path: root/dev-lang/ghc/files/ghc-9.0.2-riscv64-llvm.patch

GHC supplied all of the information necessary for a registerised build on riscv using LLVM codegen in 9.2.
Andreas Schwab backported this to 8.10.7 in OpenSUSE:  https://build.opensuse.org/package/show/openSUSE:Factory:RISCV/ghc
This is a forward-port of the backport to 9.0.

diff --git a/compiler/GHC/CmmToAsm.hs b/compiler/GHC/CmmToAsm.hs
index 23175a1cd2..3f77573495 100644
--- a/compiler/GHC/CmmToAsm.hs
+++ b/compiler/GHC/CmmToAsm.hs
@@ -182,6 +182,7 @@ nativeCodeGen dflags this_mod modLoc h us cmms
       ArchAlpha     -> panic "nativeCodeGen: No NCG for Alpha"
       ArchMipseb    -> panic "nativeCodeGen: No NCG for mipseb"
       ArchMipsel    -> panic "nativeCodeGen: No NCG for mipsel"
+      ArchRISCV64   -> panic "nativeCodeGen: No NCG for RISCV64"
       ArchUnknown   -> panic "nativeCodeGen: No NCG for unknown arch"
       ArchJavaScript-> panic "nativeCodeGen: No NCG for JavaScript"

diff --git a/compiler/GHC/CmmToAsm/Reg/Graph/TrivColorable.hs b/compiler/GHC/CmmToAsm/Reg/Graph/TrivColorable.hs
index 105e2ebdf9..07eca93ac3 100644
--- a/compiler/GHC/CmmToAsm/Reg/Graph/TrivColorable.hs
+++ b/compiler/GHC/CmmToAsm/Reg/Graph/TrivColorable.hs
@@ -120,6 +120,7 @@ trivColorable platform virtualRegSqueeze realRegSqueeze RcInteger conflicts excl
                             ArchMipseb    -> panic "trivColorable ArchMipseb"
                             ArchMipsel    -> panic "trivColorable ArchMipsel"
                             ArchS390X     -> panic "trivColorable ArchS390X"
+                            ArchRISCV64   -> panic "trivColorable ArchRISCV64"
                             ArchJavaScript-> panic "trivColorable ArchJavaScript"
                             ArchUnknown   -> panic "trivColorable ArchUnknown")
         , count2        <- accSqueeze 0 cALLOCATABLE_REGS_INTEGER
@@ -151,6 +152,7 @@ trivColorable platform virtualRegSqueeze realRegSqueeze RcFloat conflicts exclus
                             ArchMipseb    -> panic "trivColorable ArchMipseb"
                             ArchMipsel    -> panic "trivColorable ArchMipsel"
                             ArchS390X     -> panic "trivColorable ArchS390X"
+                            ArchRISCV64   -> panic "trivColorable ArchRISCV64"
                             ArchJavaScript-> panic "trivColorable ArchJavaScript"
                             ArchUnknown   -> panic "trivColorable ArchUnknown")
         , count2        <- accSqueeze 0 cALLOCATABLE_REGS_FLOAT
@@ -184,6 +186,7 @@ trivColorable platform virtualRegSqueeze realRegSqueeze RcDouble conflicts exclu
                             ArchMipseb    -> panic "trivColorable ArchMipseb"
                             ArchMipsel    -> panic "trivColorable ArchMipsel"
                             ArchS390X     -> panic "trivColorable ArchS390X"
+                            ArchRISCV64   -> panic "trivColorable ArchRISCV64"
                             ArchJavaScript-> panic "trivColorable ArchJavaScript"
                             ArchUnknown   -> panic "trivColorable ArchUnknown")
         , count2        <- accSqueeze 0 cALLOCATABLE_REGS_DOUBLE
diff --git a/compiler/GHC/CmmToAsm/Reg/Linear.hs b/compiler/GHC/CmmToAsm/Reg/Linear.hs
index 1ab353cc34..d8228c69cd 100644
--- a/compiler/GHC/CmmToAsm/Reg/Linear.hs
+++ b/compiler/GHC/CmmToAsm/Reg/Linear.hs
@@ -225,6 +225,7 @@ linearRegAlloc config entry_ids block_live sccs
       ArchAlpha      -> panic "linearRegAlloc ArchAlpha"
       ArchMipseb     -> panic "linearRegAlloc ArchMipseb"
       ArchMipsel     -> panic "linearRegAlloc ArchMipsel"
+      ArchRISCV64    -> panic "linearRegAlloc ArchRISCV64"
       ArchJavaScript -> panic "linearRegAlloc ArchJavaScript"
       ArchUnknown    -> panic "linearRegAlloc ArchUnknown"
  where
diff --git a/compiler/GHC/CmmToAsm/Reg/Linear/FreeRegs.hs b/compiler/GHC/CmmToAsm/Reg/Linear/FreeRegs.hs
index b89c056a82..1768422f5c 100644
--- a/compiler/GHC/CmmToAsm/Reg/Linear/FreeRegs.hs
+++ b/compiler/GHC/CmmToAsm/Reg/Linear/FreeRegs.hs
@@ -83,5 +83,6 @@ maxSpillSlots config = case platformArch (ncgPlatform config) of
    ArchAlpha     -> panic "maxSpillSlots ArchAlpha"
    ArchMipseb    -> panic "maxSpillSlots ArchMipseb"
    ArchMipsel    -> panic "maxSpillSlots ArchMipsel"
+   ArchRISCV64   -> panic "maxSpillSlots ArchRISCV64"
    ArchJavaScript-> panic "maxSpillSlots ArchJavaScript"
    ArchUnknown   -> panic "maxSpillSlots ArchUnknown"
diff --git a/compiler/GHC/CmmToAsm/Reg/Target.hs b/compiler/GHC/CmmToAsm/Reg/Target.hs
index ba249801ba..284df45983 100644
--- a/compiler/GHC/CmmToAsm/Reg/Target.hs
+++ b/compiler/GHC/CmmToAsm/Reg/Target.hs
@@ -51,6 +51,7 @@ targetVirtualRegSqueeze platform
       ArchAlpha     -> panic "targetVirtualRegSqueeze ArchAlpha"
       ArchMipseb    -> panic "targetVirtualRegSqueeze ArchMipseb"
       ArchMipsel    -> panic "targetVirtualRegSqueeze ArchMipsel"
+      ArchRISCV64   -> panic "targetVirtualRegSqueeze ArchRISCV64"
       ArchJavaScript-> panic "targetVirtualRegSqueeze ArchJavaScript"
       ArchUnknown   -> panic "targetVirtualRegSqueeze ArchUnknown"

