mwifiex: prevent register accesses after host is sleeping
authorAmitkumar Karwar <akarwar@marvell.com>
Wed, 28 Sep 2016 12:48:21 +0000 (18:18 +0530)
committerKalle Valo <kvalo@codeaurora.org>
Wed, 9 Nov 2016 01:33:24 +0000 (03:33 +0200)
commitec815dd2a5f110f627d7955e0027a3a008f68166
tree44bee722b840f5f53bd99e75b1f57f86f7f349b6
parentb073ac1fcf42376018f6db6acc885dfd2cc9ff02
mwifiex: prevent register accesses after host is sleeping

Following is mwifiex driver-firmware host sleep handshake.
It involves three threads. suspend handler, interrupt handler, interrupt
processing in main work queue.

1) Enter suspend handler
2) Download HS_CFG command
3) Response from firmware for HS_CFG
4) Suspend thread waits until handshake completes(i.e hs_activate becomes
   true)
5) SLEEP from firmware
6) SLEEP confirm downloaded to firmware.
7) SLEEP confirm response from firmware
8) Driver processes SLEEP confirm response and set hs_activate to wake up
suspend thread
9) Exit suspend handler
10) Read sleep cookie in loop and wait until it indicates firmware is
sleep.
11) After processing SLEEP confirm response, we are at the end of interrupt
processing routine. Recheck if there are interrupts received while we were
processing them.

During suspend-resume stress test, it's been observed that we may end up
acessing PCIe hardware(in 10 and 11) when PCIe bus is closed which leads
to a kernel crash.

This patch solves the problem with below changes.
a) action 10 above can be done before 8
b) Skip 11 if hs_activated is true. SLEEP confirm response
is the last interrupt from firmware. No need to recheck for
pending interrupts.
c) Add flush_workqueue() in suspend handler.

Signed-off-by: Amitkumar Karwar <akarwar@marvell.com>
Reviewed-by: Brian Norris <briannorris@chromium.org>
Tested-by: Brian Norris <briannorris@chromium.org>
Signed-off-by: Kalle Valo <kvalo@codeaurora.org>
drivers/net/wireless/marvell/mwifiex/pcie.c