Re: [PATCH 14/22] ARM: omap1: use pci_ioremap_io() for omap_cf

From: Tony Lindgren <tony@atomide.com>
To: Arnd Bergmann <arnd@arndb.de>
Cc: Aaro Koskinen <aaro.koskinen@iki.fi>,
	Dominik Brodowski <linux@dominikbrodowski.net>,
	linux-omap <linux-omap@vger.kernel.org>,
	Linux ARM <linux-arm-kernel@lists.infradead.org>,
	Greg Kroah-Hartman <gregkh@linuxfoundation.org>,
	Linus Walleij <linus.walleij@linaro.org>,
	Bartlomiej Zolnierkiewicz <b.zolnierkie@samsung.com>,
	Tomi Valkeinen <tomi.valkeinen@ti.com>,
	Linux Kernel Mailing List <linux-kernel@vger.kernel.org>
Subject: Re: [PATCH 14/22] ARM: omap1: use pci_ioremap_io() for omap_cf
Date: Wed, 14 Aug 2019 00:49:18 -0700	[thread overview]
Message-ID: <20190814074918.GA52127@atomide.com> (raw)
In-Reply-To: <CAK8P3a0LjKrc+7c5Ht9OL7LfYyLnG9=y7u+w24ujA1xAid_yCQ@mail.gmail.com>

* Arnd Bergmann <arnd@arndb.de> [190813 19:34]:
> On Tue, Aug 13, 2019 at 8:12 PM Aaro Koskinen <aaro.koskinen@iki.fi> wrote:
> > On Tue, Aug 13, 2019 at 01:02:16PM +0200, Arnd Bergmann wrote:
> > > On Tue, Aug 13, 2019 at 12:36 PM Tony Lindgren <tony@atomide.com> wrote:
> >
> > > - I force CONFIG_PCI to be enabled here in order to keep the
> > >   asm/io.h logic unchanged. If PCI support in itself is an issue,
> > >   then turning on CONFIG_PCI without the rest of this patch
> > >   should also break.
> >
> > The board dies early, probably in pci_reserve_io():
> >
> > Starting kernel ...
> >
> > [    0.000000] Booting Linux on physical CPU 0x0
> > [    0.000000] Linux version 5.3.0-rc4-osk-los_80efa+-00028-g09f6f22a63e9 (aaro@amd-fx-6350) (gcc version 8.3.0 (GCC)) #1 Tue Aug 13 20:50:11 EEST 2019
> > [    0.000000] CPU: ARM926EJ-S [41069263] revision 3 (ARMv5TEJ), cr=0005317f
> > [    0.000000] CPU: VIVT data cache, VIVT instruction cache
> > [    0.000000] Machine: TI-OSK
> > [    0.000000] Ignoring tag cmdline (using the default kernel command line)
> > [    0.000000] printk: bootconsole [earlycon0] enabled
> > [    0.000000] Memory policy: Data cache writeback
> > [    0.000000] Internal error: Oops - undefined instruction: 0 [#1] ARM
> > [    0.000000] CPU: 0 PID: 0 Comm: swapper Not tainted 5.3.0-rc4-osk-los_80efa+-00028-g09f6f22a63e9 #1
> > [    0.000000] Hardware name: TI-OSK
> > [    0.000000] PC is at vm_area_add_early+0x1c/0x74
> 
> That sounds like an address conflict in the virtual addres space.
> 
> In multiplatform kernels, PCI I/O space is hardwired to addresses
> 0xfee00000-0xfeffffff,
> which happened to be available on all the other machines that needed it so far.
> 
> OMAP1_IO_VIRT is 0xfefb0000-0xfefeffff, which clearly overlaps with the end of
> the PCI I/O area.
> 
> We only really need 4KB of I/O space rather than the full 2MB, but it
> would also be
> good not to make this too machine specific.
> 
> Could we change OMAP1_IO_OFFSET to stay out of that area? Something like
> 
> diff --git a/arch/arm/Kconfig.debug b/arch/arm/Kconfig.debug
> index 97c114c1ef80..3b66d203dc98 100644
> --- a/arch/arm/Kconfig.debug
> +++ b/arch/arm/Kconfig.debug
> @@ -1794,9 +1794,9 @@ config DEBUG_UART_VIRT
>         default 0xfef00000 if ARCH_IXP4XX && !CPU_BIG_ENDIAN
>         default 0xfef00003 if ARCH_IXP4XX && CPU_BIG_ENDIAN
>         default 0xfef36000 if DEBUG_HIGHBANK_UART
> -       default 0xfefb0000 if DEBUG_OMAP1UART1 || DEBUG_OMAP7XXUART1
> -       default 0xfefb0800 if DEBUG_OMAP1UART2 || DEBUG_OMAP7XXUART2
> -       default 0xfefb9800 if DEBUG_OMAP1UART3 || DEBUG_OMAP7XXUART3
> +       default 0xff000000 if DEBUG_OMAP1UART1 || DEBUG_OMAP7XXUART1
> +       default 0xff000800 if DEBUG_OMAP1UART2 || DEBUG_OMAP7XXUART2
> +       default 0xff009800 if DEBUG_OMAP1UART3 || DEBUG_OMAP7XXUART3
>         default 0xff003000 if DEBUG_U300_UART
>         default 0xffd01000 if DEBUG_HIP01_UART
>         default DEBUG_UART_PHYS if !MMU
> diff --git a/arch/arm/mach-omap1/hardware.h b/arch/arm/mach-omap1/hardware.h
> index 232b8deef907..9fc76a3c9e57 100644
> --- a/arch/arm/mach-omap1/hardware.h
> +++ b/arch/arm/mach-omap1/hardware.h
> @@ -61,7 +61,7 @@ static inline u32 omap_cs3_phys(void)
> 
>  #endif /* ifndef __ASSEMBLER__ */
> 
> -#define OMAP1_IO_OFFSET                0x01000000      /* Virtual IO
> = 0xfefb0000 */
> +#define OMAP1_IO_OFFSET                0x00fb0000      /* Virtual IO
> = 0xff000000 */
>  #define OMAP1_IO_ADDRESS(pa)   IOMEM((pa) - OMAP1_IO_OFFSET)
> 
>  #include "serial.h"

Oh OK yeah sounds like that's the issue.

> There may be additional locations that hardcode the virtual address.

Those should be in mach-omap1/io.c, and I recall innovator had some
hardcoded fpga address that should also be checked.

Regards,

Tony