[PATCH] arm64: kill flush_cache_all()

Ard Biesheuvel ard.biesheuvel at linaro.org
Mon Apr 20 10:23:00 PDT 2015


On 20 April 2015 at 11:24, Mark Rutland <mark.rutland at arm.com> wrote:
> The documented semantics of flush_cache_all are not possible to provide
> for arm64 (short of flushing the entire physical address space by VA),
> and there are currently no users; KVM uses VA maintenance exclusively,
> cpu_reset is never called, and the only two users outside of arch code
> cannot be built for arm64.
>
> While cpu_soft_reset and related functions (which call flush_cache_all)
> were thought to be useful for kexec, their current implementations only
> serve to mask bugs. For correctness kexec will need to perform
> maintenance by VA anyway to account for system caches, line migration,
> and other subtleties of the cache architecture. As the extent of this
> cache maintenance will be kexec-specific, it should probably live in the
> kexec code.
>
> This patch removes flush_cache_all, and related unused components,
> preventing further abuse.
>

While I agree fully with the general purpose of this patch, i.e., to
prevent set/way operations to be abused for managing coherency,
perhaps it would make sense to retain/repurpose some of this code as
the 'supported' way of putting the CPU in the mode that is mandated by
the boot protocol, so that it can be shared between the EFI stub and
kexec.

I think the former is not entirely safe at the moment, since the only
region we clean/invalidate [by VA] is the memory region containing the
copied kernel image, and the remainder of the efi-entry.S code itself.
I may be wrong, but I think that means that any cached software state
owned by the runtime services [including virtual remappings] could
potentially get lost when we disable the dcache without cleaning it by
set/way first.

-- 
Ard.



