MIPS: Support for hybrid FPRs

Hybrid FPRs is a scheme where scalar FP registers are 64b wide, but
accesses to odd indexed single registers use bits 63:32 of the
preceeding even indexed 64b register. In this mode all FP code
except that built for the plain FP64 ABI can execute correctly. Most
notably a combination of FP64A & FP32 code can execute correctly,
allowing for existing FP32 binaries to be linked with new FP64A binaries
that can make use of 64 bit FP & MSA.

Hybrid FPRs are implemented by setting both the FR & FRE bits, trapping
& emulating single precision FP instructions (via Reserved Instruction
exceptions) whilst allowing others to execute natively. It therefore has
a penalty in terms of execution speed, and should only be used when no
fully native mode can be. As more binaries are recompiled to use either
the FPXX or FP64(A) ABIs, the need for hybrid FPRs should diminish.
However in the short to mid term it allows for a gradual transition
towards that world, rather than a complete ABI break which is not
feasible for some users & not desirable for many.

A task will be executed using the hybrid FPR scheme when its
TIF_HYBRID_FPREGS flag is set & TIF_32BIT_FPREGS is clear. A further
patch will set the flags as necessary, this patch simply adds the
infrastructure necessary for the hybrid FPR mode to work.

Signed-off-by: Paul Burton <paul.burton@imgtec.com>
Cc: linux-mips@linux-mips.org
Cc: Alexander Viro <viro@zeniv.linux.org.uk>
Cc: linux-fsdevel@vger.kernel.org
Cc: linux-kernel@vger.kernel.org
Patchwork: https://patchwork.linux-mips.org/patch/7683/
Signed-off-by: Ralf Baechle <ralf@linux-mips.org>

5 files changed, 100 insertions, 10 deletions

diff --git a/arch/mips/include/asm/elf.h b/arch/mips/include/asm/elf.h
index 1d38fe0edd2d..9343529db7bc 100644
--- a/arch/mips/include/asm/elf.h
+++ b/arch/mips/include/asm/elf.h
@@ -269,6 +269,8 @@ do {									\
         else                                                            \
                 set_thread_flag(TIF_32BIT_FPREGS);                      \
                                                                         \
+        clear_thread_flag(TIF_HYBRID_FPREGS);                           \
+                                                                        \
         if (personality(current->personality) != PER_LINUX)             \
                 set_personality(PER_LINUX);                             \
                                                                         \
@@ -325,6 +327,7 @@ do {									\
                                                                         \
         clear_thread_flag(TIF_32BIT_REGS);                              \
         clear_thread_flag(TIF_32BIT_FPREGS);                            \
+        clear_thread_flag(TIF_HYBRID_FPREGS);                           \
         clear_thread_flag(TIF_32BIT_ADDR);                              \
                                                                         \
         if ((ex).e_ident[EI_CLASS] == ELFCLASS32)                       \
diff --git a/arch/mips/include/asm/fpu.h b/arch/mips/include/asm/fpu.h
index dd562414cd5e..994d21939676 100644
--- a/arch/mips/include/asm/fpu.h
+++ b/arch/mips/include/asm/fpu.h
@@ -36,14 +36,16 @@ extern void _restore_fp(struct task_struct *);
 
 /*
  * This enum specifies a mode in which we want the FPU to operate, for cores
- * which implement the Status.FR bit. Note that FPU_32BIT & FPU_64BIT
- * purposefully have the values 0 & 1 respectively, so that an integer value
- * of Status.FR can be trivially casted to the corresponding enum fpu_mode.
+ * which implement the Status.FR bit. Note that the bottom bit of the value
+ * purposefully matches the desired value of the Status.FR bit.
  */
 enum fpu_mode {
         FPU_32BIT = 0,          /* FR = 0 */
-        FPU_64BIT,              /* FR = 1 */
+        FPU_64BIT,              /* FR = 1, FRE = 0 */
         FPU_AS_IS,
+        FPU_HYBRID,             /* FR = 1, FRE = 1 */
+
+#define FPU_FR_MASK             0x1
 };
 
 static inline int __enable_fpu(enum fpu_mode mode)
@@ -57,6 +59,14 @@ static inline int __enable_fpu(enum fpu_mode mode)
                 enable_fpu_hazard();
                 return 0;
 
