Skip to contentSkip to conversation

FAILED: patch "[PATCH] firmware: samsung: acpm: Fix infinite loop on sequence number" failed to apply to 7.0-stable tree

PATCH New [email protected] 2 participants 2 messages Jun 15 – Jun 17

Subsystems (3)
lore
Share & tools
lore

[email protected]

author maintainer
#0 lore raw

The patch below does not apply to the 7.0-stable tree.
If someone wants it applied there, or to any other stable or longterm tree, then please email the backport, including the original git commit id to <[email protected]>.

To reproduce the conflict and resubmit, you may use the following commands:

git fetch https://git.kernel.org/pub/scm/linux/kernel/git/stable/linux.git/ linux-7.0.y
git checkout FETCH_HEAD
git cherry-pick -x 7fe40c32a33905302341797b5d12c541729dd08d
# <resolve conflicts, build, test, etc.>
git commit -s
git send-email --to '<[email protected]>' --in-reply-to '2026061542-daintily-knickers-a496@gregkh' --subject-prefix 'PATCH 7.0.y' HEAD^..

Possible dependencies:

thanks,

greg k-h

------------------ original commit in Linus's tree ------------------

From 7fe40c32a33905302341797b5d12c541729dd08d Mon Sep 17 00:00:00 2001
From: Tudor Ambarus <[email protected]>
Date: Tue, 5 May 2026 13:13:04 +0000
Subject: [PATCH] firmware: samsung: acpm: Fix infinite loop on sequence number
exhaustion

Sashiko identified a possible infinite loop [1].

ACPM IPC sequence numbers are tracked via a 64-bit bitmap. Previously, acpm_prepare_xfer() used a do...while loop to search for a free sequence number.

If all 63 available sequence numbers are leaked due to transient hardware timeouts or mailbox failures, the bitmap becomes full. The next call to acpm_prepare_xfer() would enter an infinite loop.

Fix this by utilizing the kernel's optimized bitmap search functions (find_next_zero_bit / find_first_zero_bit). If the pool is completely exhausted, log the failure and return -EBUSY to allow the kernel to fail gracefully instead of hanging.

Furthermore, drop the allocation loop entirely. Because acpm_prepare_xfer() is strictly called under the 'tx_lock' mutex, sequence number allocations are perfectly serialized. If find_next_zero_bit() locates a free bit, a single test_and_set_bit_lock() is mathematically guaranteed to succeed.

To enforce this locking invariant, wrap the allocation in a WARN_ON_ONCE. If the atomic set fails, it indicates the driver's mutex serialization is fundamentally broken. The warning generates a stack trace for debugging, while returning -EIO immediately aborts the transfer to prevent silent payload corruption.

drivers/firmware/samsung/exynos-acpm.c +31 -14
index 3fa9fe283be4..19db3674a28f 100644
@@ -12,6 +12,7 @@
 #include <linux/container_of.h>
 #include <linux/delay.h>
 #include <linux/device.h>
+#include <linux/find.h>
 #include <linux/firmware/samsung/exynos-acpm-protocol.h>
 #include <linux/io.h>
 #include <linux/iopoll.h>
@@ -390,34 +391,48 @@ static int acpm_wait_for_queue_slots(struct acpm_chan *achan, u32 next_tx_front)
  * TX queue.
  * @achan:	ACPM channel info.
  * @xfer:	reference to the transfer being prepared.
+ *
+ * Return: 0 on success, -errno otherwise.
  */
-static void acpm_prepare_xfer(struct acpm_chan *achan,
-			      const struct acpm_xfer *xfer)
+static int acpm_prepare_xfer(struct acpm_chan *achan,
+			     const struct acpm_xfer *xfer)
 {
 	struct acpm_rx_data *rx_data;
 	u32 *txd = (u32 *)xfer->txd;
+	unsigned long size = ACPM_SEQNUM_MAX - 1;
+	unsigned long bit = achan->seqnum;
+
+	bit = find_next_zero_bit(achan->bitmap_seqnum, size, bit);
+	if (bit >= size) {
+		bit = find_first_zero_bit(achan->bitmap_seqnum, size);
+		if (bit >= size) {
+			dev_err_ratelimited(achan->acpm->dev,
+					    "ACPM sequence number pool exhausted\n");
+			return -EBUSY;
+		}
+	}
 