@@ -70,6 +71,7 @@ targetRealRegSqueeze platform
       ArchAlpha     -> panic "targetRealRegSqueeze ArchAlpha"
       ArchMipseb    -> panic "targetRealRegSqueeze ArchMipseb"
       ArchMipsel    -> panic "targetRealRegSqueeze ArchMipsel"
+      ArchRISCV64   -> panic "targetRealRegSqueeze ArchRISCV64"
       ArchJavaScript-> panic "targetRealRegSqueeze ArchJavaScript"
       ArchUnknown   -> panic "targetRealRegSqueeze ArchUnknown"

@@ -88,6 +90,7 @@ targetClassOfRealReg platform
       ArchAlpha     -> panic "targetClassOfRealReg ArchAlpha"
       ArchMipseb    -> panic "targetClassOfRealReg ArchMipseb"
       ArchMipsel    -> panic "targetClassOfRealReg ArchMipsel"
+      ArchRISCV64   -> panic "targetClassOfRealReg ArchRISCV64"
       ArchJavaScript-> panic "targetClassOfRealReg ArchJavaScript"
       ArchUnknown   -> panic "targetClassOfRealReg ArchUnknown"

@@ -106,6 +109,7 @@ targetMkVirtualReg platform
       ArchAlpha     -> panic "targetMkVirtualReg ArchAlpha"
       ArchMipseb    -> panic "targetMkVirtualReg ArchMipseb"
       ArchMipsel    -> panic "targetMkVirtualReg ArchMipsel"
+      ArchRISCV64   -> panic "targetMkVirtualReg ArchRISCV64"
       ArchJavaScript-> panic "targetMkVirtualReg ArchJavaScript"
       ArchUnknown   -> panic "targetMkVirtualReg ArchUnknown"

@@ -124,6 +128,7 @@ targetRegDotColor platform
       ArchAlpha     -> panic "targetRegDotColor ArchAlpha"
       ArchMipseb    -> panic "targetRegDotColor ArchMipseb"
       ArchMipsel    -> panic "targetRegDotColor ArchMipsel"
+      ArchRISCV64   -> panic "targetRegDotColor ArchRISCV64"
       ArchJavaScript-> panic "targetRegDotColor ArchJavaScript"
       ArchUnknown   -> panic "targetRegDotColor ArchUnknown"

diff --git a/compiler/GHC/CmmToLlvm/Mangler.hs b/compiler/GHC/CmmToLlvm/Mangler.hs
index 0436dbcf07..5c71655ab3 100644
--- a/compiler/GHC/CmmToLlvm/Mangler.hs
+++ b/compiler/GHC/CmmToLlvm/Mangler.hs
@@ -42,7 +42,7 @@ llvmFixupAsm dflags f1 f2 = {-# SCC "llvm_mangler" #-}

 -- | These are the rewrites that the mangler will perform
 rewrites :: [Rewrite]
-rewrites = [rewriteSymType, rewriteAVX]
+rewrites = [rewriteSymType, rewriteAVX, rewriteCall]

 type Rewrite = DynFlags -> B.ByteString -> Maybe B.ByteString

@@ -106,6 +106,27 @@ rewriteAVX dflags s
     isVmovdqa = B.isPrefixOf (B.pack "vmovdqa")
     isVmovap = B.isPrefixOf (B.pack "vmovap")

