| lists.openwall.net | lists / announce owl-users owl-dev john-users john-dev passwdqc-users yescrypt popa3d-users / oss-security kernel-hardening musl sabotage tlsify passwords / crypt-dev xvendor / Bugtraq Full-Disclosure linux-kernel linux-netdev linux-ext4 linux-hardening linux-cve-announce PHC | |
|
Open Source and information security mailing list archives
| ||
|
Message-Id: <20211223011358.4031459-39-davidm@egauge.net>
Date: Thu, 23 Dec 2021 01:14:07 +0000 (UTC)
From: David Mosberger-Tang <davidm@...uge.net>
To: Ajay Singh <ajay.kathat@...rochip.com>
Cc: Claudiu Beznea <claudiu.beznea@...rochip.com>,
Kalle Valo <kvalo@...nel.org>,
"David S. Miller" <davem@...emloft.net>,
Jakub Kicinski <kuba@...nel.org>,
linux-wireless@...r.kernel.org, netdev@...r.kernel.org,
linux-kernel@...r.kernel.org,
David Mosberger-Tang <davidm@...uge.net>
Subject: [PATCH v2 38/50] wilc1000: take advantage of chip queue
Rather than peeking at the access-category tx queues, move packets
scheduled for transmission onto the chip queue.
Signed-off-by: David Mosberger-Tang <davidm@...uge.net>
---
.../net/wireless/microchip/wilc1000/wlan.c | 102 +++++++++---------
1 file changed, 49 insertions(+), 53 deletions(-)
diff --git a/drivers/net/wireless/microchip/wilc1000/wlan.c b/drivers/net/wireless/microchip/wilc1000/wlan.c
index 86b945e5ee076..eefc0d18c1b5c 100644
--- a/drivers/net/wireless/microchip/wilc1000/wlan.c
+++ b/drivers/net/wireless/microchip/wilc1000/wlan.c
@@ -672,74 +672,79 @@ static void set_header(struct wilc *wilc, struct sk_buff *tqe,
}
/**
- * fill_vmm_table() - Fill VMM table with packets to be sent
+ * fill_vmm_table() - fill VMM table with packets to be sent
* @wilc: Pointer to the wilc structure.
* @vmm_table: Pointer to the VMM table to fill.
- * @vmm_entries_ac: Pointer to the queue-number table to fill.
- * For each packet added to the VMM table, this will be filled in
- * with the queue-number (access-category) that the packet is coming
- * from.
*
* Fill VMM table with packets waiting to be sent. The packets are
* added based on access category (priority) but also balanced to
* provide fairness.
*
- * Context: Since this function peeks at the packet queues, the
- * txq_add_to_head_cs mutex must be acquired before calling this
- * function.
- *
* Return:
* The number of VMM entries filled in. The table is 0-terminated
* so the returned number is at most WILC_VMM_TBL_SIZE-1.
*/
-static int fill_vmm_table(const struct wilc *wilc,
- u32 vmm_table[WILC_VMM_TBL_SIZE],
- u8 vmm_entries_ac[WILC_VMM_TBL_SIZE])
+static int fill_vmm_table(struct wilc *wilc,
+ u32 vmm_table[WILC_VMM_TBL_SIZE])
{
int i;
u8 k, ac;
- u32 sum;
static const u8 ac_preserve_ratio[NQUEUES] = {1, 1, 1, 1};
u8 ac_desired_ratio[NQUEUES];
const u8 *num_pkts_to_add;
bool ac_exist = 0;
int vmm_sz = 0;
- struct sk_buff *tqe_q[NQUEUES];
+ struct sk_buff *tqe;
struct wilc_skb_tx_cb *tx_cb;
- for (ac = 0; ac < NQUEUES; ac++)
- tqe_q[ac] = skb_peek(&wilc->txq[ac]);
-
i = 0;
- sum = 0;
+
+ if (unlikely(wilc->chipq_bytes > 0)) {
+ /* fill in packets that are already on the chipq: */
+ skb_queue_walk(&wilc->chipq, tqe) {
+ tx_cb = WILC_SKB_TX_CB(tqe);
+ vmm_sz = tx_hdr_len(tx_cb->type);
+ vmm_sz += tqe->len;
+ vmm_sz = ALIGN(vmm_sz, 4);
+ vmm_table[i++] = vmm_table_entry(tqe, vmm_sz);
+ }
+ }
ac_balance(wilc, ac_desired_ratio);
num_pkts_to_add = ac_desired_ratio;
do {
ac_exist = 0;
for (ac = 0; ac < NQUEUES; ac++) {
- if (!tqe_q[ac])
+ if (skb_queue_len(&wilc->txq[ac]) < 1)
continue;
ac_exist = 1;
- for (k = 0; k < num_pkts_to_add[ac] && tqe_q[ac]; k++) {
+ for (k = 0; k < num_pkts_to_add[ac]; k++) {
if (i >= WILC_VMM_TBL_SIZE - 1)
goto out;
- tx_cb = WILC_SKB_TX_CB(tqe_q[ac]);
+ tqe = skb_dequeue(&wilc->txq[ac]);
+ if (!tqe)
+ continue;
+
+ tx_cb = WILC_SKB_TX_CB(tqe);
vmm_sz = tx_hdr_len(tx_cb->type);
- vmm_sz += tqe_q[ac]->len;
+ vmm_sz += tqe->len;
vmm_sz = ALIGN(vmm_sz, 4);
- if (sum + vmm_sz > WILC_TX_BUFF_SIZE)
+ if (wilc->chipq_bytes + vmm_sz > WILC_TX_BUFF_SIZE) {
+ /* return packet to its queue */
+ skb_queue_head(&wilc->txq[ac], tqe);
goto out;
- vmm_table[i] = vmm_table_entry(tqe_q[ac], vmm_sz);
- vmm_entries_ac[i] = ac;
+ }
+ atomic_dec(&wilc->txq_entries);
+
+ __skb_queue_tail(&wilc->chipq, tqe);
+ wilc->chipq_bytes += tqe->len;
+ vmm_table[i] = vmm_table_entry(tqe, vmm_sz);
i++;
- sum += vmm_sz;
- tqe_q[ac] = skb_peek_next(tqe_q[ac],
- &wilc->txq[ac]);
+
}
}
num_pkts_to_add = ac_preserve_ratio;
@@ -837,14 +842,11 @@ static int send_vmm_table(struct wilc *wilc,
}
/**
- * copy_packets() - Copy packets to the transmit buffer
+ * copy_packets() - copy packets to the transmit buffer
* @wilc: Pointer to the wilc structure.
