queue-5.1/nds32-avoid-iex-status-being-incorrectly-modified.patch

   1 From 7f214592e69ae8ea788b0a1413191e9de05190d3 Mon Sep 17 00:00:00 2001
   2 From: Vincent Chen <vincentc@andestech.com>
   3 Date: Mon, 20 May 2019 09:21:12 +0800
   4 Subject: nds32: Avoid IEX status being incorrectly modified
   5
   6 [ Upstream commit ed32949e0acb73e7bc054bb02e0453d4d561ceda ]
   7
   8 In order for kernel to capture each denormalized output, the UDF
   9 trapping enable bit is always raised in $fpcsr. Because underflow case will
  10 issue not an underflow exception but also an inexact exception, it causes
  11 that the IEX, IEX cumulative exception, flag in $fpcsr to be raised in each
  12 denormalized output handling. To make the emulation transparent to the
  13 user, the emulator needs to clear the IEX flag in $fpcsr if the result is a
  14 denormalized number. However, if the IEX flag has been raised before this
  15 floating point emulation, this cleanup may be incorrect. To avoid the IEX
  16 flags in $fpcsr be raised in each denormalized output handling, the IEX
  17 trap shall be always enabled.
  18
  19 Signed-off-by: Vincent Chen <vincentc@andestech.com>
  20 Acked-by: Greentime Hu <greentime@andestech.com>
  21 Signed-off-by: Greentime Hu <greentime@andestech.com>
  22 Signed-off-by: Sasha Levin <sashal@kernel.org>
  23 ---
  24  arch/nds32/include/asm/bitfield.h            |  2 +-
  25  arch/nds32/include/asm/fpu.h                 |  2 +-
  26  arch/nds32/include/asm/syscalls.h            |  2 +-
  27  arch/nds32/include/uapi/asm/fp_udfiex_crtl.h | 16 ++++++++++++
  28  arch/nds32/include/uapi/asm/sigcontext.h     | 24 ++++++++++++------
  29  arch/nds32/include/uapi/asm/udftrap.h        | 13 ----------
  30  arch/nds32/include/uapi/asm/unistd.h         |  4 +--
  31  arch/nds32/kernel/fpu.c                      | 15 +++++------
  32  arch/nds32/kernel/sys_nds32.c                | 26 +++++++++++---------
  33  9 files changed, 58 insertions(+), 46 deletions(-)
  34  create mode 100644 arch/nds32/include/uapi/asm/fp_udfiex_crtl.h
  35  delete mode 100644 arch/nds32/include/uapi/asm/udftrap.h
  36
  37 diff --git a/arch/nds32/include/asm/bitfield.h b/arch/nds32/include/asm/bitfield.h
  38 index 7414fcbbab4e..03bbb6d27828 100644
  39 --- a/arch/nds32/include/asm/bitfield.h
  40 +++ b/arch/nds32/include/asm/bitfield.h
  41 @@ -937,7 +937,7 @@
  42  #define FPCSR_mskDNIT           ( 0x1  << FPCSR_offDNIT )
  43  #define FPCSR_mskRIT           ( 0x1  << FPCSR_offRIT )
  44  #define FPCSR_mskALL           (FPCSR_mskIVO | FPCSR_mskDBZ | FPCSR_mskOVF | FPCSR_mskUDF | FPCSR_mskIEX)
  45 -#define FPCSR_mskALLE_NO_UDFE  (FPCSR_mskIVOE | FPCSR_mskDBZE | FPCSR_mskOVFE | FPCSR_mskIEXE)
  46 +#define FPCSR_mskALLE_NO_UDF_IEXE (FPCSR_mskIVOE | FPCSR_mskDBZE | FPCSR_mskOVFE)
  47  #define FPCSR_mskALLE          (FPCSR_mskIVOE | FPCSR_mskDBZE | FPCSR_mskOVFE | FPCSR_mskUDFE | FPCSR_mskIEXE)
  48  #define FPCSR_mskALLT           (FPCSR_mskIVOT | FPCSR_mskDBZT | FPCSR_mskOVFT | FPCSR_mskUDFT | FPCSR_mskIEXT |FPCSR_mskDNIT | FPCSR_mskRIT)
  49
  50 diff --git a/arch/nds32/include/asm/fpu.h b/arch/nds32/include/asm/fpu.h
  51 index 019f1bcfc5ee..8294ed4aaa2c 100644
  52 --- a/arch/nds32/include/asm/fpu.h
  53 +++ b/arch/nds32/include/asm/fpu.h
  54 @@ -36,7 +36,7 @@ extern int do_fpuemu(struct pt_regs *regs, struct fpu_struct *fpu);
  55   * enabled by default and kerenl will re-execute it by fpu emulator
  56   * when getting underflow exception.