+-- | This rewrites (tail) calls to avoid creating PLT entries for
+-- functions on riscv64. The replacement will load the address from the
+-- GOT, which is resolved to point to the real address of the function.
+rewriteCall :: Rewrite
+rewriteCall dflags l
+  | not isRISCV64 = Nothing
+  | isCall l      = Just $ replaceCall "call" "jalr" "ra" l
+  | isTail l      = Just $ replaceCall "tail" "jr" "t1" l
+  | otherwise     = Nothing
+  where
+    isRISCV64 = platformArch (targetPlatform dflags) == ArchRISCV64
+    isCall = B.isPrefixOf (B.pack "call\t")
+    isTail = B.isPrefixOf (B.pack "tail\t")
+
+    replaceCall call jump reg l =
+        appendInsn (jump ++ "\t" ++ reg) $ removePlt $
+        replaceOnce (B.pack call) (B.pack ("la\t" ++ reg ++ ",")) l
+      where
+        removePlt = replaceOnce (B.pack "@plt") (B.pack "")
+        appendInsn i = (`B.append` B.pack ("\n\t" ++ i))
+
 -- | @replaceOnce match replace bs@ replaces the first occurrence of the
 -- substring @match@ in @bs@ with @replace@.
 replaceOnce :: B.ByteString -> B.ByteString -> B.ByteString -> B.ByteString
diff --git a/compiler/GHC/Driver/Pipeline.hs b/compiler/GHC/Driver/Pipeline.hs
index 8edf49dfdd..d66404e916 100644
--- a/compiler/GHC/Driver/Pipeline.hs
+++ b/compiler/GHC/Driver/Pipeline.hs
@@ -936,6 +936,7 @@ llvmOptions dflags =
     ++ [("", "-mcpu=" ++ mcpu)   | not (null mcpu)
                                  , not (any (isInfixOf "-mcpu") (getOpts dflags opt_lc)) ]
     ++ [("", "-mattr=" ++ attrs) | not (null attrs) ]
+    ++ [("", "-target-abi=" ++ abi) | not (null abi) ]

   where target = platformMisc_llvmTarget $ platformMisc dflags
         Just (LlvmTarget _ mcpu mattr) = lookup target (llvmTargets $ llvmConfig dflags)
@@ -965,6 +966,11 @@ llvmOptions dflags =
               ++ ["+bmi"     | isBmiEnabled dflags      ]
               ++ ["+bmi2"    | isBmi2Enabled dflags     ]

+        abi :: String
+        abi = case platformArch (targetPlatform dflags) of
+             ArchRISCV64 -> "lp64d"
+             _           -> ""
+
 -- -----------------------------------------------------------------------------
 -- | Each phase in the pipeline returns the next phase to execute, and the
 -- name of the file in which the output was placed.
diff --git a/compiler/GHC/Platform/RISCV64.hs b/compiler/GHC/Platform/RISCV64.hs
new file mode 100644
index 0000000000..289e2e27a8
--- /dev/null
+++ b/compiler/GHC/Platform/RISCV64.hs
@@ -0,0 +1,9 @@
+{-# LANGUAGE CPP #-}
+
+module GHC.Platform.RISCV64 where
+
+import GHC.Prelude
+
+#define MACHREGS_NO_REGS 0
+#define MACHREGS_riscv64 1
+#include "../../../includes/CodeGen.Platform.hs"
diff --git a/compiler/GHC/Platform/Regs.hs b/compiler/GHC/Platform/Regs.hs
index 1bf21370c7..e0bdf5d6e5 100644
--- a/compiler/GHC/Platform/Regs.hs
+++ b/compiler/GHC/Platform/Regs.hs
@@ -15,6 +15,7 @@ import qualified GHC.Platform.S390X      as S390X
 import qualified GHC.Platform.SPARC      as SPARC
 import qualified GHC.Platform.X86        as X86
 import qualified GHC.Platform.X86_64     as X86_64
+import qualified GHC.Platform.RISCV64    as RISCV64
 import qualified GHC.Platform.NoRegs     as NoRegs

