Re: Fwd: RCU indicates stalls with iwlwifi, causing boot failures

[Hugh Dickins <hughd@google.com>]

Hi Dave,

On Fri, 1 Sep 2023, Ben Greear wrote:
> On 9/1/23 5:29 PM, Bagas Sanjaya wrote:
> > Hi,
> > 
> > I notice a bug report on Bugzilla [1]. Quoting from it:
> 
> Try booting with pcie=noaer ?
> 
> That fixes only known iwlwifi bug we have found in 6.5, but we are also using
> mostly
> backports iwlwifi driver...
> 
> Thanks,
> Ben
> 
> > 
> >> I'm seeing RCU warnings in Linus's current tree (like
> >> 87dfd85c38923acd9517e8df4afc908565df0961) that come from RCU:
> >>
> >> WARNING: CPU: 0 PID: 0 at kernel/rcu/tree_exp.h:787
> >> rcu_exp_handler+0x35/0xe0
> >>
> >> But they *ONLY* occur on a system with a newer iwlwifi device:
> >>
> >> aa:00.0 Network controller: Intel Corporation Wi-Fi 6 AX210/AX211/AX411
> >> 160MHz (rev 1a)
> >>
> >> and never in a VM or on an older device (like an 8260).  During a bisect
> >> the only seem to occur with the "83" version of the firmware.
> >>
> >> iwlwifi 0000:aa:00.0: loaded firmware version 83.e8f84e98.0
> >> ty-a0-gf-a0-83.ucode op_mode iwlmvm
> >>
> >> The first warning gets spit out within a millisecond of the last printk()
> >> from the iwlwifi driver.  They eventually result in a big spew of RCU
> >> messages like this:
> >>
> >> [   27.124796] rcu: INFO: rcu_preempt detected expedited stalls on
> >> CPUs/tasks: { 0-...D } 125 jiffies s: 193 root: 0x1/.
> >> [   27.126466] rcu: blocking rcu_node structures (internal RCU debug):
> >> [   27.128114] Sending NMI from CPU 3 to CPUs 0:
> >> [   27.128122] NMI backtrace for cpu 0 skipped: idling at
> >> intel_idle+0x5f/0xb0
> >> [   27.159757] loop30: detected capacity change from 0 to 8
> >> [   27.204967] rcu: INFO: rcu_preempt detected expedited stalls on
> >> CPUs/tasks: { 0-...D } 145 jiffies s: 193 root: 0x1/.
> >> [   27.206353] rcu: blocking rcu_node structures (internal RCU debug):
> >> [   27.207751] Sending NMI from CPU 3 to CPUs 0:
> >> [   27.207825] NMI backtrace for cpu 0 skipped: idling at
> >> intel_idle+0x5f/0xb0
> >>
> >> I usually see them at boot.  In that case, they usually hang the system and
> >> keep it from booting.  I've also encountered them at reboots and also seen
> >> them *not* be fatal at boot.  I suspect it has to do with which CPU gets
> >> wedged.
> > 
> > See Bugzilla for the full thread and attached full dmesg output.
> > 
> > Thanks.
> > 
> > [1]: https://bugzilla.kernel.org/show_bug.cgi?id=217856

I just took a look at your dmesg in bugzilla: I see lots of page tables
dumped, including "ESPfix Area", and think you're hitting my screwup: see

https://lore.kernel.org/linux-mm/CABXGCsNi8Tiv5zUPNXr6UJw6qV1VdaBEfGqEAMkkXE3QPvZuAQ@mail.gmail.com/

Please give the patch from the end of that thread a try:

[PATCH] mm/pagewalk: fix bootstopping regression from extra pte_unmap()

[ Commit message now written, but let's see if Dave can confirm it too ]

Signed-off-by: Hugh Dickins <hughd@google.com>
---
 mm/pagewalk.c | 2 +-
 1 file changed, 1 insertion(+), 1 deletion(-)

diff --git a/mm/pagewalk.c b/mm/pagewalk.c
index 2022333805d3..9e7d0276c38a 100644
--- a/mm/pagewalk.c
+++ b/mm/pagewalk.c
@@ -58,7 +58,7 @@ static int walk_pte_range(pmd_t *pmd, unsigned long addr, unsigned long end,
 			pte = pte_offset_map(pmd, addr);
 		if (pte) {
 			err = walk_pte_range_inner(pte, addr, end, walk);
-			if (walk->mm != &init_mm)
+			if (walk->mm != &init_mm && addr < TASK_SIZE)
 				pte_unmap(pte);
 		}
 	} else {
-- 
2.35.3