+        case FPU_HYBRID:
+                if (!cpu_has_fre)
+                        return SIGFPE;
+
+                /* set FRE */
+                write_c0_config5(read_c0_config5() | MIPS_CONF5_FRE);
+                goto fr_common;
+
         case FPU_64BIT:
 #if !(defined(CONFIG_CPU_MIPS32_R2) || defined(CONFIG_64BIT))
                 /* we only have a 32-bit FPU */
@@ -64,8 +74,11 @@ static inline int __enable_fpu(enum fpu_mode mode)
 #endif
                 /* fall through */
         case FPU_32BIT:
+                /* clear FRE */
+                write_c0_config5(read_c0_config5() & ~MIPS_CONF5_FRE);
+fr_common:
                 /* set CU1 & change FR appropriately */
-                fr = (int)mode;
+                fr = (int)mode & FPU_FR_MASK;
                 change_c0_status(ST0_CU1 | ST0_FR, ST0_CU1 | (fr ? ST0_FR : 0));
                 enable_fpu_hazard();
 
@@ -102,13 +115,17 @@ static inline int __own_fpu(void)
         enum fpu_mode mode;
         int ret;
 
-        mode = !test_thread_flag(TIF_32BIT_FPREGS);
+        if (test_thread_flag(TIF_HYBRID_FPREGS))
+                mode = FPU_HYBRID;
+        else
+                mode = !test_thread_flag(TIF_32BIT_FPREGS);
+
         ret = __enable_fpu(mode);
         if (ret)
                 return ret;
 
         KSTK_STATUS(current) |= ST0_CU1;
-        if (mode == FPU_64BIT)
+        if (mode == FPU_64BIT || mode == FPU_HYBRID)
                 KSTK_STATUS(current) |= ST0_FR;
         else /* mode == FPU_32BIT */
                 KSTK_STATUS(current) &= ~ST0_FR;
@@ -166,8 +183,24 @@ static inline int init_fpu(void)
 
         if (cpu_has_fpu) {
                 ret = __own_fpu();
-                if (!ret)
+                if (!ret) {
+                        unsigned int config5 = read_c0_config5();
+
+                        /*
+                         * Ensure FRE is clear whilst running _init_fpu, since
+                         * single precision FP instructions are used. If FRE
+                         * was set then we'll just end up initialising all 32
+                         * 64b registers.
+                         */
+                        write_c0_config5(config5 & ~MIPS_CONF5_FRE);
+                        enable_fpu_hazard();
+
                         _init_fpu();
+
+                        /* Restore FRE */
+                        write_c0_config5(config5);
+                        enable_fpu_hazard();
+                }
         } else
                 fpu_emulator_init_fpu();
 
diff --git a/arch/mips/include/asm/thread_info.h b/arch/mips/include/asm/thread_info.h
index 7de865805deb..99eea59604e9 100644
--- a/arch/mips/include/asm/thread_info.h
+++ b/arch/mips/include/asm/thread_info.h
@@ -116,6 +116,7 @@ static inline struct thread_info *current_thread_info(void)
 #define TIF_LOAD_WATCH          25      /* If set, load watch registers */
 #define TIF_SYSCALL_TRACEPOINT  26      /* syscall tracepoint instrumentation */
 #define TIF_32BIT_FPREGS        27      /* 32-bit floating point registers */
+#define TIF_HYBRID_FPREGS       28      /* 64b FP registers, odd singles in bits 63:32 of even doubles */
 #define TIF_USEDMSA             29      /* MSA has been used this quantum */
 #define TIF_MSA_CTX_LIVE        30      /* MSA context must be preserved */
 #define TIF_SYSCALL_TRACE       31      /* syscall trace active */
@@ -135,6 +136,7 @@ static inline struct thread_info *current_thread_info(void)
 #define _TIF_FPUBOUND           (1<<TIF_FPUBOUND)
 #define _TIF_LOAD_WATCH         (1<<TIF_LOAD_WATCH)
 #define _TIF_32BIT_FPREGS       (1<<TIF_32BIT_FPREGS)
+#define _TIF_HYBRID_FPREGS      (1<<TIF_HYBRID_FPREGS)
 #define _TIF_USEDMSA            (1<<TIF_USEDMSA)
 #define _TIF_MSA_CTX_LIVE       (1<<TIF_MSA_CTX_LIVE)
 #define _TIF_SYSCALL_TRACEPOINT (1<<TIF_SYSCALL_TRACEPOINT)
diff --git a/arch/mips/kernel/traps.c b/arch/mips/kernel/traps.c
index 22b19c275044..165c275163b8 100644
--- a/arch/mips/kernel/traps.c
+++ b/arch/mips/kernel/traps.c
@@ -724,6 +724,50 @@ int process_fpemu_return(int sig, void __user *fault_addr)
         }
 }
 
