edma_read_append_stats() gets called from two places in the driver.
The first place is the kernel timer that periodically updates
the statistics, so nothing gets lost due to overflows.
The second one it's part of the userspace ethtool ioctl handler
to provide up-to-date values.
For this configuration, the use of spin_lock() is not sufficient
and as per:
<https://mirrors.edge.kernel.org/pub/linux/kernel/people/rusty/kernel-locking/c214.html>
the locking has to be upgraded to spin_lock_bh().
Signed-off-by: Masafumi UTSUGI <mutsugi@allied-telesis.co.jp>
[folded patch into 710-, rewrote message]
Signed-off-by: Christian Lamparter <chunkeey@gmail.com>
+ int i;
+ u32 stat;
+
-+ spin_lock(&edma_cinfo->stats_lock);
++ spin_lock_bh(&edma_cinfo->stats_lock);
+ p = (uint32_t *)&(edma_cinfo->edma_ethstats);
+
+ for (i = 0; i < EDMA_MAX_TRANSMIT_QUEUE; i++) {
+ p++;
+ }
+
-+ spin_unlock(&edma_cinfo->stats_lock);
++ spin_unlock_bh(&edma_cinfo->stats_lock);
+}
+
+static void edma_statistics_timer(unsigned long data)
+ int i;
+ u32 stat;
+
-+ spin_lock(&edma_cinfo->stats_lock);
++ spin_lock_bh(&edma_cinfo->stats_lock);
+ p = (uint32_t *)&(edma_cinfo->edma_ethstats);
+
+ for (i = 0; i < EDMA_MAX_TRANSMIT_QUEUE; i++) {
+ p++;
+ }
+
-+ spin_unlock(&edma_cinfo->stats_lock);
++ spin_unlock_bh(&edma_cinfo->stats_lock);
+}
+
+static void edma_statistics_timer(struct timer_list *t)