looping-in the linaro-mm-sig ML.<br><br><div class="gmail_quote">On Thu, Aug 30, 2012 at 4:47 PM, Aubertin, Guillaume <span dir="ltr">&lt;<a href="mailto:g-aubertin@ti.com" target="_blank">g-aubertin@ti.com</a>&gt;</span> wrote:<br>
<blockquote class="gmail_quote" style="margin:0 0 0 .8ex;border-left:1px #ccc solid;padding-left:1ex">hi guys,<div><br></div><div>I&#39;ve been working for a few days on getting a proper rmmod with the remoteproc/rpmsg modules, and I stumbled upon an interesting issue. </div>
<div><br></div><div>when doing sucessive memory allocation and release in the CMA reservation (by loading/unloading the firmware several times),  the following message shows up  :</div>
<div><br clear="all"><div><font color="#000099"><div>[  119.908477] cma: dma_alloc_from_contiguous(cma ed10ad00, count 256, align 8)</div><div>[  119.908843] cma: dma_alloc_from_contiguous(): memory range at c0dfb000 is busy, retrying</div>

<div>[  119.909698] cma: dma_alloc_from_contiguous(): returned c0dfd000</div></font></div><div><font color="#000099"><br></font></div><div>dma_alloc_from_contiguous() tries to allocate the following range, 0xc0dfd000, succesfully this time.</div>

<div><br></div><div>In some cases, the allocation fails after trying several ranges : </div><div><br></div><div><div style="color:rgb(0,0,153)">[  119.912231] cma: dma_alloc_from_contiguous(cma ed10ad00, count 768, align 8)</div>

<div style="color:rgb(0,0,153)">[  119.912719] cma: dma_alloc_from_contiguous(): memory range at c0dff000 is busy, retrying</div><div style="color:rgb(0,0,153)">[  119.913055] cma: dma_alloc_from_contiguous(): memory range at c0e01000 is busy, retrying</div>

<div style="color:rgb(0,0,153)">[  119.913055] rproc remoteproc0: dma_alloc_coherent failed: 3145728</div><div style="color:rgb(0,0,153)"><br></div><div>Here is my understanding so far :</div><div><br></div><div>First, even if we made a CMA reservation, the kernel can still allocate pages in this area, but these pages must be movable (user process page by example).</div>

<div style="color:rgb(0,0,153)"><br></div><div>When dma_alloc_from_contiguous() is called to allocate X pages, it looks for the next X contiguous free pages in it&#39;s CMA bitmap (with respect to the memory alignment). Then, alloc_contig_range() is called to allocate the given range of pages.  Alloc_contig_range() analyses the pages we want to allocate, and if a page is already used, it is migrated to a new page outside the page array we want to reserve. this is done using isolate_migratepages_range() to list the pages to migrate, and migrate_pages() to try to migrate the pages, and that&#39;s where it fails. Below is a list of next function calls :</div>

<div><br></div><div><font color="#000066"> fallback_migrate_page() --&gt; migrate_page()  --&gt; try_to_release_page()  --&gt; try_to_free_buffer() --&gt; drop_buffers() --&gt; buffer_busy()</font></div><div><br></div><div>

I understand here that the page contains used buffers that can&#39;t be dropped, and so the page can&#39;t be migrated. Well, I must admit that once here, I&#39;m feeling a little lost in this ocean of memory management code ;). After a few researches, I found the following thread on the linux-arm-kernel ML talking about the same issue : </div>

<div><br></div><div><a href="http://lists.infradead.org/pipermail/linux-arm-kernel/2012-June/102844.html" target="_blank">http://lists.infradead.org/pipermail/linux-arm-kernel/2012-June/102844.html</a> with the following patch :</div>
<div>
<br></div><div><div><i> mm/page_alloc.c |    3 ++-</i></div><div><i> 1 files changed, 2 insertions(+), 1 deletions(-)</i></div><div><i><br></i></div><div><i>diff --git a/mm/page_alloc.c b/mm/page_alloc.c</i></div><div><i>index 0e1c6f5..c9a6483 100644</i></div>

<div><i>--- a/mm/page_alloc.c</i></div><div><i>+++ b/mm/page_alloc.c</i></div><div><i>@@ -1310,7 +1310,8 @@ void free_hot_cold_page(struct page *page, int cold)</i></div><div><i> <span style="white-space:pre-wrap">	</span> * excessively into the page allocator</i></div>

<div><i> <span style="white-space:pre-wrap">	</span> */</i></div><div><i> <span style="white-space:pre-wrap">	</span>if (migratetype &gt;= MIGRATE_PCPTYPES) {</i></div><div><i>-<span style="white-space:pre-wrap">		</span>if (unlikely(migratetype == MIGRATE_ISOLATE)) {</i></div>

<div><i>+<span style="white-space:pre-wrap">		</span>if (unlikely(migratetype == MIGRATE_ISOLATE)</i></div><div><i>+<span style="white-space:pre-wrap">		</span>   || is_migrate_cma(migratetype)) {</i></div>
<div><i> <span style="white-space:pre-wrap">			</span>free_one_page(zone, page, 0, migratetype);</i></div><div><i> <span style="white-space:pre-wrap">			</span>goto out;</i></div><div><i> <span style="white-space:pre-wrap">		</span>}</i></div>

</div><div><br></div><div>I tried the patch, and it seems to work (I didn&#39;t have any &quot;memory range busy&quot; in 5000+ tests), but I&#39;m affraid that this could have some nasty side effects. </div><div><br></div>

<div>Any idea ? </div><div><br></div><div>Thanks in advance,</div><div>Guillaume</div><span class="HOEnZb"><font color="#888888"><div><br></div><div style="color:rgb(0,0,153)"><br></div></font></span></div><span class="HOEnZb"><font color="#888888">-- <br>
<span style="font-family:arial,sans-serif;font-size:11px;border-collapse:collapse;color:rgb(166,166,166)">Texas
      Instruments France SA, 821 Avenue Jack Kilby, 06270 Villeneuve
      Loubet. 036 420 040 R.C.S Antibes. Capital de EUR 753.920</span><br>
</font></span></div>
</blockquote></div><br><br clear="all"><div><br></div>-- <br><span style="font-family:arial,sans-serif;font-size:11px;border-collapse:collapse;color:rgb(166,166,166)">Texas
      Instruments France SA, 821 Avenue Jack Kilby, 06270 Villeneuve
      Loubet. 036 420 040 R.C.S Antibes. Capital de EUR 753.920</span><br>