 -- | Returns 'True' if this global register is stored in a caller-saves
@@ -31,6 +32,7 @@ callerSaves platform
    ArchSPARC   -> SPARC.callerSaves
    ArchARM {}  -> ARM.callerSaves
    ArchAArch64 -> AArch64.callerSaves
+   ArchRISCV64 -> RISCV64.callerSaves
    arch
     | arch `elem` [ArchPPC, ArchPPC_64 ELF_V1, ArchPPC_64 ELF_V2] ->
         PPC.callerSaves
@@ -53,6 +55,7 @@ activeStgRegs platform
    ArchSPARC   -> SPARC.activeStgRegs
    ArchARM {}  -> ARM.activeStgRegs
    ArchAArch64 -> AArch64.activeStgRegs
+   ArchRISCV64 -> RISCV64.activeStgRegs
    arch
     | arch `elem` [ArchPPC, ArchPPC_64 ELF_V1, ArchPPC_64 ELF_V2] ->
         PPC.activeStgRegs
@@ -70,6 +73,7 @@ haveRegBase platform
    ArchSPARC   -> SPARC.haveRegBase
    ArchARM {}  -> ARM.haveRegBase
    ArchAArch64 -> AArch64.haveRegBase
+   ArchRISCV64 -> RISCV64.haveRegBase
    arch
     | arch `elem` [ArchPPC, ArchPPC_64 ELF_V1, ArchPPC_64 ELF_V2] ->
         PPC.haveRegBase
@@ -87,6 +91,7 @@ globalRegMaybe platform
    ArchSPARC   -> SPARC.globalRegMaybe
    ArchARM {}  -> ARM.globalRegMaybe
    ArchAArch64 -> AArch64.globalRegMaybe
+   ArchRISCV64 -> RISCV64.globalRegMaybe
    arch
     | arch `elem` [ArchPPC, ArchPPC_64 ELF_V1, ArchPPC_64 ELF_V2] ->
         PPC.globalRegMaybe
@@ -104,6 +109,7 @@ freeReg platform
    ArchSPARC   -> SPARC.freeReg
    ArchARM {}  -> ARM.freeReg
    ArchAArch64 -> AArch64.freeReg
+   ArchRISCV64 -> RISCV64.freeReg
    arch
     | arch `elem` [ArchPPC, ArchPPC_64 ELF_V1, ArchPPC_64 ELF_V2] ->
         PPC.freeReg
diff --git a/compiler/ghc.cabal.in b/compiler/ghc.cabal.in
index de53f7f548..483a601602 100644
--- a/compiler/ghc.cabal.in
+++ b/compiler/ghc.cabal.in
@@ -261,6 +261,7 @@ Library
         GHC.Platform.AArch64
         GHC.Platform.NoRegs
         GHC.Platform.PPC
+        GHC.Platform.RISCV64
         GHC.Platform.S390X
         GHC.Platform.SPARC
         GHC.Platform.X86
diff --git a/configure.ac b/configure.ac
index 20911ae43f..b017d50bd6 100644
--- a/configure.ac
+++ b/configure.ac
@@ -283,7 +283,7 @@ dnl --------------------------------------------------------------

 AC_MSG_CHECKING(whether target supports a registerised ABI)
 case "$TargetArch" in
-    i386|x86_64|powerpc|powerpc64|powerpc64le|s390x|arm|aarch64)
+    i386|x86_64|powerpc|powerpc64|powerpc64le|s390x|arm|aarch64|riscv64)
         UnregisterisedDefault=NO
         AC_MSG_RESULT([yes])
         ;;
@@ -318,7 +318,7 @@ AC_MSG_CHECKING(whether target supports tables next to code)
 case "$Unregisterised" in
     NO)
         case "$TargetArch" in
-            ia64|powerpc64|powerpc64le|s390x)
+            ia64|powerpc64|powerpc64le|s390x|riscv64)
                 TablesNextToCodeDefault=NO
                 AC_MSG_RESULT([no])
                 ;;
@@ -347,7 +347,7 @@ AC_SUBST(TablesNextToCode)
 dnl ** Does target have runtime linker support?
 dnl --------------------------------------------------------------
 case "$target" in
-    powerpc64-*|powerpc64le-*|powerpc-ibm-aix*|s390x-ibm-linux)
+    powerpc64-*|powerpc64le-*|powerpc-ibm-aix*|s390x-ibm-linux|riscv64-*)
         TargetHasRTSLinker=NO
         ;;
     *)
diff --git a/includes/CodeGen.Platform.hs b/includes/CodeGen.Platform.hs
index b00acfa38a..8c942662e6 100644
--- a/includes/CodeGen.Platform.hs
+++ b/includes/CodeGen.Platform.hs
@@ -380,6 +380,74 @@ import GHC.Platform.Reg
 # define f14 30
 # define f15 31

+#elif defined(MACHREGS_riscv64)
+
+# define zero 0
+# define ra   1
+# define sp   2
+# define gp   3
+# define tp   4
+# define t0   5
+# define t1   6
+# define t2   7
+# define s0   8
+# define s1   9
+# define a0  10
+# define a1  11
+# define a2  12
+# define a3  13
+# define a4  14
+# define a5  15
+# define a6  16
+# define a7  17
+# define s2  18
+# define s3  19
+# define s4  20
+# define s5  21
+# define s6  22
+# define s7  23
+# define s8  24
+# define s9  25
+# define s10 26
+# define s11 27
+# define t3  28
+# define t4  29
+# define t5  30
+# define t6  31
+
+# define ft0  32
+# define ft1  33
+# define ft2  34
+# define ft3  35
+# define ft4  36
+# define ft5  37
+# define ft6  38
+# define ft7  39
+# define fs0  40
+# define fs1  41
+# define fa0  42
+# define fa1  43
+# define fa2  44
+# define fa3  45
+# define fa4  46
+# define fa5  47
+# define fa6  48
+# define fa7  49
+# define fs2  50
+# define fs3  51
+# define fs4  52
+# define fs5  53
+# define fs6  54
+# define fs7  55
+# define fs8  56
+# define fs9  57
+# define fs10 58
+# define fs11 59
+# define ft8  60
+# define ft9  61
+# define ft10 62
+# define ft11 63
+
 #endif

