soundwire: cadence: recheck device0 attachment after status change

This patch adds a status check after device0 attachment to solve race
conditions observed during attachment with multiple devices per link

The sequence is the following

1) deviceA attaches as device0

2) the hardware detects a device0 status change and throws an
 interrupt.

3) the interrupt handler schedules the work function

4) the workqueue starts, we read the status
slave0 = cdns_readl(cdns, CDNS_MCP_SLAVE_INTSTAT0);
slave1 = cdns_readl(cdns, CDNS_MCP_SLAVE_INTSTAT1);

we deal with the status change and program deviceA device number to a
non-zero value.

5) deviceB attaches as device0, the device0 status seen by the
hardware does not change.

6) we clear the CDNS_MCP_SLAVE_INTSTAT0/1 registers -> we will never detect
deviceB!

This patch suggest re-checking in a loop the device0 status with a
PING frame, i.e. using the real device0 status instead of information
on status changes.
Signed-off-by: Pierre-Louis Bossart <pierre-louis.bossart@linux.intel.com>
Reviewed-by: Rander Wang <rander.wang@intel.com>
Signed-off-by: Bard Liao <yung-chuan.liao@linux.intel.com>
Link: https://lore.kernel.org/r/20220420023039.14144-1-yung-chuan.liao@linux.intel.comSigned-off-by: Vinod Koul <vkoul@kernel.org>

drivers/soundwire/cadence_master.c

@@ -959,6 +959,8 @@ static void cdns_update_slave_status_work(struct work_struct *work)
 		container_of(work, struct sdw_cdns, work);
 	u32 slave0, slave1;
 	u64 slave_intstat;
+	u32 device0_status;
+	int retry_count = 0;
 
 	slave0 = cdns_readl(cdns, CDNS_MCP_SLAVE_INTSTAT0);
 	slave1 = cdns_readl(cdns, CDNS_MCP_SLAVE_INTSTAT1);
@@ -968,10 +970,45 @@ static void cdns_update_slave_status_work(struct work_struct *work)
 
 	dev_dbg_ratelimited(cdns->dev, "Slave status change: 0x%llx\n", slave_intstat);
 
+update_status:
 	cdns_update_slave_status(cdns, slave_intstat);
 	cdns_writel(cdns, CDNS_MCP_SLAVE_INTSTAT0, slave0);
 	cdns_writel(cdns, CDNS_MCP_SLAVE_INTSTAT1, slave1);
 
+	/*
+	 * When there is more than one peripheral per link, it's
+	 * possible that a deviceB becomes attached after we deal with
+	 * the attachment of deviceA. Since the hardware does a
+	 * logical AND, the attachment of the second device does not
+	 * change the status seen by the driver.
+	 *
+	 * In that case, clearing the registers above would result in
+	 * the deviceB never being detected - until a change of status
+	 * is observed on the bus.
+	 *
+	 * To avoid this race condition, re-check if any device0 needs
+	 * attention with PING commands. There is no need to check for
+	 * ALERTS since they are not allowed until a non-zero
+	 * device_number is assigned.
+	 */
+
+	device0_status = cdns_readl(cdns, CDNS_MCP_SLAVE_STAT);
+	device0_status &= 3;
+
+	if (device0_status == SDW_SLAVE_ATTACHED) {
+		if (retry_count++ < SDW_MAX_DEVICES) {
+			dev_dbg_ratelimited(cdns->dev,
+					    "Device0 detected after clearing status, iteration %d\n",
+					    retry_count);
+			slave_intstat = CDNS_MCP_SLAVE_INTSTAT_ATTACHED;
+			goto update_status;
+		} else {
+			dev_err_ratelimited(cdns->dev,
+					    "Device0 detected after %d iterations\n",
+					    retry_count);
+		}
+	}
+
 	/* clear and unmask Slave interrupt now */
 	cdns_writel(cdns, CDNS_MCP_INTSTAT, CDNS_MCP_INT_SLAVE_MASK);
 	cdns_updatel(cdns, CDNS_MCP_INTMASK,