diff mbox series

[v2,4/5,RISCV_PM] Support pointer masking for RISC-V for i/c/f/d/a types of instructions

Message ID 20201015152139.28903-4-space.monkey.delivers@gmail.com (mailing list archive)
State New, archived
Headers show
Series [v2,1/5,RISCV_PM] Add J-extension into RISC-V | expand

Commit Message

Alexey Baturo Oct. 15, 2020, 3:21 p.m. UTC 
Signed-off-by: Alexey Baturo <space.monkey.delivers@gmail.com>
---
 target/riscv/insn_trans/trans_rva.c.inc |  3 +++
 target/riscv/insn_trans/trans_rvd.c.inc |  2 ++
 target/riscv/insn_trans/trans_rvf.c.inc |  2 ++
 target/riscv/insn_trans/trans_rvi.c.inc |  2 ++
 target/riscv/translate.c                | 14 ++++++++++++++
 5 files changed, 23 insertions(+)

Comments

Richard Henderson Oct. 15, 2020, 5 p.m. UTC | #1 
On 10/15/20 8:21 AM, Alexey Baturo wrote:
> Signed-off-by: Alexey Baturo <space.monkey.delivers@gmail.com>
> ---
>  target/riscv/insn_trans/trans_rva.c.inc |  3 +++
>  target/riscv/insn_trans/trans_rvd.c.inc |  2 ++
>  target/riscv/insn_trans/trans_rvf.c.inc |  2 ++
>  target/riscv/insn_trans/trans_rvi.c.inc |  2 ++
>  target/riscv/translate.c                | 14 ++++++++++++++
>  5 files changed, 23 insertions(+)

Looks ok.

It does occur to me to wonder how this is intended to work with unaligned
addresses, or large memory operations such as with RVV.

Without changes in the generic tcg code, an unaligned memory op that crosses
the mask will not wrap the second half.  E.g.

  upmbase = 0
  upmmask = 0xffff
  address = 0xfffe
  size    = 8

will read [0x10005 : 0xfffe] and not
[0x0005 : 0x0000] | [0xffff : 0xfffe] as a true wrapping would lead you do believe.


r~
Alexey Baturo Oct. 15, 2020, 5:30 p.m. UTC | #2 
That's a great question, but unfortunately, I don't have an answer for it
now.
Let me ask it on J WG meeting that would happen next Monday along with
extension naming and CSR numbers(hopefuly).

Thanks!

чт, 15 окт. 2020 г. в 20:00, Richard Henderson <richard.henderson@linaro.org
>:

> On 10/15/20 8:21 AM, Alexey Baturo wrote:
> > Signed-off-by: Alexey Baturo <space.monkey.delivers@gmail.com>
> > ---
> >  target/riscv/insn_trans/trans_rva.c.inc |  3 +++
> >  target/riscv/insn_trans/trans_rvd.c.inc |  2 ++
> >  target/riscv/insn_trans/trans_rvf.c.inc |  2 ++
> >  target/riscv/insn_trans/trans_rvi.c.inc |  2 ++
> >  target/riscv/translate.c                | 14 ++++++++++++++
> >  5 files changed, 23 insertions(+)
>
> Looks ok.
>
> It does occur to me to wonder how this is intended to work with unaligned
> addresses, or large memory operations such as with RVV.
>
> Without changes in the generic tcg code, an unaligned memory op that
> crosses
> the mask will not wrap the second half.  E.g.
>
>   upmbase = 0
>   upmmask = 0xffff
>   address = 0xfffe
>   size    = 8
>
> will read [0x10005 : 0xfffe] and not
> [0x0005 : 0x0000] | [0xffff : 0xfffe] as a true wrapping would lead you do
> believe.
>
>
> r~
>
diff mbox series

