mbox series

[RFC,0/4] Fix arm64 crash for accessing unmapped IO port regions (reboot)

Message ID 1610729929-188490-1-git-send-email-john.garry@huawei.com (mailing list archive)
Headers show
Series Fix arm64 crash for accessing unmapped IO port regions (reboot) | expand

Message

John Garry Jan. 15, 2021, 4:58 p.m. UTC
This is a reboot of my original series to address the problem of drivers
for legacy ISA devices accessing unmapped IO port regions on arm64 systems
and causing the system to crash.

There was another recent report of such an issue [0], and some old ones
[1] and [2] for reference.

The background is that many systems do not include PCI host controllers,
or they do and controller probe may have failed. For these cases, no IO
ports are mapped. However, loading drivers for legacy ISA devices can
crash the system as there is nothing to stop them accessing those IO
ports (which have not been io remap'ed).

My original solution tried to keep the kernel alive in these situations by
rejecting logical PIO access to PCI IO regions until PCI IO port regions
have been mapped.

This series goes one step further, by just reserving the complete legacy
IO port range in 0x0--0xffff for arm64. The motivation for doing this is
to make the request_region() calls for those drivers fail, like this:

root@ubuntu:/home/john# insmod mk712.ko
 [ 3415.575800] mk712: unable to get IO region
insmod: ERROR: could not insert module mk712.ko: No such device

Otherwise, in theory, those drivers could initiate rogue accesses to
mapped IO port regions for other devices and cause corruptions or
side-effects. Indeed, those drivers should not be allowed to access
IO ports at all in such a system.

As a secondary defence, for broken drivers who do not call
request_region(), IO port accesses in range 0--0xffff will be ignored,
again preserving the system.

I am sending as an RFC as I am not sure of any problem with reserving
first 0x10000 of IO space like this. There is reserve= commandline
argument, which does allow this already.

For reference, here's how /proc/ioports looks on my arm64 system with
this change:

root@ubuntu:/home/john# more /proc/ioports
00010000-0001ffff : PCI Bus 0002:f8
  00010000-00010fff : PCI Bus 0002:f9
    00010000-00010007 : 0002:f9:00.0
      00010000-00010007 : serial
    00010008-0001000f : 0002:f9:00.1
      00010008-0001000f : serial
    00010010-00010017 : 0002:f9:00.2
    00010018-0001001f : 0002:f9:00.2
00020000-0002ffff : PCI Bus 0004:88
00030000-0003ffff : PCI Bus 0005:78
00040000-0004ffff : PCI Bus 0006:c0
00050000-0005ffff : PCI Bus 0007:90
00060000-0006ffff : PCI Bus 000a:10
00070000-0007ffff : PCI Bus 000c:20
00080000-0008ffff : PCI Bus 000d:30

[0] https://lore.kernel.org/linux-input/20210112055129.7840-1-song.bao.hua@hisilicon.com/T/#mf86445470160c44ac110e9d200b09245169dc5b6
[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/

Difference since v4:
https://lore.kernel.org/linux-pci/1560262374-67875-1-git-send-email-john.garry@huawei.com/
- Reserve legacy ISA region

John Garry (4):
  arm64: io: Introduce IO_SPACE_BASE
  asm-generic/io.h: Add IO_SPACE_BASE
  kernel/resource: Make ioport_resource.start configurable
  logic_pio: Warn on and discard accesses to addresses below
    IO_SPACE_BASE

 arch/arm64/include/asm/io.h |  1 +
 include/asm-generic/io.h    |  4 ++++
 include/linux/logic_pio.h   |  5 +++++
 kernel/resource.c           |  2 +-
 lib/logic_pio.c             | 20 ++++++++++++++------
 5 files changed, 25 insertions(+), 7 deletions(-)

