Dear Linux kernel maintainers/developers,
I would like to discuss a modern alternate to Sysctl syscall and its
feasibility for armhf and aarch64 for supporting ioperm, iopl, inb,
outb and their siblings APIs emulation in Glibc, or even on an
external library to provide such sys/io.h API compatibility with
x86_64.
The advantage of having ioperm, iopl, inb and outb emulation for armhf
and aarch64 is that they can provide cross-compilation ability of
x86_64 applications and constitutes an easy and practical API for
simple I/O. This was not only useful for old ISA cards as it is today
for PCIe cards with I/O space, like I/O cards, LPT Parallel interface
cards, as well as many other cards that may have PCIe I/O space
(network cards, etc). PCIe LPT interface cards as well as
multi-function cards are still widely available, just search ebay or
amazon, and many suggestions will pop up.
I know for instance that LPT interface cards have their ppdev driver
which already provides an ioctl interface to access the card I/O
lines, but it is not performant for applications like bit-banging an
LPT card to load firmware into a micro-controller, for instance,
Microchip/ATMEL 8-bit micro-controllers. Plus the ioperm, iopl, inb,
outb has a simpler API for doing simple interactions with the I/O
space, like is the case with the IO lines of the LPT parallel
interface card.
There are many complaints in the web, like bug reports and
cross-compilation issues/reports regarding sys/io.h removal on armhf.
There are also people wishing for such functionality on aarch64.
I had an application working up to two years ago on two armhf systems
(armv7) on imx6q and mvebu architectures, to be concrete. The
application did and does just that, currently only for x84_64, but it
is able to bit-bang the ports to load the firmware into the chips. I
just didn't got to up-stream these quite small changes, like selecting
(in the kernel configuration):
General setup->Configure standard Kernel features (expert
users)->Sysctl syscall support->Enable
and do:
arch/arm/mach-imx/mach-imx6q.c, function __init imx6q_init_late(void)
and appended:
register_isa_ports(0xfee0000, 0x1f80000, 0);
Which is the PCIe IO memory region in the imx6q memory map. It was
just a matter of up-streaming the change. It maybe also useful to
rename register_isa_ports to register_pci_ports, or so in the kernel,
to avoid this confusion regarding to ISA bus only applicability.
So I would like to have some opinions regarding the introduction of a
modern alternate for the sysctl syscall update in order to achieve
Glibc cross-compatibility for armhf and possibly aarch64 of what
already exists in Glibc for x86_64. What do you say?
I am available to contribute with implementation, despite not being so
well familiarized with the Linux kernel code and development, but I
will do my best to provide an RFC implementation. Provided some small
guidance.
The facts:
The removal of the sysctl syscall caused Glibc to drop sys/io.h
emulation for ioperm, iopl, inb and outb for armhf.
-> commit 61a47c1ad3a4dc6882f01ebdc88138ac62d0df03
Author: Eric W. Biederman
Date: Tue Oct 1 13:01:19 2019 -0500
sysctl: Remove the sysctl system call
This system call has been deprecated almost since it was introduced, and
in a survey of the linux distributions I can no longer find any of them
that enable CONFIG_SYSCTL_SYSCALL. The only indication that I can find
that anyone might care is that a few of the defconfigs in the kernel
enable CONFIG_SYSCTL_SYSCALL. However this appears in only 31 of 414
defconfigs in the kernel, so I suspect this symbols presence is simply
because it is harmless to include rather than because it is necessary.
As there appear to be no users of the sysctl system call, remove the
code. As this removes one of the few uses of the internal kernel mount
of proc I hope this allows for even more simplifications of the proc
filesystem.
The Glibc maintainers found no evidence of current users for the
sys/io.h on armhf at the time:
-> commit 6b33f373c7b9199e00ba5fbafd94ac9bfb4337b1
Author: Florian Weimer
Date: Wed May 29 16:53:09 2019 +0200
arm: Remove ioperm/iopl/inb/inw/inl/outb/outw/outl support
Linux only supports the required ISA sysctls on StrongARM devices,
which are armv4 and no longer tested during glibc development
and probably bit-rotted by this point. (No reported test results,
and the last discussion of armv4 support was in the glibc 2.19
release notes.)