 	/*
-	 * Prevent chan->seqnum from being re-used.
-	 * test_and_set_bit_lock() provides formal LKMM Acquire semantics.
-	 * It pairs with the RX thread's clear_bit_unlock() to ensure the CPU
-	 * does not speculatively execute the rx_data buffer wipe (memset)
-	 * before the sequence number is safely claimed.
+	 * Execute the atomic set to formally claim the bit and establish
+	 * LKMM Acquire semantics against the RX thread's clear_bit_unlock().
+	 * A loop is unnecessary because allocations are strictly serialized
+	 * by tx_lock.
 	 */
-	do {
-		if (++achan->seqnum == ACPM_SEQNUM_MAX)
-			achan->seqnum = 1;
-		/* Flag the index based on seqnum. (seqnum: 1~63, bitmap: 0~62) */
-	} while (test_and_set_bit_lock(achan->seqnum - 1, achan->bitmap_seqnum));
+	if (WARN_ON_ONCE(test_and_set_bit_lock(bit, achan->bitmap_seqnum)))
+		return -EIO;
 
+	/* Flag the index based on seqnum. (seqnum: 1~63, bitmap: 0~62) */
+	achan->seqnum = bit + 1;
 	txd[0] |= FIELD_PREP(ACPM_PROTOCOL_SEQNUM, achan->seqnum);
 
 	/* Clear data for upcoming responses */
-	rx_data = &achan->rx_data[achan->seqnum - 1];
+	rx_data = &achan->rx_data[bit];
 	rx_data->completed = false;
 	memset(rx_data->cmd, 0, sizeof(*rx_data->cmd) * rx_data->n_cmd);
 	/* zero means no response expected */
 	rx_data->rxcnt = xfer->rxcnt;
+
+	return 0;
 }
 
 /**
@@ -477,7 +492,9 @@ int acpm_do_xfer(struct acpm_handle *handle, const struct acpm_xfer *xfer)
 		if (ret)
 			return ret;
 
-		acpm_prepare_xfer(achan, xfer);
+		ret = acpm_prepare_xfer(achan, xfer);
+		if (ret)
+			return ret;
 
 		/* Write TX command. */
 		__iowrite32_copy(achan->tx.base + achan->mlen * tx_front,
maintainer
#1 lore raw

PATCH 1/1 firmware: samsung: acpm: Fix infinite loop on sequence number exhaustion

From: Tudor Ambarus <[email protected]>

[ Upstream commit 7fe40c32a33905302341797b5d12c541729dd08d ]

Sashiko identified a possible infinite loop [1].

ACPM IPC sequence numbers are tracked via a 64-bit bitmap. Previously, acpm_prepare_xfer() used a do...while loop to search for a free sequence number.

If all 63 available sequence numbers are leaked due to transient hardware timeouts or mailbox failures, the bitmap becomes full. The next call to acpm_prepare_xfer() would enter an infinite loop.

Fix this by utilizing the kernel's optimized bitmap search functions (find_next_zero_bit / find_first_zero_bit). If the pool is completely exhausted, log the failure and return -EBUSY to allow the kernel to fail gracefully instead of hanging.

Furthermore, drop the allocation loop entirely. Because acpm_prepare_xfer() is strictly called under the 'tx_lock' mutex, sequence number allocations are perfectly serialized. If find_next_zero_bit() locates a free bit, a single test_and_set_bit_lock() is mathematically guaranteed to succeed.

To enforce this locking invariant, wrap the allocation in a WARN_ON_ONCE. If the atomic set fails, it indicates the driver's mutex serialization is fundamentally broken. The warning generates a stack trace for debugging, while returning -EIO immediately aborts the transfer to prevent silent payload corruption.

