Re: [PATCH 7/9] MIPS: uprobes: Flush icache via kernel address

[Leonid Yegoshin <Leonid.Yegoshin@imgtec.com>]

On 09/21/2016 06:26 AM, Ralf Baechle wrote:
> On Thu, Sep 01, 2016 at 05:30:13PM +0100, James Hogan wrote:
>
>> Update arch_uprobe_copy_ixol() to use the kmap_atomic() based kernel
>> address to flush the icache with flush_icache_range(), rather than the
>> user mapping. We have the kernel mapping available anyway and this
>> avoids having to switch to using the new __flush_icache_user_range() for
>> the sake of Enhanced Virtual Addressing (EVA) where flush_icache_range()
>> will become ineffective on user addresses.
>>
>> Signed-off-by: James Hogan <james.hogan@imgtec.com>
>> Cc: Ralf Baechle <ralf@linux-mips.org>
>> Cc: Leonid Yegoshin <leonid.yegoshin@imgtec.com>
>> Cc: linux-mips@linux-mips.org
>> ---
>>   arch/mips/kernel/uprobes.c | 13 ++++---------
>>   1 file changed, 4 insertions(+), 9 deletions(-)
>>
>> diff --git a/arch/mips/kernel/uprobes.c b/arch/mips/kernel/uprobes.c
>> index 8452d933a645..9a507ab1ea38 100644
>> --- a/arch/mips/kernel/uprobes.c
>> +++ b/arch/mips/kernel/uprobes.c
>> @@ -301,19 +301,14 @@ int set_orig_insn(struct arch_uprobe *auprobe, struct mm_struct *mm,
>>   void __weak arch_uprobe_copy_ixol(struct page *page, unsigned long vaddr,
>>   				  void *src, unsigned long len)
>>   {
>> -	void *kaddr;
>> +	void *kaddr, kstart;
>>   
>>   	/* Initialize the slot */
>>   	kaddr = kmap_atomic(page);
>> -	memcpy(kaddr + (vaddr & ~PAGE_MASK), src, len);
>> +	kstart = kaddr + (vaddr & ~PAGE_MASK);
>> +	memcpy(kstart, src, len);
>> +	flush_icache_range(kstart, kstart + len);
>>   	kunmap_atomic(kaddr);
>> -
>> -	/*
>> -	 * The MIPS version of flush_icache_range will operate safely on
>> -	 * user space addresses and more importantly, it doesn't require a
>> -	 * VMA argument.
>> -	 */
>> -	flush_icache_range(vaddr, vaddr + len);
> I can't convince myself this is right wrt. to cache aliases ...
>
>    Ralf
>
It is incorrect if there is I-cache aliasing (very rare) and there is no 
HIGHMEM cache aliasing fix (not fixed). But top tree Linux doesn't work 
with I-cache aliasing either.

- Leonid.