Then, to my surprise, I found this discussion:
http://sourceware-org.1504.n7.nabble.com/ioperm-support-for-Arm-td574053.html
"Do we still need this? I'm not sure how many kernels/boards support the
ISA bus today. "
"All that said, I don't imagine anybody would notice if the
ioperm() support in glibc were to go away, and in the unlikely
event that there is some application somewhere that relies on it,
it wouldn't be very difficult to patch ioperm() into the
application itself. It doesn't interact with anything else
inside glibc. "
Except that PCIe can also benefit from this.
Regards,
Luís Mendes
Aparapi open-source project developer (www.aparapi.com)
PhD Student and Researcher
On Wed, Jun 16, 2021 at 02:40:13PM +0100, Luís Mendes wrote:
Dear Linux kernel maintainers/developers,
I would like to discuss a modern alternate to Sysctl syscall and its
feasibility for armhf and aarch64 for supporting ioperm, iopl, inb,
outb and their siblings APIs emulation in Glibc, or even on an
external library to provide such sys/io.h API compatibility with
x86_64.
The advantage of having ioperm, iopl, inb and outb emulation for armhf
What is "armhf"?
and aarch64 is that they can provide cross-compilation ability of
x86_64 applications and constitutes an easy and practical API for
simple I/O. This was not only useful for old ISA cards as it is today
for PCIe cards with I/O space, like I/O cards, LPT Parallel interface
cards, as well as many other cards that may have PCIe I/O space
(network cards, etc). PCIe LPT interface cards as well as
multi-function cards are still widely available, just search ebay or
amazon, and many suggestions will pop up.
I know for instance that LPT interface cards have their ppdev driver
which already provides an ioctl interface to access the card I/O
lines, but it is not performant for applications like bit-banging an
LPT card to load firmware into a micro-controller, for instance,
Microchip/ATMEL 8-bit micro-controllers. Plus the ioperm, iopl, inb,
outb has a simpler API for doing simple interactions with the I/O
space, like is the case with the IO lines of the LPT parallel
interface card.
You should never use iopl, inb, and outb on modern systems, sorry.
Please use the bus-specific ways to access the devices needed, or mmap
the needed memory ranges into your program and directly access them that
way if you know what you are doing for your specific hardware type (the
UIO api provides this for you.)
There are many complaints in the web, like bug reports and
cross-compilation issues/reports regarding sys/io.h removal on armhf.
There are also people wishing for such functionality on aarch64.
I had an application working up to two years ago on two armhf systems
(armv7) on imx6q and mvebu architectures, to be concrete. The
application did and does just that, currently only for x84_64, but it
is able to bit-bang the ports to load the firmware into the chips. I
just didn't got to up-stream these quite small changes, like selecting
(in the kernel configuration):
General setup->Configure standard Kernel features (expert
users)->Sysctl syscall support->Enable
and do:
arch/arm/mach-imx/mach-imx6q.c, function __init imx6q_init_late(void)
and appended:
register_isa_ports(0xfee0000, 0x1f80000, 0);
Which is the PCIe IO memory region in the imx6q memory map. It was
just a matter of up-streaming the change. It maybe also useful to
rename register_isa_ports to register_pci_ports, or so in the kernel,
to avoid this confusion regarding to ISA bus only applicability.
As I stated above, you can access PCI IO memory regious through the UIO
api and mmap them and talk directly to them if you want to. There is no
need to do old-style inb commands anymore for stuff like this, and
hasn't been for decades now.
So I would like to have some opinions regarding the introduction of a
modern alternate for the sysctl syscall update in order to achieve
Glibc cross-compatibility for armhf and possibly aarch64 of what
already exists in Glibc for x86_64. What do you say?
sysctl should not have anything to do with i/o accesses. I don't know
how that got somehow tied together in glibc, but that seems like a glibc
issue, not a kernel one :)
thanks,
greg k-h
Greg KH [off-list ref] writes:
On Wed, Jun 16, 2021 at 02:40:13PM +0100, Luís Mendes wrote:
quoted
Dear Linux kernel maintainers/developers,
I would like to discuss a modern alternate to Sysctl syscall and its
feasibility for armhf and aarch64 for supporting ioperm, iopl, inb,
outb and their siblings APIs emulation in Glibc, or even on an
external library to provide such sys/io.h API compatibility with
x86_64.
