[v4,4/6] arm: add early_ioremap support

This patch uses the generic early_ioremap code to implement
early_ioremap for ARM. The ARM-specific bits come mostly from
an earlier patch from Leif Lindholm <leif.lindholm@linaro.org>
here:

  https://lkml.org/lkml/2013/10/3/279

Signed-off-by: Mark Salter <msalter@redhat.com>
Tested-by: Leif Lindholm <leif.lindholm@linaro.org>
Acked-by: Catalin Marinas <catalin.marinas@arm.com>
---
 arch/arm/Kconfig              | 10 +++++
 arch/arm/include/asm/Kbuild   |  1 +
 arch/arm/include/asm/fixmap.h | 20 ++++++++++
 arch/arm/include/asm/io.h     |  1 +
 arch/arm/kernel/setup.c       |  2 +
 arch/arm/mm/Makefile          |  4 ++
 arch/arm/mm/early_ioremap.c   | 93 +++++++++++++++++++++++++++++++++++++++++++
 arch/arm/mm/mmu.c             |  2 +
 8 files changed, 133 insertions(+)
 create mode 100644 arch/arm/mm/early_ioremap.c

Hi Rob,

Thanks for having a look.
Since I'm at least partially responsible for the below, I'll respond
before Mark wakes up.

On Tue, Feb 25, 2014 at 11:48:19PM -0600, Rob Herring wrote:
> On Wed, Feb 12, 2014 at 2:56 PM, Mark Salter <msalter@redhat.com> wrote:
> > This patch uses the generic early_ioremap code to implement
> > early_ioremap for ARM. The ARM-specific bits come mostly from
> > an earlier patch from Leif Lindholm <leif.lindholm@linaro.org>
> > here:
> 
> This doesn't actually work for me. The PTE flags are not correct and
> accesses to a device fault. See below.

Do they fault before paging_init()?
If not, see below.

> >
> >   https://lkml.org/lkml/2013/10/3/279
> >
> > Signed-off-by: Mark Salter <msalter@redhat.com>
> > Tested-by: Leif Lindholm <leif.lindholm@linaro.org>
> > Acked-by: Catalin Marinas <catalin.marinas@arm.com>
> > ---
> >  arch/arm/Kconfig              | 10 +++++
> >  arch/arm/include/asm/Kbuild   |  1 +
> >  arch/arm/include/asm/fixmap.h | 20 ++++++++++
> >  arch/arm/include/asm/io.h     |  1 +
> >  arch/arm/kernel/setup.c       |  2 +
> >  arch/arm/mm/Makefile          |  4 ++
> >  arch/arm/mm/early_ioremap.c   | 93 +++++++++++++++++++++++++++++++++++++++++++
> >  arch/arm/mm/mmu.c             |  2 +
> >  8 files changed, 133 insertions(+)
> >  create mode 100644 arch/arm/mm/early_ioremap.c
> >
> > diff --git a/arch/arm/Kconfig b/arch/arm/Kconfig
> > index e254198..7a35ef6 100644
> > --- a/arch/arm/Kconfig
> > +++ b/arch/arm/Kconfig
> > @@ -1874,6 +1874,16 @@ config UACCESS_WITH_MEMCPY
> >           However, if the CPU data cache is using a write-allocate mode,
> >           this option is unlikely to provide any performance gain.
> >
> > +config EARLY_IOREMAP
> > +       bool "Provide early_ioremap() support for kernel initialization"
> > +       select GENERIC_EARLY_IOREMAP
> > +       help
> > +         Provide a mechanism for kernel initialisation code to temporarily
> > +         map, in a highmem-agnostic way, memory pages in before ioremap()
> > +         and friends are available (before paging_init() has run). It uses
> > +         the same virtual memory range as kmap so all early mappings must
> > +         be unmapped before paging_init() is called.
> > +

^^

> >  config SECCOMP
> >         bool
> >         prompt "Enable seccomp to safely compute untrusted bytecode"
> > diff --git a/arch/arm/include/asm/Kbuild b/arch/arm/include/asm/Kbuild
> > index 3278afe..6fcfd7d 100644
> > --- a/arch/arm/include/asm/Kbuild
> > +++ b/arch/arm/include/asm/Kbuild
> > @@ -4,6 +4,7 @@ generic-y += auxvec.h
> >  generic-y += bitsperlong.h
> >  generic-y += cputime.h
> >  generic-y += current.h
> > +generic-y += early_ioremap.h
> >  generic-y += emergency-restart.h
> >  generic-y += errno.h
> >  generic-y += exec.h
> > diff --git a/arch/arm/include/asm/fixmap.h b/arch/arm/include/asm/fixmap.h
> > index 68ea615..ff8fa3e 100644
> > --- a/arch/arm/include/asm/fixmap.h
> > +++ b/arch/arm/include/asm/fixmap.h
> > @@ -21,8 +21,28 @@ enum fixed_addresses {
> >         FIX_KMAP_BEGIN,
> >         FIX_KMAP_END = (FIXADDR_TOP - FIXADDR_START) >> PAGE_SHIFT,
> >         __end_of_fixed_addresses
> > +/*
> > + * 224 temporary boot-time mappings, used by early_ioremap(),
> > + * before ioremap() is functional.
> > + *
> > + * (P)re-using the FIXADDR region, which is used for highmem
> > + * later on, and statically aligned to 1MB.
> > + */
> > +#define NR_FIX_BTMAPS          32
> > +#define FIX_BTMAPS_SLOTS       7
> > +#define TOTAL_FIX_BTMAPS       (NR_FIX_BTMAPS * FIX_BTMAPS_SLOTS)
> > +#define FIX_BTMAP_END          FIX_KMAP_BEGIN
> > +#define FIX_BTMAP_BEGIN                (FIX_BTMAP_END + TOTAL_FIX_BTMAPS - 1)
> 
> Why the different logic from arm64? Specifically, it doesn't make
> adding a permanent mapping simple.