 callerSaves :: GlobalReg -> Bool
@@ -667,7 +735,7 @@ globalRegMaybe :: GlobalReg -> Maybe RealReg
 #if defined(MACHREGS_i386) || defined(MACHREGS_x86_64) \
     || defined(MACHREGS_sparc) || defined(MACHREGS_powerpc) \
     || defined(MACHREGS_arm) || defined(MACHREGS_aarch64) \
-    || defined(MACHREGS_s390x)
+    || defined(MACHREGS_s390x) || defined(MACHREGS_riscv64)
 # if defined(REG_Base)
 globalRegMaybe BaseReg                  = Just (RealRegSingle REG_Base)
 # endif
diff --git a/includes/stg/MachRegs.h b/includes/stg/MachRegs.h
index be7b7d60b4..1a9012e7ef 100644
--- a/includes/stg/MachRegs.h
+++ b/includes/stg/MachRegs.h
@@ -663,6 +663,68 @@ the stack. See Note [Overlapping global registers] for implications.
 #define CALLER_SAVES_D5
 #define CALLER_SAVES_D6

+/* -----------------------------------------------------------------------------
+   The riscv64 register mapping
+
+   Register    | Role(s)                                 | Call effect
+   ------------+-----------------------------------------+-------------
+   zero        | Hard-wired zero                         | -
+   ra          | Return address                          | caller-saved
+   sp          | Stack pointer                           | callee-saved
+   gp          | Global pointer                          | callee-saved
+   tp          | Thread pointer                          | callee-saved
+   t0,t1,t2    | -                                       | caller-saved
+   s0          | Frame pointer                           | callee-saved
+   s1          | -                                       | callee-saved
+   a0,a1       | Arguments / return values               | caller-saved
+   a2..a7      | Arguments                               | caller-saved
+   s2..s11     | -                                       | callee-saved
+   t3..t6      | -                                       | caller-saved
+   ft0..ft7    | -                                       | caller-saved
+   fs0,fs1     | -                                       | callee-saved
+   fa0,fa1     | Arguments / return values               | caller-saved
+   fa2..fa7    | Arguments                               | caller-saved
+   fs2..fs11   | -                                       | callee-saved
+   ft8..ft11   | -                                       | caller-saved
+
+   Each general purpose register as well as each floating-point
+   register is 64 bits wide.
+
+   -------------------------------------------------------------------------- */
+
+#elif defined(MACHREGS_riscv64)
+
+#define REG(x) __asm__(#x)
+
+#define REG_Base        s1
+#define REG_Sp          s2
+#define REG_Hp          s3
+#define REG_R1          s4
+#define REG_R2          s5
+#define REG_R3          s6
+#define REG_R4          s7
+#define REG_R5          s8
+#define REG_R6          s9
+#define REG_R7          s10
+#define REG_SpLim       s11
+
+#define REG_F1          fs0
+#define REG_F2          fs1
+#define REG_F3          fs2
+#define REG_F4          fs3
+#define REG_F5          fs4
+#define REG_F6          fs5
+
+#define REG_D1          fs6
+#define REG_D2          fs7
+#define REG_D3          fs8
+#define REG_D4          fs9
+#define REG_D5          fs10
+#define REG_D6          fs11
+
+#define MAX_REAL_FLOAT_REG   6
+#define MAX_REAL_DOUBLE_REG  6
+
 #else

 #error Cannot find platform to give register info for
diff --git a/includes/stg/MachRegsForHost.h b/includes/stg/MachRegsForHost.h
index 3597b2be90..e902d528f6 100644
--- a/includes/stg/MachRegsForHost.h
+++ b/includes/stg/MachRegsForHost.h
@@ -71,6 +71,10 @@
 #define MACHREGS_s390x    1
 #endif

