UPSTREAM: MIPS: math-emu: Add FP emu debugfs stats for individual instructions

Add FP emulation debugfs statistics for individual instructions. The
debugfs files that contain counter values are placed in a separate
directory called "instructions". This means that the default path for
these new stat is "/sys/kernel/debug/mips/fpuemustats/instructions".

Each instruction counter is mapped to the debugfs file that has the
same name as instruction name. The lowercase is choosen as more
commonly used case for instruction names.

One example of usage:

mips_host::/sys/kernel/debug/mips/fpuemustats/instructions # grep "" *

The shortened output of this command is:

abs.d:34
abs.s:5711
add.d:10401
add.s:399307
bc1eqz:3199
...
...
...
sub.s:167211
trunc.l.d:375
trunc.l.s:8054
trunc.w.d:421
trunc.w.s:27032

The limitation of this patch is that it handles R6 FP emulation
instructions only. There are altogether 114 handled instructions.

Signed-off-by: Miodrag Dinic <miodrag.dinic@imgtec.com>
Signed-off-by: Goran Ferenc <goran.ferenc@imgtec.com>
Signed-off-by: Aleksandar Markovic <aleksandar.markovic@imgtec.com>
Cc: Douglas Leung <douglas.leung@imgtec.com>
Cc: James Hogan <james.hogan@imgtec.com>
Cc: Maciej W. Rozycki <macro@imgtec.com>
Cc: Masahiro Yamada <yamada.masahiro@socionext.com>
Cc: Paul Burton <paul.burton@imgtec.com>
Cc: Petar Jovanovic <petar.jovanovic@imgtec.com>
Cc: Raghu Gandham <raghu.gandham@imgtec.com>
Cc: linux-mips@linux-mips.org
Cc: linux-kernel@vger.kernel.org
Patchwork: https://patchwork.linux-mips.org/patch/17145/
Signed-off-by: Ralf Baechle <ralf@linux-mips.org>
(cherry picked from commit 454854ace22f5a9fdd369a4e428493159a02f029)
Signed-off-by: Greg Kroah-Hartman <gregkh@google.com>

2 files changed, 535 insertions, 4 deletions

diff --git a/arch/mips/math-emu/cp1emu.c b/arch/mips/math-emu/cp1emu.c
index 008372d19a6a..aa44e16db1ab 100644
--- a/arch/mips/math-emu/cp1emu.c
+++ b/arch/mips/math-emu/cp1emu.c
@@ -1196,9 +1196,11 @@ emul:
 			bit0 = get_fpr32(fpr, 0) & 0x1;
 			switch (MIPSInst_RS(ir)) {
 			case bc1eqz_op:
+				MIPS_FPU_EMU_INC_STATS(bc1eqz);
 				cond = bit0 == 0;
 				break;
 			case bc1nez_op:
+				MIPS_FPU_EMU_INC_STATS(bc1nez);
 				cond = bit0 != 0;
 				break;
 			}
@@ -1685,15 +1687,19 @@ static int fpu_emu(struct pt_regs *xcp, struct mips_fpu_struct *ctx,
 		switch (MIPSInst_FUNC(ir)) {
 			/* binary ops */
 		case fadd_op:
+			MIPS_FPU_EMU_INC_STATS(add_s);
 			handler.b = ieee754sp_add;
 			goto scopbop;
 		case fsub_op:
+			MIPS_FPU_EMU_INC_STATS(sub_s);
 			handler.b = ieee754sp_sub;
 			goto scopbop;
 		case fmul_op:
+			MIPS_FPU_EMU_INC_STATS(mul_s);
 			handler.b = ieee754sp_mul;
 			goto scopbop;
 		case fdiv_op:
+			MIPS_FPU_EMU_INC_STATS(div_s);
 			handler.b = ieee754sp_div;
 			goto scopbop;
 
@@ -1702,6 +1708,7 @@ static int fpu_emu(struct pt_regs *xcp, struct mips_fpu_struct *ctx,
 			if (!cpu_has_mips_2_3_4_5_r)
 				return SIGILL;
 
+			MIPS_FPU_EMU_INC_STATS(sqrt_s);
 			handler.u = ieee754sp_sqrt;
 			goto scopuop;
 
@@ -1714,6 +1721,7 @@ static int fpu_emu(struct pt_regs *xcp, struct mips_fpu_struct *ctx,
 			if (!cpu_has_mips_4_5_64_r2_r6)
 				return SIGILL;
 
+			MIPS_FPU_EMU_INC_STATS(rsqrt_s);
 			handler.u = fpemu_sp_rsqrt;
 			goto scopuop;
 
@@ -1721,6 +1729,7 @@ static int fpu_emu(struct pt_regs *xcp, struct mips_fpu_struct *ctx,
 			if (!cpu_has_mips_4_5_64_r2_r6)
 				return SIGILL;
 
+			MIPS_FPU_EMU_INC_STATS(recip_s);
 			handler.u = fpemu_sp_recip;
 			goto scopuop;
 
@@ -1757,6 +1766,7 @@ static int fpu_emu(struct pt_regs *xcp, struct mips_fpu_struct *ctx,
 			if (!cpu_has_mips_r6)
 				return SIGILL;
 
+			MIPS_FPU_EMU_INC_STATS(seleqz_s);
 			SPFROMREG(rv.s, MIPSInst_FT(ir));
 			if (rv.w & 0x1)
 				rv.w = 0;
@@ -1768,6 +1778,7 @@ static int fpu_emu(struct pt_regs *xcp, struct mips_fpu_struct *ctx,
 			if (!cpu_has_mips_r6)
 				return SIGILL;
 
+			MIPS_FPU_EMU_INC_STATS(selnez_s);
 			SPFROMREG(rv.s, MIPSInst_FT(ir));
 			if (rv.w & 0x1)
 				SPFROMREG(rv.s, MIPSInst_FS(ir));
@@ -1781,6 +1792,7 @@ static int fpu_emu(struct pt_regs *xcp, struct mips_fpu_struct *ctx,
 			if (!cpu_has_mips_r6)
 				return SIGILL;
 
+			MIPS_FPU_EMU_INC_STATS(maddf_s);
 			SPFROMREG(ft, MIPSInst_FT(ir));
 			SPFROMREG(fs, MIPSInst_FS(ir));
 			SPFROMREG(fd, MIPSInst_FD(ir));
@@ -1794,6 +1806,7 @@ static int fpu_emu(struct pt_regs *xcp, struct mips_fpu_struct *ctx,
 			if (!cpu_has_mips_r6)
 				return SIGILL;
 
