[PATCH 0/4] [SPARC] Workarounds for UT699, UT700, and GR712RC errata

[PATCH 0/4] [SPARC] Workarounds for UT699, UT700, and GR712RC errata

[Daniel Cederman]

Hello,

This patch series adds workarounds for the newly discovered errata for UT699, UT700, and GR712RC. The errata and possible workarounds are described in the following documents available at http://www.gaisler.com/index.php/information/app-tech-notes:

GRLIB-TN-0010 - LEON3/FT AHB Deadlock After Sequence of Load and Atomic Instructions
GRLIB-TN-0011 - LEON3/FT AHB Lock Release during Atomic Operation
GRLIB-TN-0012 - GR712RC Incorrect Annulation of Floating-point Operation on Instruction Cache Parity Error
GRLIB-TN-0013 - GRFPU Floating-point controller: Missing FDIV/FSQRT Result

Daniel Cederman (4):
  [SPARC] Errata workaround for GRLIB-TN-0012
  [SPARC] Errata workaround for GRLIB-TN-0011
  [SPARC] Errata workaround for GRLIB-TN-0010
  [SPARC] Errata workaround for GRLIB-TN-0013

 gcc/config/sparc/sparc.c   | 205 ++++++++++++++++++++++++++++++++++++++++++++-
 gcc/config/sparc/sparc.md  |  24 ++++++
 gcc/config/sparc/sparc.opt |   4 +
 gcc/config/sparc/sync.md   |  18 +++-
 4 files changed, 245 insertions(+), 6 deletions(-)

-- 
2.9.3

[PATCH 2/4] [SPARC] Errata workaround for GRLIB-TN-0011

[Daniel Cederman]

This patch provides a workaround for the errata described in GRLIB-TN-0011.

If the workaround is enabled it will:

* Insert .align 16 before atomic instructions (swap, ldstub, casa).

It is applicable to GR712RC.

gcc/ChangeLog:

2017-11-17  Daniel Cederman  <cederman@gaisler.com>

	* config/sparc/sync.md (swapsi): 16-byte align if sparc_fix_gr712rc.
	(atomic_compare_and_swap_leon3_1): Likewise.
	(ldstub): Likewise.
---
 gcc/config/sparc/sync.md | 16 ++++++++++++++--
 1 file changed, 14 insertions(+), 2 deletions(-)

diff --git a/gcc/config/sparc/sync.md b/gcc/config/sparc/sync.md
index 1593bde..d5c505b 100644
--- a/gcc/config/sparc/sync.md
+++ b/gcc/config/sparc/sync.md
@@ -222,6 +222,8 @@
 	  UNSPECV_CAS))]
   "TARGET_LEON3"
 {
+  if (sparc_fix_gr712rc)
+    output_asm_insn (".align\t16", operands);
   if (TARGET_SV_MODE)
     return "casa\t%1 0xb, %2, %0"; /* ASI for supervisor data space.  */
   else
@@ -275,7 +277,12 @@
    (set (match_dup 1)
 	(match_operand:SI 2 "register_operand" "0"))]
   "(TARGET_V8 || TARGET_V9) && !sparc_fix_ut699"
-  "swap\t%1, %0"
+{
+  if (sparc_fix_gr712rc)
+    return ".align\t16\n\tswap\t%1, %0";
+  else
+    return "swap\t%1, %0";
+}
   [(set_attr "type" "multi")])
 
 (define_expand "atomic_test_and_set"
@@ -307,5 +314,10 @@
 			    UNSPECV_LDSTUB))
    (set (match_dup 1) (const_int -1))]
   "!sparc_fix_ut699"
-  "ldstub\t%1, %0"
+{
+  if (sparc_fix_gr712rc)
+    return ".align\t16\n\tldstub\t%1, %0";
+  else
+    return "ldstub\t%1, %0";
+}
   [(set_attr "type" "multi")])
-- 
2.9.3

Re: [PATCH 2/4] [SPARC] Errata workaround for GRLIB-TN-0011

[Eric Botcazou]

> 2017-11-17  Daniel Cederman  <cederman@gaisler.com>
> 
> 	* config/sparc/sync.md (swapsi): 16-byte align if sparc_fix_gr712rc.
> 	(atomic_compare_and_swap_leon3_1): Likewise.
> 	(ldstub): Likewise.