+#if defined(riscv64_HOST_ARCH)
+#define MACHREGS_riscv64  1
+#endif
+
 #endif

 #include "MachRegs.h"
diff --git a/includes/stg/SMP.h b/includes/stg/SMP.h
index 389dd95c88..f12fd3f71b 100644
--- a/includes/stg/SMP.h
+++ b/includes/stg/SMP.h
@@ -377,6 +377,8 @@ write_barrier(void) {
     __asm__ __volatile__ ("" : : : "memory");
 #elif defined(arm_HOST_ARCH) || defined(aarch64_HOST_ARCH)
     __asm__ __volatile__ ("dmb  st" : : : "memory");
+#elif defined(riscv64_HOST_ARCH)
+    __asm__ __volatile__ ("fence w,w" : : : "memory");
 #else
 #error memory barriers unimplemented on this architecture
 #endif
@@ -401,6 +403,8 @@ store_load_barrier(void) {
     __asm__ __volatile__ ("dmb" : : : "memory");
 #elif defined(aarch64_HOST_ARCH)
     __asm__ __volatile__ ("dmb sy" : : : "memory");
+#elif defined(riscv64_HOST_ARCH)
+    __asm__ __volatile__ ("fence w,r" : : : "memory");
 #else
 #error memory barriers unimplemented on this architecture
 #endif
@@ -426,6 +430,8 @@ load_load_barrier(void) {
     __asm__ __volatile__ ("dmb" : : : "memory");
 #elif defined(aarch64_HOST_ARCH)
     __asm__ __volatile__ ("dmb sy" : : : "memory");
+#elif defined(riscv64_HOST_ARCH)
+    __asm__ __volatile__ ("fence w,r" : : : "memory");
 #else
 #error memory barriers unimplemented on this architecture
 #endif
diff --git a/libraries/ghc-boot/GHC/Platform.hs b/libraries/ghc-boot/GHC/Platform.hs
index 186f8c8af6..44d2c68c0a 100644
--- a/libraries/ghc-boot/GHC/Platform.hs
+++ b/libraries/ghc-boot/GHC/Platform.hs
@@ -130,6 +130,7 @@ data Arch
         | ArchAlpha
         | ArchMipseb
         | ArchMipsel
+        | ArchRISCV64
         | ArchJavaScript
         deriving (Read, Show, Eq)

@@ -171,6 +172,7 @@ stringEncodeArch = \case
   ArchAlpha -> "alpha"
   ArchMipseb -> "mipseb"
   ArchMipsel -> "mipsel"
+  ArchRISCV64 -> "riscv64"
   ArchJavaScript -> "js"

 isARM :: Arch -> Bool
@@ -338,6 +340,7 @@ platformCConvNeedsExtension :: Platform -> Bool
 platformCConvNeedsExtension platform = case platformArch platform of
   ArchPPC_64 _ -> True
   ArchS390X    -> True
+  ArchRISCV64  -> True
   _            -> False


diff --git a/libraries/ghci/GHCi/InfoTable.hsc b/libraries/ghci/GHCi/InfoTable.hsc
index ad4eb4d6a8..1d1f8cd074 100644
--- a/libraries/ghci/GHCi/InfoTable.hsc
+++ b/libraries/ghci/GHCi/InfoTable.hsc
@@ -73,6 +73,7 @@ data Arch = ArchSPARC
           | ArchPPC64
           | ArchPPC64LE
           | ArchS390X
+          | ArchRISCV64
  deriving Show

 mkJumpToAddr :: MonadFail m => EntryFunPtr-> m ItblCodes
@@ -109,6 +110,8 @@ mArch =
        Just ArchPPC64LE
 #elif defined(s390x_HOST_ARCH)
        Just ArchS390X
+#elif defined(riscv64_HOST_ARCH)
+        Just ArchRISCV64
 #else
        Nothing
 #endif
@@ -288,6 +291,15 @@ mkJumpToAddr' platform a = case platform of
                   0xC0, 0x19, byte3 w64, byte2 w64, byte1 w64, byte0 w64,
                   0x07, 0xF1 ]

+    ArchRISCV64 ->
+        let w64 = fromIntegral (funPtrToInt a) :: Word64
+        in Right [ 0x00000297          -- auipc t0,0
+                 , 0x01053283          -- ld    t0,16(t0)
+                 , 0x00028067          -- jr    t0
+                 , 0x00000013          -- nop
+                 , fromIntegral w64
+                 , fromIntegral (w64 `shiftR` 32) ]
+
 byte0 :: (Integral w) => w -> Word8
 byte0 w = fromIntegral w

