Re: [PATCH 2/4] arm64/mm: wire up CONFIG_ARCH_HAS_SET_DIRECT_MAP

[Ard Biesheuvel <ard.biesheuvel@arm.com>]

On 5/28/19 10:10 AM, Anshuman Khandual wrote:
> 
> 
> On 05/23/2019 03:52 PM, Ard Biesheuvel wrote:
>> Wire up the special helper functions to manipulate aliases of vmalloc
>> regions in the linear map.
> 
> IMHO the commit message here could be bit more descriptive because of the
> amount of changes this patch brings in.
> 
>>
>> Signed-off-by: Ard Biesheuvel <ard.biesheuvel@arm.com>
>> ---
>>   arch/arm64/Kconfig                  |  1 +
>>   arch/arm64/include/asm/cacheflush.h |  3 ++
>>   arch/arm64/mm/pageattr.c            | 48 ++++++++++++++++----
>>   mm/vmalloc.c                        | 11 -----
>>   4 files changed, 44 insertions(+), 19 deletions(-)
>>
>> diff --git a/arch/arm64/Kconfig b/arch/arm64/Kconfig
>> index ca9c175fb949..4ab32180eabd 100644
>> --- a/arch/arm64/Kconfig
>> +++ b/arch/arm64/Kconfig
>> @@ -26,6 +26,7 @@ config ARM64
>>   	select ARCH_HAS_MEMBARRIER_SYNC_CORE
>>   	select ARCH_HAS_PTE_SPECIAL
>>   	select ARCH_HAS_SETUP_DMA_OPS
>> +	select ARCH_HAS_SET_DIRECT_MAP
>>   	select ARCH_HAS_SET_MEMORY
>>   	select ARCH_HAS_STRICT_KERNEL_RWX
>>   	select ARCH_HAS_STRICT_MODULE_RWX
>> diff --git a/arch/arm64/include/asm/cacheflush.h b/arch/arm64/include/asm/cacheflush.h
>> index 19844211a4e6..b9ee5510067f 100644
>> --- a/arch/arm64/include/asm/cacheflush.h
>> +++ b/arch/arm64/include/asm/cacheflush.h
>> @@ -187,4 +187,7 @@ static inline void flush_cache_vunmap(unsigned long start, unsigned long end)
>>   
>>   int set_memory_valid(unsigned long addr, int numpages, int enable);
>>   
>> +int set_direct_map_invalid_noflush(struct page *page);
>> +int set_direct_map_default_noflush(struct page *page);
>> +
>>   #endif
>> diff --git a/arch/arm64/mm/pageattr.c b/arch/arm64/mm/pageattr.c
>> index 6cd645edcf35..9c6b9039ec8f 100644
>> --- a/arch/arm64/mm/pageattr.c
>> +++ b/arch/arm64/mm/pageattr.c
>> @@ -159,17 +159,48 @@ int set_memory_valid(unsigned long addr, int numpages, int enable)
>>   					__pgprot(PTE_VALID));
>>   }
>>   
>> -#ifdef CONFIG_DEBUG_PAGEALLOC
>> +int set_direct_map_invalid_noflush(struct page *page)
>> +{
>> +	struct page_change_data data = {
>> +		.set_mask = __pgprot(0),
>> +		.clear_mask = __pgprot(PTE_VALID),
>> +	};
>> +
>> +	if (!rodata_full)
>> +		return 0;
> 
> Why rodata_full needs to be probed here ? Should not we still require the following
> transition even if rodata_full is not enabled. Just wondering whether we can use
> VM_FLUSH_RESET_PERMS feature without these required transitions.
> 
>          /*
>           * Set direct map to something invalid so that it won't be cached if
>           * there are any accesses after the TLB flush, then flush the TLB and
>           * reset the direct map permissions to the default.
>           */
>          set_area_direct_map(area, set_direct_map_invalid_noflush);
>          _vm_unmap_aliases(start, end, 1);
>          set_area_direct_map(area, set_direct_map_default_noflush);
> 
>   > +

How would that work? With rodata_full disabled, the linear region is not 
mapped down to pages, and so there is no way we can manipulate linear 
aliases at page granularity.

