[PATCH V2 11/17] i3c: mipi-i3c-hci: Stop rings gracefully when suspending

Adrian Hunter adrian.hunter at intel.com
Thu Sep 17 12:13:50 PDT 2026


hci_dma_suspend() tore the rings down with a single write of zero to
RH_RING_CONTROL, clearing the RS and ENABLE fields together and without
waiting for the ring to stop.  I3C HCI v1.1 section 6.1.2 separates those
steps: clear RS for all running Ring Bundles, and only then clear ENABLE
for all enabled Ring Bundles.

The ring registers were also written outside hci->lock, while the
interrupt handler, which takes that lock, could still be running on
another CPU.  i3c_hci_sync_irq_inactive() was called only afterwards.

Finally, an IBI can still be sitting in the IBI Status Ring when suspend
runs.  Clearing HC_CONTROL.BUS_ENABLE is deferred: per the description of
that field, if a disable request occurs while receiving an IBI, the
actual disabling does not occur until reception of the IBI is complete.

Instead, clear RS under hci->lock, wait for RING_STATUS_RUNNING to clear,
and make the interrupt handler inactive.  Only then disable the ring
interrupt signals, drain anything left in the IBI ring, and clear ENABLE.

Fixes: 816958720443 ("i3c: mipi-i3c-hci: Add DMA suspend and resume support")
Signed-off-by: Adrian Hunter <adrian.hunter at intel.com>
---


Changes in V2:

	Added comments explaining the choice of the RING_STOP_TIMEOUT_US
	and RING_STOP_SLEEP_US values.


 drivers/i3c/master/mipi-i3c-hci/dma.c | 57 +++++++++++++++++++++++++--
 1 file changed, 53 insertions(+), 4 deletions(-)

diff --git a/drivers/i3c/master/mipi-i3c-hci/dma.c b/drivers/i3c/master/mipi-i3c-hci/dma.c
index ec4b469abd33..b8fcea0870ab 100644
--- a/drivers/i3c/master/mipi-i3c-hci/dma.c
+++ b/drivers/i3c/master/mipi-i3c-hci/dma.c
@@ -15,6 +15,7 @@
 #include <linux/errno.h>
 #include <linux/i3c/master.h>
 #include <linux/io.h>
+#include <linux/iopoll.h>
 
 #include "hci.h"
 #include "cmd.h"
@@ -1052,19 +1053,67 @@ static bool hci_dma_irq_handler(struct i3c_hci *hci)
 	return handled;
 }
 
+/*
+ * With the bus disabled, a ring should stop within a few microseconds. The
+ * timeout is therefore only expected to expire if the hardware is stuck.
+ * Allow sufficient margin for slow systems, but keep the delay acceptable
+ * during suspend.
+ */
+#define RING_STOP_TIMEOUT_US	(100 * USEC_PER_MSEC)
+/*
+ * The ring is usually already stopped, so polling typically completes on the
+ * first iteration. Use a modest sleep interval to avoid busy-waiting without
+ * adding excessive latency.
+ */
+#define RING_STOP_SLEEP_US	100
+
 static void hci_dma_suspend(struct i3c_hci *hci)
 {
 	struct hci_rings_data *rings = hci->io_data;
 	int n = rings ? rings->total : 0;
+	struct hci_rh_data *rh;
+	u32 regval;
 
-	for (int i = 0; i < n; i++) {
-		struct hci_rh_data *rh = &rings->headers[i];
+	/* Gracefully stop the rings */
+	scoped_guard(spinlock_irqsave, &hci->lock) {
+		for (int i = 0; i < n; i++) {
+			rh = &rings->headers[i];
+			regval = rh_reg_read(RING_CONTROL);
+			if (regval & RING_CTRL_RUN_STOP)
+				rh_reg_write(RING_CONTROL, regval & ~RING_CTRL_RUN_STOP);
+		}
+	}
 
-		rh_reg_write(INTR_SIGNAL_ENABLE, 0);
-		rh_reg_write(RING_CONTROL, 0);
+	/* Wait for actual stop */
+	for (int i = 0; i < n; i++) {
+		rh = &rings->headers[i];
+		if (readx_poll_timeout(readl, rh->regs + RH_RING_STATUS, regval,
+				       !(regval & RING_STATUS_RUNNING),
+				       RING_STOP_SLEEP_US, RING_STOP_TIMEOUT_US))
+			dev_err(&hci->master.dev, "%s: Ring did not stop, status %#x\n",
+				__func__, regval);
 	}
 
+	/*
+	 * With the rings stopped, no more IBIs can be received. Flush and make
+	 * the interrupt handler inactive.
+	 */
 	i3c_hci_sync_irq_inactive(hci);
+
+	/* Disable interrupt signals and disable the rings */
+	scoped_guard(spinlock_irqsave, &hci->lock)
+		for (int i = 0; i < n; i++) {
+			rh = &rings->headers[i];
+			rh_reg_write(INTR_SIGNAL_ENABLE, 0);
+			/*
+			 * Be absolutely certain there is no unprocessed IBI.
+			 * hci_dma_drain_ibi_ring() will do nothing if there is
+			 * none.
+			 */
+			if (i < IBI_RINGS)
+				hci_dma_drain_ibi_ring(hci, rh);
+			rh_reg_write(RING_CONTROL, 0);
+		}
 }
 
 static void hci_dma_resume(struct i3c_hci *hci)
-- 
2.53.0




More information about the linux-i3c mailing list