Patch

diff --git a/target/riscv/insn_trans/trans_rva.c.inc b/target/riscv/insn_trans/trans_rva.c.inc
index be8a9f06dd..5559e347ba 100644
--- a/target/riscv/insn_trans/trans_rva.c.inc
+++ b/target/riscv/insn_trans/trans_rva.c.inc
@@ -26,6 +26,7 @@  static inline bool gen_lr(DisasContext *ctx, arg_atomic *a, MemOp mop)
     if (a->rl) {
         tcg_gen_mb(TCG_MO_ALL | TCG_BAR_STRL);
     }
+    gen_pm_adjust_address(ctx, src1, src1);
     tcg_gen_qemu_ld_tl(load_val, src1, ctx->mem_idx, mop);
     if (a->aq) {
         tcg_gen_mb(TCG_MO_ALL | TCG_BAR_LDAQ);
@@ -46,6 +47,7 @@  static inline bool gen_sc(DisasContext *ctx, arg_atomic *a, MemOp mop)
     TCGLabel *l2 = gen_new_label();
 
     gen_get_gpr(src1, a->rs1);
+    gen_pm_adjust_address(ctx, src1, src1);
     tcg_gen_brcond_tl(TCG_COND_NE, load_res, src1, l1);
 
     gen_get_gpr(src2, a->rs2);
@@ -91,6 +93,7 @@  static bool gen_amo(DisasContext *ctx, arg_atomic *a,
     gen_get_gpr(src1, a->rs1);
     gen_get_gpr(src2, a->rs2);
 
+    gen_pm_adjust_address(ctx, src1, src1);
     (*func)(src2, src1, src2, ctx->mem_idx, mop);
 
     gen_set_gpr(a->rd, src2);
diff --git a/target/riscv/insn_trans/trans_rvd.c.inc b/target/riscv/insn_trans/trans_rvd.c.inc
index 4f832637fa..935342f66d 100644
--- a/target/riscv/insn_trans/trans_rvd.c.inc
+++ b/target/riscv/insn_trans/trans_rvd.c.inc
@@ -25,6 +25,7 @@  static bool trans_fld(DisasContext *ctx, arg_fld *a)
     TCGv t0 = tcg_temp_new();
     gen_get_gpr(t0, a->rs1);
     tcg_gen_addi_tl(t0, t0, a->imm);
+    gen_pm_adjust_address(ctx, t0, t0);
 
     tcg_gen_qemu_ld_i64(cpu_fpr[a->rd], t0, ctx->mem_idx, MO_TEQ);
 
@@ -40,6 +41,7 @@  static bool trans_fsd(DisasContext *ctx, arg_fsd *a)
     TCGv t0 = tcg_temp_new();
     gen_get_gpr(t0, a->rs1);
     tcg_gen_addi_tl(t0, t0, a->imm);
+    gen_pm_adjust_address(ctx, t0, t0);
 
     tcg_gen_qemu_st_i64(cpu_fpr[a->rs2], t0, ctx->mem_idx, MO_TEQ);
 
diff --git a/target/riscv/insn_trans/trans_rvf.c.inc b/target/riscv/insn_trans/trans_rvf.c.inc
index 3dfec8211d..04b3c3eb3d 100644
--- a/target/riscv/insn_trans/trans_rvf.c.inc
+++ b/target/riscv/insn_trans/trans_rvf.c.inc
@@ -30,6 +30,7 @@  static bool trans_flw(DisasContext *ctx, arg_flw *a)
     TCGv t0 = tcg_temp_new();
     gen_get_gpr(t0, a->rs1);
     tcg_gen_addi_tl(t0, t0, a->imm);
+    gen_pm_adjust_address(ctx, t0, t0);
 
     tcg_gen_qemu_ld_i64(cpu_fpr[a->rd], t0, ctx->mem_idx, MO_TEUL);
     gen_nanbox_s(cpu_fpr[a->rd], cpu_fpr[a->rd]);
@@ -47,6 +48,7 @@  static bool trans_fsw(DisasContext *ctx, arg_fsw *a)
     gen_get_gpr(t0, a->rs1);
 
     tcg_gen_addi_tl(t0, t0, a->imm);
+    gen_pm_adjust_address(ctx, t0, t0);
 
     tcg_gen_qemu_st_i64(cpu_fpr[a->rs2], t0, ctx->mem_idx, MO_TEUL);
 