Making a permanent mapping using this would require either:
- Not using the fixmap region.
- Rewriting arm kmap.

> >  };
> >
> > +#define FIXMAP_PAGE_COMMON (L_PTE_YOUNG | L_PTE_PRESENT | L_PTE_XN)
> > +
> > +#define FIXMAP_PAGE_NORMAL (FIXMAP_PAGE_COMMON | L_PTE_MT_WRITEBACK)
> > +#define FIXMAP_PAGE_IO    (FIXMAP_PAGE_COMMON | L_PTE_MT_DEV_NONSHARED)
> 
> This should be L_PTE_MT_DEV_SHARED and also needs L_PTE_DIRTY and
> L_PTE_SHARED as we want this to match MT_DEVICE.
> 
> These should also be wrapped with __pgprot().

Ok.

> > +
> > +extern void __early_set_fixmap(enum fixed_addresses idx,
> > +                              phys_addr_t phys, pgprot_t flags);
> > +
> >  #include <asm-generic/fixmap.h>
> >
> >  #endif
> > diff --git a/arch/arm/include/asm/io.h b/arch/arm/include/asm/io.h
> > index 8aa4cca..637e0cd 100644
> > --- a/arch/arm/include/asm/io.h
> > +++ b/arch/arm/include/asm/io.h
> > @@ -28,6 +28,7 @@
> >  #include <asm/byteorder.h>
> >  #include <asm/memory.h>
> >  #include <asm-generic/pci_iomap.h>
> > +#include <asm/early_ioremap.h>
> >  #include <xen/xen.h>
> >
> >  /*
> > diff --git a/arch/arm/kernel/setup.c b/arch/arm/kernel/setup.c
> > index b0df976..9c8b751 100644
> > --- a/arch/arm/kernel/setup.c
> > +++ b/arch/arm/kernel/setup.c
> > @@ -30,6 +30,7 @@
> >  #include <linux/bug.h>
> >  #include <linux/compiler.h>
> >  #include <linux/sort.h>
> > +#include <linux/io.h>
> >
> >  #include <asm/unified.h>
> >  #include <asm/cp15.h>
> > @@ -880,6 +881,7 @@ void __init setup_arch(char **cmdline_p)
> >         const struct machine_desc *mdesc;
> >
> >         setup_processor();
> > +       early_ioremap_init();
> >         mdesc = setup_machine_fdt(__atags_pointer);
> >         if (!mdesc)
> >                 mdesc = setup_machine_tags(__atags_pointer, __machine_arch_type);
> > diff --git a/arch/arm/mm/Makefile b/arch/arm/mm/Makefile
> > index 7f39ce2..501af98 100644
> > --- a/arch/arm/mm/Makefile
> > +++ b/arch/arm/mm/Makefile
> > @@ -12,6 +12,10 @@ ifneq ($(CONFIG_MMU),y)
> >  obj-y                          += nommu.o
> >  endif
> >
> > +ifeq ($(CONFIG_MMU),y)
> > +obj-$(CONFIG_EARLY_IOREMAP)    += early_ioremap.o
> > +endif
> > +
> >  obj-$(CONFIG_ARM_PTDUMP)       += dump.o
> >  obj-$(CONFIG_MODULES)          += proc-syms.o
> >
> > diff --git a/arch/arm/mm/early_ioremap.c b/arch/arm/mm/early_ioremap.c
> > new file mode 100644
> > index 0000000..c3e2bf2
> > --- /dev/null
> > +++ b/arch/arm/mm/early_ioremap.c
> > @@ -0,0 +1,93 @@
> > +/*
> > + * early_ioremap() support for ARM
> > + *
> > + * Based on existing support in arch/x86/mm/ioremap.c
> > + *
> > + * Restrictions: currently only functional before paging_init()
> > + */
> > +
> > +#include <linux/init.h>
> > +#include <linux/io.h>
> 
> io.h doesn't appear to be needed.

No, not in this version.

