RTEMS | Fix SJA1000 driver to use proper PCI/PCIe I/O access routines (i.e. inb, outb, ioread8, iowrite8) (#5563)

[Pavel Pisa (@ppisa)]

Pavel Pisa created an issue: https://gitlab.rtems.org/rtems/rtos/rtems/-/issues/5563



## Summary

When SJA1000 support has been tested on x86 it has been done out-of-mainline tree against RTEMS build i386/pc686. The header file defining `inb` and `outb` is installed in
```
/opt/rtems/7/i386-rtems7/pc686/lib/include/i386_io.h
```
and code builds OK. I have proposed to Michal to include change
in his mainline patch but he has tested it only with AMD/Xilinx
Zynq.

Problem is that rtems/cpukit is build in separation from BSPs and that is why located in BSPs section `rtems/bsps/i386/include/i386_io.h` is not seen by build
```
/opt/rtems/7/bin/i386-rtems7-gcc -MMD', '-std=gnu23 -D_BSD_SOURCE \
-Wall -Wextra -Werror -Wmissing-prototypes -Wimplicit-function-declaration \
-Wstrict-prototypes -Wnested-externs -Wold-style-declaration \
-mtune=pentiumpro -march=pentium -ffunction-sections -fdata-sections \
-O2 -g -Icpukit/include \
-I../../../../../git/rtems/cpukit/include \
-Icpukit/score/cpu/i386/include \
-I../../../../../git/rtems/cpukit/score/cpu/i386/include \
-Icpukit/contrib/include \
-I../../../../../git/rtems/cpukit/contrib/include \
-Icontrib/cpukit/include \
-I../../../../../git/rtems/contrib/cpukit/include \
-Icontrib/cpukit/fastlz \
-I../../../../../git/rtems/contrib/cpukit/fastlz \
-Icontrib/cpukit/utf8proc \
-I../../../../../git/rtems/contrib/cpukit/utf8proc \
-Icontrib/cpukit/uuid \
-I../../../../../git/rtems/contrib/cpukit/uuid \
-Icontrib/cpukit/zlib \
-I../../../../../git/rtems/contrib/cpukit/zlib \
-Icontrib/cpukit/xz-embedded/userspace \
-I../../../../../git/rtems/contrib/cpukit/xz-embedded/userspace \
../../../../../git/rtems/cpukit/dev/can/sja1000/sja1000.c \
-c -o..../cpukit/dev/can/sja1000/sja1000.c.60.o \
-DHAVE_CONFIG_H=1
```

The fast fix is to disable option to support I/O mapped SJA1000 chips

```
#if defined(__i386__)
#define RTEMS_CAN_SJA1000_ARCH_WITH_IO_PORTS 1
#include "i386_io.h"
#endif
```

The proper correction is more problematic because I have not found some portable I/O access primitives in general for memory mapped peripheral accesses with proper barriers on the architectures which require that and for IO mapped devices. Linux kernel provides methods for IO mapped (`inb` and `outb`) and MEM mapped devices accesses (`readb` and `writeb`). But there are even iowrite8 and ioread8 which allows to receive not only memory addresses but even I/O cookie on architectures that require that for PCI/PCIe I/O space access. Which these guaranteed
functions over all architectures it is easy to write drivers which work on PCI with every architecture

https://docs.kernel.org/driver-api/device-io.html

## Steps to reproduce

```
$RTEMS_DIR/waf bspdefaults --rtems-bsps=i386/pc686 -t "$RTEMS_DIR" -o "$MY_DIR" --prefix "/opt/rtems/7" >config.ini || exit 1
$RTEMS_DIR/waf configure -t "$RTEMS_DIR" -o "$MY_DIR" --rtems-config "$MY_DIR/config.ini" --prefix "/opt/rtems/7" || exit 1
$RTEMS_DIR/waf -v || exit 1
```

-- 
View it on GitLab: https://gitlab.rtems.org/rtems/rtos/rtems/-/issues/5563
You're receiving this email because of your account on gitlab.rtems.org.


-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://lists.rtems.org/pipermail/bugs/attachments/20260429/91f421f9/attachment.htm>