@@ -182,7 +182,7 @@ int qed_hw_prepare(struct qed_dev *cdev,
void qed_hw_remove(struct qed_dev *cdev);
/**
- * qed_ptt_acquire(): Allocate a PTT window.
+ * qed_ptt_acquire(): Allocate a PTT window in sleepable context.
*
* @p_hwfn: HW device data.
*
@@ -193,6 +193,16 @@ void qed_hw_remove(struct qed_dev *cdev);
*/
struct qed_ptt *qed_ptt_acquire(struct qed_hwfn *p_hwfn);
+/**
+ * _qed_ptt_acquire(): Allocate a PTT window based on the context.
+ *
+ * @p_hwfn: HW device data.
+ * @is_atomic: context (sleepable or unsleepable) based on which ptt is acquired.
+ *
+ * Return: struct qed_ptt.
+ */
+struct qed_ptt *_qed_ptt_acquire(struct qed_hwfn *p_hwfn, bool is_atomic);
+
/**
* qed_ptt_release(): Release PTT Window.
*
@@ -23,7 +23,10 @@
#include "qed_reg_addr.h"
#include "qed_sriov.h"
-#define QED_BAR_ACQUIRE_TIMEOUT 1000
+#define QED_BAR_ACQUIRE_TIMEOUT_USLEEP_CNT 1000
+#define QED_BAR_ACQUIRE_TIMEOUT_USLEEP 1000
+#define QED_BAR_ACQUIRE_TIMEOUT_UDELAY_CNT 100000
+#define QED_BAR_ACQUIRE_TIMEOUT_UDELAY 10
/* Invalid values */
#define QED_BAR_INVALID_OFFSET (cpu_to_le32(-1))
@@ -83,13 +86,18 @@ void qed_ptt_pool_free(struct qed_hwfn *p_hwfn)
p_hwfn->p_ptt_pool = NULL;
}
-struct qed_ptt *qed_ptt_acquire(struct qed_hwfn *p_hwfn)
+struct qed_ptt *_qed_ptt_acquire(struct qed_hwfn *p_hwfn, bool is_atomic)
{
struct qed_ptt *p_ptt;
- unsigned int i;
+ unsigned int i, count;
+
+ if (is_atomic)
+ count = QED_BAR_ACQUIRE_TIMEOUT_UDELAY_CNT;
+ else
+ count = QED_BAR_ACQUIRE_TIMEOUT_USLEEP_CNT;
/* Take the free PTT from the list */
- for (i = 0; i < QED_BAR_ACQUIRE_TIMEOUT; i++) {
+ for (i = 0; i < count; i++) {
spin_lock_bh(&p_hwfn->p_ptt_pool->lock);
if (!list_empty(&p_hwfn->p_ptt_pool->free_list)) {
@@ -105,13 +113,23 @@ struct qed_ptt *qed_ptt_acquire(struct qed_hwfn *p_hwfn)
}
spin_unlock_bh(&p_hwfn->p_ptt_pool->lock);
- usleep_range(1000, 2000);
+
+ if (is_atomic)
+ udelay(QED_BAR_ACQUIRE_TIMEOUT_UDELAY);
+ else
+ usleep_range(QED_BAR_ACQUIRE_TIMEOUT_USLEEP,
+ QED_BAR_ACQUIRE_TIMEOUT_USLEEP * 2);
}
DP_NOTICE(p_hwfn, "PTT acquire timeout - failed to allocate PTT\n");
return NULL;
}
+struct qed_ptt *qed_ptt_acquire(struct qed_hwfn *p_hwfn)
+{
+ return _qed_ptt_acquire(p_hwfn, false);
+}
+
void qed_ptt_release(struct qed_hwfn *p_hwfn, struct qed_ptt *p_ptt)
{
spin_lock_bh(&p_hwfn->p_ptt_pool->lock);
@@ -1863,7 +1863,7 @@ static void __qed_get_vport_stats(struct qed_hwfn *p_hwfn,
}
static void _qed_get_vport_stats(struct qed_dev *cdev,
- struct qed_eth_stats *stats)
+ struct qed_eth_stats *stats, bool is_atomic)
{
u8 fw_vport = 0;
int i;
@@ -1872,7 +1872,7 @@ static void _qed_get_vport_stats(struct qed_dev *cdev,
for_each_hwfn(cdev, i) {
struct qed_hwfn *p_hwfn = &cdev->hwfns[i];
- struct qed_ptt *p_ptt = IS_PF(cdev) ? qed_ptt_acquire(p_hwfn)
+ struct qed_ptt *p_ptt = IS_PF(cdev) ? _qed_ptt_acquire(p_hwfn, is_atomic)
: NULL;
bool b_get_port_stats;
@@ -1899,7 +1899,8 @@ static void _qed_get_vport_stats(struct qed_dev *cdev,
}
}
-void qed_get_vport_stats(struct qed_dev *cdev, struct qed_eth_stats *stats)
+void qed_get_vport_stats(struct qed_dev *cdev, struct qed_eth_stats *stats,
+ bool is_atomic)
{
u32 i;
@@ -1908,7 +1909,7 @@ void qed_get_vport_stats(struct qed_dev *cdev, struct qed_eth_stats *stats)
return;
}
- _qed_get_vport_stats(cdev, stats);