You need to adjust the "length" attribute accordingly (and conservatively), 
see atomic_compare_and_swapdi_v8plus for reference.

-- 
Eric Botcazou

[PATCH 1/4] [SPARC] Errata workaround for GRLIB-TN-0012

[Daniel Cederman]

This patch provides a workaround for the errata described in GRLIB-TN-0012.

If the workaround is enabled it will:

* Prevent any floating-point operation from being placed in the
  delay slot of an annulled integer branch.

* Place a NOP at the branch target of an integer branch if it is
  a floating-point operation or a floating-point branch.

It is applicable to GR712RC.

gcc/ChangeLog:

2017-11-17  Daniel Cederman  <cederman@gaisler.com>

	* config/sparc/sparc.c (fpop_insn_p): New function.
	(sparc_do_work_around_errata): Insert NOP instructions to
	prevent sequences that could trigger the TN-0012 errata for
	GR712RC.
	(pass_work_around_errata::gate): Also test sparc_fix_gr712rc.
	* config/sparc/sparc.md (fix_gr712rc): New attribute.
	(in_branch_annul_delay): Prevent floating-point instructions
	in delay slot of annulled integer branch.
---
 gcc/config/sparc/sparc.c  | 50 +++++++++++++++++++++++++++++++++++++++++++++--
 gcc/config/sparc/sparc.md | 17 ++++++++++++++++
 2 files changed, 65 insertions(+), 2 deletions(-)

diff --git a/gcc/config/sparc/sparc.c b/gcc/config/sparc/sparc.c
index a9945e2..8f6eb48 100644
--- a/gcc/config/sparc/sparc.c
+++ b/gcc/config/sparc/sparc.c
@@ -945,6 +945,31 @@ mem_ref (rtx x)
   return NULL_RTX;
 }
 
+/* True if floating-point instruction.  */
+
+static int
+fpop_insn_p (rtx_insn *insn)
+{
+  if ( GET_CODE (PATTERN (insn)) != SET)
+    return false;
+
+  switch (get_attr_type (insn))
+    {
+    case TYPE_FPMOVE:
+    case TYPE_FPCMOVE:
+    case TYPE_FP:
+    case TYPE_FPCMP:
+    case TYPE_FPMUL:
+    case TYPE_FPDIVS:
+    case TYPE_FPSQRTS:
+    case TYPE_FPDIVD:
+    case TYPE_FPSQRTD:
+      return true;
+    default:
+      return false;
+    }
+}
+
 /* We use a machine specific pass to enable workarounds for errata.
 
    We need to have the (essentially) final form of the insn stream in order
@@ -970,11 +995,31 @@ sparc_do_work_around_errata (void)
     {
       bool insert_nop = false;
       rtx set;
+      rtx_insn *jump = 0;
 
       /* Look into the instruction in a delay slot.  */
       if (NONJUMP_INSN_P (insn))
 	if (rtx_sequence *seq = dyn_cast <rtx_sequence *> (PATTERN (insn)))
-	  insn = seq->insn (1);
+	  {
+	    jump = seq->insn (0);
+	    insn = seq->insn (1);
+	  }
+
+      /* Place a NOP at the branch target of an integer branch if it is
+	 a floating-point operation or a floating-point branch.  */
+      if (sparc_fix_gr712rc
+	  && (JUMP_P (insn) || jump)
+	  && get_attr_branch_type (jump ? jump : insn) == BRANCH_TYPE_ICC)
+	{
+	  rtx_insn *target;
+
+	  target = next_active_insn (JUMP_LABEL_AS_INSN (jump ? jump : insn));
+	  if (target
+	      && (fpop_insn_p (target)
+		  || ((JUMP_P (target)
+		       && get_attr_branch_type (target) == BRANCH_TYPE_FCC))))
+	    emit_insn_before (gen_nop (), target);
+	}
 
       /* Look for either of these two sequences:
 
