<html> <head> <meta http-equiv="Content-Type" content="text/html; charset=utf-8"> <meta name="Generator" content="Microsoft Exchange Server">  <style></style> </head> <body> <font face="Calibri" size="2"><span style="font-size:11pt;"> <div>aubinator.c: In function ‘parse_commands’:</div> <div>aubinator.c:768:19: warning: suggest parentheses around comparison in operand of ‘&’ [-Wparentheses]</div> <div> if (p[0] & (1 << 22) == 0)</div> <a name="_MailEndCompose"></a> <div> </div> <div>-----Original Message-----<br> From: mesa-dev [<a href="mailto:mesa-dev-bounces@lists.freedesktop.org">mailto:mesa-dev-bounces@lists.freedesktop.org</a>] On Behalf Of Jason Ekstrand<br> Sent: Thursday, September 08, 2016 9:12 PM<br> To: mesa-dev@lists.freedesktop.org<br> Cc: Ekstrand, Jason <jason.ekstrand@intel.com><br> Subject: [Mesa-dev] [PATCH] intel/aubinator: Properly handle batch buffer chaining</div> <div> </div> <div>The original aubinator that Kristian wrote had a bug in the handling of MI_BATCH_BUFFER_START that propagated into the version in upstream mesa.</div> <div>Say you have two batch buffers A and B where A calls MI_BATCH_BUFFER_START to jump to B. Now suppose that A and B are placed consecutively in the address space with A before B. What can happen is that aubinator will process A, and start processing B when it should. When it gets done with B, it returns and continues to process A. Because A doesn't have any more data after the MI_BATCH_BUFFER_START, it will just process a bunch of NOPs until it gets to the next buffer in memory which is B again. In this scenario B gets processed twice which can be very confusing. If you place things in memory just right, you can also end up with infinite loops which are all sorts of fun.</div> <div> </div> <div>The root problem here is that it continues to process commands even after an MI_BATCH_BUFFER_START. By simply checking the 2nd level we can detect whether or not the command buffer we are jumping to will return here and stop processing commands if it won't.</div> <div> </div> <div>Signed-off-by: Jason Ekstrand <<a href="mailto:jason@jlekstrand.net">jason@jlekstrand.net</a>></div> <div>---</div> <div> src/intel/tools/aubinator.c | 7 +++++++</div> <div> 1 file changed, 7 insertions(+)</div> <div> </div> <div>diff --git a/src/intel/tools/aubinator.c b/src/intel/tools/aubinator.c index fe1f369..73a7f21 100644</div> <div>--- a/src/intel/tools/aubinator.c</div> <div>+++ b/src/intel/tools/aubinator.c</div> <div>@@ -766,6 +766,13 @@ parse_commands(struct gen_spec *spec, uint32_t *cmds, int size, int engine)</div> <div> start = p[1];</div> <div> </div> <div> parse_commands(spec, gtt + start, 1 << 20, engine);</div> <div>+</div> <div>+ /* MI_BATCH_BUFFER_START with "2nd Level Batch Buffer" unset acts</div> <div>+ * like a goto. No commands after such a MI_BATCH_BUFFER_START will</div> <div>+ * get processed so we should bail as well.</div> <div>+ */</div> <div>+ if (p[0] & (1 << 22) == 0)</div> <div> </div> <div><font color="#7030A0">[SG]: The above line might need extra pair of parenthesis around comparison to get rid of the compile time warning I was seeing as below. Otherwise, Patch looks good to me.</font></div> <div><font color="#7030A0">aubinator.c: In function ‘parse_commands’:</font></div> <div><font color="#7030A0">aubinator.c:768:19: warning: suggest parentheses around comparison in operand of ‘&’ [-Wparentheses]</font></div> <div><font color="#7030A0"> if (p[0] & (1 << 22) == 0)</font></div> <div> </div> <div>+ break;</div> <div> } else if ((p[0] & 0xffff0000) == AUB_MI_BATCH_BUFFER_END) {</div> <div> break;</div> <div> }</div> <div>--</div> <div>2.5.0.400.gff86faf</div> <div> </div> <div>_______________________________________________</div> <div>mesa-dev mailing list</div> <div><a href="mailto:mesa-dev@lists.freedesktop.org">mesa-dev@lists.freedesktop.org</a></div> <div><a href="https://lists.freedesktop.org/mailman/listinfo/mesa-dev">https://lists.freedesktop.org/mailman/listinfo/mesa-dev</a></div> <div> </div> </span></font> </body> </html>