Re: [PATCH v4 03/15] xenctx: Add -m (--multiple_pages) option to output larger stack

[George Dunlap <george.dunlap@eu.citrix.com>]

On 03/20/2014 01:19 AM, Don Slutz wrote:
> On 03/19/14 11:34, George Dunlap wrote:
>> On 03/18/2014 10:15 PM, Don Slutz wrote:
>>> From: Don Slutz <Don@CloudSwitch.com>
>>>
>>> Important: This is the stack size to display not the configured
>>> stack size.
>>>
>>> Using pictures (for a 3 page configured system):
>>>
>>>         +------------------+
>>>         |                  |
>>>         |                  |
>>>         |                  |
>>>         |                  |
>>>         |                  |
>>>         +------------------+
>>>         |                  |
>>>         |                  |
>>>         |                  |
>>>         |                  |
>>>         |                  |
>>>         +------------------+
>>>         |                  |
>>>         |                  |
>>>         |                  |
>>> SP --> |                  |
>>>         |                  |
>>>         +------------------+
>>
>> Sorry, what is this a picture of?  I can't make any sense out of it.  
>> Shouldn't this have only one box, the next have two, and the bottom 
>> one have three?
>>
>
> This is various pictures of a 3 page stack, and where the SP currently 
> is.  Each box is a page.  So here the "stack limit" is the end of 1 
> page.  In the sense of how much stack is used, you are right it is 1, 
> 2 and then 3.  This tracks with the value passed for "-m" (see next 
> line).

Oh, right: so you're saying that the developer running xenctx has to 
*guess* how many pages are currently in use?  In other words, this 
really means, "Display N stack pages", and if you guessed too high, 
you'll get some pages worth of garbage.

[snip]
>>>     int main(int argc, char **argv)
>>>   {
>>>       int ch;
>>>       int ret;
>>> -    static const char *sopts = "fs:hak:SC";
>>> +    static const char *sopts = "fs:hak:SCm:";
>>>       static const struct option lopts[] = {
>>>           {"stack-trace", 0, NULL, 'S'},
>>>           {"symbol-table", 1, NULL, 's'},
>>>           {"frame-pointers", 0, NULL, 'f'},
>>>           {"kernel-start", 1, NULL, 'k'},
>>> +        {"multiple-pages", 0, NULL, 'm'},
>>
>> I think I would call the long option "kernel-stack-pages" or 
>> something like that.  "Multiple pages" doesn't really convey much 
>> meaning.  -m is probably a fine short option, but -n might be more 
>> memorable.
>>
>
> The issue with "kernel-stack-pages" is that it leads to configured 
> kernel stack pages (which for the pictures above is 3).  3 is most 
> likely not the number to use here.
>
> A big part of this is that how a "kernel" knows where it is in the 
> stack can be simple like for a 2 page stack, 1 page is odd, 2nd page 
> is even.  (3 pages is most likely more complex, but fence page(s) may 
> help here.)
>
> Maybe stack-limit-in-pages is better?

Hmm -- "--display-stack-pages", and then say, "Display N pages from the 
stack pointer"?

  -George