[PATCH net-next 10/12] net: mvpp2: map the CPUs to threads

From: Antoine Tenart
Date: Wed Sep 19 2018 - 05:30:27 EST


This patch maps all uses of the CPU to threads. All this_cpu calls are
replaced, and all smp_processor_id() calls are wrapped into the
indirection.

Signed-off-by: Antoine Tenart <antoine.tenart@xxxxxxxxxxx>
---
drivers/net/ethernet/marvell/mvpp2/mvpp2.h | 5 +-
.../net/ethernet/marvell/mvpp2/mvpp2_main.c | 90 +++++++++++--------
2 files changed, 55 insertions(+), 40 deletions(-)

diff --git a/drivers/net/ethernet/marvell/mvpp2/mvpp2.h b/drivers/net/ethernet/marvell/mvpp2/mvpp2.h
index 87ac86a87bd5..0280856ff6ec 100644
--- a/drivers/net/ethernet/marvell/mvpp2/mvpp2.h
+++ b/drivers/net/ethernet/marvell/mvpp2/mvpp2.h
@@ -814,6 +814,9 @@ struct mvpp2_port {
void __iomem *base;
void __iomem *stats_base;

+ /* Number of threads used on the port */
+ unsigned int nthreads;
+
struct mvpp2_rx_queue **rxqs;
unsigned int nrxqs;
struct mvpp2_tx_queue **txqs;
@@ -971,7 +974,7 @@ struct mvpp2_txq_pcpu_buf {

/* Per-CPU Tx queue control */
struct mvpp2_txq_pcpu {
- unsigned int cpu;
+ unsigned int thread;

/* Number of Tx DMA descriptors in the descriptor ring */
int size;
diff --git a/drivers/net/ethernet/marvell/mvpp2/mvpp2_main.c b/drivers/net/ethernet/marvell/mvpp2/mvpp2_main.c
index 0c51bd4f9354..b46ffbca35af 100644
--- a/drivers/net/ethernet/marvell/mvpp2/mvpp2_main.c
+++ b/drivers/net/ethernet/marvell/mvpp2/mvpp2_main.c
@@ -1681,13 +1681,14 @@ static int mvpp2_txq_alloc_reserved_desc(struct mvpp2 *priv,
/* Check if there are enough reserved descriptors for transmission.
* If not, request chunk of reserved descriptors and check again.
*/
-static int mvpp2_txq_reserved_desc_num_proc(struct mvpp2 *priv,
+static int mvpp2_txq_reserved_desc_num_proc(struct mvpp2_port *port,
struct mvpp2_tx_queue *txq,
struct mvpp2_txq_pcpu *txq_pcpu,
int num)
{
+ struct mvpp2 *priv = port->priv;
int req, desc_count;
- unsigned int cpu;
+ unsigned int thread;

if (txq_pcpu->reserved_num >= num)
return 0;
@@ -1698,10 +1699,10 @@ static int mvpp2_txq_reserved_desc_num_proc(struct mvpp2 *priv,

desc_count = 0;
/* Compute total of used descriptors */
- for_each_present_cpu(cpu) {
+ for (thread = 0; thread < port->nthreads; thread++) {
struct mvpp2_txq_pcpu *txq_pcpu_aux;

- txq_pcpu_aux = per_cpu_ptr(txq->pcpu, cpu);
+ txq_pcpu_aux = per_cpu_ptr(txq->pcpu, thread);
desc_count += txq_pcpu_aux->count;
desc_count += txq_pcpu_aux->reserved_num;
}
@@ -1710,7 +1711,7 @@ static int mvpp2_txq_reserved_desc_num_proc(struct mvpp2 *priv,
desc_count += req;

if (desc_count >
- (txq->size - (num_present_cpus() * MVPP2_CPU_DESC_CHUNK)))
+ (txq->size - (MVPP2_MAX_THREADS * MVPP2_CPU_DESC_CHUNK)))
return -ENOMEM;

txq_pcpu->reserved_num += mvpp2_txq_alloc_reserved_desc(priv, txq, req);
@@ -1984,7 +1985,7 @@ static void mvpp2_txq_done(struct mvpp2_port *port, struct mvpp2_tx_queue *txq,
struct netdev_queue *nq = netdev_get_tx_queue(port->dev, txq->log_id);
int tx_done;

