[PATCH] wifi: iwlwifi: stop peripheral polling when NIC access fails

[Jared Hooker]

When an AX211 loses power across s2idle, RX shutdown can poll an
inaccessible peripheral register during PCI resume. The register-read
helper returns 0x5a5a5a5a when NIC access fails, hiding the failure from
the polling loop. Each iteration can then incur another NIC-access
timeout, although only the polling delay is charged to the timeout
budget.

On a ThinkPad L14 Gen 3 running 7.2.8-9-omarchy, function-graph tracing
measured 7.91 seconds in iwl_poll_prph_bit() for the RX stop caller's
1000-us timeout. The surrounding PCI resume callback took 8.26 seconds,
leaving the desktop visibly frozen after opening the lid.

Acquire NIC access explicitly for each peripheral read and return
-ETIMEDOUT immediately when acquisition fails. Preserve the existing
acquire/read/release sequence, mask comparison, polling delay and retry
budget while register access succeeds. This also avoids treating the
failure placeholder as a matching register value.

With the equivalent change backported to the affected 7.2.8 driver,
the poll took 76.24 ms and PCI resume took 400.79 ms. Wake responsiveness
and Wi-Fi connectivity were confirmed on the affected machine.

Link: https://github.com/omacom/omarchy-pkgs/pull/1493
Signed-off-by: Jared Hooker <jaredroberthooker@xxxxxxxxx>
---
Submission notes:
- Rebased onto iwlwifi-next/next at 3627924ed81f, where the helper is
  named iwl_pcie_poll_prph_bit() in pcie/utils.c.
- Hardware testing used the equivalent iwl-io.c backport on
  7.2.8-9-omarchy with an Intel AX211 (8086:51f0), firmware
  89.d2579d43.0, and ThinkPad L14 Gen 3 BIOS R1XET52W (1.34).
- Before/after function-graph timings:
                          before          after
  peripheral poll     7,914,548 us      76,235.97 us
  RX stop             7,994,772 us     152,336.40 us
  PCI resume          8,257,866 us     400,792.70 us
- Six mocked-transport C regression cases pass against the actual
  adapted helper. The original helper fails the access-failure case.
- The affected driver module subtree builds against the installed
  7.2.8 headers, as does the updated pcie/utils.o from this upstream tree.
  A complete build/boot of iwlwifi-next has not been performed.
- checkpatch.pl reports zero errors and warnings.
- This fixes repeated polling after access failure; it does not establish
  why the device lost power or identify the original version regression.

 drivers/net/wireless/intel/iwlwifi/pcie/utils.c | 11 ++++++++++-
 1 file changed, 10 insertions(+), 1 deletion(-)

diff --git a/drivers/net/wireless/intel/iwlwifi/pcie/utils.c
b/drivers/net/wireless/intel/iwlwifi/pcie/utils.c
index 8aa43e5..0962df4 100644
--- a/drivers/net/wireless/intel/iwlwifi/pcie/utils.c
+++ b/drivers/net/wireless/intel/iwlwifi/pcie/utils.c
@@ -139,7 +139,16 @@ int iwl_pcie_poll_prph_bit(struct iwl_trans
*trans, u32 addr,
 	int t = 0;

 	do {
-		if ((iwl_trans_read_prph(trans, addr) & mask) == (bits & mask))
+		u32 val;
+
+		/* Do not retry register reads when the NIC is inaccessible. */
+		if (!iwl_trans_grab_nic_access(trans))
+			return -ETIMEDOUT;
+
+		val = iwl_trans_read_prph_no_grab(trans, addr);
+		iwl_trans_release_nic_access(trans);
+
+		if ((val & mask) == (bits & mask))
 			return 0;
 		udelay(IWL_PCIE_POLL_INTERVAL);
 		t += IWL_PCIE_POLL_INTERVAL;

base-commit: 3627924ed81fd7783394fb2555ea179c6606a231
-- 
2.56.0