[DTrace-devel] [PATCH 31/61] Remove the "default tuple" area

[eugene.loh at oracle.com]

From: Eugene Loh <eugene.loh at oracle.com>

All we need is a bunch of zeroes, which are available from the newly
introduced block of zeroes.  Using that block is better because:

1)  The new block is global, while the older area was per-CPU.
    We only need one copy of this read-only memory.

2)  The new block will be overlapped with other uses of zeroes.

Signed-off-by: Eugene Loh <eugene.loh at oracle.com>
---
 libdtrace/dt_bpf.c  |  4 ++++
 libdtrace/dt_cg.c   |  6 ++++--
 libdtrace/dt_dctx.h | 16 ++++------------
 3 files changed, 12 insertions(+), 14 deletions(-)

diff --git a/libdtrace/dt_bpf.c b/libdtrace/dt_bpf.c
index ee9aa846..f0d1a99e 100644
--- a/libdtrace/dt_bpf.c
+++ b/libdtrace/dt_bpf.c
@@ -351,6 +351,10 @@ dt_bpf_gmap_create(dtrace_hdl_t *dtp)
 	if (pr_mapfd == -1)
 		return -1;		/* dt_errno is set for us */
 
+	/* figure out how big the region of zeroes should be */
+	if (dtp->dt_zerosize < dtp->dt_maxtuplesize)
+		dtp->dt_zerosize = dtp->dt_maxtuplesize;
+
 	/* global variables, including a region of zeroes */
 	dtp->dt_zerooffset = P2ROUNDUP(dt_idhash_datasize(dtp->dt_globals), 8);
 	sz = dtp->dt_zerooffset + dtp->dt_zerosize;
diff --git a/libdtrace/dt_cg.c b/libdtrace/dt_cg.c
index f48452ee..11a9c8c5 100644
--- a/libdtrace/dt_cg.c
+++ b/libdtrace/dt_cg.c
@@ -2631,6 +2631,7 @@ dt_cg_arglist(dt_ident_t *idp, dt_node_t *args, dt_irlist_t *dlp,
 	const dt_idsig_t	*isp = idp->di_data;
 	dt_node_t		*dnp;
 	dt_ident_t		*maxtupsz = dt_dlib_get_var(dtp, "TUPSZ");
+	dt_ident_t		*zero_off = dt_dlib_get_var(dtp, "ZERO_OFF");
 	int			i;
 	int			treg, areg;
 	uint_t			tuplesize;
@@ -2676,8 +2677,9 @@ dt_cg_arglist(dt_ident_t *idp, dt_node_t *args, dt_irlist_t *dlp,
 
 	emit(dlp,  BPF_MOV_REG(BPF_REG_1, treg));
 	emite(dlp, BPF_MOV_IMM(BPF_REG_2, -1), maxtupsz);
-	emit(dlp,  BPF_MOV_REG(BPF_REG_3, BPF_REG_1));
-	emite(dlp, BPF_ALU64_IMM(BPF_ADD, BPF_REG_3, -1), maxtupsz);
+	emit(dlp,  BPF_LOAD(BPF_DW, BPF_REG_3, BPF_REG_FP, DT_STK_DCTX));
+	emit(dlp,  BPF_LOAD(BPF_DW, BPF_REG_3, BPF_REG_3, DCTX_GVARS));
+	emite(dlp, BPF_ALU64_IMM(BPF_ADD, BPF_REG_3, -1), zero_off);
 	dt_regset_xalloc(drp, BPF_REG_0);
 	emit(dlp,  BPF_CALL_HELPER(BPF_FUNC_probe_read));
 	dt_regset_free_args(drp);
diff --git a/libdtrace/dt_dctx.h b/libdtrace/dt_dctx.h
index ea4758b6..c93fff2b 100644
--- a/libdtrace/dt_dctx.h
+++ b/libdtrace/dt_dctx.h
@@ -79,12 +79,10 @@ typedef struct dt_dctx {
  *                  0 -> | Stack          :     tstring    | \
  *                       |   trace     (shared)   storage  |  |
  *                       |     storage    :                |  |
- *                       +----------------+----------------+  |
- *        DMEM_STRTOK -> |     strtok() internal state     |   > DMEM_SIZE
+ *                       +----------------+----------------+   > DMEM_SIZE
+ *        DMEM_STRTOK -> |     strtok() internal state     |  |
  *                       +---------------------------------+  |
- *        DMEM_TUPLE  -> |       tuple assembly area       |  |
- *                       +---------------------------------+  |
- *   DMEM_TUPLE_DFLT  -> |       default empty tuple       | /
+ *        DMEM_TUPLE  -> |       tuple assembly area       | /
  *                       +---------------------------------+
  */
 
@@ -107,17 +105,11 @@ typedef struct dt_dctx {
 
 /*
  * Macros to determine the offset of the components of dctx->mem.
- *
- * CAUTION: DMEM_TUPLE(dtp) and DMEM_TUPLE_DFLT(dtp) depend on data collected
- *	    during code generation and therefore cannot be used until all code
- *	    generation has completed.
  */
 #define DMEM_STRTOK(dtp) \
 		MAX(DMEM_STACK_SZ(dtp), DMEM_TSTR_SZ(dtp))
 #define DMEM_TUPLE(dtp) \
 		(DMEM_STRTOK(dtp) + DMEM_STRTOK_SZ(dtp))
-#define DMEM_TUPLE_DFLT(dtp) \
-		(DMEM_TUPLE(dtp) + DMEM_TUPLE_SZ(dtp))
 
 /*
  * Macro to determine the total size of the mem area.
@@ -126,7 +118,7 @@ typedef struct dt_dctx {
  *	    and therefore cannot be used until all code generation has
  *	    completed.
  */
-#define DMEM_SIZE(dtp)	(DMEM_TUPLE_DFLT(dtp) + DMEM_TUPLE_SZ(dtp))
+#define DMEM_SIZE(dtp)	(DMEM_TUPLE(dtp) + DMEM_TUPLE_SZ(dtp))
 
 /*
  * The stack layout for BPF programs that are generated as trampolines for
-- 
2.18.4