---
 drivers/firmware/samsung/exynos-acpm.c | 43 +++++++++++++++++++-------
 1 file changed, 32 insertions(+), 11 deletions(-)
drivers/firmware/samsung/exynos-acpm.c +32 -11
index 987b59778ffc4a..075aadb10d23a3 100644
@@ -12,6 +12,7 @@
 #include <linux/container_of.h>
 #include <linux/delay.h>
 #include <linux/device.h>
+#include <linux/find.h>
 #include <linux/firmware/samsung/exynos-acpm-protocol.h>
 #include <linux/io.h>
 #include <linux/iopoll.h>
@@ -364,29 +365,47 @@ static int acpm_wait_for_queue_slots(struct acpm_chan *achan, u32 next_tx_front)
  * TX queue.
  * @achan:	ACPM channel info.
  * @xfer:	reference to the transfer being prepared.
+ *
+ * Return: 0 on success, -errno otherwise.
  */
-static void acpm_prepare_xfer(struct acpm_chan *achan,
-			      const struct acpm_xfer *xfer)
+static int acpm_prepare_xfer(struct acpm_chan *achan,
+			     const struct acpm_xfer *xfer)
 {
 	struct acpm_rx_data *rx_data;
 	u32 *txd = (u32 *)xfer->txd;
+	unsigned long size = ACPM_SEQNUM_MAX - 1;
+	unsigned long bit = achan->seqnum;
+
+	bit = find_next_zero_bit(achan->bitmap_seqnum, size, bit);
+	if (bit >= size) {
+		bit = find_first_zero_bit(achan->bitmap_seqnum, size);
+		if (bit >= size) {
+			dev_err_ratelimited(achan->acpm->dev,
+					    "ACPM sequence number pool exhausted\n");
+			return -EBUSY;
+		}
+	}
 
-	/* Prevent chan->seqnum from being re-used */
-	do {
-		if (++achan->seqnum == ACPM_SEQNUM_MAX)
-			achan->seqnum = 1;
-	} while (test_bit(achan->seqnum - 1, achan->bitmap_seqnum));
+	/*
+	 * Execute the atomic set to formally claim the bit and establish
+	 * LKMM Acquire semantics against the RX thread's clear_bit_unlock().
+	 * A loop is unnecessary because allocations are strictly serialized
+	 * by tx_lock.
+	 */
+	if (WARN_ON_ONCE(test_and_set_bit_lock(bit, achan->bitmap_seqnum)))
+		return -EIO;
 
+	/* Flag the index based on seqnum. (seqnum: 1~63, bitmap: 0~62) */
+	achan->seqnum = bit + 1;
 	txd[0] |= FIELD_PREP(ACPM_PROTOCOL_SEQNUM, achan->seqnum);
 
 	/* Clear data for upcoming responses */
-	rx_data = &achan->rx_data[achan->seqnum - 1];
+	rx_data = &achan->rx_data[bit];
 	memset(rx_data->cmd, 0, sizeof(*rx_data->cmd) * rx_data->n_cmd);
 	if (xfer->rxd)
 		rx_data->response = true;
 
-	/* Flag the index based on seqnum. (seqnum: 1~63, bitmap: 0~62) */
-	set_bit(achan->seqnum - 1, achan->bitmap_seqnum);
+	return 0;
 }
 
 /**
@@ -444,7 +463,9 @@ int acpm_do_xfer(struct acpm_handle *handle, const struct acpm_xfer *xfer)
 		if (ret)
 			return ret;
 
-		acpm_prepare_xfer(achan, xfer);
+		ret = acpm_prepare_xfer(achan, xfer);
+		if (ret)
+			return ret;
 
 		/* Write TX command. */
 		__iowrite32_copy(achan->tx.base + achan->mlen * tx_front,
Signature · 1 line
2.53.0