diff --git a/llvm-targets b/llvm-targets
index c328b2e001..0f809b2418 100644
--- a/llvm-targets
+++ b/llvm-targets
@@ -38,6 +38,8 @@
 ,("powerpc64le-unknown-linux-musl", ("e-m:e-i64:64-n32:64", "ppc64le", "+secure-plt"))
 ,("powerpc64le-unknown-linux", ("e-m:e-i64:64-n32:64", "ppc64le", ""))
 ,("s390x-ibm-linux", ("E-m:e-i1:8:16-i8:8:16-i64:64-f128:64-a:8:16-n32:64", "z10", ""))
+,("riscv64-unknown-linux-gnu", ("e-m:e-p:64:64-i64:64-i128:128-n64-S128", "", "+m +a +f +d +c +relax"))
+,("riscv64-unknown-linux", ("e-m:e-p:64:64-i64:64-i128:128-n64-S128", "", "+m +a +f +d +c +relax"))
 ,("i386-apple-darwin", ("e-m:o-p:32:32-p270:32:32-p271:32:32-p272:64:64-f64:32:64-f80:128-n8:16:32-S128", "penryn", ""))
 ,("x86_64-apple-darwin", ("e-m:o-p270:32:32-p271:32:32-p272:64:64-i64:64-f80:128-n8:16:32:64-S128", "penryn", ""))
 ,("arm64-apple-darwin", ("e-m:o-i64:64-i128:128-n32:64-S128", "apple-a12", "+v8.3a +fp-armv8 +neon +crc +crypto +fullfp16 +ras +lse +rdm +rcpc +zcm +zcz +sha2 +aes"))
diff --git a/m4/fptools.m4 b/m4/fptools.m4
index 8c6c9b7062..7dbe53d3b4 100644
--- a/m4/fptools.m4
+++ b/m4/fptools.m4
@@ -231,7 +231,10 @@ AC_DEFUN([FPTOOLS_SET_HASKELL_PLATFORM_VARS],
         mipsel)
             test -z "[$]2" || eval "[$]2=ArchMipsel"
             ;;
-        hppa|hppa1_1|ia64|m68k|nios2|riscv32|riscv64|rs6000|s390|sh4|vax)
+        riscv64)
+            test -z "[$]2" || eval "[$]2=ArchRISCV64"
+            ;;
+        hppa|hppa1_1|ia64|m68k|nios2|riscv32|rs6000|s390|sh4|vax)
             test -z "[$]2" || eval "[$]2=ArchUnknown"
             ;;
         *)
diff --git a/mk/config.mk.in b/mk/config.mk.in
index 48b8e23145..e260730991 100644
--- a/mk/config.mk.in
+++ b/mk/config.mk.in
@@ -179,7 +179,7 @@ ifeq "$(TargetArch_CPP)" "arm"
 # We don't support load/store barriers pre-ARMv7. See #10433.
 ArchSupportsSMP=$(if $(filter $(ARM_ISA),ARMv5 ARMv6),NO,YES)
 else
-ArchSupportsSMP=$(strip $(patsubst $(TargetArch_CPP), YES, $(findstring $(TargetArch_CPP), i386 x86_64 sparc powerpc powerpc64 powerpc64le s390x aarch64)))
+ArchSupportsSMP=$(strip $(patsubst $(TargetArch_CPP), YES, $(findstring $(TargetArch_CPP), i386 x86_64 sparc powerpc powerpc64 powerpc64le s390x aarch64 riscv64)))
 endif

 # The THREADED_RTS requires `BaseReg` to be in a register and the
diff --git a/rts/StgCRunAsm.S b/rts/StgCRunAsm.S
index 60f1bf9a13..16b125a7e8 100644
--- a/rts/StgCRunAsm.S
+++ b/rts/StgCRunAsm.S
@@ -232,6 +232,150 @@ StgReturn:
 	.size StgReturn, .-StgReturn