- if (txq_pcpu->cpu != smp_processor_id())
+ if (txq_pcpu->thread != mvpp2_cpu_to_thread(smp_processor_id()))
netdev_err(port->dev, "wrong cpu on the end of Tx processing\n");

tx_done = mvpp2_txq_sent_desc_proc(port, txq);
@@ -2168,7 +2169,7 @@ static int mvpp2_txq_init(struct mvpp2_port *port,
struct mvpp2_tx_queue *txq)
{
u32 val;
- unsigned int cpu, thread;
+ unsigned int thread;
int desc, desc_per_txq, tx_port_num;
struct mvpp2_txq_pcpu *txq_pcpu;

@@ -2225,8 +2226,8 @@ static int mvpp2_txq_init(struct mvpp2_port *port,
mvpp2_write(port->priv, MVPP2_TXQ_SCHED_TOKEN_SIZE_REG(txq->log_id),
val);

- for_each_present_cpu(cpu) {
- txq_pcpu = per_cpu_ptr(txq->pcpu, cpu);
+ for (thread = 0; thread < port->nthreads; thread++) {
+ txq_pcpu = per_cpu_ptr(txq->pcpu, thread);
txq_pcpu->size = txq->size;
txq_pcpu->buffs = kmalloc_array(txq_pcpu->size,
sizeof(*txq_pcpu->buffs),
@@ -2260,10 +2261,10 @@ static void mvpp2_txq_deinit(struct mvpp2_port *port,
struct mvpp2_tx_queue *txq)
{
struct mvpp2_txq_pcpu *txq_pcpu;
- unsigned int cpu, thread;
+ unsigned int thread;

- for_each_present_cpu(cpu) {
- txq_pcpu = per_cpu_ptr(txq->pcpu, cpu);
+ for (thread = 0; thread < port->nthreads; thread++) {
+ txq_pcpu = per_cpu_ptr(txq->pcpu, thread);
kfree(txq_pcpu->buffs);

if (txq_pcpu->tso_headers)
@@ -2301,7 +2302,7 @@ static void mvpp2_txq_clean(struct mvpp2_port *port, struct mvpp2_tx_queue *txq)
{
struct mvpp2_txq_pcpu *txq_pcpu;
int delay, pending;
- unsigned int cpu, thread = mvpp2_cpu_to_thread(get_cpu());
+ unsigned int thread = mvpp2_cpu_to_thread(get_cpu());
u32 val;

mvpp2_percpu_write(port->priv, thread, MVPP2_TXQ_NUM_REG, txq->id);
@@ -2332,8 +2333,8 @@ static void mvpp2_txq_clean(struct mvpp2_port *port, struct mvpp2_tx_queue *txq)
mvpp2_percpu_write(port->priv, thread, MVPP2_TXQ_PREF_BUF_REG, val);
put_cpu();

- for_each_present_cpu(cpu) {
- txq_pcpu = per_cpu_ptr(txq->pcpu, cpu);
+ for (thread = 0; thread < port->nthreads; thread++) {
+ txq_pcpu = per_cpu_ptr(txq->pcpu, thread);

/* Release all packets */
mvpp2_txq_bufs_free(port, txq, txq_pcpu, txq_pcpu->count);
@@ -2514,16 +2515,20 @@ static void mvpp2_tx_proc_cb(unsigned long data)
{
struct net_device *dev = (struct net_device *)data;
struct mvpp2_port *port = netdev_priv(dev);
- struct mvpp2_port_pcpu *port_pcpu = this_cpu_ptr(port->pcpu);
+ struct mvpp2_port_pcpu *port_pcpu;
unsigned int tx_todo, cause;

+ port_pcpu = per_cpu_ptr(port->pcpu,
+ mvpp2_cpu_to_thread(smp_processor_id()));
+
if (!netif_running(dev))
return;
port_pcpu->timer_scheduled = false;

/* Process all the Tx queues */
cause = (1 << port->ntxqs) - 1;
- tx_todo = mvpp2_tx_done(port, cause, smp_processor_id());
+ tx_todo = mvpp2_tx_done(port, cause,
+ mvpp2_cpu_to_thread(smp_processor_id()));