Comments

Jiaxun Yang Jan. 18, 2021, 1:59 a.m. UTC | #1
在 2021/1/16 上午12:58, John Garry 写道:
> This is a reboot of my original series to address the problem of drivers
> for legacy ISA devices accessing unmapped IO port regions on arm64 systems
> and causing the system to crash.
>
> There was another recent report of such an issue [0], and some old ones
> [1] and [2] for reference.
>
> The background is that many systems do not include PCI host controllers,
> or they do and controller probe may have failed. For these cases, no IO
> ports are mapped. However, loading drivers for legacy ISA devices can
> crash the system as there is nothing to stop them accessing those IO
> ports (which have not been io remap'ed).
>
> My original solution tried to keep the kernel alive in these situations by
> rejecting logical PIO access to PCI IO regions until PCI IO port regions
> have been mapped.
>
> This series goes one step further, by just reserving the complete legacy
> IO port range in 0x0--0xffff for arm64. The motivation for doing this is
> to make the request_region() calls for those drivers fail, like this:
>
> root@ubuntu:/home/john# insmod mk712.ko
>   [ 3415.575800] mk712: unable to get IO region
> insmod: ERROR: could not insert module mk712.ko: No such device
>
> Otherwise, in theory, those drivers could initiate rogue accesses to
> mapped IO port regions for other devices and cause corruptions or
> side-effects. Indeed, those drivers should not be allowed to access
> IO ports at all in such a system.
>
> As a secondary defence, for broken drivers who do not call
> request_region(), IO port accesses in range 0--0xffff will be ignored,
> again preserving the system.
>
> I am sending as an RFC as I am not sure of any problem with reserving
> first 0x10000 of IO space like this. There is reserve= commandline
> argument, which does allow this already.

Hi John,

Is it ok with ACPI? I'm not really familiar with ACPI on arm64 but my 
impression
is ACPI would use legacy I/O ports to communicate with kbd controller, 
EC and
power management facilities.

We'd better have a method to detect if ISA bus is not present on the system
instead of reserve them unconditionally.

Thanks.

- Jiaxun

>
> For reference, here's how /proc/ioports looks on my arm64 system with
> this change:
>
> root@ubuntu:/home/john# more /proc/ioports
> 00010000-0001ffff : PCI Bus 0002:f8
>    00010000-00010fff : PCI Bus 0002:f9
>      00010000-00010007 : 0002:f9:00.0
>        00010000-00010007 : serial
>      00010008-0001000f : 0002:f9:00.1
>        00010008-0001000f : serial
>      00010010-00010017 : 0002:f9:00.2
>      00010018-0001001f : 0002:f9:00.2
> 00020000-0002ffff : PCI Bus 0004:88
> 00030000-0003ffff : PCI Bus 0005:78
> 00040000-0004ffff : PCI Bus 0006:c0
> 00050000-0005ffff : PCI Bus 0007:90
> 00060000-0006ffff : PCI Bus 000a:10
> 00070000-0007ffff : PCI Bus 000c:20
> 00080000-0008ffff : PCI Bus 000d:30
>
> [0] https://lore.kernel.org/linux-input/20210112055129.7840-1-song.bao.hua@hisilicon.com/T/#mf86445470160c44ac110e9d200b09245169dc5b6
> [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/
>
> Difference since v4:
> https://lore.kernel.org/linux-pci/1560262374-67875-1-git-send-email-john.garry@huawei.com/
> - Reserve legacy ISA region
>
> John Garry (4):
>    arm64: io: Introduce IO_SPACE_BASE
>    asm-generic/io.h: Add IO_SPACE_BASE
>    kernel/resource: Make ioport_resource.start configurable
>    logic_pio: Warn on and discard accesses to addresses below
>      IO_SPACE_BASE
>
>   arch/arm64/include/asm/io.h |  1 +
>   include/asm-generic/io.h    |  4 ++++
>   include/linux/logic_pio.h   |  5 +++++
>   kernel/resource.c           |  2 +-
>   lib/logic_pio.c             | 20 ++++++++++++++------
>   5 files changed, 25 insertions(+), 7 deletions(-)
>
John Garry Jan. 18, 2021, 9:42 a.m. UTC | #2
On 18/01/2021 01:59, Jiaxun Yang wrote:
> 在 2021/1/16 上午12:58, John Garry 写道:
>> This is a reboot of my original series to address the problem of drivers
>> for legacy ISA devices accessing unmapped IO port regions on arm64 
>> systems
>> and causing the system to crash.
>>
>> There was another recent report of such an issue [0], and some old ones
>> [1] and [2] for reference.
>>
>> The background is that many systems do not include PCI host controllers,
>> or they do and controller probe may have failed. For these cases, no IO
>> ports are mapped. However, loading drivers for legacy ISA devices can
>> crash the system as there is nothing to stop them accessing those IO
>> ports (which have not been io remap'ed).
>>
>> My original solution tried to keep the kernel alive in these 
>> situations by
>> rejecting logical PIO access to PCI IO regions until PCI IO port regions
>> have been mapped.
>>
>> This series goes one step further, by just reserving the complete legacy
>> IO port range in 0x0--0xffff for arm64. The motivation for doing this is
>> to make the request_region() calls for those drivers fail, like this:
>>
>> root@ubuntu:/home/john# insmod mk712.ko
>>   [ 3415.575800] mk712: unable to get IO region
>> insmod: ERROR: could not insert module mk712.ko: No such device
>>
>> Otherwise, in theory, those drivers could initiate rogue accesses to
>> mapped IO port regions for other devices and cause corruptions or
>> side-effects. Indeed, those drivers should not be allowed to access
>> IO ports at all in such a system.
>>
>> As a secondary defence, for broken drivers who do not call
>> request_region(), IO port accesses in range 0--0xffff will be ignored,
>> again preserving the system.
>>
>> I am sending as an RFC as I am not sure of any problem with reserving
>> first 0x10000 of IO space like this. There is reserve= commandline
>> argument, which does allow this already.
> 