+ _qed_get_vport_stats(cdev, stats, is_atomic);
if (!cdev->reset_stats)
return;
@@ -1960,7 +1961,7 @@ void qed_reset_vport_stats(struct qed_dev *cdev)
if (!cdev->reset_stats) {
DP_INFO(cdev, "Reset stats not allocated\n");
} else {
- _qed_get_vport_stats(cdev, cdev->reset_stats);
+ _qed_get_vport_stats(cdev, cdev->reset_stats, false);
cdev->reset_stats->common.link_change_count = 0;
}
}
@@ -249,7 +249,8 @@ qed_sp_eth_rx_queues_update(struct qed_hwfn *p_hwfn,
enum spq_mode comp_mode,
struct qed_spq_comp_cb *p_comp_data);
-void qed_get_vport_stats(struct qed_dev *cdev, struct qed_eth_stats *stats);
+void qed_get_vport_stats(struct qed_dev *cdev,
+ struct qed_eth_stats *stats, bool is_atomic);
void qed_reset_vport_stats(struct qed_dev *cdev);
@@ -3101,7 +3101,7 @@ void qed_get_protocol_stats(struct qed_dev *cdev,
switch (type) {
case QED_MCP_LAN_STATS:
- qed_get_vport_stats(cdev, ð_stats);
+ qed_get_vport_stats(cdev, ð_stats, false);
stats->lan_stats.ucast_rx_pkts =
eth_stats.common.rx_ucast_pkts;
stats->lan_stats.ucast_tx_pkts =
@@ -3161,7 +3161,7 @@ qed_fill_generic_tlv_data(struct qed_dev *cdev, struct qed_mfw_tlv_generic *tlv)
}
}
- qed_get_vport_stats(cdev, &stats);
+ qed_get_vport_stats(cdev, &stats, false);
p_common = &stats.common;
tlv->rx_frames = p_common->rx_ucast_pkts + p_common->rx_mcast_pkts +
p_common->rx_bcast_pkts;
@@ -569,7 +569,7 @@ void qede_set_udp_tunnels(struct qede_dev *edev);
void qede_reload(struct qede_dev *edev,
struct qede_reload_args *args, bool is_locked);
int qede_change_mtu(struct net_device *dev, int new_mtu);
-void qede_fill_by_demand_stats(struct qede_dev *edev);
+void qede_fill_by_demand_stats(struct qede_dev *edev, bool is_atomic);
void __qede_lock(struct qede_dev *edev);
void __qede_unlock(struct qede_dev *edev);
bool qede_has_rx_work(struct qede_rx_queue *rxq);
@@ -408,7 +408,7 @@ static void qede_get_ethtool_stats(struct net_device *dev,
struct qede_fastpath *fp;
int i;
- qede_fill_by_demand_stats(edev);
+ qede_fill_by_demand_stats(edev, false);
/* Need to protect the access to the fastpath array */
__qede_lock(edev);
@@ -301,12 +301,12 @@ module_exit(qede_cleanup);
static int qede_open(struct net_device *ndev);
static int qede_close(struct net_device *ndev);
-void qede_fill_by_demand_stats(struct qede_dev *edev)
+void qede_fill_by_demand_stats(struct qede_dev *edev, bool is_atomic)
{
struct qede_stats_common *p_common = &edev->stats.common;
struct qed_eth_stats stats;
- edev->ops->get_vport_stats(edev->cdev, &stats);
+ edev->ops->get_vport_stats(edev->cdev, &stats, is_atomic);
p_common->no_buff_discards = stats.common.no_buff_discards;
p_common->packet_too_big_discard = stats.common.packet_too_big_discard;
@@ -413,7 +413,7 @@ static void qede_get_stats64(struct net_device *dev,
struct qede_dev *edev = netdev_priv(dev);
struct qede_stats_common *p_common;
- qede_fill_by_demand_stats(edev);
+ qede_fill_by_demand_stats(edev, true);
p_common = &edev->stats.common;
stats->rx_packets = p_common->rx_ucast_pkts + p_common->rx_mcast_pkts +
@@ -319,7 +319,7 @@ struct qed_eth_ops {
struct eth_slow_path_rx_cqe *cqe);
void (*get_vport_stats)(struct qed_dev *cdev,
- struct qed_eth_stats *stats);
+ struct qed_eth_stats *stats, bool is_atomic);
int (*tunn_config)(struct qed_dev *cdev,
struct qed_tunn_params *params);