@@ -1303,7 +1348,8 @@ public:
   /* opt_pass methods: */
   virtual bool gate (function *)
     {
-      return sparc_fix_at697f || sparc_fix_ut699 || sparc_fix_b2bst;
+      return sparc_fix_at697f || sparc_fix_ut699 || sparc_fix_b2bst
+	|| sparc_fix_gr712rc;
     }
 
   virtual unsigned int execute (function *)
diff --git a/gcc/config/sparc/sparc.md b/gcc/config/sparc/sparc.md
index d9cbd4f..ef789e2 100644
--- a/gcc/config/sparc/sparc.md
+++ b/gcc/config/sparc/sparc.md
@@ -430,6 +430,10 @@
    (symbol_ref "(sparc_fix_b2bst != 0
 		 ? FIX_B2BST_TRUE : FIX_B2BST_FALSE)"))
 
+(define_attr "fix_gr712rc" "false,true"
+   (symbol_ref "(sparc_fix_gr712rc != 0
+		 ? FIX_GR712RC_TRUE : FIX_GR712RC_FALSE)"))
+
 ;; Length (in # of insns).
 ;; Beware that setting a length greater or equal to 3 for conditional branches
 ;; has a side-effect (see output_cbranch and output_v9branch).
@@ -590,6 +594,15 @@
 	   (const_string "true")
 	] (const_string "false")))
 
+(define_attr "in_branch_annul_delay" "false,true"
+  (cond[(and (eq_attr "fix_gr712rc" "true")
+	     (eq_attr "type" "fp,fpcmp,fpmove,fpcmove,fpmul,
+			      fpdivs,fpsqrts,fpdivd,fpsqrtd"))
+	   (const_string "false")
+	(eq_attr "in_branch_delay" "true")
+	   (const_string "true")
+	] (const_string "false")))
+
 (define_delay (eq_attr "type" "call")
   [(eq_attr "in_call_delay" "true") (nil) (nil)])
 
@@ -602,6 +615,10 @@
 (define_delay (eq_attr "type" "branch")
   [(eq_attr "in_branch_delay" "true") (nil) (eq_attr "in_branch_delay" "true")])
 
+(define_delay (and (eq_attr "type" "branch") (eq_attr "branch_type" "icc"))
+  [(eq_attr "in_branch_delay" "true") (nil)
+  (eq_attr "in_branch_annul_delay" "true")])
+
 (define_delay (eq_attr "type" "uncond_branch")
   [(eq_attr "in_branch_delay" "true") (nil) (nil)])
 
-- 
2.9.3

Re: [PATCH 1/4] [SPARC] Errata workaround for GRLIB-TN-0012

[Eric Botcazou]

> 2017-11-17  Daniel Cederman  <cederman@gaisler.com>
> 
> 	* config/sparc/sparc.c (fpop_insn_p): New function.
> 	(sparc_do_work_around_errata): Insert NOP instructions to
> 	prevent sequences that could trigger the TN-0012 errata for
> 	GR712RC.
> 	(pass_work_around_errata::gate): Also test sparc_fix_gr712rc.
> 	* config/sparc/sparc.md (fix_gr712rc): New attribute.
> 	(in_branch_annul_delay): Prevent floating-point instructions
> 	in delay slot of annulled integer branch.

OK for mainline and 7 branch modulo the following nits:

> +/* True if floating-point instruction.  */
> +
> +static int
> +fpop_insn_p (rtx_insn *insn)

'bool' instead of 'int'.  "True if INSN is a floating-point instruction."

> +{
> +  if ( GET_CODE (PATTERN (insn)) != SET)
> +    return false;

No space before GET_CODE.

>  /* We use a machine specific pass to enable workarounds for errata.
> 
>     We need to have the (essentially) final form of the insn stream in order
> @@ -970,11 +995,31 @@ sparc_do_work_around_errata (void)
>      {
>        bool insert_nop = false;
>        rtx set;
> +      rtx_insn *jump = 0;
> 
>        /* Look into the instruction in a delay slot.  */
>        if (NONJUMP_INSN_P (insn))
>  	if (rtx_sequence *seq = dyn_cast <rtx_sequence *> (PATTERN (insn)))
> -	  insn = seq->insn (1);
> +	  {
> +	    jump = seq->insn (0);
> +	    insn = seq->insn (1);
> +	  }

This should be changed into:

      rtx_insn jump;

      /* Look into the instruction in a delay slot.  */
      if (NONJUMP_INSN_P (insn)
	  && (rtx_sequence *seq = dyn_cast <rtx_sequence *> (PATTERN (insn)))
	{
	  jump = seq->insn (0);
	  insn = seq->insn (1);
	}
      else if (JUMP_P (insn))
	jump = insn
      else
	jump = NULL_RTX;

and the block below simplified accordingly.

> +      /* Place a NOP at the branch target of an integer branch if it is
> +	 a floating-point operation or a floating-point branch.  */
> +      if (sparc_fix_gr712rc
> +	  && (JUMP_P (insn) || jump)
> +	  && get_attr_branch_type (jump ? jump : insn) == BRANCH_TYPE_ICC)
> +	{
> +	  rtx_insn *target;
> +
> +	  target = next_active_insn (JUMP_LABEL_AS_INSN (jump ? jump : insn));

	  rtx_insn *target = next_active_insn (JUMP_LABEL_AS_INSN (jump));

>        /* Look for either of these two sequences:
> 
> @@ -1303,7 +1348,8 @@ public:
>    /* opt_pass methods: */
>    virtual bool gate (function *)
>      {
> -      return sparc_fix_at697f || sparc_fix_ut699 || sparc_fix_b2bst;
> +      return sparc_fix_at697f || sparc_fix_ut699 || sparc_fix_b2bst
> +	|| sparc_fix_gr712rc;
>      }

"|| sparc_fix_gr712rc" lined up under sparc_fix_at697f.

> @@ -602,6 +615,10 @@
>  (define_delay (eq_attr "type" "branch")
>    [(eq_attr "in_branch_delay" "true") (nil) (eq_attr "in_branch_delay"
> "true")])
> 
> +(define_delay (and (eq_attr "type" "branch") (eq_attr "branch_type" "icc"))
> +  [(eq_attr "in_branch_delay" "true") (nil)
> +  (eq_attr "in_branch_annul_delay" "true")])
> +

I think that we'd better keep the various define_delay's mutually exclusive.

-- 
Eric Botcazou

[PATCH 4/4] [SPARC] Errata workaround for GRLIB-TN-0013

[Daniel Cederman]

This patch provides a workaround for the errata described in GRLIB-TN-0013.

If the workaround is enabled it will:

* Prevent div and sqrt instructions in the delay slot.

* Insert NOPs to prevent the sequence (div/sqrt) -> (two or three floating
  point operations or loads) -> (div/sqrt).

* Not insert NOPs if any of the floating point operations have a dependency
  on the destination register of the first (div/sqrt).

* Not insert NOPs if one of the floating point operations is a (div/sqrt).

* Insert NOPs to prevent (div/sqrt) followed by a branch.

It is applicable to GR712RC, UT700, and UT699.

gcc/ChangeLog:

2017-11-17  Daniel Cederman  <cederman@gaisler.com>

	* config/sparc/sparc.c (fpop_reg_depend_p): New function.
	(div_sqrt_insn_p): New function.
	(sparc_do_work_around_errata): Insert NOP instructions to
	prevent sequences that could trigger the TN-0013 errata for
	certain LEON3 processors.
	(pass_work_around_errata::gate): Also test sparc_fix_tn0013.
	(sparc_option_override): Set sparc_fix_tn0013 appropriately.
	* config/sparc/sparc.md (fix_tn0013): New attribute.
	(in_branch_delay): Prevent div and sqrt in delay slot if fix_tn0013.
	* config/sparc/sparc.opt (sparc_fix_tn0013: New variable.
---
 gcc/config/sparc/sparc.c   | 114 ++++++++++++++++++++++++++++++++++++++++++++-
 gcc/config/sparc/sparc.md  |   7 +++
 gcc/config/sparc/sparc.opt |   4 ++
 3 files changed, 123 insertions(+), 2 deletions(-)

diff --git a/gcc/config/sparc/sparc.c b/gcc/config/sparc/sparc.c
index 9faf774..3da642e 100644
--- a/gcc/config/sparc/sparc.c
+++ b/gcc/config/sparc/sparc.c
@@ -945,6 +945,39 @@ mem_ref (rtx x)
   return NULL_RTX;
 }
 
+/* True if any of the floating-point instruction's source
+   registers is the same as the provided register.  */
+
+static int
+fpop_reg_depend_p (rtx_insn *insn, unsigned int reg)
+{
+  extract_insn (insn);
+  return (REGNO (recog_data.operand[1]) == reg
+	  || (recog_data.n_operands == 3
+	      && REGNO (recog_data.operand[2]) == reg));
+}
+
+/* True if the instruction is floating-point division or
+   floating-point square-root.  */
+
+static int
+div_sqrt_insn_p (rtx_insn *insn)
+{
+  if ( GET_CODE (PATTERN (insn)) != SET)
+    return false;
+
+  switch (get_attr_type (insn))
+    {
+    case TYPE_FPDIVS:
+    case TYPE_FPSQRTS:
+    case TYPE_FPDIVD:
+    case TYPE_FPSQRTD:
+      return true;
+    default:
+      return false;
+    }
+}
+
 /* True if floating-point instruction.  */
 
 static int
@@ -1064,6 +1097,80 @@ sparc_do_work_around_errata (void)
 	    insert_nop = true;
 	}
 
+      /* Look for sequences that could trigger the TN-0013 errata.  */
+      if (sparc_fix_tn0013
+	  && NONJUMP_INSN_P (insn)
+	  && div_sqrt_insn_p (insn))
+	{
+	  int i;
+	  int fp_found = 0;
+	  unsigned int dest_reg;
+	  rtx_insn *after;
+
+	  dest_reg = REGNO (SET_DEST (single_set (insn)));
+
+	  next = next_active_insn (insn);
+	  if (!next)
+	    break;
+
+	  for (after = next, i = 0; i < 4; i++)
+	    {
+	      /* Count floating-point operations.  */
+	      if (i != 3
+		  && fpop_insn_p (after))
+		{
+		  /* If the insn uses the destination register of
+		     the div/sqrt, then it cannot be problematic.  */
+		  if (fpop_reg_depend_p (after, dest_reg))
+		    break;
+		  fp_found++;
+		}
+
+	      /* Count floating-point loads.  */
+	      if (i != 3
+		  && (set = single_set (after)) != NULL_RTX
+		  && REG_P (SET_DEST (set))
+		  && REGNO (SET_DEST (set)) > 31)
+		{
+		  /* If the insn uses the destination register of
+		     the div/sqrt, then it cannot be problematic.  */
+		  if (REGNO (SET_DEST (set)) == dest_reg)
+		    break;
+		  fp_found++;
+		}
+
+	      /* Check if this is a problematic sequence.  */
+	      if (i > 1
+		  && fp_found >= 2
+		  && div_sqrt_insn_p (after))
+		{
+		  /* Add extra NOP to prevent second version of
+		     problematic sequence.  */
+		  if (i == 2)
+		    emit_insn_before (gen_nop (), next);
+		  insert_nop = true;
+		  break;
+		}
+
+	      /* No need to scan past a second div/sqrt.  */
+	      if (div_sqrt_insn_p (after))
+		break;
+
+	      /* Insert NOP before branch.  */
+	      if (i < 3
+		  && (!NONJUMP_INSN_P (after)
+		      || GET_CODE (PATTERN (after)) == SEQUENCE))
+		{
+		  insert_nop = true;
+		  break;
+		}
+
+	      after = next_active_insn (after);
+	      if (!after)
+		break;
+	    }
+	}
+
       /* Look for either of these two sequences:
 
 	 Sequence A:
@@ -1392,7 +1499,7 @@ public:
   virtual bool gate (function *)
     {
       return sparc_fix_at697f || sparc_fix_ut699 || sparc_fix_b2bst
-	|| sparc_fix_gr712rc;
+	|| sparc_fix_gr712rc || sparc_fix_tn0013;
     }
 
   virtual unsigned int execute (function *)
@@ -1762,9 +1869,12 @@ sparc_option_override (void)
   if (!(target_flags_explicit & MASK_LRA))
     target_flags |= MASK_LRA;
 
-  /* Enable the back-to-back store errata workaround for LEON3FT.  */
+  /* Enable applicable errata workarounds for LEON3FT.  */
   if (sparc_fix_ut699 || sparc_fix_ut700 || sparc_fix_gr712rc)
+    {
     sparc_fix_b2bst = 1;
+    sparc_fix_tn0013 = 1;
+    }
 
   /* Disable FsMULd for the UT699 since it doesn't work correctly.  */
   if (sparc_fix_ut699)
diff --git a/gcc/config/sparc/sparc.md b/gcc/config/sparc/sparc.md
index ef789e2..ec2b476 100644
--- a/gcc/config/sparc/sparc.md
+++ b/gcc/config/sparc/sparc.md
@@ -430,6 +430,10 @@
    (symbol_ref "(sparc_fix_b2bst != 0
 		 ? FIX_B2BST_TRUE : FIX_B2BST_FALSE)"))
 
+(define_attr "fix_tn0013" "false,true"
+   (symbol_ref "(sparc_fix_tn0013 != 0
+		 ? FIX_TN0013_TRUE : FIX_TN0013_FALSE)"))
+
 (define_attr "fix_gr712rc" "false,true"
    (symbol_ref "(sparc_fix_gr712rc != 0
 		 ? FIX_GR712RC_TRUE : FIX_GR712RC_FALSE)"))
@@ -581,6 +585,9 @@
 (define_attr "in_branch_delay" "false,true"
   (cond [(eq_attr "type" "uncond_branch,branch,cbcond,uncond_cbcond,call,sibcall,call_no_delay_slot,multi")
 	   (const_string "false")
+	 (and (eq_attr "fix_tn0013" "true")
+	      (eq_attr "type" "fpdivs,fpsqrts,fpdivd,fpsqrtd"))
+	   (const_string "false")
 	 (and (eq_attr "fix_b2bst" "true") (eq_attr "type" "store,fpstore"))
 	   (const_string "false")
 	 (and (eq_attr "fix_ut699" "true") (eq_attr "type" "load,sload"))
diff --git a/gcc/config/sparc/sparc.opt b/gcc/config/sparc/sparc.opt
index 22267f5..43cd964 100644
--- a/gcc/config/sparc/sparc.opt
+++ b/gcc/config/sparc/sparc.opt
@@ -253,6 +253,10 @@ Enable workarounds for the errata of the GR712RC processor.
 TargetVariable
 unsigned int sparc_fix_b2bst
 
+;; Enable workaround for TN-0013 errata
+TargetVariable
+unsigned int sparc_fix_tn0013
+
 Mask(LONG_DOUBLE_128)
 ;; Use 128-bit long double
 
-- 
2.9.3

Re: [PATCH 4/4] [SPARC] Errata workaround for GRLIB-TN-0013

[Eric Botcazou]

> 2017-11-17  Daniel Cederman  <cederman@gaisler.com>
> 
> 	* config/sparc/sparc.c (fpop_reg_depend_p): New function.
> 	(div_sqrt_insn_p): New function.
> 	(sparc_do_work_around_errata): Insert NOP instructions to
> 	prevent sequences that could trigger the TN-0013 errata for
> 	certain LEON3 processors.
> 	(pass_work_around_errata::gate): Also test sparc_fix_tn0013.
> 	(sparc_option_override): Set sparc_fix_tn0013 appropriately.
> 	* config/sparc/sparc.md (fix_tn0013): New attribute.
> 	(in_branch_delay): Prevent div and sqrt in delay slot if fix_tn0013.
> 	* config/sparc/sparc.opt (sparc_fix_tn0013: New variable.

OK for mainline and 7 branch modulo the following nits:

> +/* True if any of the floating-point instruction's source
> +   registers is the same as the provided register.  */
> +
> +static int
> fpop_reg_depend_p (rtx_insn *insn, unsigned int reg)

'bool' instead of 'int'.  "True if any of INSN's source register(s) is REG".

The function's name is awkward.  insn_uses_reg_p?  insn_reads_reg_p?

> +/* True if the instruction is floating-point division or
> +   floating-point square-root.  */
> +
> +static int
> +div_sqrt_insn_p (rtx_insn *insn)

'bool' instead of 'int'. "True if INSN is a floating-point ..."

> +  if ( GET_CODE (PATTERN (insn)) != SET)
> +    return false;

No space before GET_CODE.

> @@ -1064,6 +1097,80 @@ sparc_do_work_around_errata (void)
>  	    insert_nop = true;
>  	}
> 
> +      /* Look for sequences that could trigger the TN-0013 errata.  */
> +      if (sparc_fix_tn0013
> +	  && NONJUMP_INSN_P (insn)
> +	  && div_sqrt_insn_p (insn))
> +	{
> +	  int i;
> +	  int fp_found = 0;
> +	  unsigned int dest_reg;
> +	  rtx_insn *after;
> +
> +	  dest_reg = REGNO (SET_DEST (single_set (insn)));

	  const unsigned int dest_reg = REGNO (SET_DEST (single_set (insn)));

> +	  next = next_active_insn (insn);
> +	  if (!next)
> +	    break;
> +
> +	  for (after = next, i = 0; i < 4; i++)
> +	    {
> +	      /* Count floating-point operations.  */
> +	      if (i != 3
> +		  && fpop_insn_p (after))

On a single line.

> +		  /* If the insn uses the destination register of
> +		     the div/sqrt, then it cannot be problematic.  */
> +		  if (fpop_reg_depend_p (after, dest_reg))
> +		    break;
> +		  fp_found++;
> +		}
> +
> +	      /* Count floating-point loads.  */
> +	      if (i != 3
> +		  && (set = single_set (after)) != NULL_RTX
> +		  && REG_P (SET_DEST (set))
> +		  && REGNO (SET_DEST (set)) > 31)
> +		{
> +		  /* If the insn uses the destination register of
> +		     the div/sqrt, then it cannot be problematic.  */
> +		  if (REGNO (SET_DEST (set)) == dest_reg)
> +		    break;
> +		  fp_found++;
> +		}
> +
> +	      /* Check if this is a problematic sequence.  */
> +	      if (i > 1
> +		  && fp_found >= 2
> +		  && div_sqrt_insn_p (after))
> +		{
> +		  /* Add extra NOP to prevent second version of
> +		     problematic sequence.  */
> +		  if (i == 2)
> +		    emit_insn_before (gen_nop (), next);
> +		  insert_nop = true;
> +		  break;
> +		}

So you emit 2 NOPs in a row if i == 2?

> diff --git a/gcc/config/sparc/sparc.opt b/gcc/config/sparc/sparc.opt
> index 22267f5..43cd964 100644
> --- a/gcc/config/sparc/sparc.opt
> +++ b/gcc/config/sparc/sparc.opt
> @@ -253,6 +253,10 @@ Enable workarounds for the errata of the GR712RC
> processor. TargetVariable
>  unsigned int sparc_fix_b2bst
> 
> +;; Enable workaround for TN-0013 errata
> +TargetVariable
> +unsigned int sparc_fix_tn0013

Can we find a more evocative name for the variable?

-- 
Eric Botcazou

[PATCH 3/4] [SPARC] Errata workaround for GRLIB-TN-0010

[Daniel Cederman]

This patch provides a workaround for the errata described in GRLIB-TN-0010.

If the workaround is enabled it will:

* Insert a NOP between load instruction and atomic
  instruction (swap, ldstub, casa).

* Insert a NOP at branch target if load in delay slot
  and atomic instruction at branch target.

It is applicable to UT700.

gcc/ChangeLog:

2017-11-17  Daniel Cederman  <cederman@gaisler.com>

	* config/sparc/sparc.c (atomic_insn_p): New function.
	(sparc_do_work_around_errata): Insert NOP instructions to
	prevent sequences that could trigger the TN-0010 errata for
	UT700.
	* config/sparc/sync.md (atomic_compare_and_swap_leon3_1): Make
	instruction referable in atomic_insns_p.
---
 gcc/config/sparc/sparc.c | 43 +++++++++++++++++++++++++++++++++++++++++++
 gcc/config/sparc/sync.md |  2 +-
 2 files changed, 44 insertions(+), 1 deletion(-)

diff --git a/gcc/config/sparc/sparc.c b/gcc/config/sparc/sparc.c
index 8f6eb48..9faf774 100644
--- a/gcc/config/sparc/sparc.c
+++ b/gcc/config/sparc/sparc.c
@@ -970,6 +970,22 @@ fpop_insn_p (rtx_insn *insn)
     }
 }
 
+/* True if complex atomic instruction.  */
+
+static int
+atomic_insn_p (rtx_insn *insn)
+{
+  switch (INSN_CODE (insn))
+    {
+    case CODE_FOR_swapsi:
+    case CODE_FOR_ldstub:
+    case CODE_FOR_atomic_compare_and_swap_leon3_1:
+      return true;
+    default:
+      return false;
+    }
+}
+
 /* We use a machine specific pass to enable workarounds for errata.
 