Hi Jiaxun,

> 
> Is it ok with ACPI? I'm not really familiar with ACPI on arm64 but my 
> impression
> is ACPI would use legacy I/O ports to communicate with kbd controller, 
> EC and
> power management facilities.

I tested for ACPI. As far as I'm concerned, fixed IO ports should not be 
used on arm64 systems.

Indeed, ACPI spec says IO port addresses should be CPU addressable, and 
it is the job of the kernel to io remap those correctly for systems 
which do not support IO ports natively, i.e. those that define PCI_IOBASE.

> 
> We'd better have a method to detect if ISA bus is not present on the system
> instead of reserve them unconditionally.
> 

For ISA bus or any IO ports region, they would/should be behind PCI host 
bridge or modeled as INDIRECT IO host and we should allocate logic PIO 
region for them, and there should be no assumption on the IO port 
address in drivers, i.e. not fixed.

Thanks,
John

> 
>>
>> For reference, here's how /proc/ioports looks on my arm64 system with
>> this change:
>>
>> root@ubuntu:/home/john# more /proc/ioports
>> 00010000-0001ffff : PCI Bus 0002:f8
>>    00010000-00010fff : PCI Bus 0002:f9
>>      00010000-00010007 : 0002:f9:00.0
>>        00010000-00010007 : serial
>>      00010008-0001000f : 0002:f9:00.1
>>        00010008-0001000f : serial
>>      00010010-00010017 : 0002:f9:00.2
>>      00010018-0001001f : 0002:f9:00.2
>> 00020000-0002ffff : PCI Bus 0004:88
>> 00030000-0003ffff : PCI Bus 0005:78
>> 00040000-0004ffff : PCI Bus 0006:c0
>> 00050000-0005ffff : PCI Bus 0007:90
>> 00060000-0006ffff : PCI Bus 000a:10
>> 00070000-0007ffff : PCI Bus 000c:20
>> 00080000-0008ffff : PCI Bus 000d:30
>>
>> [0] 
>> https://lore.kernel.org/linux-input/20210112055129.7840-1-song.bao.hua@hisilicon.com/T/#mf86445470160c44ac110e9d200b09245169dc5b6 
>>
>> [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/ 
>>
>>
>> Difference since v4:
>> https://lore.kernel.org/linux-pci/1560262374-67875-1-git-send-email-john.garry@huawei.com/ 
>>
>> - Reserve legacy ISA region
>>
>> John Garry (4):
>>    arm64: io: Introduce IO_SPACE_BASE
>>    asm-generic/io.h: Add IO_SPACE_BASE
>>    kernel/resource: Make ioport_resource.start configurable
>>    logic_pio: Warn on and discard accesses to addresses below
>>      IO_SPACE_BASE
>>
>>   arch/arm64/include/asm/io.h |  1 +
>>   include/asm-generic/io.h    |  4 ++++
>>   include/linux/logic_pio.h   |  5 +++++
>>   kernel/resource.c           |  2 +-
>>   lib/logic_pio.c             | 20 ++++++++++++++------
>>   5 files changed, 25 insertions(+), 7 deletions(-)
>>
> 
> .
Arnd Bergmann Jan. 26, 2021, 3:16 p.m. UTC | #3
On Fri, Jan 15, 2021 at 5:58 PM John Garry <john.garry@huawei.com> wrote:
>
> This is a reboot of my original series to address the problem of drivers
> for legacy ISA devices accessing unmapped IO port regions on arm64 systems
