At&nbsp;2011-03-07&nbsp;22:44:58，"Rabin&nbsp;Vincent"&nbsp;&lt;rabin@rab.in&gt;&nbsp;wrote:

&gt;2011/3/5&nbsp;rocky&nbsp;&lt;bill_carson@126.com&gt;:

&gt;pgd_free()&nbsp;on&nbsp;the&nbsp;other&nbsp;hand&nbsp;is&nbsp;safe&nbsp;because&nbsp;it's&nbsp;called&nbsp;when&nbsp;we've

&gt;switched&nbsp;away&nbsp;from&nbsp;this&nbsp;mm,&nbsp;and&nbsp;the&nbsp;page&nbsp;table&nbsp;is&nbsp;no&nbsp;longer&nbsp;in&nbsp;use

&gt;(mm-&gt;mm_count&nbsp;==&nbsp;0).

&gt;

&gt;Have&nbsp;a&nbsp;look&nbsp;at&nbsp;exit_mm(),&nbsp;which&nbsp;increases&nbsp;the&nbsp;mm_count&nbsp;before&nbsp;calling

&gt;mm_put(),&nbsp;and&nbsp;finish_task_switch(),&nbsp;which&nbsp;does&nbsp;the&nbsp;mmdrop().<br><br>

crystal clear:)<br>

your explanation DOES make sense<br><br>I can say when mm-&gt;mm_count ==0 , current_thread_info is not the thread info corresponding with mm used in mmdrop()<br><br>currently I am working on hugetlb for ARM <br>I put huge page table in thread_info when calling pgd_alloc<br>and wants to free it in pgd_free which is doomed to fail.<br>now I move those cookies into mm_struct , definitely not right place.<br>but I cannot find a desired place!<br><br>Thanks Rabin<br><br>rocky<br><br><br><br><br><br><br><br><br><br><br><br><br><br><br><br><br><span title="neteasefooter"><span id="netease_mail_footer"></span></span>