Re: [PATCH v2 4/7] ARM: mm: print out correct page table entries

["Russell King (Oracle)" <linux@armlinux.org.uk>]

On Wed, Jun 02, 2021 at 03:02:43PM +0800, Kefeng Wang wrote:
> Like commit 67ce16ec15ce ("arm64: mm: print out correct page table entries")
> does, drop the struct mm_struct argument of show_pte(), print the tables
> based on the faulting address.
> 
> Signed-off-by: Kefeng Wang <wangkefeng.wang@huawei.com>

This can be misleading on 32-bit ARM.

The effective page tables for each thread are the threads *own* page
tables. There is no hardware magic for addresses above PAGE_OFFSET being
directed to the init_mm page tables.

So, when we hit a fault in kernel space, we need to be printing the
currently in-use page tables associated with the running thread.

Hence:

>  /*
> - * This is useful to dump out the page tables associated with
> - * 'addr' in mm 'mm'.
> + * Dump out the page tables associated with 'addr' in the currently active mm
>   */
> -void show_pte(const char *lvl, struct mm_struct *mm, unsigned long addr)
> +void show_pte(const char *lvl, unsigned long addr)
>  {
>  	pgd_t *pgd;
> -
> -	if (!mm)
> +	struct mm_struct *mm;
> +
> +	if (addr < TASK_SIZE) {
> +		mm = current->active_mm;
> +		if (mm == &init_mm) {
> +			printk("%s[%08lx] user address but active_mm is swapper\n",
> +				lvl, addr);
> +			return;
> +		}
> +	} else {
>  		mm = &init_mm;
> +	}

is incorrect here.

It's completely fine for architectures where kernel accesses always go
to the init_mm page tables, but for 32-bit ARM that is not the case.

-- 
RMK's Patch system: https://www.armlinux.org.uk/developer/patches/
FTTP is here! 40Mbps down 10Mbps up. Decent connectivity at last!