> Signed-off-by: Mark Rutland <mark.rutland at arm.com>
> Cc: AKASHI Takahiro <takahiro.akashi at linaro.org>
> Cc: Ard Biesheuvel <ard.biesheuvel at linaro.org>
> Cc: Catalin Marinas <catalin.marinas at arm.com>
> Cc: Geoff Levand <geoff at infradead.org>
> Cc: Lorenzo Pieralisi <lorenzo.pieralisi at arm.com>
> Cc: Marc Zyngier <marc.zyngier at arm.com>
> Cc: Will Deacon <will.deacon at arm.com>
> ---
>  arch/arm64/include/asm/cacheflush.h  |  5 ---
>  arch/arm64/include/asm/proc-fns.h    |  4 --
>  arch/arm64/include/asm/system_misc.h |  1 -
>  arch/arm64/kernel/process.c          | 12 +-----
>  arch/arm64/mm/cache.S                | 73 ------------------------------------
>  arch/arm64/mm/flush.c                |  1 -
>  arch/arm64/mm/proc.S                 | 46 -----------------------
>  7 files changed, 1 insertion(+), 141 deletions(-)
>
> diff --git a/arch/arm64/include/asm/cacheflush.h b/arch/arm64/include/asm/cacheflush.h
> index 67d309c..c75b8d0 100644
> --- a/arch/arm64/include/asm/cacheflush.h
> +++ b/arch/arm64/include/asm/cacheflush.h
> @@ -40,10 +40,6 @@
>   *     the implementation assumes non-aliasing VIPT D-cache and (aliasing)
>   *     VIPT or ASID-tagged VIVT I-cache.
>   *
> - *     flush_cache_all()
> - *
> - *             Unconditionally clean and invalidate the entire cache.
> - *
>   *     flush_cache_mm(mm)
>   *
>   *             Clean and invalidate all user space cache entries
> @@ -69,7 +65,6 @@
>   *             - kaddr  - page address
>   *             - size   - region size
>   */
> -extern void flush_cache_all(void);
>  extern void flush_cache_range(struct vm_area_struct *vma, unsigned long start, unsigned long end);
>  extern void flush_icache_range(unsigned long start, unsigned long end);
>  extern void __flush_dcache_area(void *addr, size_t len);
> diff --git a/arch/arm64/include/asm/proc-fns.h b/arch/arm64/include/asm/proc-fns.h
> index 4d9ede7..06732c8 100644
> --- a/arch/arm64/include/asm/proc-fns.h
> +++ b/arch/arm64/include/asm/proc-fns.h
> @@ -28,12 +28,8 @@
>  struct mm_struct;
>  struct cpu_suspend_ctx;
>
> -extern void cpu_cache_off(void);
>  extern void cpu_do_idle(void);
>  extern void cpu_do_switch_mm(unsigned long pgd_phys, struct mm_struct *mm);
> -extern void cpu_reset(unsigned long addr) __attribute__((noreturn));
> -void cpu_soft_restart(phys_addr_t cpu_reset,
> -               unsigned long addr) __attribute__((noreturn));
>  extern void cpu_do_suspend(struct cpu_suspend_ctx *ptr);
>  extern u64 cpu_do_resume(phys_addr_t ptr, u64 idmap_ttbr);
>
> diff --git a/arch/arm64/include/asm/system_misc.h b/arch/arm64/include/asm/system_misc.h
> index 7a18fab..659fbf5 100644
> --- a/arch/arm64/include/asm/system_misc.h
> +++ b/arch/arm64/include/asm/system_misc.h
> @@ -41,7 +41,6 @@ struct mm_struct;
>  extern void show_pte(struct mm_struct *mm, unsigned long addr);
>  extern void __show_regs(struct pt_regs *);
>
> -void soft_restart(unsigned long);
>  extern void (*arm_pm_restart)(enum reboot_mode reboot_mode, const char *cmd);
>
>  #define UDBG_UNDEFINED (1 << 0)
> diff --git a/arch/arm64/kernel/process.c b/arch/arm64/kernel/process.c
> index c6b1f3b9..c506bee 100644
> --- a/arch/arm64/kernel/process.c
> +++ b/arch/arm64/kernel/process.c
> @@ -58,14 +58,6 @@ unsigned long __stack_chk_guard __read_mostly;
>  EXPORT_SYMBOL(__stack_chk_guard);
>  #endif
>
> -void soft_restart(unsigned long addr)
> -{
> -       setup_mm_for_reboot();
> -       cpu_soft_restart(virt_to_phys(cpu_reset), addr);
> -       /* Should never get here */
> -       BUG();
> -}
> -
>  /*
>   * Function pointers to optional machine specific functions
>   */
> @@ -136,9 +128,7 @@ void machine_power_off(void)
>
>  /*
>   * Restart requires that the secondary CPUs stop performing any activity
> - * while the primary CPU resets the system. Systems with a single CPU can
> - * use soft_restart() as their machine descriptor's .restart hook, since that
> - * will cause the only available CPU to reset. Systems with multiple CPUs must
> + * while the primary CPU resets the system. Systems with multiple CPUs must
>   * provide a HW restart implementation, to ensure that all CPUs reset at once.
>   * This is required so that any code running after reset on the primary CPU
>   * doesn't have to co-ordinate with other CPUs to ensure they aren't still
> diff --git a/arch/arm64/mm/cache.S b/arch/arm64/mm/cache.S
> index 2560e1e..f563e9a 100644
> --- a/arch/arm64/mm/cache.S
> +++ b/arch/arm64/mm/cache.S
> @@ -27,79 +27,6 @@
>  #include "proc-macros.S"
>
>  /*
> - *     __flush_dcache_all()
> - *
> - *     Flush the whole D-cache.
> - *
> - *     Corrupted registers: x0-x7, x9-x11
> - */
> -__flush_dcache_all:
> -       dmb     sy                              // ensure ordering with previous memory accesses
> -       mrs     x0, clidr_el1                   // read clidr
> -       and     x3, x0, #0x7000000              // extract loc from clidr
> -       lsr     x3, x3, #23                     // left align loc bit field
> -       cbz     x3, finished                    // if loc is 0, then no need to clean
> -       mov     x10, #0                         // start clean at cache level 0
