Thread (6 messages) 6 messages, 2 authors, 2021-08-31

Re: Re: [PATCH v3] Fix one error in mthca_alloc

From: 康宁 <hidden>
Date: 2021-08-31 10:41:06

&gt; 
&gt; On Fri, Aug 27, 2021 at 08:52:28AM +0800, kangning wrote:
&gt; &gt; drivers/infiniband/hw/mthca/mthca_allocator.c: alloc-&gt;last left unchanged in mthca_alloc, which
&gt; &gt; has impact on performance of function find_next_zero_bit in mthca_alloc.
&gt; 
&gt; I don't know what the sentence above means, but the change is unlikely
&gt; to be correct.
&gt; 
&gt; When alloc-&gt;last starts to be equal to alloc-&gt;max, the
&gt; find_next_zero_bit() will always return alloc-&gt;max. which will ensure
&gt; that the following code is executed.
&gt; 
&gt;    48         if (obj &gt;= alloc-&gt;max) {
&gt;    49                 alloc-&gt;top = (alloc-&gt;top + alloc-&gt;max) &amp; alloc-&gt;mask;
&gt;    50                 obj = find_first_zero_bit(alloc-&gt;table, alloc-&gt;max);
&gt;    51         }
&gt; 

Thanks for your review.
Yes, your analysis is right. However, this is a bitmap allocator for resource id allocation (like QP, and CQ). 
We first look at the situation where no resource id is released.
When the value of alloc-&gt;last starts to reaches alloc-&gt;max, the bitmap will 
be full. In this case, it's normal to return an invalid value.

Now, let's add resource releasing into consideration. 
The following code is part of mthca_free:

70	spin_lock_irqsave(&amp;alloc-&gt;lock, flags);
71
72	clear_bit(obj, alloc-&gt;table);
73	alloc-&gt;last = min(alloc-&gt;last, obj);
74	alloc-&gt;top = (alloc-&gt;top + alloc-&gt;max) &amp; alloc-&gt;mask;

mthca_free() is used to release the allocated resource id. It will update alloc-&gt;last when obj is freed. 
So, if the bitmap has space for allocation, my modification can continuously work.

&gt; 
&gt; However the mthca_alloc() function has other error, it returns -1 while
&gt; based on its declaration it needs to be unsigned,

I think you are right about this. But obj is the return value of u32 type, which is the requirement of resource id 
(though it may not fully use 32 bits). I have no idea of how to fix it.

Thanks.

&gt; 
&gt; Thanks
&gt; 
&gt; &gt; 
&gt; &gt; Signed-off-by: kangning [off-list ref]
&gt; &gt; ---
&gt; &gt;  
&gt; &gt;  I squashed two commits into one in this version.
&gt; &gt;  
&gt; &gt;  drivers/infiniband/hw/mthca/mthca_allocator.c | 3 +++
&gt; &gt;  1 file changed, 3 insertions(+)
&gt; &gt; 
&gt; &gt; diff --git a/drivers/infiniband/hw/mthca/mthca_allocator.c b/drivers/infiniband/hw/mthca/mthca_allocator.c
&gt; &gt; index aef1d274a14e..1141695093e7 100644
&gt; &gt; --- a/drivers/infiniband/hw/mthca/mthca_allocator.c
&gt; &gt; +++ b/drivers/infiniband/hw/mthca/mthca_allocator.c
&gt; &gt; @@ -51,6 +51,9 @@ u32 mthca_alloc(struct mthca_alloc *alloc)
&gt; &gt;  	}
&gt; &gt;  
&gt; &gt;  	if (obj &lt; alloc-&gt;max) {
&gt; &gt; +		alloc-&gt;last = obj + 1;
&gt; &gt; +		if (alloc-&gt;last == alloc-&gt;max)
&gt; &gt; +			alloc-&gt;last = 0;
&gt; &gt;  		set_bit(obj, alloc-&gt;table);
&gt; &gt;  		obj |= alloc-&gt;top;
&gt; &gt;  	} else
&gt; &gt; -- 
&gt; &gt; 2.17.1
&gt; &gt; 
[off-list ref]
Keyboard shortcuts
hback out one level
jnext message in thread
kprevious message in thread
ldrill in
Escclose help / fold thread tree
?toggle this help