>> +	return apply_to_page_range(&init_mm,
>> +				   (unsigned long)page_address(page),
>> +				   PAGE_SIZE, change_page_range, &data);
>> +}
>> +
>> +int set_direct_map_default_noflush(struct page *page)
>> +{
>> +	struct page_change_data data = {
>> +		.set_mask = __pgprot(PTE_VALID | PTE_WRITE),
>> +		.clear_mask = __pgprot(PTE_RDONLY),
> 
> Replace __pgprot(PTE_VALID | PTE_WRITE) with PAGE_KERNEL instead !
> 

This is a delta mask, so no need to pull in the PTE_MAYBE_NG or other 
attributes that we know we haven't changed.

>> +	};
>> +
>> +	if (!rodata_full)
>> +		return 0;
>> +
>> +	return apply_to_page_range(&init_mm,
>> +				   (unsigned long)page_address(page),
>> +				   PAGE_SIZE, change_page_range, &data);
>> +}
>> +
> 
> IIUC set_direct_map_invalid_noflush() and set_direct_map_default_noflush()
> should set *appropriate* permissions as seen fit from platform perspective
> to implement this transition.
> 
> In here set_direct_map_invalid_noflush() makes the entry invalid preventing
> further MMU walks (hence new TLB entries). set_direct_map_default_noflush()
> makes it a valid write entry. Though it looks similar to PAGE_KERNEL which
> is the default permission for linear mapping on arm64 via __map_memblock().
> Should not PAGE_KERNEL be used explicitly as suggested above.
> 

No. We should restore the attributes that we cleared when manipulating 
the linear aliases. There isn't a lot of code that does that, so I don't 
see the need to use PAGE_KERNEL here.

>>   void __kernel_map_pages(struct page *page, int numpages, int enable)
>>   {
>> +	if (!debug_pagealloc_enabled() && !rodata_full)
>> +		return;
>> +
> 
> I guess this is not related to CONFIG_ARCH_HAS_SET_DIRECT_MAP here and should
> be a fix or an enhancement to CONFIG_DEBUG_PAGEALLOC implementation. Just
> curious, !rodata_full check here to ensure that linear mapping does not have
> block or contig mappings and should be backed by regular pages only ?
> 

It is related. CONFIG_ARCH_HAS_SET_DIRECT_MAP introduces references to 
__kernel_map_pages() in generic code, so enabling 
CONFIG_ARCH_HAS_SET_DIRECT_MAP should make __kernel_map_pages() 
available unconditionally as well.

>>   	set_memory_valid((unsigned long)page_address(page), numpages, enable);
>>   }
>> -#ifdef CONFIG_HIBERNATION
>> +
>>   /*
>> - * When built with CONFIG_DEBUG_PAGEALLOC and CONFIG_HIBERNATION, this function
>> - * is used to determine if a linear map page has been marked as not-valid by
>> - * CONFIG_DEBUG_PAGEALLOC. Walk the page table and check the PTE_VALID bit.
>> - * This is based on kern_addr_valid(), which almost does what we need.
>> + * This function is used to determine if a linear map page has been marked as
>> + * not-valid. Walk the page table and check the PTE_VALID bit. This is based
>> + * on kern_addr_valid(), which almost does what we need.
>>    *
>>    * Because this is only called on the kernel linear map,  p?d_sect() implies
>>    * p?d_present(). When debug_pagealloc is enabled, sections mappings are
>> @@ -183,6 +214,9 @@ bool kernel_page_present(struct page *page)
>>   	pte_t *ptep;
>>   	unsigned long addr = (unsigned long)page_address(page);
>>   
>> +	if (!debug_pagealloc_enabled() && !rodata_full)
>> +		return true;
>> +
> 
> Ditto, not related to CONFIG_ARCH_HAS_SET_DIRECT_MAP.
> 

It is related.