/* Set the timer in case not all the packets were processed */
if (tx_todo)
@@ -2739,7 +2744,8 @@ static inline void
tx_desc_unmap_put(struct mvpp2_port *port, struct mvpp2_tx_queue *txq,
struct mvpp2_tx_desc *desc)
{
- struct mvpp2_txq_pcpu *txq_pcpu = this_cpu_ptr(txq->pcpu);
+ unsigned int thread = mvpp2_cpu_to_thread(smp_processor_id());
+ struct mvpp2_txq_pcpu *txq_pcpu = per_cpu_ptr(txq->pcpu, thread);

dma_addr_t buf_dma_addr =
mvpp2_txdesc_dma_addr_get(port, desc);
@@ -2756,7 +2762,8 @@ static int mvpp2_tx_frag_process(struct mvpp2_port *port, struct sk_buff *skb,
struct mvpp2_tx_queue *aggr_txq,
struct mvpp2_tx_queue *txq)
{
- struct mvpp2_txq_pcpu *txq_pcpu = this_cpu_ptr(txq->pcpu);
+ unsigned int thread = mvpp2_cpu_to_thread(smp_processor_id());
+ struct mvpp2_txq_pcpu *txq_pcpu = per_cpu_ptr(txq->pcpu, thread);
struct mvpp2_tx_desc *tx_desc;
int i;
dma_addr_t buf_dma_addr;
@@ -2877,7 +2884,7 @@ static int mvpp2_tx_tso(struct sk_buff *skb, struct net_device *dev,
/* Check number of available descriptors */
if (mvpp2_aggr_desc_num_check(port->priv, aggr_txq,
tso_count_descs(skb)) ||
- mvpp2_txq_reserved_desc_num_proc(port->priv, txq, txq_pcpu,
+ mvpp2_txq_reserved_desc_num_proc(port, txq, txq_pcpu,
tso_count_descs(skb)))
return 0;

@@ -2924,14 +2931,17 @@ static int mvpp2_tx(struct sk_buff *skb, struct net_device *dev)
struct mvpp2_txq_pcpu *txq_pcpu;
struct mvpp2_tx_desc *tx_desc;
dma_addr_t buf_dma_addr;
+ unsigned int thread;
int frags = 0;
u16 txq_id;
u32 tx_cmd;

+ thread = mvpp2_cpu_to_thread(smp_processor_id());
+
txq_id = skb_get_queue_mapping(skb);
txq = port->txqs[txq_id];
- txq_pcpu = this_cpu_ptr(txq->pcpu);
- aggr_txq = &port->priv->aggr_txqs[smp_processor_id()];
+ txq_pcpu = per_cpu_ptr(txq->pcpu, thread);
+ aggr_txq = &port->priv->aggr_txqs[thread];

if (skb_is_gso(skb)) {
frags = mvpp2_tx_tso(skb, dev, txq, aggr_txq, txq_pcpu);
@@ -2941,8 +2951,7 @@ static int mvpp2_tx(struct sk_buff *skb, struct net_device *dev)

/* Check number of available descriptors */
if (mvpp2_aggr_desc_num_check(port->priv, aggr_txq, frags) ||
- mvpp2_txq_reserved_desc_num_proc(port->priv, txq,
- txq_pcpu, frags)) {
+ mvpp2_txq_reserved_desc_num_proc(port, txq, txq_pcpu, frags)) {
frags = 0;
goto out;
}
@@ -2984,7 +2993,7 @@ static int mvpp2_tx(struct sk_buff *skb, struct net_device *dev)

out:
if (frags > 0) {
- struct mvpp2_pcpu_stats *stats = this_cpu_ptr(port->stats);
+ struct mvpp2_pcpu_stats *stats = per_cpu_ptr(port->stats, thread);
struct netdev_queue *nq = netdev_get_tx_queue(dev, txq_id);

txq_pcpu->reserved_num -= frags;
@@ -3014,7 +3023,7 @@ static int mvpp2_tx(struct sk_buff *skb, struct net_device *dev)
/* Set the timer in case not all frags were processed */
if (!port->has_tx_irqs && txq_pcpu->count <= frags &&
txq_pcpu->count > 0) {
- struct mvpp2_port_pcpu *port_pcpu = this_cpu_ptr(port->pcpu);
+ struct mvpp2_port_pcpu *port_pcpu = per_cpu_ptr(port->pcpu, thread);

mvpp2_timer_set(port_pcpu);
}
@@ -3407,7 +3416,7 @@ static int mvpp2_stop(struct net_device *dev)
{
struct mvpp2_port *port = netdev_priv(dev);
struct mvpp2_port_pcpu *port_pcpu;
- unsigned int cpu;
+ unsigned int thread;

mvpp2_stop_dev(port);