The advantage of having ioperm, iopl, inb and outb emulation for armhf
What is "armhf"?
quoted
and aarch64 is that they can provide cross-compilation ability of
x86_64 applications and constitutes an easy and practical API for
simple I/O. This was not only useful for old ISA cards as it is today
for PCIe cards with I/O space, like I/O cards, LPT Parallel interface
cards, as well as many other cards that may have PCIe I/O space
(network cards, etc). PCIe LPT interface cards as well as
multi-function cards are still widely available, just search ebay or
amazon, and many suggestions will pop up.
I know for instance that LPT interface cards have their ppdev driver
which already provides an ioctl interface to access the card I/O
lines, but it is not performant for applications like bit-banging an
LPT card to load firmware into a micro-controller, for instance,
Microchip/ATMEL 8-bit micro-controllers. Plus the ioperm, iopl, inb,
outb has a simpler API for doing simple interactions with the I/O
space, like is the case with the IO lines of the LPT parallel
interface card.
You should never use iopl, inb, and outb on modern systems, sorry.
Please use the bus-specific ways to access the devices needed, or mmap
the needed memory ranges into your program and directly access them that
way if you know what you are doing for your specific hardware type (the
UIO api provides this for you.)
quoted
There are many complaints in the web, like bug reports and
cross-compilation issues/reports regarding sys/io.h removal on armhf.
There are also people wishing for such functionality on aarch64.
I had an application working up to two years ago on two armhf systems
(armv7) on imx6q and mvebu architectures, to be concrete. The
application did and does just that, currently only for x84_64, but it
is able to bit-bang the ports to load the firmware into the chips. I
just didn't got to up-stream these quite small changes, like selecting
(in the kernel configuration):
General setup->Configure standard Kernel features (expert
users)->Sysctl syscall support->Enable
and do:
arch/arm/mach-imx/mach-imx6q.c, function __init imx6q_init_late(void)
and appended:
register_isa_ports(0xfee0000, 0x1f80000, 0);
Which is the PCIe IO memory region in the imx6q memory map. It was
just a matter of up-streaming the change. It maybe also useful to
rename register_isa_ports to register_pci_ports, or so in the kernel,
to avoid this confusion regarding to ISA bus only applicability.
As I stated above, you can access PCI IO memory regious through the UIO
api and mmap them and talk directly to them if you want to. There is no
need to do old-style inb commands anymore for stuff like this, and
hasn't been for decades now.
quoted
So I would like to have some opinions regarding the introduction of a
modern alternate for the sysctl syscall update in order to achieve
Glibc cross-compatibility for armhf and possibly aarch64 of what
already exists in Glibc for x86_64. What do you say?
sysctl should not have anything to do with i/o accesses. I don't know
how that got somehow tied together in glibc, but that seems like a glibc
issue, not a kernel one :)
There is somewhere a sysctl that provides information needed for the
emulation. Base addresses or something like that. I have only seen it
mentioned briefly. This was one of the uses of sys_sysctl that
was mentioned in the first conversation about removing it.
The information should still be exported through /proc/sys/
Greg I will agree with you that whatever userspace is doing it should get
the information in a modern way that supports multiple busses or
at a bare minimum userspace should read the values it needs from
the files under /proc/sys.
I do sympathize as this appears to be a regression. Even if not all of
the code was merged.
The use case is one I am familiar with bring up code wanting easy access
from userspace so people can experiment and figure out what is really
going on with the hardware.
Luis please see if you can use the interfaces that support multiple
busses. That should be more robust in the future. Especially if you
are bit-banging to load firmware or other userspace drivers, the is a
reasonable chance your hardware will appear on a system with multiple
busses someday and will actually need that support.
Eric