> and causing the system to crash.
>
> There was another recent report of such an issue [0], and some old ones
> [1] and [2] for reference.
>
> The background is that many systems do not include PCI host controllers,
> or they do and controller probe may have failed. For these cases, no IO
> ports are mapped. However, loading drivers for legacy ISA devices can
> crash the system as there is nothing to stop them accessing those IO
> ports (which have not been io remap'ed).
>
> My original solution tried to keep the kernel alive in these situations by
> rejecting logical PIO access to PCI IO regions until PCI IO port regions
> have been mapped.
>
> This series goes one step further, by just reserving the complete legacy
> IO port range in 0x0--0xffff for arm64. The motivation for doing this is
> to make the request_region() calls for those drivers fail, like this:
>
> root@ubuntu:/home/john# insmod mk712.ko
>  [ 3415.575800] mk712: unable to get IO region
> insmod: ERROR: could not insert module mk712.ko: No such device
>
> Otherwise, in theory, those drivers could initiate rogue accesses to
> mapped IO port regions for other devices and cause corruptions or
> side-effects. Indeed, those drivers should not be allowed to access
> IO ports at all in such a system.
>
> As a secondary defence, for broken drivers who do not call
> request_region(), IO port accesses in range 0--0xffff will be ignored,
> again preserving the system.
>
> I am sending as an RFC as I am not sure of any problem with reserving
> first 0x10000 of IO space like this. There is reserve= commandline
> argument, which does allow this already.
>
> For reference, here's how /proc/ioports looks on my arm64 system with
> this change:
>
> root@ubuntu:/home/john# more /proc/ioports
> 00010000-0001ffff : PCI Bus 0002:f8
>   00010000-00010fff : PCI Bus 0002:f9
>     00010000-00010007 : 0002:f9:00.0
>       00010000-00010007 : serial
>     00010008-0001000f : 0002:f9:00.1
>       00010008-0001000f : serial
>     00010010-00010017 : 0002:f9:00.2
>     00010018-0001001f : 0002:f9:00.2
> 00020000-0002ffff : PCI Bus 0004:88
> 00030000-0003ffff : PCI Bus 0005:78
> 00040000-0004ffff : PCI Bus 0006:c0
> 00050000-0005ffff : PCI Bus 0007:90
> 00060000-0006ffff : PCI Bus 000a:10
> 00070000-0007ffff : PCI Bus 000c:20
> 00080000-0008ffff : PCI Bus 000d:30

Doesn't this mean we lose the ability to access PCI devices
with legacy ISA compatibility? Most importantly, any GPU today
should in theory still support VGA frame buffer mode or text
console, but both of these stop working if the low I/O ports are
not mapped to the corresponding PCI bus. There is of course
already a problem if you have multiple PCI host bridges, and
each one gets its own PIO range, which means that only one
of them can have an ISA bridge with working PIO behind it.