- * @entries: The number of packets to send from the VMM table.
- * @vmm_table: The VMM table to send.
- * @vmm_entries_ac: Table index i contains the number of the queue to
- * take the i-th packet from.
+ * @entries: The number of packets to copy from the chip queue.
*
- * Copy a set of packets to the transmit buffer.
+ * Copy a number of packets to the transmit buffer.
*
* Context: The txq_add_to_head_cs mutex must still be held when
* calling this function.
@@ -852,8 +854,7 @@ static int send_vmm_table(struct wilc *wilc,
* Return: Number of bytes copied to the transmit buffer (always
* non-negative).
*/
-static int copy_packets(struct wilc *wilc, int entries, u32 *vmm_table,
- u8 *vmm_entries_ac)
+static int copy_packets(struct wilc *wilc, int entries)
{
u8 ac_pkt_num_to_chip[NQUEUES] = {0, 0, 0, 0};
struct wilc_skb_tx_cb *tx_cb;
@@ -867,21 +868,19 @@ static int copy_packets(struct wilc *wilc, int entries, u32 *vmm_table,
struct sk_buff *tqe;
u32 buffer_offset;
- tqe = skb_dequeue(&wilc->txq[vmm_entries_ac[i]]);
- if (!tqe)
+ tqe = __skb_dequeue(&wilc->chipq);
+ if (WARN_ON(!tqe))
break;
+ wilc->chipq_bytes -= tqe->len;
- atomic_dec(&wilc->txq_entries);
- ac_pkt_num_to_chip[vmm_entries_ac[i]]++;
tx_cb = WILC_SKB_TX_CB(tqe);
- if (vmm_table[i] == 0)
- break;
-
- le32_to_cpus(&vmm_table[i]);
- vmm_sz = FIELD_GET(WILC_VMM_BUFFER_SIZE, vmm_table[i]);
- vmm_sz *= 4;
+ ac_pkt_num_to_chip[tx_cb->q_num]++;
buffer_offset = tx_hdr_len(tx_cb->type);
+ vmm_sz = buffer_offset;
+ vmm_sz += tqe->len;
+ vmm_sz = ALIGN(vmm_sz, 4);
+
set_header(wilc, tqe, vmm_sz, txb + offset);
memcpy(&txb[offset + buffer_offset], tqe->data, tqe->len);
offset += vmm_sz;
@@ -916,13 +915,11 @@ static int send_packets(struct wilc *wilc, int len)
return func->hif_block_tx_ext(wilc, 0, wilc->tx_buffer, len);
}
-static int copy_and_send_packets(struct wilc *wilc, int entries,
- u32 vmm_table[WILC_VMM_TBL_SIZE],
- u8 vmm_entries_ac[WILC_VMM_TBL_SIZE])
+static int copy_and_send_packets(struct wilc *wilc, int entries)
{
int len, ret;
- len = copy_packets(wilc, entries, vmm_table, vmm_entries_ac);
+ len = copy_packets(wilc, entries);
if (len <= 0)
return len;
@@ -935,7 +932,6 @@ static int copy_and_send_packets(struct wilc *wilc, int entries,
int wilc_wlan_handle_txq(struct wilc *wilc, u32 *txq_count)
{
int vmm_table_len, entries;
- u8 vmm_entries_ac[WILC_VMM_TBL_SIZE];
int ret = 0;
u32 vmm_table[WILC_VMM_TBL_SIZE];
int srcu_idx;
@@ -951,7 +947,7 @@ int wilc_wlan_handle_txq(struct wilc *wilc, u32 *txq_count)
wilc_wlan_txq_filter_dup_tcp_ack(vif->ndev);
srcu_read_unlock(&wilc->srcu, srcu_idx);
- vmm_table_len = fill_vmm_table(wilc, vmm_table, vmm_entries_ac);
+ vmm_table_len = fill_vmm_table(wilc, vmm_table);
if (vmm_table_len == 0)
goto out_unlock;
@@ -966,7 +962,7 @@ int wilc_wlan_handle_txq(struct wilc *wilc, u32 *txq_count)
if (entries <= 0) {
ret = entries;
} else {
- ret = copy_and_send_packets(wilc, entries, vmm_table, vmm_entries_ac);
+ ret = copy_and_send_packets(wilc, entries);
}
if (ret >= 0 && entries < vmm_table_len)
ret = WILC_VMM_ENTRY_FULL_RETRY;
--
2.25.1
Powered by blists - more mailing lists