On Tue, May 30, 2017 at 05:55:20PM +0000, Sanchez, Sebastian wrote:
> >Out of curiosity, do you have pahole data before vs. after to share?
>
> Yes, below is the pahole output for this patch.

Thanks

>
> Sebastian
>
> ---------
>
> Before:
>
>
> struct user_sdma_request {
> 	struct sdma_req_info       info;                 /*     0     8 */
> 	struct hfi1_user_sdma_pkt_q * pq;                /*     8     8 */
> 	struct hfi1_user_sdma_comp_q * cq;               /*    16     8 */
> 	struct hfi1_pkt_header     hdr;                  /*    24    64 */
> 	/* --- cacheline 1 boundary (64 bytes) was 24 bytes ago --- */
> 	struct sdma_engine *       sde;                  /*    88     8 */
> 	s8                         ahg_idx;              /*    96     1 */
>
> 	/* XXX 3 bytes hole, try to pack */
>
> 	u32                        ahg[9];               /*   100    36 */
> 	/* --- cacheline 2 boundary (128 bytes) was 8 bytes ago --- */
> 	u32                        koffset;              /*   136     4 */
> 	u32                        tidoffset;            /*   140     4 */
> 	unsigned int               data_iovs;            /*   144     4 */
> 	u32                        data_len;             /*   148     4 */
> 	unsigned int               iov_idx;              /*   152     4 */
>
> 	/* XXX 4 bytes hole, try to pack */
>
> 	struct user_sdma_iovec     iovs[8];              /*   160   512 */
> 	/* --- cacheline 10 boundary (640 bytes) was 32 bytes ago --- */
> 	u16                        n_tids;               /*   672     2 */
>
> 	/* XXX 6 bytes hole, try to pack */
>
> 	u32 *                      tids;                 /*   680     8 */
> 	u16                        tididx;               /*   688     2 */
>
> 	/* XXX 2 bytes hole, try to pack */
>
> 	u32                        sent;                 /*   692     4 */
> 	u64                        seqnum;               /*   696     8 */
> 	/* --- cacheline 11 boundary (704 bytes) --- */
> 	u64                        seqcomp;              /*   704     8 */
> 	u64                        seqsubmitted;         /*   712     8 */
> 	struct list_head           txps;                 /*   720    16 */
> 	long unsigned int          flags;                /*   736     8 */
> 	int                        status;               /*   744     4 */
>
> 	/* size: 752, cachelines: 12, members: 23 */
> 	/* sum members: 733, holes: 4, sum holes: 15 */
> 	/* padding: 4 */
> 	/* last cacheline: 48 bytes */
> };
>
> After:
>
> struct user_sdma_request {
> 	struct hfi1_pkt_header     hdr;                  /*     0    64 */
> 	/* --- cacheline 1 boundary (64 bytes) --- */
> 	struct hfi1_user_sdma_pkt_q * pq;                /*    64     8 */
> 	struct hfi1_user_sdma_comp_q * cq;               /*    72     8 */
> 	struct sdma_engine *       sde;                  /*    80     8 */
> 	struct sdma_req_info       info;                 /*    88     8 */
> 	u32 *                      tids;                 /*    96     8 */
> 	u32                        data_len;             /*   104     4 */
> 	u16                        n_tids;               /*   108     2 */
> 	u8                         data_iovs;            /*   110     1 */
> 	s8                         ahg_idx;              /*   111     1 */
>
> 	/* XXX 16 bytes hole, try to pack */
>
> 	/* --- cacheline 2 boundary (128 bytes) --- */
> 	u64                        seqcomp;              /*   128     8 */
> 	u64                        seqsubmitted;         /*   136     8 */
> 	long unsigned int          flags;                /*   144     8 */
> 	int                        status;               /*   152     4 */
>
> 	/* XXX 36 bytes hole, try to pack */
>
> 	/* --- cacheline 3 boundary (192 bytes) --- */
> 	struct list_head           txps;                 /*   192    16 */
> 	u64                        seqnum;               /*   208     8 */
> 	u32                        tidoffset;            /*   216     4 */
> 	u32                        koffset;              /*   220     4 */
> 	u32                        sent;                 /*   224     4 */
> 	u16                        tididx;               /*   228     2 */
> 	u8                         iov_idx;              /*   230     1 */
>
> 	/* XXX 1 byte hole, try to pack */
>
> 	struct user_sdma_iovec     iovs[8];              /*   232   512 */
> 	/* --- cacheline 11 boundary (704 bytes) was 40 bytes ago --- */
>
> 	/* size: 768, cachelines: 12, members: 22 */
> 	/* sum members: 691, holes: 3, sum holes: 53 */
> 	/* padding: 24 */
> };