@@ -3422,8 +3431,8 @@ static int mvpp2_stop(struct net_device *dev)

mvpp2_irqs_deinit(port);
if (!port->has_tx_irqs) {
- for_each_present_cpu(cpu) {
- port_pcpu = per_cpu_ptr(port->pcpu, cpu);
+ for (thread = 0; thread < port->nthreads; thread++) {
+ port_pcpu = per_cpu_ptr(port->pcpu, thread);

hrtimer_cancel(&port_pcpu->tx_done_timer);
port_pcpu->timer_scheduled = false;
@@ -4104,7 +4113,7 @@ static int mvpp2_port_init(struct mvpp2_port *port)
struct device *dev = port->dev->dev.parent;
struct mvpp2 *priv = port->priv;
struct mvpp2_txq_pcpu *txq_pcpu;
- unsigned int cpu;
+ unsigned int thread;
int queue, err;

/* Checks for hardware constraints */
@@ -4149,9 +4158,9 @@ static int mvpp2_port_init(struct mvpp2_port *port)
txq->id = queue_phy_id;
txq->log_id = queue;
txq->done_pkts_coal = MVPP2_TXDONE_COAL_PKTS_THRESH;
- for_each_present_cpu(cpu) {
- txq_pcpu = per_cpu_ptr(txq->pcpu, cpu);
- txq_pcpu->cpu = cpu;
+ for (thread = 0; thread < port->nthreads; thread++) {
+ txq_pcpu = per_cpu_ptr(txq->pcpu, thread);
+ txq_pcpu->thread = thread;
}

port->txqs[queue] = txq;
@@ -4645,7 +4654,7 @@ static int mvpp2_port_probe(struct platform_device *pdev,
struct resource *res;
struct phylink *phylink;
char *mac_from = "";
- unsigned int ntxqs, nrxqs, cpu;
+ unsigned int ntxqs, nrxqs, thread;
unsigned long flags = 0;
bool has_tx_irqs;
u32 id;
@@ -4709,6 +4718,9 @@ static int mvpp2_port_probe(struct platform_device *pdev,
port->has_tx_irqs = has_tx_irqs;
port->flags = flags;

+ port->nthreads = min_t(unsigned int, num_present_cpus(),
+ MVPP2_MAX_THREADS);
+
err = mvpp2_queue_vectors_init(port, port_node);
if (err)
goto err_free_netdev;
@@ -4804,8 +4816,8 @@ static int mvpp2_port_probe(struct platform_device *pdev,
}

if (!port->has_tx_irqs) {
- for_each_present_cpu(cpu) {
- port_pcpu = per_cpu_ptr(port->pcpu, cpu);
+ for (thread = 0; thread < port->nthreads; thread++) {
+ port_pcpu = per_cpu_ptr(port->pcpu, thread);

hrtimer_init(&port_pcpu->tx_done_timer, CLOCK_MONOTONIC,
HRTIMER_MODE_REL_PINNED);
@@ -5089,13 +5101,13 @@ static int mvpp2_init(struct platform_device *pdev, struct mvpp2 *priv)
}

/* Allocate and initialize aggregated TXQs */
- priv->aggr_txqs = devm_kcalloc(&pdev->dev, num_present_cpus(),
+ priv->aggr_txqs = devm_kcalloc(&pdev->dev, MVPP2_MAX_THREADS,
sizeof(*priv->aggr_txqs),
GFP_KERNEL);
if (!priv->aggr_txqs)
return -ENOMEM;

- for_each_present_cpu(i) {
+ for (i = 0; i < MVPP2_MAX_THREADS; i++) {
priv->aggr_txqs[i].id = i;
priv->aggr_txqs[i].size = MVPP2_AGGR_TXQ_SIZE;
err = mvpp2_aggr_txq_init(pdev, &priv->aggr_txqs[i], i, priv);
@@ -5390,7 +5402,7 @@ static int mvpp2_remove(struct platform_device *pdev)
mvpp2_bm_pool_destroy(pdev, priv, bm_pool);
}

- for_each_present_cpu(i) {
+ for (i = 0; i < MVPP2_MAX_THREADS; i++) {
struct mvpp2_tx_queue *aggr_txq = &priv->aggr_txqs[i];

dma_free_coherent(&pdev->dev,
--
2.17.1