[<prev] [next>] [day] [month] [year] [list]
Message-ID: <20220302183245.xzxaf6zz2yxmnsbm@bsd-mbp.dhcp.thefacebook.com>
Date: Wed, 2 Mar 2022 10:32:45 -0800
From: Jonathan Lemon <jonathan.lemon@...il.com>
To: Thomas Gleixner <tglx@...utronix.de>
Cc: linux-kernel@...r.kernel.org, kernel-team@...com
Subject: IRTE entry not present errors for serial driver
I'm running into an issue with the ptp_ocp driver that I can use some
help with.
This driver is for a PCIe board which has an FPGA providing UARTs
for communication with onboard chips. The design uses MSI interrupts.
When the driver is initially loaded, tracepoints show that the
the interrupt vector is activated and allocated:
kworker/0:2-2827 [000] ..... 76495.343455: vector_activate: irq=82 is_managed=0 can_reserve=1 reserve=0
kworker/0:2-2827 [000] d..1. 76495.343456: vector_alloc: irq=82 vector=37 reserved=1 ret=0
kworker/0:2-2827 [000] d..1. 76495.343456: vector_update: irq=82 vector=37 cpu=3 prev_vector=0 prev_cpu=0
kworker/0:2-2827 [000] d..1. 76495.343457: vector_config: irq=82 vector=37 cpu=3 apicdest=0x00000006
Later, when the serial device is /first/ opened, there are no errors
and serial communication proceeds without a problem. When the tty
is closed, the MSI interrupt entry is deactivated:
tio-3209 [008] d..1. 76557.424960: vector_deactivate: irq=82 is_managed=0 can_reserve=0 reserve=0
On the second (and subsequent) opens, a DMAR fault is observed, which
seems to indicate that an IRQ from the serial port was taken before the
IRTE vector was updated.
[76561.103090] DMAR: DRHD: handling fault status reg 2
[76561.105092] DMAR: [INTR-REMAP] Request device [65:00.0] fault index 0x25 [fault reason 0x22] Present field in the IRTE entry is clear
tio-3211 [002] d..1. 76561.108431: vector_activate: irq=82 is_managed=0 can_reserve=0 reserve=0
(serial port closed)
tio-3211 [002] d..1. 76564.596150: vector_deactivate: irq=82 is_managed=0 can_reserve=0 reserve=0
Why is the vector being deactivated when the tty device is closed,
if it is left activated on module load?
This is on net-next (5.17.0-rc4+)
--
Jonathan
Powered by blists - more mailing lists