We are able to capture all TLB-misses of an application using HMTT. The trick is UCUPT: preserving a special uncached memroy region and relocate the page table of an application to this region. Then every time a TLB miss occurs, page-walk process will access the page table in uncached memory and captured by HMTT.
The prototype is worked now.
Jiutian Zhang, Yuhang Liu, Xiaojing Zhu, Yuan Ruan and Mingyu Chen,Poster: "PTAT: An Efficient and Precise Tool for Collecting Detailed TLB Miss Traces",IEEE International Symposium on Performance Analysis of Systems and Software(ISPASS),April 24-25, 2017,Santa Rosa, California