dev, cnt, paged),
TP_STRUCT__entry(
DEV_ENTRY
__field(u8, cnt)
+ __field(bool, paged)
),
TP_fast_assign(
DEV_ASSIGN;
__entry->cnt = cnt;
+ __entry->paged = paged;
),
- TP_printk(DEV_PR_FMT "%d", DEV_PR_ARG, __entry->cnt)
+ TP_printk(DEV_PR_FMT "cnt:%d paged:%d",
+ DEV_PR_ARG, __entry->cnt, __entry->paged)
);

DEFINE_EVENT(dev_simple_evt, set_key,