Another such case would be a BMC that has legacy ISA devices
behind a (real or emulated) LPC bus, e.g. a 8250 UART, ps2
keyboard, RTC, or an ATA CDROM. Not sure if any of those are
ever used on Arm machines.

Regarding the size of the reservation, does this actually need
to cover the 0x0fff...0xffff range or just 0x0000...0x0fff? I don't
think there are any drivers that hardcode I/O ports beyond 0x0fff
because those would not work on ISA buses but require PCI
assigned BARs.

One more thought: There are two common ways in which PCI
host bridges map their PIO ports: either each host bridge has
its own 0x0...0xffff BAR range but gets remapped to an
arbitrary range of port numbers in the kernel, or each host bridge
uses a distinct range of port numbers, and the kernel can use
a 1:1 mapping between hardware and software port numbers,
i.e. the number in the BAR is the same as in the kernel.

If all numbers are shifted by 0x10000, that second case no
longer works, and there is always an offset.

        Arnd
John Garry Jan. 26, 2021, 5:56 p.m. UTC | #4
>>
>> For reference, here's how /proc/ioports looks on my arm64 system with
>> this change:
>>
>> root@ubuntu:/home/john# more /proc/ioports
>> 00010000-0001ffff : PCI Bus 0002:f8
>>    00010000-00010fff : PCI Bus 0002:f9
>>      00010000-00010007 : 0002:f9:00.0
>>        00010000-00010007 : serial
>>      00010008-0001000f : 0002:f9:00.1
>>        00010008-0001000f : serial
>>      00010010-00010017 : 0002:f9:00.2
>>      00010018-0001001f : 0002:f9:00.2
>> 00020000-0002ffff : PCI Bus 0004:88
>> 00030000-0003ffff : PCI Bus 0005:78
>> 00040000-0004ffff : PCI Bus 0006:c0
>> 00050000-0005ffff : PCI Bus 0007:90
>> 00060000-0006ffff : PCI Bus 000a:10
>> 00070000-0007ffff : PCI Bus 000c:20
>> 00080000-0008ffff : PCI Bus 000d:30

Hi Arnd,

> Doesn't this mean we lose the ability to access PCI devices
> with legacy ISA compatibility? Most importantly, any GPU today
> should in theory still support VGA frame buffer mode or text
> console, but both of these stop working if the low I/O ports are
> not mapped to the corresponding PCI bus.

Hmmm.. so are you saying that there is an expectation that the kernel 
PIO region assigned for these devices must start at 0x0? If so, I assume 
it's because fixed IO ports are used.

> There is of course
> already a problem if you have multiple PCI host bridges, and
> each one gets its own PIO range, which means that only one
> of them can have an ISA bridge with working PIO behind it.

The answer to my question, above, seems to be 'yes' now.

> 
> Another such case would be a BMC that has legacy ISA devices
> behind a (real or emulated) LPC bus, e.g. a 8250 UART, ps2
> keyboard, RTC, or an ATA CDROM. Not sure if any of those are
> ever used on Arm machines.
> 
> Regarding the size of the reservation, does this actually need
> to cover the 0x0fff...0xffff range or just 0x0000...0x0fff? I don't
> think there are any drivers that hardcode I/O ports beyond 0x0fff
> because those would not work on ISA buses but require PCI
> assigned BARs.

I just chose the complete legacy IO port range, that being 0x0--0xffff. 
If there would be no hardcoded ports beyond 0x0fff, then reserving 
0x0--0xfff would work.

> 
> One more thought: There are two common ways in which PCI
> host bridges map their PIO ports: either each host bridge has
> its own 0x0...0xffff BAR range but gets remapped to an
> arbitrary range of port numbers in the kernel, or each host bridge
> uses a distinct range of port numbers, and the kernel can use
> a 1:1 mapping between hardware and software port numbers,
> i.e. the number in the BAR is the same as in the kernel.
> 
> If all numbers are shifted by 0x10000, that second case no
> longer works, and there is always an offset.

Yes, this change would definitely break the second. But does - or could 
- anyone use it on arm64? I didn't think that it was possible.

Thanks,
John