<html>
<head>
<meta http-equiv="Content-Type" content="text/html; charset=UTF-8">
</head>
<body>
<div class="moz-cite-prefix">On 25/07/2020 02:33, Umesh Nerlige
Ramappa wrote:<br>
</div>
<blockquote type="cite"
cite="mid:20200724233319.GA45897@orsosgc001.amr.corp.intel.com">
<blockquote type="cite" style="color: #000000;"><br>
16Mb * 256 = 4Gb
<br>
<br>
That way you verify that we're not leaking GGTT space when
closing the perf fd.
<br>
<br>
You might want to tweak the noa_wait sysfs value before/after
the loop.
<br>
<br>
This might also only work on !32bits machines with enough
memory...
<br>
<br>
</blockquote>
<br>
Looks like calling close() in the above sequence will not result
in a call to i915_perf_release (because mmap holds a reference to
the file). <br>
Based on the latest patch series in the mailing list, if you see
something we can add, please let me know. Note that we block
mremap of an mmap-ped address by setting VM_DONTEXPAND in i915
perf mmap implementation.
<br>
<br>
Thanks,
<br>
Umesh
</blockquote>
<p>Oh thanks for reminding me :)</p>
<p>Since mmap holds a ref that means nobody will be able to open the
stream once more until munmap is called.</p>
<p>So there won't be any GGTT space leakage and it's all good.</p>
<p><br>
</p>
<p>Forget my request then.</p>
<p>Thanks!</p>
<p><br>
</p>
<p>-Lionel<br>
</p>
</body>
</html>