+			MIPS_FPU_EMU_INC_STATS(msubf_s);
 			SPFROMREG(ft, MIPSInst_FT(ir));
 			SPFROMREG(fs, MIPSInst_FS(ir));
 			SPFROMREG(fd, MIPSInst_FD(ir));
@@ -1807,6 +1820,7 @@ static int fpu_emu(struct pt_regs *xcp, struct mips_fpu_struct *ctx,
 			if (!cpu_has_mips_r6)
 				return SIGILL;
 
+			MIPS_FPU_EMU_INC_STATS(rint_s);
 			SPFROMREG(fs, MIPSInst_FS(ir));
 			rv.s = ieee754sp_rint(fs);
 			goto copcsr;
@@ -1818,6 +1832,7 @@ static int fpu_emu(struct pt_regs *xcp, struct mips_fpu_struct *ctx,
 			if (!cpu_has_mips_r6)
 				return SIGILL;
 
+			MIPS_FPU_EMU_INC_STATS(class_s);
 			SPFROMREG(fs, MIPSInst_FS(ir));
 			rv.w = ieee754sp_2008class(fs);
 			rfmt = w_fmt;
@@ -1830,6 +1845,7 @@ static int fpu_emu(struct pt_regs *xcp, struct mips_fpu_struct *ctx,
 			if (!cpu_has_mips_r6)
 				return SIGILL;
 
+			MIPS_FPU_EMU_INC_STATS(min_s);
 			SPFROMREG(ft, MIPSInst_FT(ir));
 			SPFROMREG(fs, MIPSInst_FS(ir));
 			rv.s = ieee754sp_fmin(fs, ft);
@@ -1842,6 +1858,7 @@ static int fpu_emu(struct pt_regs *xcp, struct mips_fpu_struct *ctx,
 			if (!cpu_has_mips_r6)
 				return SIGILL;
 
+			MIPS_FPU_EMU_INC_STATS(mina_s);
 			SPFROMREG(ft, MIPSInst_FT(ir));
 			SPFROMREG(fs, MIPSInst_FS(ir));
 			rv.s = ieee754sp_fmina(fs, ft);
@@ -1854,6 +1871,7 @@ static int fpu_emu(struct pt_regs *xcp, struct mips_fpu_struct *ctx,
 			if (!cpu_has_mips_r6)
 				return SIGILL;
 
+			MIPS_FPU_EMU_INC_STATS(max_s);
 			SPFROMREG(ft, MIPSInst_FT(ir));
 			SPFROMREG(fs, MIPSInst_FS(ir));
 			rv.s = ieee754sp_fmax(fs, ft);
@@ -1866,6 +1884,7 @@ static int fpu_emu(struct pt_regs *xcp, struct mips_fpu_struct *ctx,
 			if (!cpu_has_mips_r6)
 				return SIGILL;
 
+			MIPS_FPU_EMU_INC_STATS(maxa_s);
 			SPFROMREG(ft, MIPSInst_FT(ir));
 			SPFROMREG(fs, MIPSInst_FS(ir));
 			rv.s = ieee754sp_fmaxa(fs, ft);
@@ -1873,15 +1892,18 @@ static int fpu_emu(struct pt_regs *xcp, struct mips_fpu_struct *ctx,
 		}
 
 		case fabs_op:
+			MIPS_FPU_EMU_INC_STATS(abs_s);
 			handler.u = ieee754sp_abs;
 			goto scopuop;
 
 		case fneg_op:
+			MIPS_FPU_EMU_INC_STATS(neg_s);
 			handler.u = ieee754sp_neg;
 			goto scopuop;
 
 		case fmov_op:
 			/* an easy one */
+			MIPS_FPU_EMU_INC_STATS(mov_s);
 			SPFROMREG(rv.s, MIPSInst_FS(ir));
 			goto copcsr;
 
@@ -1924,12 +1946,14 @@ copcsr:
 			return SIGILL;	/* not defined */
 
 		case fcvtd_op:
+			MIPS_FPU_EMU_INC_STATS(cvt_d_s);
 			SPFROMREG(fs, MIPSInst_FS(ir));
 			rv.d = ieee754dp_fsp(fs);
 			rfmt = d_fmt;
 			goto copcsr;
 
 		case fcvtw_op:
+			MIPS_FPU_EMU_INC_STATS(cvt_w_s);
 			SPFROMREG(fs, MIPSInst_FS(ir));
 			rv.w = ieee754sp_tint(fs);
 			rfmt = w_fmt;
@@ -1942,6 +1966,15 @@ copcsr:
 			if (!cpu_has_mips_2_3_4_5_r)
 				return SIGILL;
 
+			if (MIPSInst_FUNC(ir) == fceil_op)
+				MIPS_FPU_EMU_INC_STATS(ceil_w_s);
+			if (MIPSInst_FUNC(ir) == ffloor_op)
+				MIPS_FPU_EMU_INC_STATS(floor_w_s);
+			if (MIPSInst_FUNC(ir) == fround_op)
+				MIPS_FPU_EMU_INC_STATS(round_w_s);
+			if (MIPSInst_FUNC(ir) == ftrunc_op)
+				MIPS_FPU_EMU_INC_STATS(trunc_w_s);
+
 			oldrm = ieee754_csr.rm;
 			SPFROMREG(fs, MIPSInst_FS(ir));
 			ieee754_csr.rm = MIPSInst_FUNC(ir);
@@ -1954,6 +1987,7 @@ copcsr:
 			if (!cpu_has_mips_r6)
 				return SIGILL;
 
+			MIPS_FPU_EMU_INC_STATS(sel_s);
 			SPFROMREG(fd, MIPSInst_FD(ir));
 			if (fd.bits & 0x1)
 				SPFROMREG(rv.s, MIPSInst_FT(ir));
@@ -1965,6 +1999,7 @@ copcsr:
 			if (!cpu_has_mips_3_4_5_64_r2_r6)
 				return SIGILL;
 
+			MIPS_FPU_EMU_INC_STATS(cvt_l_s);
 			SPFROMREG(fs, MIPSInst_FS(ir));
 			rv.l = ieee754sp_tlong(fs);
 			rfmt = l_fmt;
@@ -1977,6 +2012,15 @@ copcsr:
 			if (!cpu_has_mips_3_4_5_64_r2_r6)
 				return SIGILL;
 
