Re: [PATCH] x86/split_lock: Sanitize userspace and guest error output

[Xiaoyao Li <xiaoyao.li@intel.com>]

On 6/6/2020 12:42 AM, Prarit Bhargava wrote:
> 
> 
> On 6/5/20 11:29 AM, Xiaoyao Li wrote:
>> On 6/5/2020 7:44 PM, Prarit Bhargava wrote:
>>> There are two problems with kernel messages in fatal mode that
>>> were found during testing of guests and userspace programs.
>>>
>>> The first is that no kernel message is output when the split lock detector
>>> is triggered with a userspace program.  As a result the userspace process
>>> dies from receiving SIGBUS with no indication to the user of what caused
>>> the process to die.
>>>
>>> The second problem is that only the first triggering guest causes a kernel
>>> message to be output because the message is output with pr_warn_once().
>>> This also results in a loss of information to the user.
>>>
>>> While fixing these I noticed that the same message was being output
>>> three times so I'm cleaning that up too.
>>>
>>> Fix fatal mode output, and use consistent messages for fatal and
>>> warn modes for both userspace and guests.
>>>
>>> Signed-off-by: Prarit Bhargava <prarit@redhat.com>
>>> Cc: Thomas Gleixner <tglx@linutronix.de>
>>> Cc: Ingo Molnar <mingo@redhat.com>
>>> Cc: Borislav Petkov <bp@alien8.de>
>>> Cc: x86@kernel.org
>>> Cc: "H. Peter Anvin" <hpa@zytor.com>
>>> Cc: Tony Luck <tony.luck@intel.com>
>>> Cc: "Peter Zijlstra (Intel)" <peterz@infradead.org>
>>> Cc: Sean Christopherson <sean.j.christopherson@intel.com>
>>> Cc: Rahul Tanwar <rahul.tanwar@linux.intel.com>
>>> Cc: Xiaoyao Li <xiaoyao.li@intel.com>
>>> Cc: Ricardo Neri <ricardo.neri-calderon@linux.intel.com>
>>> Cc: Dave Hansen <dave.hansen@linux.intel.com>
>>> ---
>>>    arch/x86/kernel/cpu/intel.c | 24 ++++++++++--------------
>>>    1 file changed, 10 insertions(+), 14 deletions(-)
>>>
>>> diff --git a/arch/x86/kernel/cpu/intel.c b/arch/x86/kernel/cpu/intel.c
>>> index 166d7c355896..463022aa9b7a 100644
>>> --- a/arch/x86/kernel/cpu/intel.c
>>> +++ b/arch/x86/kernel/cpu/intel.c
>>> @@ -1074,10 +1074,14 @@ static void split_lock_init(void)
>>>        split_lock_verify_msr(sld_state != sld_off);
>>>    }
>>>    -static void split_lock_warn(unsigned long ip)
>>> +static bool split_lock_warn(unsigned long ip, int fatal)
>>>    {
>>> -    pr_warn_ratelimited("#AC: %s/%d took a split_lock trap at address: 0x%lx\n",
>>> -                current->comm, current->pid, ip);
>>> +    pr_warn_ratelimited("#AC: %s/%d %ssplit_lock trap at address: 0x%lx\n",
>>> +                current->comm, current->pid,
>>> +                sld_state == sld_fatal ? "fatal " : "", ip);
>>> +
>>> +    if (sld_state == sld_fatal || fatal)
>>> +        return false;
>>>          /*
>>>         * Disable the split lock detection for this task so it can make
>>> @@ -1086,18 +1090,13 @@ static void split_lock_warn(unsigned long ip)
>>>         */
>>>        sld_update_msr(false);
>>>        set_tsk_thread_flag(current, TIF_SLD);
>>> +    return true;
>>>    }
>>>      bool handle_guest_split_lock(unsigned long ip)
>>>    {
>>> -    if (sld_state == sld_warn) {
>>> -        split_lock_warn(ip);
>>> +    if (split_lock_warn(ip, 0))
>>>            return true;
>>> -    }
>>> -
>>> -    pr_warn_once("#AC: %s/%d %s split_lock trap at address: 0x%lx\n",
>>> -             current->comm, current->pid,
>>> -             sld_state == sld_fatal ? "fatal" : "bogus", ip);
>>>          current->thread.error_code = 0;
>>>        current->thread.trap_nr = X86_TRAP_AC;
>>> @@ -1108,10 +1107,7 @@ EXPORT_SYMBOL_GPL(handle_guest_split_lock);
>>>      bool handle_user_split_lock(struct pt_regs *regs, long error_code)
>>>    {
>>> -    if ((regs->flags & X86_EFLAGS_AC) || sld_state == sld_fatal)
>>> -        return false;
>>> -    split_lock_warn(regs->ip);
>>> -    return true;
>>> +    return split_lock_warn(regs->ip, regs->flags & X86_EFLAGS_AC);
>>
>> It's incorrect. You change the behavior that it will print the split lock
>> warning even when CPL 3 Alignment Check is turned on.
> 
> Do you want the message to be displayed in the fatal case of CPL 3 Alignment check?
> 

No. It should never be displayed if #AC happens in CPL 3 and 
X86_EFLAGS_AC is set. In this case, an unaligned access triggers #AC 
regardless of #LOCK prefix. What's more, even there is a #LOCK prefix, 
we still cannot tell the cause because we don't know the priority of 
legacy alignment check #AC and split lock #AC.

If you do want a message, we can only say "unaligned access at address xxx".