diff --git a/target/riscv/insn_trans/trans_rvi.c.inc b/target/riscv/insn_trans/trans_rvi.c.inc
index d04ca0394c..bee7f6be46 100644
--- a/target/riscv/insn_trans/trans_rvi.c.inc
+++ b/target/riscv/insn_trans/trans_rvi.c.inc
@@ -141,6 +141,7 @@  static bool gen_load(DisasContext *ctx, arg_lb *a, MemOp memop)
     TCGv t1 = tcg_temp_new();
     gen_get_gpr(t0, a->rs1);
     tcg_gen_addi_tl(t0, t0, a->imm);
+    gen_pm_adjust_address(ctx, t0, t0);
 
     tcg_gen_qemu_ld_tl(t1, t0, ctx->mem_idx, memop);
     gen_set_gpr(a->rd, t1);
@@ -180,6 +181,7 @@  static bool gen_store(DisasContext *ctx, arg_sb *a, MemOp memop)
     TCGv dat = tcg_temp_new();
     gen_get_gpr(t0, a->rs1);
     tcg_gen_addi_tl(t0, t0, a->imm);
+    gen_pm_adjust_address(ctx, t0, t0);
     gen_get_gpr(dat, a->rs2);
 
     tcg_gen_qemu_st_tl(dat, t0, ctx->mem_idx, memop);
diff --git a/target/riscv/translate.c b/target/riscv/translate.c
index 79dca2291b..a7cbf909f3 100644
--- a/target/riscv/translate.c
+++ b/target/riscv/translate.c
@@ -101,6 +101,16 @@  static void gen_nanbox_s(TCGv_i64 out, TCGv_i64 in)
     tcg_gen_ori_i64(out, in, MAKE_64BIT_MASK(32, 32));
 }
 
+/*
+ * Temp stub: generates address adjustment for PointerMasking
+ */
+static void gen_pm_adjust_address(DisasContext *s,
+                                  TCGv_i64      dst,
+                                  TCGv_i64      src)
+{
+    tcg_gen_mov_i64(dst, src);
+}
+
 /*
  * A narrow n-bit operation, where n < FLEN, checks that input operands
  * are correctly Nan-boxed, i.e., all upper FLEN - n bits are 1.
@@ -380,6 +390,7 @@  static void gen_load_c(DisasContext *ctx, uint32_t opc, int rd, int rs1,
     TCGv t1 = tcg_temp_new();
     gen_get_gpr(t0, rs1);
     tcg_gen_addi_tl(t0, t0, imm);
+    gen_pm_adjust_address(ctx, t0, t0);
     int memop = tcg_memop_lookup[(opc >> 12) & 0x7];
 
     if (memop < 0) {
@@ -400,6 +411,7 @@  static void gen_store_c(DisasContext *ctx, uint32_t opc, int rs1, int rs2,
     TCGv dat = tcg_temp_new();
     gen_get_gpr(t0, rs1);
     tcg_gen_addi_tl(t0, t0, imm);
+    gen_pm_adjust_address(ctx, t0, t0);
     gen_get_gpr(dat, rs2);
     int memop = tcg_memop_lookup[(opc >> 12) & 0x7];
 
@@ -459,6 +471,7 @@  static void gen_fp_load(DisasContext *ctx, uint32_t opc, int rd,
     t0 = tcg_temp_new();
     gen_get_gpr(t0, rs1);
     tcg_gen_addi_tl(t0, t0, imm);
+    gen_pm_adjust_address(ctx, t0, t0);
 
     switch (opc) {
     case OPC_RISC_FLW:
@@ -498,6 +511,7 @@  static void gen_fp_store(DisasContext *ctx, uint32_t opc, int rs1,
     t0 = tcg_temp_new();
     gen_get_gpr(t0, rs1);
     tcg_gen_addi_tl(t0, t0, imm);
+    gen_pm_adjust_address(ctx, t0, t0);
 
     switch (opc) {
     case OPC_RISC_FSW: