Hexagon: add support for new v4+ registers

Add support for a couple new v4+ registers, along with
newer save/restore pt_regs.

Signed-off-by: Richard Kuo <rkuo@codeaurora.org>

8 files changed, 206 insertions, 48 deletions

diff --git a/arch/hexagon/include/uapi/asm/registers.h b/arch/hexagon/include/uapi/asm/registers.h
index c20406f63b5..80504155ca3 100644
--- a/arch/hexagon/include/uapi/asm/registers.h
+++ b/arch/hexagon/include/uapi/asm/registers.h
@@ -57,10 +57,17 @@ struct pt_regs {
 	};
 	union {
 		struct {
-			unsigned long gp;
 			unsigned long ugp;
+			unsigned long gp;
+		};
+		long long int gpugp;
+	};
+	union {
+		struct {
+			unsigned long cs0;
+			unsigned long cs1;
 		};
-		long long int ugpgp;
+		long long int cs1cs0;
 	};
 	/*
 	* Be extremely careful with rearranging these, if at all.  Some code
diff --git a/arch/hexagon/include/uapi/asm/user.h b/arch/hexagon/include/uapi/asm/user.h
index cef13ee1413..3dae94d9ced 100644
--- a/arch/hexagon/include/uapi/asm/user.h
+++ b/arch/hexagon/include/uapi/asm/user.h
@@ -55,9 +55,15 @@ struct user_regs_struct {
 	unsigned long pc;
 	unsigned long cause;
 	unsigned long badva;
+#if CONFIG_HEXAGON_ARCH_VERSION < 4
 	unsigned long pad1;  /* pad out to 48 words total */
 	unsigned long pad2;  /* pad out to 48 words total */
 	unsigned long pad3;  /* pad out to 48 words total */
+#else
+	unsigned long cs0;
+	unsigned long cs1;
+	unsigned long pad1;  /* pad out to 48 words total */
+#endif
 };
 
 #endif
diff --git a/arch/hexagon/kernel/asm-offsets.c b/arch/hexagon/kernel/asm-offsets.c
index 2d5e84d3b00..181515ba825 100644
--- a/arch/hexagon/kernel/asm-offsets.c
+++ b/arch/hexagon/kernel/asm-offsets.c
@@ -44,7 +44,8 @@ int main(void)
 
 	COMMENT("Hexagon pt_regs definitions");
 	OFFSET(_PT_SYSCALL_NR, pt_regs, syscall_nr);
-	OFFSET(_PT_UGPGP, pt_regs, ugpgp);
+	OFFSET(_PT_GPUGP, pt_regs, gpugp);
+	OFFSET(_PT_CS1CS0, pt_regs, cs1cs0);
 	OFFSET(_PT_R3130, pt_regs, r3130);
 	OFFSET(_PT_R2928, pt_regs, r2928);
 	OFFSET(_PT_R2726, pt_regs, r2726);
diff --git a/arch/hexagon/kernel/kgdb.c b/arch/hexagon/kernel/kgdb.c
index 34464537064..19f9e6ca5e3 100644
--- a/arch/hexagon/kernel/kgdb.c
+++ b/arch/hexagon/kernel/kgdb.c
@@ -70,6 +70,8 @@ struct dbg_reg_def_t dbg_reg_def[DBG_MAX_REG_NUM] = {
 	{ "lc1", GDB_SIZEOF_REG, offsetof(struct pt_regs, lc1)},
 	{ " gp", GDB_SIZEOF_REG, offsetof(struct pt_regs, gp)},
 	{ "ugp", GDB_SIZEOF_REG, offsetof(struct pt_regs, ugp)},
+	{ "cs0", GDB_SIZEOF_REG, offsetof(struct pt_regs, cs0)},
+	{ "cs1", GDB_SIZEOF_REG, offsetof(struct pt_regs, cs1)},
 	{ "psp", GDB_SIZEOF_REG, offsetof(struct pt_regs, hvmer.vmpsp)},
 	{ "elr", GDB_SIZEOF_REG, offsetof(struct pt_regs, hvmer.vmel)},
 	{ "est", GDB_SIZEOF_REG, offsetof(struct pt_regs, hvmer.vmest)},
diff --git a/arch/hexagon/kernel/ptrace.c b/arch/hexagon/kernel/ptrace.c
index 670b1b0bee6..3982d9c9ec2 100644
--- a/arch/hexagon/kernel/ptrace.c
+++ b/arch/hexagon/kernel/ptrace.c
@@ -76,6 +76,10 @@ static int genregs_get(struct task_struct *target,
 	dummy = pt_cause(regs);
 	ONEXT(&dummy, cause);
 	ONEXT(&pt_badva(regs), badva);
+#if CONFIG_HEXAGON_ARCH_VERSION >=4
+	ONEXT(&regs->cs0, cs0);
+	ONEXT(&regs->cs1, cs1);
+#endif
 
 	/* Pad the rest with zeros, if needed */
 	if (!ret)
@@ -123,6 +127,11 @@ static int genregs_set(struct task_struct *target,
 	INEXT(&bucket, cause);
 	INEXT(&bucket, badva);
 
+#if CONFIG_HEXAGON_ARCH_VERSION >=4
+	INEXT(&regs->cs0, cs0);
+	INEXT(&regs->cs1, cs1);
+#endif
+
 	/* Ignore the rest, if needed */
 	if (!ret)
 		ret = user_regset_copyin_ignore(&pos, &count, &kbuf, &ubuf,
diff --git a/arch/hexagon/kernel/signal.c b/arch/hexagon/kernel/signal.c
index 60fa2ca3202..a1492a69752 100644
--- a/arch/hexagon/kernel/signal.c
+++ b/arch/hexagon/kernel/signal.c
@@ -66,7 +66,10 @@ static int setup_sigcontext(struct pt_regs *regs, struct sigcontext __user *sc)
 	err |= __put_user(regs->preds, &sc->sc_regs.p3_0);
 	err |= __put_user(regs->gp, &sc->sc_regs.gp);
 	err |= __put_user(regs->ugp, &sc->sc_regs.ugp);
-
+#if CONFIG_HEXAGON_ARCH_VERSION >= 4
+	err |= __put_user(regs->cs0, &sc->sc_regs.cs0);
+	err |= __put_user(regs->cs1, &sc->sc_regs.cs1);
+#endif
 	tmp = pt_elr(regs); err |= __put_user(tmp, &sc->sc_regs.pc);
 	tmp = pt_cause(regs); err |= __put_user(tmp, &sc->sc_regs.cause);
 	tmp = pt_badva(regs); err |= __put_user(tmp, &sc->sc_regs.badva);
@@ -93,7 +96,10 @@ static int restore_sigcontext(struct pt_regs *regs,
 	err |= __get_user(regs->preds, &sc->sc_regs.p3_0);
 	err |= __get_user(regs->gp, &sc->sc_regs.gp);
 	err |= __get_user(regs->ugp, &sc->sc_regs.ugp);
-
+#if CONFIG_HEXAGON_ARCH_VERSION >= 4
+	err |= __get_user(regs->cs0, &sc->sc_regs.cs0);
+	err |= __get_user(regs->cs1, &sc->sc_regs.cs1);
+#endif
 	err |= __get_user(tmp, &sc->sc_regs.pc); pt_set_elr(regs, tmp);
 
 	return err;
diff --git a/arch/hexagon/kernel/vm_entry.S b/arch/hexagon/kernel/vm_entry.S
index 425e50c694f..ffe4a424b70 100644
--- a/arch/hexagon/kernel/vm_entry.S
+++ b/arch/hexagon/kernel/vm_entry.S
@@ -45,48 +45,88 @@
  * number in the case where we decode a system call (trap0(#1)).
  */
 
+#if CONFIG_HEXAGON_ARCH_VERSION < 4
 #define save_pt_regs()\
-	memd(R0 + #_PT_R3130) = R31:30; \
+ memd(R0 + #_PT_R3130) = R31:30; \
+ { memw(R0 + #_PT_R2928) = R28; \
+   R31 = memw(R0 + #_PT_ER_VMPSP); }\
+ { memw(R0 + #(_PT_R2928 + 4)) = R31; \
+   R31 = ugp; } \
+ { memd(R0 + #_PT_R2726) = R27:26; \
+   R30 = gp ; } \
+ memd(R0 + #_PT_R2524) = R25:24; \
+ memd(R0 + #_PT_R2322) = R23:22; \
+ memd(R0 + #_PT_R2120) = R21:20; \
+ memd(R0 + #_PT_R1918) = R19:18; \
+ memd(R0 + #_PT_R1716) = R17:16; \
+ memd(R0 + #_PT_R1514) = R15:14; \
+ memd(R0 + #_PT_R1312) = R13:12; \
+ { memd(R0 + #_PT_R1110) = R11:10; \
+   R15 = lc0; } \
+ { memd(R0 + #_PT_R0908) = R9:8; \
+   R14 = sa0; } \
+ { memd(R0 + #_PT_R0706) = R7:6; \
+   R13 = lc1; } \
+ { memd(R0 + #_PT_R0504) = R5:4; \
+   R12 = sa1; } \
+ { memd(R0 + #_PT_GPUGP) = R31:30; \
+   R11 = m1; \
+   R2.H = #HI(_THREAD_SIZE); } \
+ { memd(R0 + #_PT_LC0SA0) = R15:14; \
+   R10 = m0; \
+   R2.L = #LO(_THREAD_SIZE); } \
+ { memd(R0 + #_PT_LC1SA1) = R13:12; \
+   R15 = p3:0; \
+   R2 = neg(R2); } \
+ { memd(R0 + #_PT_M1M0) = R11:10; \
+   R14  = usr; \
+   R2 = and(R0,R2); } \
+ { memd(R0 + #_PT_PREDSUSR) =  R15:14; \
+   THREADINFO_REG = R2; } \
+ { r24 = memw(THREADINFO_REG + #_THREAD_INFO_PT_REGS); \
+   memw(THREADINFO_REG + #_THREAD_INFO_PT_REGS) = R0; \
+   R2 = #-1; } \
+ { memw(R0 + #_PT_SYSCALL_NR) = R2; \
+   R30 = #0; }
+#else
+/* V4+ */
+/* the # ## # syntax inserts a literal ## */
+#define save_pt_regs()\
+	{ memd(R0 + #_PT_R3130) = R31:30; \
+		R30 = memw(R0 + #_PT_ER_VMPSP); }\
 	{ memw(R0 + #_PT_R2928) = R28; \
-	  R31 = memw(R0 + #_PT_ER_VMPSP); }\
-	{ memw(R0 + #(_PT_R2928 + 4)) = R31; \
-	  R31 = ugp; } \
-	{ memd(R0 + #_PT_R2726) = R27:26; \
-	  R30 = gp ; } \
-	memd(R0 + #_PT_R2524) = R25:24; \
-	memd(R0 + #_PT_R2322) = R23:22; \
-	memd(R0 + #_PT_R2120) = R21:20; \
-	memd(R0 + #_PT_R1918) = R19:18; \
-	memd(R0 + #_PT_R1716) = R17:16; \
-	memd(R0 + #_PT_R1514) = R15:14; \
-	memd(R0 + #_PT_R1312) = R13:12; \
+		memw(R0 + #(_PT_R2928 + 4)) = R30; }\
+	{ R31:30 = C11:10; \
+		memd(R0 + #_PT_R2726) = R27:26; \
+		memd(R0 + #_PT_R2524) = R25:24; }\
+	{ memd(R0 + #_PT_R2322) = R23:22; \
+		memd(R0 + #_PT_R2120) = R21:20; }\
+	{ memd(R0 + #_PT_R1918) = R19:18; \
+		memd(R0 + #_PT_R1716) = R17:16; }\
+	{ memd(R0 + #_PT_R1514) = R15:14; \
+		memd(R0 + #_PT_R1312) = R13:12; \
+		R17:16 = C13:12; }\
 	{ memd(R0 + #_PT_R1110) = R11:10; \
-	  R15 = lc0; } \
-	{ memd(R0 + #_PT_R0908) = R9:8; \
-	  R14 = sa0; } \
+		memd(R0 + #_PT_R0908) = R9:8; \
+	  R15:14 = C1:0; } \
 	{ memd(R0 + #_PT_R0706) = R7:6; \
-	  R13 = lc1; } \
-	{ memd(R0 + #_PT_R0504) = R5:4; \
-	  R12 = sa1; } \
-	{ memd(R0 + #_PT_UGPGP) = R31:30; \
-	  R11 = m1; \
-	  R2.H = #HI(_THREAD_SIZE); } \
-	{ memd(R0 + #_PT_LC0SA0) = R15:14; \
-	  R10 = m0; \
-	  R2.L = #LO(_THREAD_SIZE); } \
-	{ memd(R0 + #_PT_LC1SA1) = R13:12; \
-	  R15 = p3:0; \
-	  R2 = neg(R2); } \
+		memd(R0 + #_PT_R0504) = R5:4; \
+    R13:12 = C3:2; } \
+	{ memd(R0 + #_PT_GPUGP) = R31:30; \
+		memd(R0 + #_PT_LC0SA0) = R15:14; \
+	  R11:10 = C7:6; }\
+	{	THREADINFO_REG = and(R0, # ## #-_THREAD_SIZE); \
+		memd(R0 + #_PT_LC1SA1) = R13:12; \
+	  R15 = p3:0; }\
 	{ memd(R0 + #_PT_M1M0) = R11:10; \
-	  R14  = usr; \
-	  R2 = and(R0,R2); } \
-	{ memd(R0 + #_PT_PREDSUSR) =  R15:14; \
-	  THREADINFO_REG = R2; } \
+		memw(R0 + #_PT_PREDSUSR + 4) =  R15; }\
 	{ r24 = memw(THREADINFO_REG + #_THREAD_INFO_PT_REGS); \
 	  memw(THREADINFO_REG + #_THREAD_INFO_PT_REGS) = R0; \
 	  R2 = #-1; } \
 	{ memw(R0 + #_PT_SYSCALL_NR) = R2; \
+		memd(R0 + #_PT_CS1CS0) = R17:16; \
 	  R30 = #0; }
+#endif
 
 /*
  * Restore registers and thread_info.regs state. THREADINFO_REG
@@ -94,6 +134,7 @@
  * preserved. Don't restore R29 (SP) until later.
  */
 
+#if CONFIG_HEXAGON_ARCH_VERSION < 4
 #define restore_pt_regs() \
 	{ memw(THREADINFO_REG + #_THREAD_INFO_PT_REGS) = R24; \
 	  R15:14 = memd(R0 + #_PT_PREDSUSR); } \
@@ -121,11 +162,44 @@
 	  R23:22 = memd(R0 + #_PT_R2322); } \
 	{ R25:24 = memd(R0 + #_PT_R2524); \
 	  R27:26 = memd(R0 + #_PT_R2726); } \
-	R31:30 = memd(R0 + #_PT_UGPGP); \
+	R31:30 = memd(R0 + #_PT_GPUGP); \
 	{ R28 = memw(R0 + #_PT_R2928); \
 	  ugp = R31; } \
 	{ R31:30 = memd(R0 + #_PT_R3130); \
 	  gp = R30; }
+#else
+/* V4+ */
+#define restore_pt_regs() \
+	{ memw(THREADINFO_REG + #_THREAD_INFO_PT_REGS) = R24; \
+	  R15:14 = memd(R0 + #_PT_PREDSUSR); } \
+	{ R11:10 = memd(R0 + #_PT_M1M0); \
+		R13:12 = memd(R0 + #_PT_LC1SA1); \
+		p3:0 = R15; } \
+	{ R15:14 = memd(R0 + #_PT_LC0SA0); \
+		R3:2 = memd(R0 + #_PT_R0302); \
+		usr = R14; } \
+	{ R5:4 = memd(R0 + #_PT_R0504); \
+		R7:6 = memd(R0 + #_PT_R0706); \
+		C7:6 = R11:10; }\
+	{ R9:8 = memd(R0 + #_PT_R0908); \
+		R11:10 = memd(R0 + #_PT_R1110); \
+    C3:2 = R13:12; }\
+	{ R13:12 = memd(R0 + #_PT_R1312); \
+	  R15:14 = memd(R0 + #_PT_R1514); \
+		C1:0 = R15:14; }\
+	{ R17:16 = memd(R0 + #_PT_R1716); \
+	  R19:18 = memd(R0 + #_PT_R1918); } \
+	{ R21:20 = memd(R0 + #_PT_R2120); \
+	  R23:22 = memd(R0 + #_PT_R2322); } \
+	{ R25:24 = memd(R0 + #_PT_R2524); \
+	  R27:26 = memd(R0 + #_PT_R2726); } \
+	R31:30 = memd(R0 + #_PT_CS1CS0); \
+	{ C13:12 = R31:30; \
+		R31:30 = memd(R0 + #_PT_GPUGP) ; \
+		R28 = memw(R0 + #_PT_R2928); }\
+	{ C11:10 = R31:30; \
+		R31:30 = memd(R0 + #_PT_R3130); }
+#endif
 
 	/*
 	 * Clears off enough space for the rest of pt_regs; evrec is a part
@@ -139,6 +213,7 @@
  * Need to save off R0, R1, R2, R3 immediately.
  */
 
+#if CONFIG_HEXAGON_ARCH_VERSION < 4
 #define	vm_event_entry(CHandler) \
 	{ \
 		R29 = add(R29, #-(_PT_REGS_SIZE)); \
@@ -158,6 +233,34 @@
 		R1.H = #HI(CHandler); \
 		jump event_dispatch; \
 	}
+#else
+/* V4+ */
+/* turn on I$ prefetch early */
+/* the # ## # syntax inserts a literal ## */
+#define	vm_event_entry(CHandler) \
+	{ \
+		R29 = add(R29, #-(_PT_REGS_SIZE)); \
+		memd(R29 + #(_PT_R0100 + -_PT_REGS_SIZE)) = R1:0; \
+		memd(R29 + #(_PT_R0302 + -_PT_REGS_SIZE)) = R3:2; \
+		R0 = usr; \
+	} \
+	{ \
+		memw(R29 + #_PT_PREDSUSR) = R0; \
+		R0 = setbit(R0, #16); \
+	} \
+	usr = R0; \
+	R1:0 = G1:0; \
+	{ \
+		memd(R29 + #_PT_ER_VMEL) = R1:0; \
+		R1 = # ## #(CHandler); \
+		R3:2 = G3:2; \
+	} \
+	{ \
+		R0 = R29; \
+		memd(R29 + #_PT_ER_VMPSP) = R3:2; \
+		jump event_dispatch; \
+	}
+#endif
 
 .text
 	/*
@@ -184,8 +287,10 @@ event_dispatch:
 
 	/*  "Nested control path" -- if the previous mode was kernel  */
 	R0 = memw(R29 + #_PT_ER_VMEST);
-	P0 = tstbit(R0, #HVM_VMEST_UM_SFT);
-	if !P0 jump restore_all;
+	{
+		P0 = tstbit(R0, #HVM_VMEST_UM_SFT);
+		if (!P0.new) jump:nt restore_all;
+	}
 	/*
 	 * Returning from system call, normally coming back from user mode
 	 */
@@ -198,11 +303,18 @@ return_from_syscall:
 	 * Coming back from the C-world, our thread info pointer
 	 * should be in the designated register (usually R19)
 	 */
+#if CONFIG_HEXAGON_ARCH_VERSION < 4
 	R1.L = #LO(_TIF_ALLWORK_MASK)
 	{
 		R1.H = #HI(_TIF_ALLWORK_MASK);
 		R0 = memw(THREADINFO_REG + #_THREAD_INFO_FLAGS);
 	}
+#else
+	{
+		R1 = ##_TIF_ALLWORK_MASK;
+		R0 = memw(THREADINFO_REG + #_THREAD_INFO_FLAGS);
+	}
+#endif
 
 	/*
 	 * Compare against the "return to userspace" _TIF_WORK_MASK
@@ -222,10 +334,14 @@ work_pending:
 work_notifysig:
 	/*  this is the part that's kind of fuzzy.  */
 	R1 = and(R0, #(_TIF_SIGPENDING | _TIF_NOTIFY_RESUME));
-	P0 = cmp.eq(R1, #0);
-	if P0 jump restore_all
-	R1 = R0; 	/* unsigned long thread_info_flags */
-	R0 = R29;	/* regs should still be at top of stack  */
+	{
+		P0 = cmp.eq(R1, #0);
+		if P0.new jump:t restore_all;
+	}
+	{
+		R1 = R0; 	/* unsigned long thread_info_flags */
+		R0 = R29;	/* regs should still be at top of stack  */
+	}
 	call do_notify_resume
 
 restore_all:
@@ -235,14 +351,23 @@ restore_all:
 
 	/*  do the setregs here for VM 0.5  */
 	/*  R29 here should already be pointing at pt_regs  */
-	R1:0 = memd(R29 + #_PT_ER_VMEL);
-	R3:2 = memd(R29 + #_PT_ER_VMPSP);
+	{
+		R1:0 = memd(R29 + #_PT_ER_VMEL);
+		R3:2 = memd(R29 + #_PT_ER_VMPSP);
+	}
+#if CONFIG_HEXAGON_ARCH_VERSION < 4
 	trap1(#HVM_TRAP1_VMSETREGS);
+#else
+	G1:0 = R1:0;
+	G3:2 = R3:2;
+#endif
 
 	R0 = R29
 	restore_pt_regs()
-	R1:0 = memd(R29 + #_PT_R0100);
-	R29 = add(R29, #_PT_REGS_SIZE);
+	{
+		R1:0 = memd(R29 + #_PT_R0100);
+		R29 = add(R29, #_PT_REGS_SIZE);
+	}
 	trap1(#HVM_TRAP1_VMRTE)
 	/* Notreached */
 
diff --git a/arch/hexagon/kernel/vm_events.c b/arch/hexagon/kernel/vm_events.c
index 9b5a4a295a6..5b81bacf712 100644
--- a/arch/hexagon/kernel/vm_events.c
+++ b/arch/hexagon/kernel/vm_events.c
@@ -42,6 +42,8 @@ void show_regs(struct pt_regs *regs)
 	       regs->lc1, regs->sa1, regs->m1);
 	printk(KERN_EMERG "gp: \t0x%08lx   ugp: 0x%08lx   usr: 0x%08lx\n",
 	       regs->gp, regs->ugp, regs->usr);
+	printk(KERN_EMERG "cs0: \t0x%08lx   cs1: 0x%08lx\n",
+	       regs->cs0, regs->cs1);
 	printk(KERN_EMERG "r0: \t0x%08lx %08lx %08lx %08lx\n", regs->r00,
 		regs->r01,
 		regs->r02,