Bonding module collects the statistics while holding the spinlock, beneath that qede->qed driver statistics flow gets scheduled out due to usleep_range() used in PTT acquire logic which results into below bug and traces - [ 3673.988874] Hardware name: HPE ProLiant DL365 Gen10 Plus/ProLiant DL365 Gen10 Plus, BIOS A42 10/29/2021 [ 3673.988878] Call Trace: [ 3673.988891] dump_stack_lvl+0x34/0x44 [ 3673.988908] __schedule_bug.cold+0x47/0x53 [ 3673.988918] __schedule+0x3fb/0x560 [ 3673.988929] schedule+0x43/0xb0 [ 3673.988932] schedule_hrtimeout_range_clock+0xbf/0x1b0 [ 3673.988937] ? __hrtimer_init+0xc0/0xc0 [ 3673.988950] usleep_range+0x5e/0x80 [ 3673.988955] qed_ptt_acquire+0x2b/0xd0 [qed] [ 3673.988981] _qed_get_vport_stats+0x141/0x240 [qed] [ 3673.989001] qed_get_vport_stats+0x18/0x80 [qed] [ 3673.989016] qede_fill_by_demand_stats+0x37/0x400 [qede] [ 3673.989028] qede_get_stats64+0x19/0xe0 [qede] [ 3673.989034] dev_get_stats+0x5c/0xc0 [ 3673.989045] netstat_show.constprop.0+0x52/0xb0 [ 3673.989055] dev_attr_show+0x19/0x40 [ 3673.989065] sysfs_kf_seq_show+0x9b/0xf0 [ 3673.989076] seq_read_iter+0x120/0x4b0 [ 3673.989087] new_sync_read+0x118/0x1a0 [ 3673.989095] vfs_read+0xf3/0x180 [ 3673.989099] ksys_read+0x5f/0xe0 [ 3673.989102] do_syscall_64+0x3b/0x90 [ 3673.989109] entry_SYSCALL_64_after_hwframe+0x44/0xae [ 3673.989115] RIP: 0033:0x7f8467d0b082 [ 3673.989119] Code: c0 e9 b2 fe ff ff 50 48 8d 3d ca 05 08 00 e8 35 e7 01 00 0f 1f 44 00 00 f3 0f 1e fa 64 8b 04 25 18 00 00 00 85 c0 75 10 0f 05 <48> 3d 00 f0 ff ff 77 56 c3 0f 1f 44 00 00 48 83 ec 28 48 89 54 24 [ 3673.989121] RSP: 002b:00007ffffb21fd08 EFLAGS: 00000246 ORIG_RAX: 0000000000000000 [ 3673.989127] RAX: ffffffffffffffda RBX: 000000000100eca0 RCX: 00007f8467d0b082 [ 3673.989128] RDX: 00000000000003ff RSI: 00007ffffb21fdc0 RDI: 0000000000000003 [ 3673.989130] RBP: 00007f8467b96028 R08: 0000000000000010 R09: 00007ffffb21ec00 [ 3673.989132] R10: 00007ffffb27b170 R11: 0000000000000246 R12: 00000000000000f0 [ 3673.989134] R13: 0000000000000003 R14: 00007f8467b92000 R15: 0000000000045a05 [ 3673.989139] CPU: 30 PID: 285188 Comm: read_all Kdump: loaded Tainted: G W OE Fix this by having caller (QEDE driver flows) to provide the context whether it could be in atomic context flow or not when getting the vport stats from QED driver. QED driver based on the context provided decide to schedule out or not when acquiring the PTT BAR window. Fixes: 133fac0eedc3 ("qede: Add basic ethtool support") Cc: Sudarsana Kalluru <skalluru@marvell.com> Cc: David S. Miller <davem@davemloft.net> Signed-off-by: Manish Chopra <manishc@marvell.com> Signed-off-by: Ariel Elior <aelior@marvell.com> Signed-off-by: Alok Prasad <palok@marvell.com> --- v1->v2: - Fixed checkpatch and kdoc warnings. v2->v3: - Moving the changelog after tags. --- drivers/net/ethernet/qlogic/qed/qed_dev_api.h | 12 +++++++- drivers/net/ethernet/qlogic/qed/qed_hw.c | 28 +++++++++++++++---- drivers/net/ethernet/qlogic/qed/qed_l2.c | 11 ++++---- drivers/net/ethernet/qlogic/qed/qed_l2.h | 3 +- drivers/net/ethernet/qlogic/qed/qed_main.c | 4 +-- drivers/net/ethernet/qlogic/qede/qede.h | 2 +- .../net/ethernet/qlogic/qede/qede_ethtool.c | 2 +- drivers/net/ethernet/qlogic/qede/qede_main.c | 6 ++-- include/linux/qed/qed_eth_if.h | 2 +- 9 files changed, 50 insertions(+), 20 deletions(-)