| Message ID | 20200814124245.29692-1-geert+renesas@glider.be (mailing list archive) |
|---|---|
| State | New, archived |
| Headers | show |
| Series | sh: landisk: Add missing initialization of sh_io_port_base | expand |
Hi Geert! On 8/14/20 2:42 PM, Geert Uytterhoeven wrote: > The Landisk setup code maps the CF IDE area using ioremap_prot(), and > passes the resulting virtual addresses to the pata_platform driver, > disguising them as I/O port addresses. Hence the pata_platform driver > translates them again using ioport_map(). > As CONFIG_GENERIC_IOMAP=n, and CONFIG_HAS_IOPORT_MAP=y, the > SuperH-specific mapping code in arch/sh/kernel/ioport.c translates > I/O port addresses to virtual addresses by adding sh_io_port_base, which > defaults to -1, thus breaking the assumption of an identity mapping. > > Fix this by setting sh_io_port_base to zero. > > Fixes: 37b7a97884ba64bf ("sh: machvec IO death.") > Signed-off-by: Geert Uytterhoeven <geert+renesas@glider.be> Wow, that was fast :-). > --- > Tested on USL-5P, using either > CONFIG_ATA/CONFIG_ATA_SFF/CONFIG_PATA_PLATFORM, or legacy > CONFIG_IDE/CONFIG_BLK_DEV_PLATFORM. > > Sato-san: As you said IDE works fine on Landisk using > CONFIG_BLK_DEV_AEC62XX, I assume you have a different model > than the USL-5P, with PCI IDE? I think Yoshinori is using one which has the HDD built in [1] and not the USL-5P boards which are for external USB disks only. I also got one of the models with an HDD built in, but I have never tried to install Linux on it yet. Adrian > [1] https://www.iodata.jp/lib/product/h/1049.htm
diff --git a/arch/sh/boards/mach-landisk/setup.c b/arch/sh/boards/mach-landisk/setup.c index 16b4d8b0bb8503d9..2c44b94f82fb2740 100644 --- a/arch/sh/boards/mach-landisk/setup.c +++ b/arch/sh/boards/mach-landisk/setup.c @@ -82,6 +82,9 @@ device_initcall(landisk_devices_setup); static void __init landisk_setup(char **cmdline_p) { + /* I/O port identity mapping */ + __set_io_port_base(0); + /* LED ON */ __raw_writeb(__raw_readb(PA_LED) | 0x03, PA_LED);
The Landisk setup code maps the CF IDE area using ioremap_prot(), and passes the resulting virtual addresses to the pata_platform driver, disguising them as I/O port addresses. Hence the pata_platform driver translates them again using ioport_map(). As CONFIG_GENERIC_IOMAP=n, and CONFIG_HAS_IOPORT_MAP=y, the SuperH-specific mapping code in arch/sh/kernel/ioport.c translates I/O port addresses to virtual addresses by adding sh_io_port_base, which defaults to -1, thus breaking the assumption of an identity mapping. Fix this by setting sh_io_port_base to zero. Fixes: 37b7a97884ba64bf ("sh: machvec IO death.") Signed-off-by: Geert Uytterhoeven <geert+renesas@glider.be> --- Tested on USL-5P, using either CONFIG_ATA/CONFIG_ATA_SFF/CONFIG_PATA_PLATFORM, or legacy CONFIG_IDE/CONFIG_BLK_DEV_PLATFORM. Sato-san: As you said IDE works fine on Landisk using CONFIG_BLK_DEV_AEC62XX, I assume you have a different model than the USL-5P, with PCI IDE? --- arch/sh/boards/mach-landisk/setup.c | 3 +++ 1 file changed, 3 insertions(+)