 	.section	.note.GNU-stack,"",@progbits
+
+#elif defined(riscv64_HOST_ARCH)
+# define STACK_FRAME_SIZE (RESERVED_C_STACK_BYTES+208)
+       .text
+       .align 1
+       .globl StgRun
+       .type StgRun, @function
+StgRun:
+       .cfi_startproc
+       addi    sp,sp,-208
+       .cfi_def_cfa_offset 208
+       /* save callee-saved registers */
+       sd ra,200(sp)
+       sd s0,192(sp)
+       sd s1,184(sp)
+       sd s2,176(sp)
+       sd s3,168(sp)
+       sd s4,160(sp)
+       sd s5,152(sp)
+       sd s6,144(sp)
+       sd s7,136(sp)
+       sd s8,128(sp)
+       sd s9,120(sp)
+       sd s10,112(sp)
+       sd s11,104(sp)
+       fsd fs0,88(sp)
+       fsd fs1,80(sp)
+       fsd fs2,72(sp)
+       fsd fs3,64(sp)
+       fsd fs4,56(sp)
+       fsd fs5,48(sp)
+       fsd fs6,40(sp)
+       fsd fs7,32(sp)
+       fsd fs8,24(sp)
+       fsd fs9,16(sp)
+       fsd fs10,8(sp)
+       fsd fs11,0(sp)
+       /* allocate stack frame */
+       li t0,RESERVED_C_STACK_BYTES
+       sub sp,sp,t0
+       .cfi_def_cfa_offset STACK_FRAME_SIZE
+       .cfi_offset 1, -8
+       .cfi_offset 8, -16
+       .cfi_offset 9, -24
+       .cfi_offset 18, -32
+       .cfi_offset 19, -40
+       .cfi_offset 20, -48
+       .cfi_offset 21, -56
+       .cfi_offset 22, -64
+       .cfi_offset 23, -72
+       .cfi_offset 24, -80
+       .cfi_offset 25, -88
+       .cfi_offset 26, -96
+       .cfi_offset 27, -104
+       .cfi_offset 40, -120
+       .cfi_offset 41, -128
+       .cfi_offset 50, -136
+       .cfi_offset 51, -144
+       .cfi_offset 52, -152
+       .cfi_offset 53, -160
+       .cfi_offset 54, -168
+       .cfi_offset 55, -176
+       .cfi_offset 56, -184
+       .cfi_offset 57, -192
+       .cfi_offset 58, -200
+       .cfi_offset 59, -208
+       /* set STGs BaseReg from RISCV a1 */
+        mv s1,a1
+       /* jump to STG function */
+        jr a0
+       .cfi_endproc
+       .size StgRun, .-StgRun
+
+       .text
+       .align 1
+       .globl StgReturn
+       .type StgReturn, @function
+StgReturn:
+       .cfi_startproc
+       /* set return value from STGs R1 (RISCV s4) */
+       mv a0,s4
+       /* deallocate stack frame */
+       li t0,RESERVED_C_STACK_BYTES
+       add sp,sp,t0
+       .cfi_def_cfa_offset 208
+       /* restore callee-saved registers */
+       ld      ra,200(sp)
+       .cfi_restore 1
+       ld      s0,192(sp)
+       .cfi_restore 8
+       ld      s1,184(sp)
+       .cfi_restore 9
+       ld      s2,176(sp)
+       .cfi_restore 18
+       ld      s3,168(sp)
+       .cfi_restore 19
+       ld      s4,160(sp)
+       .cfi_restore 20
+       ld      s5,152(sp)
+       .cfi_restore 21
+       ld      s6,144(sp)
+       .cfi_restore 22
+       ld      s7,136(sp)
+       .cfi_restore 23
+       ld      s8,128(sp)
+       .cfi_restore 24
+       ld      s9,120(sp)
+       .cfi_restore 25
+       ld      s10,112(sp)
+       .cfi_restore 26
+       ld      s11,104(sp)
+       .cfi_restore 27
+       fld     fs0,88(sp)
+       .cfi_restore 40
+       fld     fs1,80(sp)
+       .cfi_restore 41
+       fld     fs2,72(sp)
+       .cfi_restore 50
+       fld     fs3,64(sp)
+       .cfi_restore 51
+       fld     fs4,56(sp)
+       .cfi_restore 52
+       fld     fs5,48(sp)
+       .cfi_restore 53
+       fld     fs6,40(sp)
+       .cfi_restore 54
+       fld     fs7,32(sp)
+       .cfi_restore 55
+       fld     fs8,24(sp)
+       .cfi_restore 56
+       fld     fs9,16(sp)
+       .cfi_restore 57
+       fld     fs10,8(sp)
+       .cfi_restore 58
+       fld     fs11,0(sp)
+       .cfi_restore 59
+       addi    sp,sp,208
+       .cfi_def_cfa_offset 0
+       /* jump back to caller of StgRun() */
+       ret
+       .cfi_endproc
+       .size StgReturn, .-StgReturn
+
+       .section        .note.GNU-stack,"",@progbits
 #endif

 #endif /* !USE_MINIINTERPRETER */
diff --git a/rts/linker/Elf.c b/rts/linker/Elf.c
index 1da69d03fa..0aa1c8dd3f 100644
--- a/rts/linker/Elf.c
+++ b/rts/linker/Elf.c
@@ -412,6 +412,12 @@ ocVerifyImage_ELF ( ObjectCode* oc )
 #endif
 #if defined(EM_AARCH64)
       case EM_AARCH64: IF_DEBUG(linker,debugBelch( "aarch64" )); break;
+#endif
+#if defined(EM_RISCV)
+      case EM_RISCV:   IF_DEBUG(linker,debugBelch( "riscv" ));
+          errorBelch("%s: RTS linker not implemented on riscv",
+                     oc->fileName);
+          return 0;
 #endif
        default:       IF_DEBUG(linker,debugBelch( "unknown" ));
                      errorBelch("%s: unknown architecture (e_machine == %d)"
diff --git a/utils/genapply/Main.hs b/utils/genapply/Main.hs
index a6d944efa4..a5697b427f 100644
--- a/utils/genapply/Main.hs
+++ b/utils/genapply/Main.hs
@@ -17,6 +17,8 @@ module Main(main) where
 -- We improperly include *HOST* macros for our target...
 #include "../../includes/ghcconfig.h"

+#undef UnregisterisedCompiler
+
 -- ...so that this header defines the right stuff.  It is the RTS's host, but
 -- our target, as we are generating code that uses that RTS.
 #include "../../includes/stg/MachRegsForHost.h"
diff --git a/utils/llvm-targets/gen-data-layout.sh b/utils/llvm-targets/gen-data-layout.sh
index 32032cb97e..092e0892e5 100755
--- a/utils/llvm-targets/gen-data-layout.sh
+++ b/utils/llvm-targets/gen-data-layout.sh
@@ -76,6 +76,9 @@ TARGETS=(
     "powerpc64le-unknown-linux"
     # Linux s390x
     "s390x-ibm-linux"
+    # Linux riscv64
+    "riscv64-unknown-linux-gnu"
+    "riscv64-unknown-linux"

     #########################
     # Darwin