+			if (MIPSInst_FUNC(ir) == fceill_op)
+				MIPS_FPU_EMU_INC_STATS(ceil_l_s);
+			if (MIPSInst_FUNC(ir) == ffloorl_op)
+				MIPS_FPU_EMU_INC_STATS(floor_l_s);
+			if (MIPSInst_FUNC(ir) == froundl_op)
+				MIPS_FPU_EMU_INC_STATS(round_l_s);
+			if (MIPSInst_FUNC(ir) == ftruncl_op)
+				MIPS_FPU_EMU_INC_STATS(trunc_l_s);
+
 			oldrm = ieee754_csr.rm;
 			SPFROMREG(fs, MIPSInst_FS(ir));
 			ieee754_csr.rm = MIPSInst_FUNC(ir);
@@ -2018,15 +2062,19 @@ copcsr:
 		switch (MIPSInst_FUNC(ir)) {
 			/* binary ops */
 		case fadd_op:
+			MIPS_FPU_EMU_INC_STATS(add_d);
 			handler.b = ieee754dp_add;
 			goto dcopbop;
 		case fsub_op:
+			MIPS_FPU_EMU_INC_STATS(sub_d);
 			handler.b = ieee754dp_sub;
 			goto dcopbop;
 		case fmul_op:
+			MIPS_FPU_EMU_INC_STATS(mul_d);
 			handler.b = ieee754dp_mul;
 			goto dcopbop;
 		case fdiv_op:
+			MIPS_FPU_EMU_INC_STATS(div_d);
 			handler.b = ieee754dp_div;
 			goto dcopbop;
 
@@ -2035,6 +2083,7 @@ copcsr:
 			if (!cpu_has_mips_2_3_4_5_r)
 				return SIGILL;
 
+			MIPS_FPU_EMU_INC_STATS(sqrt_d);
 			handler.u = ieee754dp_sqrt;
 			goto dcopuop;
 		/*
@@ -2046,12 +2095,14 @@ copcsr:
 			if (!cpu_has_mips_4_5_64_r2_r6)
 				return SIGILL;
 
+			MIPS_FPU_EMU_INC_STATS(rsqrt_d);
 			handler.u = fpemu_dp_rsqrt;
 			goto dcopuop;
 		case frecip_op:
 			if (!cpu_has_mips_4_5_64_r2_r6)
 				return SIGILL;
 
+			MIPS_FPU_EMU_INC_STATS(recip_d);
 			handler.u = fpemu_dp_recip;
 			goto dcopuop;
 		case fmovc_op:
@@ -2085,6 +2136,7 @@ copcsr:
 			if (!cpu_has_mips_r6)
 				return SIGILL;
 
+			MIPS_FPU_EMU_INC_STATS(seleqz_d);
 			DPFROMREG(rv.d, MIPSInst_FT(ir));
 			if (rv.l & 0x1)
 				rv.l = 0;
@@ -2096,6 +2148,7 @@ copcsr:
 			if (!cpu_has_mips_r6)
 				return SIGILL;
 
+			MIPS_FPU_EMU_INC_STATS(selnez_d);
 			DPFROMREG(rv.d, MIPSInst_FT(ir));
 			if (rv.l & 0x1)
 				DPFROMREG(rv.d, MIPSInst_FS(ir));
@@ -2109,6 +2162,7 @@ copcsr:
 			if (!cpu_has_mips_r6)
 				return SIGILL;
 
+			MIPS_FPU_EMU_INC_STATS(maddf_d);
 			DPFROMREG(ft, MIPSInst_FT(ir));
 			DPFROMREG(fs, MIPSInst_FS(ir));
 			DPFROMREG(fd, MIPSInst_FD(ir));
@@ -2122,6 +2176,7 @@ copcsr:
 			if (!cpu_has_mips_r6)
 				return SIGILL;
 
+			MIPS_FPU_EMU_INC_STATS(msubf_d);
 			DPFROMREG(ft, MIPSInst_FT(ir));
 			DPFROMREG(fs, MIPSInst_FS(ir));
 			DPFROMREG(fd, MIPSInst_FD(ir));
@@ -2135,6 +2190,7 @@ copcsr:
 			if (!cpu_has_mips_r6)
 				return SIGILL;
 
+			MIPS_FPU_EMU_INC_STATS(rint_d);
 			DPFROMREG(fs, MIPSInst_FS(ir));
 			rv.d = ieee754dp_rint(fs);
 			goto copcsr;
@@ -2146,6 +2202,7 @@ copcsr:
 			if (!cpu_has_mips_r6)
 				return SIGILL;
 
+			MIPS_FPU_EMU_INC_STATS(class_d);
 			DPFROMREG(fs, MIPSInst_FS(ir));
 			rv.l = ieee754dp_2008class(fs);
 			rfmt = l_fmt;
@@ -2158,6 +2215,7 @@ copcsr:
 			if (!cpu_has_mips_r6)
 				return SIGILL;
 
+			MIPS_FPU_EMU_INC_STATS(min_d);
 			DPFROMREG(ft, MIPSInst_FT(ir));
 			DPFROMREG(fs, MIPSInst_FS(ir));
 			rv.d = ieee754dp_fmin(fs, ft);
@@ -2170,6 +2228,7 @@ copcsr:
 			if (!cpu_has_mips_r6)
 				return SIGILL;
 
+			MIPS_FPU_EMU_INC_STATS(mina_d);
 			DPFROMREG(ft, MIPSInst_FT(ir));
 			DPFROMREG(fs, MIPSInst_FS(ir));
 			rv.d = ieee754dp_fmina(fs, ft);
@@ -2182,6 +2241,7 @@ copcsr:
 			if (!cpu_has_mips_r6)
 				return SIGILL;
 
+			MIPS_FPU_EMU_INC_STATS(max_d);
 			DPFROMREG(ft, MIPSInst_FT(ir));
 			DPFROMREG(fs, MIPSInst_FS(ir));
 			rv.d = ieee754dp_fmax(fs, ft);
@@ -2194,6 +2254,7 @@ copcsr:
 			if (!cpu_has_mips_r6)
 				return SIGILL;
 