> -loop1:
> -       add     x2, x10, x10, lsr #1            // work out 3x current cache level
> -       lsr     x1, x0, x2                      // extract cache type bits from clidr
> -       and     x1, x1, #7                      // mask of the bits for current cache only
> -       cmp     x1, #2                          // see what cache we have at this level
> -       b.lt    skip                            // skip if no cache, or just i-cache
> -       save_and_disable_irqs x9                // make CSSELR and CCSIDR access atomic
> -       msr     csselr_el1, x10                 // select current cache level in csselr
> -       isb                                     // isb to sych the new cssr&csidr
> -       mrs     x1, ccsidr_el1                  // read the new ccsidr
> -       restore_irqs x9
> -       and     x2, x1, #7                      // extract the length of the cache lines
> -       add     x2, x2, #4                      // add 4 (line length offset)
> -       mov     x4, #0x3ff
> -       and     x4, x4, x1, lsr #3              // find maximum number on the way size
> -       clz     w5, w4                          // find bit position of way size increment
> -       mov     x7, #0x7fff
> -       and     x7, x7, x1, lsr #13             // extract max number of the index size
> -loop2:
> -       mov     x9, x4                          // create working copy of max way size
> -loop3:
> -       lsl     x6, x9, x5
> -       orr     x11, x10, x6                    // factor way and cache number into x11
> -       lsl     x6, x7, x2
> -       orr     x11, x11, x6                    // factor index number into x11
> -       dc      cisw, x11                       // clean & invalidate by set/way
> -       subs    x9, x9, #1                      // decrement the way
> -       b.ge    loop3
> -       subs    x7, x7, #1                      // decrement the index
> -       b.ge    loop2
> -skip:
> -       add     x10, x10, #2                    // increment cache number
> -       cmp     x3, x10
> -       b.gt    loop1
> -finished:
> -       mov     x10, #0                         // swith back to cache level 0
> -       msr     csselr_el1, x10                 // select current cache level in csselr
> -       dsb     sy
> -       isb
> -       ret
> -ENDPROC(__flush_dcache_all)
> -
> -/*
> - *     flush_cache_all()
> - *
> - *     Flush the entire cache system.  The data cache flush is now achieved
> - *     using atomic clean / invalidates working outwards from L1 cache. This
> - *     is done using Set/Way based cache maintainance instructions.  The
> - *     instruction cache can still be invalidated back to the point of
> - *     unification in a single instruction.
> - */
> -ENTRY(flush_cache_all)
> -       mov     x12, lr
> -       bl      __flush_dcache_all
> -       mov     x0, #0
> -       ic      ialluis                         // I+BTB cache invalidate
> -       ret     x12
> -ENDPROC(flush_cache_all)
> -
> -/*
>   *     flush_icache_range(start,end)
>   *
>   *     Ensure that the I and D caches are coherent within specified region.
> diff --git a/arch/arm64/mm/flush.c b/arch/arm64/mm/flush.c
> index b6f14e8..4dfa397 100644
> --- a/arch/arm64/mm/flush.c
> +++ b/arch/arm64/mm/flush.c
> @@ -102,7 +102,6 @@ EXPORT_SYMBOL(flush_dcache_page);
>  /*
>   * Additional functions defined in assembly.
>   */
> -EXPORT_SYMBOL(flush_cache_all);
>  EXPORT_SYMBOL(flush_icache_range);
>
>  #ifdef CONFIG_TRANSPARENT_HUGEPAGE
> diff --git a/arch/arm64/mm/proc.S b/arch/arm64/mm/proc.S
> index cdd754e..39139a3 100644
> --- a/arch/arm64/mm/proc.S
> +++ b/arch/arm64/mm/proc.S
> @@ -46,52 +46,6 @@
>  #define MAIR(attr, mt) ((attr) << ((mt) * 8))
>
>  /*
> - *     cpu_cache_off()
> - *
> - *     Turn the CPU D-cache off.
> - */
> -ENTRY(cpu_cache_off)
> -       mrs     x0, sctlr_el1
> -       bic     x0, x0, #1 << 2                 // clear SCTLR.C
> -       msr     sctlr_el1, x0
> -       isb
> -       ret
> -ENDPROC(cpu_cache_off)
> -
> -/*
> - *     cpu_reset(loc)
> - *
> - *     Perform a soft reset of the system.  Put the CPU into the same state
> - *     as it would be if it had been reset, and branch to what would be the
> - *     reset vector. It must be executed with the flat identity mapping.
> - *
> - *     - loc   - location to jump to for soft reset
> - */
> -       .align  5
> -ENTRY(cpu_reset)
> -       mrs     x1, sctlr_el1
> -       bic     x1, x1, #1
> -       msr     sctlr_el1, x1                   // disable the MMU
> -       isb
> -       ret     x0
> -ENDPROC(cpu_reset)
> -
> -ENTRY(cpu_soft_restart)
> -       /* Save address of cpu_reset() and reset address */
> -       mov     x19, x0
> -       mov     x20, x1
> -
> -       /* Turn D-cache off */
> -       bl      cpu_cache_off
> -
> -       /* Push out all dirty data, and ensure cache is empty */
> -       bl      flush_cache_all
> -
> -       mov     x0, x20
> -       ret     x19
> -ENDPROC(cpu_soft_restart)
> -
> -/*
>   *     cpu_do_idle()
>   *
>   *     Idle the processor (wait for interrupt).
> --
> 1.9.1
>



More information about the linux-arm-kernel mailing list