Commit ef17bd7c authored by Ioana Radulescu's avatar Ioana Radulescu Committed by David S. Miller

dpaa2-eth: Avoid unbounded while loops

Throughout the driver there are several places where we wait
indefinitely for DPIO portal commands to be executed, while
the portal returns a busy response code.

Even though in theory we are guaranteed the portals become
available eventually, in practice the QBMan hardware module
may become unresponsive in various corner cases.

Make sure we can never get stuck in an infinite while loop
by adding a retry counter for all portal commands.
Signed-off-by: default avatarIoana Radulescu <ruxandra.radulescu@nxp.com>
Signed-off-by: default avatarIoana Ciornei <ioana.ciornei@nxp.com>
Signed-off-by: default avatarDavid S. Miller <davem@davemloft.net>
parent 4b177f06
......@@ -221,6 +221,7 @@ static void xdp_release_buf(struct dpaa2_eth_priv *priv,
struct dpaa2_eth_channel *ch,
dma_addr_t addr)
{
int retries = 0;
int err;
ch->xdp.drop_bufs[ch->xdp.drop_cnt++] = addr;
......@@ -229,8 +230,11 @@ static void xdp_release_buf(struct dpaa2_eth_priv *priv,
while ((err = dpaa2_io_service_release(ch->dpio, priv->bpid,
ch->xdp.drop_bufs,
ch->xdp.drop_cnt)) == -EBUSY)
ch->xdp.drop_cnt)) == -EBUSY) {
if (retries++ >= DPAA2_ETH_SWP_BUSY_RETRIES)
break;
cpu_relax();
}
if (err) {
free_bufs(priv, ch->xdp.drop_bufs, ch->xdp.drop_cnt);
......@@ -458,7 +462,7 @@ static int consume_frames(struct dpaa2_eth_channel *ch,
struct dpaa2_eth_fq *fq = NULL;
struct dpaa2_dq *dq;
const struct dpaa2_fd *fd;
int cleaned = 0;
int cleaned = 0, retries = 0;
int is_last;
do {
......@@ -469,6 +473,11 @@ static int consume_frames(struct dpaa2_eth_channel *ch,
* the store until we get some sort of valid response
* token (either a valid frame or an "empty dequeue")
*/
if (retries++ >= DPAA2_ETH_SWP_BUSY_RETRIES) {
netdev_err_once(priv->net_dev,
"Unable to read a valid dequeue response\n");
return -ETIMEDOUT;
}
continue;
}
......@@ -477,6 +486,7 @@ static int consume_frames(struct dpaa2_eth_channel *ch,
fq->consume(priv, ch, fd, fq);
cleaned++;
retries = 0;
} while (!is_last);
if (!cleaned)
......@@ -949,6 +959,7 @@ static int add_bufs(struct dpaa2_eth_priv *priv,
u64 buf_array[DPAA2_ETH_BUFS_PER_CMD];
struct page *page;
dma_addr_t addr;
int retries = 0;
int i, err;
for (i = 0; i < DPAA2_ETH_BUFS_PER_CMD; i++) {
......@@ -980,8 +991,11 @@ static int add_bufs(struct dpaa2_eth_priv *priv,
release_bufs:
/* In case the portal is busy, retry until successful */
while ((err = dpaa2_io_service_release(ch->dpio, bpid,
buf_array, i)) == -EBUSY)
buf_array, i)) == -EBUSY) {
if (retries++ >= DPAA2_ETH_SWP_BUSY_RETRIES)
break;
cpu_relax();
}
/* If release command failed, clean up and bail out;
* not much else we can do about it
......@@ -1032,16 +1046,21 @@ static int seed_pool(struct dpaa2_eth_priv *priv, u16 bpid)
static void drain_bufs(struct dpaa2_eth_priv *priv, int count)
{
u64 buf_array[DPAA2_ETH_BUFS_PER_CMD];
int retries = 0;
int ret;
do {
ret = dpaa2_io_service_acquire(NULL, priv->bpid,
buf_array, count);
if (ret < 0) {
if (ret == -EBUSY &&
retries++ >= DPAA2_ETH_SWP_BUSY_RETRIES)
continue;
netdev_err(priv->net_dev, "dpaa2_io_service_acquire() failed\n");
return;
}
free_bufs(priv, buf_array, ret);
retries = 0;
} while (ret);
}
......@@ -1094,7 +1113,7 @@ static int pull_channel(struct dpaa2_eth_channel *ch)
ch->store);
dequeues++;
cpu_relax();
} while (err == -EBUSY);
} while (err == -EBUSY && dequeues < DPAA2_ETH_SWP_BUSY_RETRIES);
ch->stats.dequeue_portal_busy += dequeues;
if (unlikely(err))
......@@ -1118,6 +1137,7 @@ static int dpaa2_eth_poll(struct napi_struct *napi, int budget)
struct netdev_queue *nq;
int store_cleaned, work_done;
struct list_head rx_list;
int retries = 0;
int err;
ch = container_of(napi, struct dpaa2_eth_channel, napi);
......@@ -1136,7 +1156,7 @@ static int dpaa2_eth_poll(struct napi_struct *napi, int budget)
refill_pool(priv, ch, priv->bpid);
store_cleaned = consume_frames(ch, &fq);
if (!store_cleaned)
if (store_cleaned <= 0)
break;
if (fq->type == DPAA2_RX_FQ) {
rx_cleaned += store_cleaned;
......@@ -1163,7 +1183,7 @@ static int dpaa2_eth_poll(struct napi_struct *napi, int budget)
do {
err = dpaa2_io_service_rearm(ch->dpio, &ch->nctx);
cpu_relax();
} while (err == -EBUSY);
} while (err == -EBUSY && retries++ < DPAA2_ETH_SWP_BUSY_RETRIES);
WARN_ONCE(err, "CDAN notifications rearm failed on core %d",
ch->nctx.desired_cpu);
......
......@@ -245,6 +245,14 @@ static inline struct dpaa2_faead *dpaa2_get_faead(void *buf_addr, bool swa)
*/
#define DPAA2_ETH_ENQUEUE_RETRIES 10
/* Number of times to retry DPIO portal operations while waiting
* for portal to finish executing current command and become
* available. We want to avoid being stuck in a while loop in case
* hardware becomes unresponsive, but not give up too easily if
* the portal really is busy for valid reasons
*/
#define DPAA2_ETH_SWP_BUSY_RETRIES 1000
/* Driver statistics, other than those in struct rtnl_link_stats64.
* These are usually collected per-CPU and aggregated by ethtool.
*/
......
Markdown is supported
0%
or
You are about to add 0 people to the discussion. Proceed with caution.
Finish editing this message first!
Please register or to comment