+			MIPS_FPU_EMU_INC_STATS(maxa_d);
 			DPFROMREG(ft, MIPSInst_FT(ir));
 			DPFROMREG(fs, MIPSInst_FS(ir));
 			rv.d = ieee754dp_fmaxa(fs, ft);
@@ -2201,15 +2262,18 @@ copcsr:
 		}
 
 		case fabs_op:
+			MIPS_FPU_EMU_INC_STATS(abs_d);
 			handler.u = ieee754dp_abs;
 			goto dcopuop;
 
 		case fneg_op:
+			MIPS_FPU_EMU_INC_STATS(neg_d);
 			handler.u = ieee754dp_neg;
 			goto dcopuop;
 
 		case fmov_op:
 			/* an easy one */
+			MIPS_FPU_EMU_INC_STATS(mov_d);
 			DPFROMREG(rv.d, MIPSInst_FS(ir));
 			goto copcsr;
 
@@ -2229,6 +2293,7 @@ dcopuop:
 		 * unary conv ops
 		 */
 		case fcvts_op:
+			MIPS_FPU_EMU_INC_STATS(cvt_s_d);
 			DPFROMREG(fs, MIPSInst_FS(ir));
 			rv.s = ieee754sp_fdp(fs);
 			rfmt = s_fmt;
@@ -2238,6 +2303,7 @@ dcopuop:
 			return SIGILL;	/* not defined */
 
 		case fcvtw_op:
+			MIPS_FPU_EMU_INC_STATS(cvt_w_d);
 			DPFROMREG(fs, MIPSInst_FS(ir));
 			rv.w = ieee754dp_tint(fs);	/* wrong */
 			rfmt = w_fmt;
@@ -2250,6 +2316,15 @@ dcopuop:
 			if (!cpu_has_mips_2_3_4_5_r)
 				return SIGILL;
 
+			if (MIPSInst_FUNC(ir) == fceil_op)
+				MIPS_FPU_EMU_INC_STATS(ceil_w_d);
+			if (MIPSInst_FUNC(ir) == ffloor_op)
+				MIPS_FPU_EMU_INC_STATS(floor_w_d);
+			if (MIPSInst_FUNC(ir) == fround_op)
+				MIPS_FPU_EMU_INC_STATS(round_w_d);
+			if (MIPSInst_FUNC(ir) == ftrunc_op)
+				MIPS_FPU_EMU_INC_STATS(trunc_w_d);
+
 			oldrm = ieee754_csr.rm;
 			DPFROMREG(fs, MIPSInst_FS(ir));
 			ieee754_csr.rm = MIPSInst_FUNC(ir);
@@ -2262,6 +2337,7 @@ dcopuop:
 			if (!cpu_has_mips_r6)
 				return SIGILL;
 
+			MIPS_FPU_EMU_INC_STATS(sel_d);
 			DPFROMREG(fd, MIPSInst_FD(ir));
 			if (fd.bits & 0x1)
 				DPFROMREG(rv.d, MIPSInst_FT(ir));
@@ -2273,6 +2349,7 @@ dcopuop:
 			if (!cpu_has_mips_3_4_5_64_r2_r6)
 				return SIGILL;
 
+			MIPS_FPU_EMU_INC_STATS(cvt_l_d);
 			DPFROMREG(fs, MIPSInst_FS(ir));
 			rv.l = ieee754dp_tlong(fs);
 			rfmt = l_fmt;
@@ -2285,6 +2362,15 @@ dcopuop:
 			if (!cpu_has_mips_3_4_5_64_r2_r6)
 				return SIGILL;
 
+			if (MIPSInst_FUNC(ir) == fceill_op)
+				MIPS_FPU_EMU_INC_STATS(ceil_l_d);
+			if (MIPSInst_FUNC(ir) == ffloorl_op)
+				MIPS_FPU_EMU_INC_STATS(floor_l_d);
+			if (MIPSInst_FUNC(ir) == froundl_op)
+				MIPS_FPU_EMU_INC_STATS(round_l_d);
+			if (MIPSInst_FUNC(ir) == ftruncl_op)
+				MIPS_FPU_EMU_INC_STATS(trunc_l_d);
+
 			oldrm = ieee754_csr.rm;
 			DPFROMREG(fs, MIPSInst_FS(ir));
 			ieee754_csr.rm = MIPSInst_FUNC(ir);
@@ -2326,12 +2412,14 @@ dcopuop:
 		switch (MIPSInst_FUNC(ir)) {
 		case fcvts_op:
 			/* convert word to single precision real */
+			MIPS_FPU_EMU_INC_STATS(cvt_s_w);
 			SPFROMREG(fs, MIPSInst_FS(ir));
 			rv.s = ieee754sp_fint(fs.bits);
 			rfmt = s_fmt;
 			goto copcsr;
 		case fcvtd_op:
 			/* convert word to double precision real */
+			MIPS_FPU_EMU_INC_STATS(cvt_d_w);
 			SPFROMREG(fs, MIPSInst_FS(ir));
 			rv.d = ieee754dp_fint(fs.bits);
 			rfmt = d_fmt;
@@ -2351,6 +2439,90 @@ dcopuop:
 			    (MIPSInst_FUNC(ir) & 0x20))
 				return SIGILL;
 