    We need to have the (essentially) final form of the insn stream in order
@@ -1021,6 +1037,33 @@ sparc_do_work_around_errata (void)
 	    emit_insn_before (gen_nop (), target);
 	}
 
+      /* Insert a NOP between load instruction and atomic
+	 instruction. Insert a NOP at branch target if load
+	 in delay slot and atomic instruction at branch target.  */
+      if (sparc_fix_ut700
+	  && NONJUMP_INSN_P (insn)
+	  && (set = single_set (insn)) != NULL_RTX
+	  && MEM_P (SET_SRC (set))
+	  && REG_P (SET_DEST (set)))
+	{
+	  if (jump)
+	    {
+	      rtx_insn *target;
+
+	      target = next_active_insn (JUMP_LABEL_AS_INSN (jump));
+	      if (target
+		  && atomic_insn_p (target))
+		emit_insn_before (gen_nop (), target);
+	    }
+
+	  next = next_active_insn (insn);
+	  if (!next)
+	    break;
+
+	  if (atomic_insn_p (next))
+	    insert_nop = true;
+	}
+
       /* Look for either of these two sequences:
 
 	 Sequence A:
diff --git a/gcc/config/sparc/sync.md b/gcc/config/sparc/sync.md
index d5c505b..b1baa73 100644
--- a/gcc/config/sparc/sync.md
+++ b/gcc/config/sparc/sync.md
@@ -212,7 +212,7 @@
   "cas<modesuffix>\t%1, %2, %0"
   [(set_attr "type" "multi")])
 
-(define_insn "*atomic_compare_and_swap_leon3_1"
+(define_insn "atomic_compare_and_swap_leon3_1"
   [(set (match_operand:SI 0 "register_operand" "=r")
 	(match_operand:SI 1 "mem_noofs_operand" "+w"))
    (set (match_dup 1)
-- 
2.9.3

Re: [PATCH 3/4] [SPARC] Errata workaround for GRLIB-TN-0010

[Eric Botcazou]

> 2017-11-17  Daniel Cederman  <cederman@gaisler.com>
> 
> 	* config/sparc/sparc.c (atomic_insn_p): New function.
> 	(sparc_do_work_around_errata): Insert NOP instructions to
> 	prevent sequences that could trigger the TN-0010 errata for
> 	UT700.
> 	* config/sparc/sync.md (atomic_compare_and_swap_leon3_1): Make
> 	instruction referable in atomic_insns_p.

OK for mainline and 7 branch modulo the following nits:

> +/* True if complex atomic instruction.  */
> +
> +static int
> +atomic_insn_p (rtx_insn *insn)

'bool' instead of 'int'.  "True if INSN is an atomic instruction."

I'd rename the function into "atomic_insn_for_leon3_p" since you're handling 
only the LEON3 instructions.

> @@ -1021,6 +1037,33 @@ sparc_do_work_around_errata (void
>  	    emit_insn_before (gen_nop (), target);
>  	}
> 
> +      /* Insert a NOP between load instruction and atomic
> +	 instruction. Insert a NOP at branch target if load
> +	 in delay slot and atomic instruction at branch target.  */

Double space after colon.

> +      if (sparc_fix_ut700
> +	  && NONJUMP_INSN_P (insn)
> +	  && (set = single_set (insn)) != NULL_RTX
> +	  && MEM_P (SET_SRC (set))
> +	  && REG_P (SET_DEST (set)))
> +	{
> +	  if (jump)
> +	    {
> +	      rtx_insn *target;
> +
> +	      target = next_active_insn (JUMP_LABEL_AS_INSN (jump));

	      rtx_insn *target
		= next_active_insn (JUMP_LABEL_AS_INSN (jump));


> +	      if (target
> +		  && atomic_insn_p (target))

On a single line.

-- 
Eric Botcazou