Re: [RFC 3/4] trace_uprobe: Support SDT markers having semaphore

[Ravi Bangoria]

On 03/01/2018 07:37 PM, Masami Hiramatsu wrote:
> On Wed, 28 Feb 2018 13:23:44 +0530
> Ravi Bangoria <ravi.bangoria@xxxxxxxxxxxxxxxxxx> wrote:
>
>> @@ -502,6 +504,16 @@ static int create_trace_uprobe(int argc, char **argv)
>>  	for (i = 0; i < argc && i < MAX_TRACE_ARGS; i++) {
>>  		struct probe_arg *parg = &tu->tp.args[i];
>>  
>> +		/* This is not really an argument. */
>> +		if (argv[i][0] == '*') {
>> +			ret = kstrtoul(&(argv[i][1]), 0, &tu->sdt_offset);
>> +			if (ret) {
>> +				pr_info("Invalid semaphore address.\n");
>> +				goto error;
>> +			}
>> +			continue;
>> +		}
> So, this part should be done with parsing probe-point as I pointed.

Yes, will change it.

>> +static void sdt_increment_sem(struct trace_uprobe *tu)
>> +{
>> +	struct uprobe_map_info *info;
>> +	struct vm_area_struct *vma;
>> +	unsigned long vaddr;
>> +
>> +	uprobe_start_dup_mmap();
>> +	info = build_uprobe_map_info(tu->inode->i_mapping, tu->sdt_offset, false);
>> +	if (IS_ERR(info))
>> +		goto out;
>> +
>> +	while (info) {
>> +		down_write(&info->mm->mmap_sem);
>> +		vma = sdt_find_vma(info->mm, tu);
>> +		if (!vma)
>> +			goto cont;
>> +
>> +		vaddr = offset_to_vaddr(vma, tu->sdt_offset);
>> +		sdt_update_sem(info->mm, vaddr, 1);
>> +
>> +cont:
> Why would you use goto here?

Hmm.. sdt_find_vma() must return vma. Sure, will remove the goto.

Should I add a WARN_ON(!vma) ?

Thanks for the review,
Ravi