+			if (!sig) {
+				if (!(MIPSInst_FUNC(ir) & PREDICATE_BIT)) {
+					switch (cmpop) {
+					case 0:
+					MIPS_FPU_EMU_INC_STATS(cmp_af_s);
+					break;
+					case 1:
+					MIPS_FPU_EMU_INC_STATS(cmp_un_s);
+					break;
+					case 2:
+					MIPS_FPU_EMU_INC_STATS(cmp_eq_s);
+					break;
+					case 3:
+					MIPS_FPU_EMU_INC_STATS(cmp_ueq_s);
+					break;
+					case 4:
+					MIPS_FPU_EMU_INC_STATS(cmp_lt_s);
+					break;
+					case 5:
+					MIPS_FPU_EMU_INC_STATS(cmp_ult_s);
+					break;
+					case 6:
+					MIPS_FPU_EMU_INC_STATS(cmp_le_s);
+					break;
+					case 7:
+					MIPS_FPU_EMU_INC_STATS(cmp_ule_s);
+					break;
+					}
+				} else {
+					switch (cmpop) {
+					case 1:
+					MIPS_FPU_EMU_INC_STATS(cmp_or_s);
+					break;
+					case 2:
+					MIPS_FPU_EMU_INC_STATS(cmp_une_s);
+					break;
+					case 3:
+					MIPS_FPU_EMU_INC_STATS(cmp_ne_s);
+					break;
+					}
+				}
+			} else {
+				if (!(MIPSInst_FUNC(ir) & PREDICATE_BIT)) {
+					switch (cmpop) {
+					case 0:
+					MIPS_FPU_EMU_INC_STATS(cmp_saf_s);
+					break;
+					case 1:
+					MIPS_FPU_EMU_INC_STATS(cmp_sun_s);
+					break;
+					case 2:
+					MIPS_FPU_EMU_INC_STATS(cmp_seq_s);
+					break;
+					case 3:
+					MIPS_FPU_EMU_INC_STATS(cmp_sueq_s);
+					break;
+					case 4:
+					MIPS_FPU_EMU_INC_STATS(cmp_slt_s);
+					break;
+					case 5:
+					MIPS_FPU_EMU_INC_STATS(cmp_sult_s);
+					break;
+					case 6:
+					MIPS_FPU_EMU_INC_STATS(cmp_sle_s);
+					break;
+					case 7:
+					MIPS_FPU_EMU_INC_STATS(cmp_sule_s);
+					break;
+					}
+				} else {
+					switch (cmpop) {
+					case 1:
+					MIPS_FPU_EMU_INC_STATS(cmp_sor_s);
+					break;
+					case 2:
+					MIPS_FPU_EMU_INC_STATS(cmp_sune_s);
+					break;
+					case 3:
+					MIPS_FPU_EMU_INC_STATS(cmp_sne_s);
+					break;
+					}
+				}
+			}
+
 			/* fmt is w_fmt for single precision so fix it */
 			rfmt = s_fmt;
 			/* default to false */
@@ -2407,11 +2579,13 @@ dcopuop:
 		switch (MIPSInst_FUNC(ir)) {
 		case fcvts_op:
 			/* convert long to single precision real */
+			MIPS_FPU_EMU_INC_STATS(cvt_s_l);
 			rv.s = ieee754sp_flong(bits);
 			rfmt = s_fmt;
 			goto copcsr;
 		case fcvtd_op:
 			/* convert long to double precision real */
+			MIPS_FPU_EMU_INC_STATS(cvt_d_l);
 			rv.d = ieee754dp_flong(bits);
 			rfmt = d_fmt;
 			goto copcsr;
@@ -2425,6 +2599,90 @@ dcopuop:
 			    (MIPSInst_FUNC(ir) & 0x20))
 				return SIGILL;
 
+			if (!sig) {
+				if (!(MIPSInst_FUNC(ir) & PREDICATE_BIT)) {
+					switch (cmpop) {
+					case 0:
+					MIPS_FPU_EMU_INC_STATS(cmp_af_d);
+					break;
+					case 1:
+					MIPS_FPU_EMU_INC_STATS(cmp_un_d);
+					break;
+					case 2:
+					MIPS_FPU_EMU_INC_STATS(cmp_eq_d);
+					break;
+					case 3:
+					MIPS_FPU_EMU_INC_STATS(cmp_ueq_d);
+					break;
+					case 4:
+					MIPS_FPU_EMU_INC_STATS(cmp_lt_d);
+					break;
+					case 5:
+					MIPS_FPU_EMU_INC_STATS(cmp_ult_d);
+					break;
+					case 6:
+					MIPS_FPU_EMU_INC_STATS(cmp_le_d);
+					break;
+					case 7:
+					MIPS_FPU_EMU_INC_STATS(cmp_ule_d);
+					break;
+					}
+				} else {
+					switch (cmpop) {
+					case 1:
+					MIPS_FPU_EMU_INC_STATS(cmp_or_d);
+					break;
+					case 2:
+					MIPS_FPU_EMU_INC_STATS(cmp_une_d);
+					break;
+					case 3:
+					MIPS_FPU_EMU_INC_STATS(cmp_ne_d);
+					break;
+					}
+				}
+			} else {
+				if (!(MIPSInst_FUNC(ir) & PREDICATE_BIT)) {
+					switch (cmpop) {
+					case 0:
+					MIPS_FPU_EMU_INC_STATS(cmp_saf_d);
+					break;
+					case 1:
+					MIPS_FPU_EMU_INC_STATS(cmp_sun_d);
+					break;
+					case 2:
+					MIPS_FPU_EMU_INC_STATS(cmp_seq_d);
+					break;
+					case 3:
+					MIPS_FPU_EMU_INC_STATS(cmp_sueq_d);
+					break;
+					case 4:
+					MIPS_FPU_EMU_INC_STATS(cmp_slt_d);
+					break;
+					case 5:
+					MIPS_FPU_EMU_INC_STATS(cmp_sult_d);
+					break;
+					case 6:
+					MIPS_FPU_EMU_INC_STATS(cmp_sle_d);
+					break;
+					case 7:
+					MIPS_FPU_EMU_INC_STATS(cmp_sule_d);
+					break;
+					}
+				} else {
+					switch (cmpop) {
+					case 1:
+					MIPS_FPU_EMU_INC_STATS(cmp_sor_d);
+					break;
+					case 2:
+					MIPS_FPU_EMU_INC_STATS(cmp_sune_d);
+					break;
+					case 3:
+					MIPS_FPU_EMU_INC_STATS(cmp_sne_d);
+					break;
+					}
+				}
+			}
+
 			/* fmt is l_fmt for double precision so fix it */
 			rfmt = d_fmt;
 			/* default to false */
diff --git a/arch/mips/math-emu/me-debugfs.c b/arch/mips/math-emu/me-debugfs.c
index f080493ba743..8c0ec154aecc 100644
--- a/arch/mips/math-emu/me-debugfs.c
+++ b/arch/mips/math-emu/me-debugfs.c
@@ -28,6 +28,26 @@ static int fpuemu_stat_get(void *data, u64 *val)
 }
 DEFINE_SIMPLE_ATTRIBUTE(fops_fpuemu_stat, fpuemu_stat_get, NULL, "%llu\n");
 
+/*
+ * Used to obtain names for a debugfs instruction counter, given field name
+ * in fpuemustats structure. For example, for input "cmp_sueq_d", the output
+ * would be "cmp.sueq.d". This is needed since dots are not allowed to be
+ * used in structure field names, and are, on the other hand, desired to be
+ * used in debugfs item names to be clearly associated to corresponding
+ * MIPS FPU instructions.
+ */
+static void adjust_instruction_counter_name(char *out_name, char *in_name)
+{
+	int i = 0;
+
+	strcpy(out_name, in_name);
+	while (in_name[i] != '\0') {
+		if (out_name[i] == '_')
+			out_name[i] = '.';
+		i++;
+	}
+}
+
 static int fpuemustats_clear_show(struct seq_file *s, void *unused)
 {
 	__this_cpu_write((fpuemustats).emulated, 0);
@@ -44,6 +64,121 @@ static int fpuemustats_clear_show(struct seq_file *s, void *unused)
 	__this_cpu_write((fpuemustats).ieee754_invalidop, 0);
 	__this_cpu_write((fpuemustats).ds_emul, 0);
 
+	__this_cpu_write((fpuemustats).abs_s, 0);
+	__this_cpu_write((fpuemustats).abs_d, 0);
+	__this_cpu_write((fpuemustats).add_s, 0);
+	__this_cpu_write((fpuemustats).add_d, 0);
+	__this_cpu_write((fpuemustats).bc1eqz, 0);
+	__this_cpu_write((fpuemustats).bc1nez, 0);
+	__this_cpu_write((fpuemustats).ceil_w_s, 0);
+	__this_cpu_write((fpuemustats).ceil_w_d, 0);
+	__this_cpu_write((fpuemustats).ceil_l_s, 0);
+	__this_cpu_write((fpuemustats).ceil_l_d, 0);
+	__this_cpu_write((fpuemustats).class_s, 0);
+	__this_cpu_write((fpuemustats).class_d, 0);
+	__this_cpu_write((fpuemustats).cmp_af_s, 0);
+	__this_cpu_write((fpuemustats).cmp_af_d, 0);
+	__this_cpu_write((fpuemustats).cmp_eq_s, 0);
+	__this_cpu_write((fpuemustats).cmp_eq_d, 0);
+	__this_cpu_write((fpuemustats).cmp_le_s, 0);
+	__this_cpu_write((fpuemustats).cmp_le_d, 0);
+	__this_cpu_write((fpuemustats).cmp_lt_s, 0);
+	__this_cpu_write((fpuemustats).cmp_lt_d, 0);
+	__this_cpu_write((fpuemustats).cmp_ne_s, 0);
+	__this_cpu_write((fpuemustats).cmp_ne_d, 0);
+	__this_cpu_write((fpuemustats).cmp_or_s, 0);
+	__this_cpu_write((fpuemustats).cmp_or_d, 0);
+	__this_cpu_write((fpuemustats).cmp_ueq_s, 0);
+	__this_cpu_write((fpuemustats).cmp_ueq_d, 0);
+	__this_cpu_write((fpuemustats).cmp_ule_s, 0);
+	__this_cpu_write((fpuemustats).cmp_ule_d, 0);
+	__this_cpu_write((fpuemustats).cmp_ult_s, 0);
+	__this_cpu_write((fpuemustats).cmp_ult_d, 0);
+	__this_cpu_write((fpuemustats).cmp_un_s, 0);
+	__this_cpu_write((fpuemustats).cmp_un_d, 0);
+	__this_cpu_write((fpuemustats).cmp_une_s, 0);
+	__this_cpu_write((fpuemustats).cmp_une_d, 0);
+	__this_cpu_write((fpuemustats).cmp_saf_s, 0);
+	__this_cpu_write((fpuemustats).cmp_saf_d, 0);
+	__this_cpu_write((fpuemustats).cmp_seq_s, 0);
+	__this_cpu_write((fpuemustats).cmp_seq_d, 0);
+	__this_cpu_write((fpuemustats).cmp_sle_s, 0);
+	__this_cpu_write((fpuemustats).cmp_sle_d, 0);
+	__this_cpu_write((fpuemustats).cmp_slt_s, 0);
+	__this_cpu_write((fpuemustats).cmp_slt_d, 0);
+	__this_cpu_write((fpuemustats).cmp_sne_s, 0);
+	__this_cpu_write((fpuemustats).cmp_sne_d, 0);
+	__this_cpu_write((fpuemustats).cmp_sor_s, 0);
+	__this_cpu_write((fpuemustats).cmp_sor_d, 0);
+	__this_cpu_write((fpuemustats).cmp_sueq_s, 0);
+	__this_cpu_write((fpuemustats).cmp_sueq_d, 0);
+	__this_cpu_write((fpuemustats).cmp_sule_s, 0);
+	__this_cpu_write((fpuemustats).cmp_sule_d, 0);
+	__this_cpu_write((fpuemustats).cmp_sult_s, 0);
+	__this_cpu_write((fpuemustats).cmp_sult_d, 0);
+	__this_cpu_write((fpuemustats).cmp_sun_s, 0);
+	__this_cpu_write((fpuemustats).cmp_sun_d, 0);
+	__this_cpu_write((fpuemustats).cmp_sune_s, 0);
+	__this_cpu_write((fpuemustats).cmp_sune_d, 0);
+	__this_cpu_write((fpuemustats).cvt_d_l, 0);
+	__this_cpu_write((fpuemustats).cvt_d_s, 0);
+	__this_cpu_write((fpuemustats).cvt_d_w, 0);
+	__this_cpu_write((fpuemustats).cvt_l_s, 0);
+	__this_cpu_write((fpuemustats).cvt_l_d, 0);
+	__this_cpu_write((fpuemustats).cvt_s_d, 0);
+	__this_cpu_write((fpuemustats).cvt_s_l, 0);
+	__this_cpu_write((fpuemustats).cvt_s_w, 0);
+	__this_cpu_write((fpuemustats).cvt_w_s, 0);
+	__this_cpu_write((fpuemustats).cvt_w_d, 0);
+	__this_cpu_write((fpuemustats).div_s, 0);
+	__this_cpu_write((fpuemustats).div_d, 0);
+	__this_cpu_write((fpuemustats).floor_w_s, 0);
+	__this_cpu_write((fpuemustats).floor_w_d, 0);
+	__this_cpu_write((fpuemustats).floor_l_s, 0);
+	__this_cpu_write((fpuemustats).floor_l_d, 0);
+	__this_cpu_write((fpuemustats).maddf_s, 0);
+	__this_cpu_write((fpuemustats).maddf_d, 0);
+	__this_cpu_write((fpuemustats).max_s, 0);
+	__this_cpu_write((fpuemustats).max_d, 0);
+	__this_cpu_write((fpuemustats).maxa_s, 0);
+	__this_cpu_write((fpuemustats).maxa_d, 0);
+	__this_cpu_write((fpuemustats).min_s, 0);
+	__this_cpu_write((fpuemustats).min_d, 0);
+	__this_cpu_write((fpuemustats).mina_s, 0);
+	__this_cpu_write((fpuemustats).mina_d, 0);
+	__this_cpu_write((fpuemustats).mov_s, 0);
+	__this_cpu_write((fpuemustats).mov_d, 0);
+	__this_cpu_write((fpuemustats).msubf_s, 0);
+	__this_cpu_write((fpuemustats).msubf_d, 0);
+	__this_cpu_write((fpuemustats).mul_s, 0);
+	__this_cpu_write((fpuemustats).mul_d, 0);
+	__this_cpu_write((fpuemustats).neg_s, 0);
+	__this_cpu_write((fpuemustats).neg_d, 0);
+	__this_cpu_write((fpuemustats).recip_s, 0);
+	__this_cpu_write((fpuemustats).recip_d, 0);
+	__this_cpu_write((fpuemustats).rint_s, 0);
+	__this_cpu_write((fpuemustats).rint_d, 0);
+	__this_cpu_write((fpuemustats).round_w_s, 0);
+	__this_cpu_write((fpuemustats).round_w_d, 0);
+	__this_cpu_write((fpuemustats).round_l_s, 0);
+	__this_cpu_write((fpuemustats).round_l_d, 0);
+	__this_cpu_write((fpuemustats).rsqrt_s, 0);
+	__this_cpu_write((fpuemustats).rsqrt_d, 0);
+	__this_cpu_write((fpuemustats).sel_s, 0);
+	__this_cpu_write((fpuemustats).sel_d, 0);
+	__this_cpu_write((fpuemustats).seleqz_s, 0);
+	__this_cpu_write((fpuemustats).seleqz_d, 0);
+	__this_cpu_write((fpuemustats).selnez_s, 0);
+	__this_cpu_write((fpuemustats).selnez_d, 0);
+	__this_cpu_write((fpuemustats).sqrt_s, 0);
+	__this_cpu_write((fpuemustats).sqrt_d, 0);
+	__this_cpu_write((fpuemustats).sub_s, 0);
+	__this_cpu_write((fpuemustats).sub_d, 0);
+	__this_cpu_write((fpuemustats).trunc_w_s, 0);
+	__this_cpu_write((fpuemustats).trunc_w_d, 0);
+	__this_cpu_write((fpuemustats).trunc_l_s, 0);
+	__this_cpu_write((fpuemustats).trunc_l_d, 0);
+
 	return 0;
 }
 
@@ -61,13 +196,18 @@ static const struct file_operations fpuemustats_clear_fops = {
 
 static int __init debugfs_fpuemu(void)
 {
-	struct dentry *d, *dir, *reset_file;
+	struct dentry *fpuemu_debugfs_base_dir;
+	struct dentry *fpuemu_debugfs_inst_dir;
+	struct dentry *d, *reset_file;
 
 	if (!mips_debugfs_dir)
 		return -ENODEV;
-	dir = debugfs_create_dir("fpuemustats", mips_debugfs_dir);
-	if (!dir)
+
+	fpuemu_debugfs_base_dir = debugfs_create_dir("fpuemustats",
+						     mips_debugfs_dir);
+	if (!fpuemu_debugfs_base_dir)
 		return -ENOMEM;
+
 	reset_file = debugfs_create_file("fpuemustats_clear", 0444,
 					 mips_debugfs_dir, NULL,
 					 &fpuemustats_clear_fops);
@@ -79,7 +219,7 @@ static int __init debugfs_fpuemu(void)
 
 #define FPU_STAT_CREATE(m)						\
 do {									\
-	d = debugfs_create_file(#m , S_IRUGO, dir,			\
+	d = debugfs_create_file(#m, 0444, fpuemu_debugfs_base_dir,	\
 				(void *)FPU_EMU_STAT_OFFSET(m),		\
 				&fops_fpuemu_stat);			\
 	if (!d)								\
@@ -100,6 +240,139 @@ do {									\
 	FPU_STAT_CREATE(ieee754_invalidop);
 	FPU_STAT_CREATE(ds_emul);
 
+	fpuemu_debugfs_inst_dir = debugfs_create_dir("instructions",
+						     fpuemu_debugfs_base_dir);
+	if (!fpuemu_debugfs_inst_dir)
+		return -ENOMEM;
+
+#define FPU_STAT_CREATE_EX(m)						\
+do {									\
+	char name[32];							\
+									\
+	adjust_instruction_counter_name(name, #m);			\
+									\
+	d = debugfs_create_file(name, 0444, fpuemu_debugfs_inst_dir,	\
+				(void *)FPU_EMU_STAT_OFFSET(m),		\
+				&fops_fpuemu_stat);			\
+	if (!d)								\
+		return -ENOMEM;						\
+} while (0)
+
+	FPU_STAT_CREATE_EX(abs_s);
+	FPU_STAT_CREATE_EX(abs_d);
+	FPU_STAT_CREATE_EX(add_s);
+	FPU_STAT_CREATE_EX(add_d);
+	FPU_STAT_CREATE_EX(bc1eqz);
+	FPU_STAT_CREATE_EX(bc1nez);
+	FPU_STAT_CREATE_EX(ceil_w_s);
+	FPU_STAT_CREATE_EX(ceil_w_d);
+	FPU_STAT_CREATE_EX(ceil_l_s);
+	FPU_STAT_CREATE_EX(ceil_l_d);
+	FPU_STAT_CREATE_EX(class_s);
+	FPU_STAT_CREATE_EX(class_d);
+	FPU_STAT_CREATE_EX(cmp_af_s);
+	FPU_STAT_CREATE_EX(cmp_af_d);
+	FPU_STAT_CREATE_EX(cmp_eq_s);
+	FPU_STAT_CREATE_EX(cmp_eq_d);
+	FPU_STAT_CREATE_EX(cmp_le_s);
+	FPU_STAT_CREATE_EX(cmp_le_d);
+	FPU_STAT_CREATE_EX(cmp_lt_s);
+	FPU_STAT_CREATE_EX(cmp_lt_d);
+	FPU_STAT_CREATE_EX(cmp_ne_s);
+	FPU_STAT_CREATE_EX(cmp_ne_d);
+	FPU_STAT_CREATE_EX(cmp_or_s);
+	FPU_STAT_CREATE_EX(cmp_or_d);
+	FPU_STAT_CREATE_EX(cmp_ueq_s);
+	FPU_STAT_CREATE_EX(cmp_ueq_d);
+	FPU_STAT_CREATE_EX(cmp_ule_s);
+	FPU_STAT_CREATE_EX(cmp_ule_d);
+	FPU_STAT_CREATE_EX(cmp_ult_s);
+	FPU_STAT_CREATE_EX(cmp_ult_d);
+	FPU_STAT_CREATE_EX(cmp_un_s);
+	FPU_STAT_CREATE_EX(cmp_un_d);
+	FPU_STAT_CREATE_EX(cmp_une_s);
+	FPU_STAT_CREATE_EX(cmp_une_d);
+	FPU_STAT_CREATE_EX(cmp_saf_s);
+	FPU_STAT_CREATE_EX(cmp_saf_d);
+	FPU_STAT_CREATE_EX(cmp_seq_s);
+	FPU_STAT_CREATE_EX(cmp_seq_d);
+	FPU_STAT_CREATE_EX(cmp_sle_s);
+	FPU_STAT_CREATE_EX(cmp_sle_d);
+	FPU_STAT_CREATE_EX(cmp_slt_s);
+	FPU_STAT_CREATE_EX(cmp_slt_d);
+	FPU_STAT_CREATE_EX(cmp_sne_s);
+	FPU_STAT_CREATE_EX(cmp_sne_d);
+	FPU_STAT_CREATE_EX(cmp_sor_s);
+	FPU_STAT_CREATE_EX(cmp_sor_d);
+	FPU_STAT_CREATE_EX(cmp_sueq_s);
+	FPU_STAT_CREATE_EX(cmp_sueq_d);
+	FPU_STAT_CREATE_EX(cmp_sule_s);
+	FPU_STAT_CREATE_EX(cmp_sule_d);
+	FPU_STAT_CREATE_EX(cmp_sult_s);
+	FPU_STAT_CREATE_EX(cmp_sult_d);
+	FPU_STAT_CREATE_EX(cmp_sun_s);
+	FPU_STAT_CREATE_EX(cmp_sun_d);
+	FPU_STAT_CREATE_EX(cmp_sune_s);
+	FPU_STAT_CREATE_EX(cmp_sune_d);
+	FPU_STAT_CREATE_EX(cvt_d_l);
+	FPU_STAT_CREATE_EX(cvt_d_s);
+	FPU_STAT_CREATE_EX(cvt_d_w);
+	FPU_STAT_CREATE_EX(cvt_l_s);
+	FPU_STAT_CREATE_EX(cvt_l_d);
+	FPU_STAT_CREATE_EX(cvt_s_d);
+	FPU_STAT_CREATE_EX(cvt_s_l);
+	FPU_STAT_CREATE_EX(cvt_s_w);
+	FPU_STAT_CREATE_EX(cvt_w_s);
+	FPU_STAT_CREATE_EX(cvt_w_d);
+	FPU_STAT_CREATE_EX(div_s);
+	FPU_STAT_CREATE_EX(div_d);
+	FPU_STAT_CREATE_EX(floor_w_s);
+	FPU_STAT_CREATE_EX(floor_w_d);
+	FPU_STAT_CREATE_EX(floor_l_s);
+	FPU_STAT_CREATE_EX(floor_l_d);
+	FPU_STAT_CREATE_EX(maddf_s);
+	FPU_STAT_CREATE_EX(maddf_d);
+	FPU_STAT_CREATE_EX(max_s);
+	FPU_STAT_CREATE_EX(max_d);
+	FPU_STAT_CREATE_EX(maxa_s);
+	FPU_STAT_CREATE_EX(maxa_d);
+	FPU_STAT_CREATE_EX(min_s);
+	FPU_STAT_CREATE_EX(min_d);
+	FPU_STAT_CREATE_EX(mina_s);
+	FPU_STAT_CREATE_EX(mina_d);
+	FPU_STAT_CREATE_EX(mov_s);
+	FPU_STAT_CREATE_EX(mov_d);
+	FPU_STAT_CREATE_EX(msubf_s);
+	FPU_STAT_CREATE_EX(msubf_d);
+	FPU_STAT_CREATE_EX(mul_s);
+	FPU_STAT_CREATE_EX(mul_d);
+	FPU_STAT_CREATE_EX(neg_s);
+	FPU_STAT_CREATE_EX(neg_d);
+	FPU_STAT_CREATE_EX(recip_s);
+	FPU_STAT_CREATE_EX(recip_d);
+	FPU_STAT_CREATE_EX(rint_s);
+	FPU_STAT_CREATE_EX(rint_d);
+	FPU_STAT_CREATE_EX(round_w_s);
+	FPU_STAT_CREATE_EX(round_w_d);
+	FPU_STAT_CREATE_EX(round_l_s);
+	FPU_STAT_CREATE_EX(round_l_d);
+	FPU_STAT_CREATE_EX(rsqrt_s);
+	FPU_STAT_CREATE_EX(rsqrt_d);
+	FPU_STAT_CREATE_EX(sel_s);
+	FPU_STAT_CREATE_EX(sel_d);
+	FPU_STAT_CREATE_EX(seleqz_s);
+	FPU_STAT_CREATE_EX(seleqz_d);
+	FPU_STAT_CREATE_EX(selnez_s);
+	FPU_STAT_CREATE_EX(selnez_d);
+	FPU_STAT_CREATE_EX(sqrt_s);
+	FPU_STAT_CREATE_EX(sqrt_d);
+	FPU_STAT_CREATE_EX(sub_s);
+	FPU_STAT_CREATE_EX(sub_d);
+	FPU_STAT_CREATE_EX(trunc_w_s);
+	FPU_STAT_CREATE_EX(trunc_w_d);
+	FPU_STAT_CREATE_EX(trunc_l_s);
+	FPU_STAT_CREATE_EX(trunc_l_d);
+
 	return 0;
 }
 arch_initcall(debugfs_fpuemu);