Re: [v4 PATCH] arm64: mm: force write fault for atomic RMW instructions

[Yang Shi <yang@os.amperecomputing.com>]

On 6/14/24 5:20 AM, Catalin Marinas wrote:
> On Wed, Jun 05, 2024 at 01:37:23PM -0700, Yang Shi wrote:
>> +static __always_inline bool aarch64_insn_is_class_cas(u32 insn)
>> +{
>> +	return aarch64_insn_is_cas(insn) ||
>> +	       aarch64_insn_is_casp(insn);
>> +}
>> +
>> +/*
>> + * Exclude unallocated atomic instructions and LD64B/LDAPR.
>> + * The masks and values were generated by using Python sympy module.
>> + */
>> +static __always_inline bool aarch64_atomic_insn_has_wr_perm(u32 insn)
>> +{
>> +	return ((insn & 0x3f207c00) == 0x38200000) ||
>> +	       ((insn & 0x3f208c00) == 0x38200000) ||
>> +	       ((insn & 0x7fe06c00) == 0x78202000) ||
>> +	       ((insn & 0xbf204c00) == 0x38200000);
>> +}
> This is still pretty opaque if we want to modify it in the future. I
> guess we could add more tests on top but it would be nice to have a way
> to re-generate these masks. I'll think about, for now these tests would
> do.

Sorry for the late reply, just came back from vacation and tried to 
catch up all the emails and TODOs. We should be able to share the tool 
used by us to generate the tests. But it may take some time.

>
>> @@ -511,6 +539,7 @@ static int __kprobes do_page_fault(unsigned long far, unsigned long esr,
>>   	unsigned long addr = untagged_addr(far);
>>   	struct vm_area_struct *vma;
>>   	int si_code;
>> +	bool may_force_write = false;
>>   
>>   	if (kprobe_page_fault(regs, esr))
>>   		return 0;
>> @@ -547,6 +576,7 @@ static int __kprobes do_page_fault(unsigned long far, unsigned long esr,
>>   		/* If EPAN is absent then exec implies read */
>>   		if (!alternative_has_cap_unlikely(ARM64_HAS_EPAN))
>>   			vm_flags |= VM_EXEC;
>> +		may_force_write = true;
>>   	}
>>   
>>   	if (is_ttbr0_addr(addr) && is_el1_permission_fault(addr, esr, regs)) {
>> @@ -568,6 +598,12 @@ static int __kprobes do_page_fault(unsigned long far, unsigned long esr,
>>   	if (!vma)
>>   		goto lock_mmap;
>>   
>> +	if (may_force_write && (vma->vm_flags & VM_WRITE) &&
>> +	    is_el0_atomic_instr(regs)) {
>> +		vm_flags = VM_WRITE;
>> +		mm_flags |= FAULT_FLAG_WRITE;
>> +	}
> I think we can get rid of may_force_write and just test (vm_flags &
> VM_READ).

Yes, will fix it in v5.
>