+static int simulate_fp(struct pt_regs *regs, unsigned int opcode,
+                       unsigned long old_epc, unsigned long old_ra)
+{
+        union mips_instruction inst = { .word = opcode };
+        void __user *fault_addr = NULL;
+        int sig;
+
+        /* If it's obviously not an FP instruction, skip it */
+        switch (inst.i_format.opcode) {
+        case cop1_op:
+        case cop1x_op:
+        case lwc1_op:
+        case ldc1_op:
+        case swc1_op:
+        case sdc1_op:
+                break;
+
+        default:
+                return -1;
+        }
+
+        /*
+         * do_ri skipped over the instruction via compute_return_epc, undo
+         * that for the FPU emulator.
+         */
+        regs->cp0_epc = old_epc;
+        regs->regs[31] = old_ra;
+
+        /* Save the FP context to struct thread_struct */
+        lose_fpu(1);
+
+        /* Run the emulator */
+        sig = fpu_emulator_cop1Handler(regs, &current->thread.fpu, 1,
+                                       &fault_addr);
+
+        /* If something went wrong, signal */
+        process_fpemu_return(sig, fault_addr);
+
+        /* Restore the hardware register state */
+        own_fpu(1);
+
+        return 0;
+}
+
 /*
  * XXX Delayed fp exceptions when doing a lazy ctx switch XXX
  */
@@ -1016,6 +1060,9 @@ asmlinkage void do_ri(struct pt_regs *regs)
 
                 if (status < 0)
                         status = simulate_sync(regs, opcode);
+
+                if (status < 0)
+                        status = simulate_fp(regs, opcode, old_epc, old31);
         }
 
         if (status < 0)
diff --git a/arch/mips/math-emu/cp1emu.c b/arch/mips/math-emu/cp1emu.c
index cac529a405b8..9dfcd7fc1bc3 100644
--- a/arch/mips/math-emu/cp1emu.c
+++ b/arch/mips/math-emu/cp1emu.c
@@ -643,9 +643,14 @@ static inline int cop1_64bit(struct pt_regs *xcp)
         return !test_thread_flag(TIF_32BIT_FPREGS);
 }
 
+static inline bool hybrid_fprs(void)
+{
+        return test_thread_flag(TIF_HYBRID_FPREGS);
+}
+
 #define SIFROMREG(si, x)                                                \
 do {                                                                    \
-        if (cop1_64bit(xcp))                                            \
+        if (cop1_64bit(xcp) && !hybrid_fprs())                          \
                 (si) = (int)get_fpr32(&ctx->fpr[x], 0);                 \
         else                                                            \
                 (si) = (int)get_fpr32(&ctx->fpr[(x) & ~1], (x) & 1);    \
@@ -653,7 +658,7 @@ do {									\
 
 #define SITOREG(si, x)                                                  \
 do {                                                                    \
-        if (cop1_64bit(xcp)) {                                          \
+        if (cop1_64bit(xcp) && !hybrid_fprs()) {                        \
                 unsigned i;                                             \
                 set_fpr32(&ctx->fpr[x], 0, si);                         \
                 for (i = 1; i < ARRAY_SIZE(ctx->fpr[x].val32); i++)     \