Re: [PATCH] arm: flush: don't abuse pfn_valid() to check if pfn is in RAM

[Robin Murphy <robin.murphy@arm.com>]

On 31/01/2024 12:59 pm, Yongqiang Liu wrote:
> Since commit a4d5613c4dc6 ("arm: extend pfn_valid to take into account
> freed memory map alignment") changes the semantics of pfn_valid() to check
> presence of the memory map for a PFN. __sync_icache_dcache() should use
> memblock_is_map_memory() instead of pfn_valid() to check if a PFN is in
> RAM or not.In Some uio case we will get a crash on a system with the
> following memory layout:
> 
>   node   0: [mem 0x00000000c0a00000-0x00000000cc8fffff]
>   node   0: [mem 0x00000000d0000000-0x00000000da1fffff]
>   the uio layout is：0xc0900000, 0x100000
> 
> the crash backtrace like:
> 
>    Unable to handle kernel paging request at virtual address bff00000
>    [...]
>    CPU: 1 PID: 465 Comm: startapp.bin Tainted: G           O      5.10.0 #1
>    Hardware name: Generic DT based system
>    PC is at b15_flush_kern_dcache_area+0x24/0x3c
>    LR is at __sync_icache_dcache+0x6c/0x98
>    [...]
>     (b15_flush_kern_dcache_area) from (__sync_icache_dcache+0x6c/0x98)
>     (__sync_icache_dcache) from (set_pte_at+0x28/0x54)
>     (set_pte_at) from (remap_pfn_range+0x1a0/0x274)
>     (remap_pfn_range) from (uio_mmap+0x184/0x1b8 [uio])
>     (uio_mmap [uio]) from (__mmap_region+0x264/0x5f4)
>     (__mmap_region) from (__do_mmap_mm+0x3ec/0x440)
>     (__do_mmap_mm) from (do_mmap+0x50/0x58)
>     (do_mmap) from (vm_mmap_pgoff+0xfc/0x188)
>     (vm_mmap_pgoff) from (ksys_mmap_pgoff+0xac/0xc4)
>     (ksys_mmap_pgoff) from (ret_fast_syscall+0x0/0x5c)
>    Code: e0801001 e2423001 e1c00003 f57ff04f (ee070f3e)
>    ---[ end trace 09cf0734c3805d52 ]---
>    Kernel panic - not syncing: Fatal exception
> 
> Fixes: a4d5613c4dc6 ("arm: extend pfn_valid to take into account freed memory map alignment")
> Signed-off-by: Yongqiang Liu <liuyongqiang13@huawei.com>
> ---
>   arch/arm/mm/flush.c | 3 ++-
>   1 file changed, 2 insertions(+), 1 deletion(-)
> 
> diff --git a/arch/arm/mm/flush.c b/arch/arm/mm/flush.c
> index d19d140a10c7..11ec6c5ff5fc 100644
> --- a/arch/arm/mm/flush.c
> +++ b/arch/arm/mm/flush.c
> @@ -15,6 +15,7 @@
>   #include <asm/smp_plat.h>
>   #include <asm/tlbflush.h>
>   #include <linux/hugetlb.h>
> +#include <linux/memblock.h>
>   
>   #include "mm.h"
>   
> @@ -292,7 +293,7 @@ void __sync_icache_dcache(pte_t pteval)
>   		/* only flush non-aliasing VIPT caches for exec mappings */
>   		return;
>   	pfn = pte_pfn(pteval);
> -	if (!pfn_valid(pfn))
> +	if (!memblock_is_map_memory(PFN_PHYS(pfn)))
>   		return;
>   
>   	folio = page_folio(pfn_to_page(pfn));

Hmm, it's a bit odd in context, since pfn_valid() obviously pairs with 
this pfn_to_page(), whereas it's not necessarily clear that 
memblock_is_map_memory() implies pfn_valid().

However, in this case we're starting from a PTE - rather than going off 
to do a slow scan of memblock to determine whether a round-trip through 
page_address() is going to give back a mapped VA, can we not trivially 
identify that from whether the PTE itself is valid?

Thanks,
Robin.