[<prev] [next>] [thread-next>] [day] [month] [year] [list]
Date: Thu, 4 Apr 2019 23:59:58 +0800
From: John Garry <john.garry@...wei.com>
To: <bhelgaas@...gle.com>, <rafael@...nel.org>, <arnd@...db.de>,
<lorenzo.pieralisi@....com>
CC: <linux@...ck-us.net>, <linux-kernel@...r.kernel.org>,
<linux-pci@...r.kernel.org>, <bp@...e.de>,
<wangkefeng.wang@...wei.com>, <linuxarm@...wei.com>,
<andy.shevchenko@...il.com>,
<linux-arm-kernel@...ts.infradead.org>, <catalin.marinas@....com>,
<will.deacon@....com>, John Garry <john.garry@...wei.com>
Subject: [PATCH v3 0/4] Fix system crash for accessing unmapped IO port regions
It was reported some time ago that arm64 systems will crash if a driver
attempts to access IO port addresses when the PCI IO port region has not
been mapped [1].
More recently, a similar crash is where the system PCI host probe fails,
and the IPMI driver crashes the system while attempting to do some IO port
accesses [2].
This patchset attempts to keep the kernel alive in such situations by 2
complementary methods:
1. Rejecting IO port resource requests until PCI IO port regions have been
mapped (in a pci_remap_iospace() call).
2. Rejecting logic PIO access to PCI IO regions until, again, PCI IO port
regions have been mapped
About 1:
Currently the PCI IO port region is initialized to the full range,
{0, IO_SPACE_LIMIT}. As such, any IO port region requests would not fail
because of PCI IO port regions not being mapped.
Patch 1/4 looks to remedy this issue by ensuring IO port requests are
made to direct children of ioport_resource (PCI host IO port regions),
similar to Arnd's solution, mentioned in [1]:
"I see that ioport_resource gets initialized to the {0, IO_SPACE_LIMIT}
range. If we could change it so that pci_remap_iospace() hooks up
to ioport_resource and extends it whenever something gets mapped
there up to IO_SPACE_LIMIT, we can change the default range to
{0,0}, which would fail for any request_region call before the
first pci_remap_iospace."
I didn't use this solution, as logical PIO space is sparse in
{0, IO_SPACE_LIMIT}, so we cannot simply grow the region.
*As discussed with Bjorn in v2 series, we doubt that this approach is
sound, as legacy ISA devices do not necessarily reply on PCI.*
However I will keep the patch in the series as a reference and as a topic
of debate.
It's also an RFC as the implementation solution is not idea.
About 2:
Some drivers - like f71805f hwmon driver - do not call
request_{muxed_}region() prior to accessing IO port regions, as they
should do.
So patches 2-3/4 adds a safeguard against this, in that unwarranted PIO IO
accesses will be discarded in the low-level accessors.
About the issue of f71805f driver not requesting the IO port region -
many drivers do this, and need to be fixed up separately.
1. https://lore.kernel.org/linux-pci/56F209A9.4040304@huawei.com
2. https://lore.kernel.org/linux-arm-kernel/e6995b4a-184a-d8d4-f4d4-9ce75d8f47c0@huawei.com/
Differences to v2 patchset:
https://lkml.org/lkml/2019/3/20/788
- Add a patch to use logical PIO accessors for !CONFIG_INDIRECT_PIO
- Some tidy-up according to Andy's review
Differences to v1 patchset:
https://lkml.org/lkml/2019/3/14/630
- Drop f71805f fix - it can be done in a separate patchset
- Change implementation in resource.c patch to check if parent of region
is ioport_resource
- Add patch to fix some logic_pio.c prints
John Garry (4):
resource: Request IO port regions from children of ioport_resource
lib: logic_pio: Use logical PIO low-level accessors for
!CONFIG_INDIRECT_PIO
lib: logic_pio: Reject accesses to unregistered CPU MMIO regions
lib: logic_pio: Fix up some prints
include/linux/ioport.h | 11 ++-
include/linux/logic_pio.h | 7 +-
kernel/resource.c | 28 +++++++
lib/logic_pio.c | 157 +++++++++++++++++++++++++++++---------
4 files changed, 159 insertions(+), 44 deletions(-)
--
2.17.1
Powered by blists - more mailing lists