> > +
> > +#include <asm/fixmap.h>
> > +#include <asm/pgalloc.h>
> > +#include <asm/pgtable.h>
> > +#include <asm/tlbflush.h>
> > +
> > +#include <asm/mach/map.h>
> > +
> > +static pte_t bm_pte[PTRS_PER_PTE] __aligned(PTE_HWTABLE_SIZE) __initdata;
> > +
> > +static inline pmd_t * __init early_ioremap_pmd(unsigned long addr)
> > +{
> > +       unsigned int index = pgd_index(addr);
> > +       pgd_t *pgd = cpu_get_pgd() + index;
> > +       pud_t *pud = pud_offset(pgd, addr);
> > +       pmd_t *pmd = pmd_offset(pud, addr);
> > +
> > +       return pmd;
> > +}
> > +
> > +static inline pte_t * __init early_ioremap_pte(unsigned long addr)
> > +{
> > +       return &bm_pte[pte_index(addr)];
> > +}
> > +
> > +void __init early_ioremap_init(void)
> > +{
> > +       pmd_t *pmd;
> > +
> > +       pmd = early_ioremap_pmd(fix_to_virt(FIX_BTMAP_BEGIN));
> > +
> > +       pmd_populate_kernel(NULL, pmd, bm_pte);
> > +
> > +       /*
> > +        * Make sure we don't span multiple pmds.
> > +        */
> > +       BUILD_BUG_ON((__fix_to_virt(FIX_BTMAP_BEGIN) >> PMD_SHIFT)
> > +                    != (__fix_to_virt(FIX_BTMAP_END) >> PMD_SHIFT));
> > +
> > +       if (pmd != early_ioremap_pmd(fix_to_virt(FIX_BTMAP_END))) {
> > +               WARN_ON(1);
> > +               pr_warn("pmd %p != %p\n",
> > +                       pmd, early_ioremap_pmd(fix_to_virt(FIX_BTMAP_END)));
> > +               pr_warn("fix_to_virt(FIX_BTMAP_BEGIN): %08lx\n",
> > +                       fix_to_virt(FIX_BTMAP_BEGIN));
> > +               pr_warn("fix_to_virt(FIX_BTMAP_END):   %08lx\n",
> > +                       fix_to_virt(FIX_BTMAP_END));
> > +               pr_warn("FIX_BTMAP_END:       %d\n", FIX_BTMAP_END);
> > +               pr_warn("FIX_BTMAP_BEGIN:     %d\n", FIX_BTMAP_BEGIN);
> > +       }
> > +
> > +       early_ioremap_setup();
> > +}
> > +
> > +void __init __early_set_fixmap(enum fixed_addresses idx,
> > +                              phys_addr_t phys, pgprot_t flags)
> > +{
> > +       unsigned long addr = __fix_to_virt(idx);
> > +       pte_t *pte;
> > +       u64 desc;
> > +
> > +       if (idx > FIX_KMAP_END) {
> > +               BUG();
> > +               return;
> > +       }
> > +       pte = early_ioremap_pte(addr);
> > +
> > +       if (pgprot_val(flags))
> > +               set_pte_at(NULL, 0xfff00000, pte,
> 
> Couldn't you use addr here instead of 0xfff00000? It's not really used
> other than a check against TASK_SIZE.

Sure.

> > +                          pfn_pte(phys >> PAGE_SHIFT, flags));
> 
> phys_to_pfn(phys)

Stolen like that from x86 :)
Sure.

> > +       else
> > +               pte_clear(NULL, addr, pte);
> > +       flush_tlb_kernel_range(addr, addr + PAGE_SIZE);
> > +       desc = *pte;
> > +}
> > +
> > +void __init
> > +early_ioremap_shutdown(void)
> > +{
> > +       pmd_t *pmd;
> > +       pmd = early_ioremap_pmd(fix_to_virt(FIX_BTMAP_BEGIN));
> > +       pmd_clear(pmd);
> 
> This is redundant with the clearing done in devicemaps_init. Not a big
> deal but this is probably something we don't want with permanent
> mappings. I'm still trying to figure out how to do those. Leaving this
> page table in place doesn't seem to work, so I think we'll have to
> copy mappings to the new page tables.

As described in the Kconfig option, and more explicitly in the
documentation included with my last submission (last summer), these
mappings don't stick around.

/
    Leif

On Tue, 2014-02-25 at 23:48 -0600, Rob Herring wrote:
> > +#define NR_FIX_BTMAPS          32
> > +#define FIX_BTMAPS_SLOTS       7
> > +#define TOTAL_FIX_BTMAPS       (NR_FIX_BTMAPS * FIX_BTMAPS_SLOTS)
> > +#define FIX_BTMAP_END          FIX_KMAP_BEGIN
> > +#define FIX_BTMAP_BEGIN                (FIX_BTMAP_END +
> TOTAL_FIX_BTMAPS - 1)
> 
> Why the different logic from arm64? Specifically, it doesn't make
> adding a permanent mapping simple.

I looked at adding support for permanent mappings but it was going to
take more time than I had. Also on ARM, we have to deal with kmap's
needs as well. Working that out was going to take more time than I
had. I think getting the patch in now to support early_ioremap is
the way to go. Adding support for permanent mappings can be done
later with the early console support for which it is needed.