  57   */
  58 -#define FPCSR_INIT  FPCSR_mskUDFE
  59 +#define FPCSR_INIT  (FPCSR_mskUDFE | FPCSR_mskIEXE)
  60  #else
  61  #define FPCSR_INIT  0x0UL
  62  #endif
  63 diff --git a/arch/nds32/include/asm/syscalls.h b/arch/nds32/include/asm/syscalls.h
  64 index da32101b455d..b9c9becce5d6 100644
  65 --- a/arch/nds32/include/asm/syscalls.h
  66 +++ b/arch/nds32/include/asm/syscalls.h
  67 @@ -7,7 +7,7 @@
  68  asmlinkage long sys_cacheflush(unsigned long addr, unsigned long len, unsigned int op);
  69  asmlinkage long sys_fadvise64_64_wrapper(int fd, int advice, loff_t offset, loff_t len);
  70  asmlinkage long sys_rt_sigreturn_wrapper(void);
  71 -asmlinkage long sys_udftrap(int option);
  72 +asmlinkage long sys_fp_udfiex_crtl(int cmd, int act);
  73
  74  #include <asm-generic/syscalls.h>
  75
  76 diff --git a/arch/nds32/include/uapi/asm/fp_udfiex_crtl.h b/arch/nds32/include/uapi/asm/fp_udfiex_crtl.h
  77 new file mode 100644
  78 index 000000000000..d54a5d6c6538
  79 --- /dev/null
  80 +++ b/arch/nds32/include/uapi/asm/fp_udfiex_crtl.h
  81 @@ -0,0 +1,16 @@
  82 +/* SPDX-License-Identifier: GPL-2.0 */
  83 +/* Copyright (C) 2005-2019 Andes Technology Corporation */
  84 +#ifndef        _FP_UDF_IEX_CRTL_H
  85 +#define        _FP_UDF_IEX_CRTL_H
  86 +
  87 +/*
  88 + * The cmd list of sys_fp_udfiex_crtl()
  89 + */
  90 +/* Disable UDF or IEX trap based on the content of parameter act */
  91 +#define DISABLE_UDF_IEX_TRAP   0
  92 +/* Enable UDF or IEX trap based on the content of parameter act */
  93 +#define ENABLE_UDF_IEX_TRAP    1
  94 +/* Get current status of UDF and IEX trap */
  95 +#define GET_UDF_IEX_TRAP       2
  96 +
  97 +#endif /* _FP_UDF_IEX_CRTL_H */
  98 diff --git a/arch/nds32/include/uapi/asm/sigcontext.h b/arch/nds32/include/uapi/asm/sigcontext.h
  99 index 58afc416473e..b53634033e32 100644
 100 --- a/arch/nds32/include/uapi/asm/sigcontext.h
 101 +++ b/arch/nds32/include/uapi/asm/sigcontext.h
 102 @@ -13,14 +13,24 @@ struct fpu_struct {
 103         unsigned long long fd_regs[32];
 104         unsigned long fpcsr;
 105         /*
 106 -        * UDF_trap is used to recognize whether underflow trap is enabled
 107 -        * or not. When UDF_trap == 1, this process will be traped and then
 108 -        * get a SIGFPE signal when encountering an underflow exception.
 109 -        * UDF_trap is only modified through setfputrap syscall. Therefore,
 110 -        * UDF_trap needn't be saved or loaded to context in each context
 111 -        * switch.
 112 +        * When CONFIG_SUPPORT_DENORMAL_ARITHMETIC is defined, kernel prevents
 113 +        * hardware from treating the denormalized output as an underflow case
 114 +        * and rounding it to a normal number. Hence kernel enables the UDF and
 115 +        * IEX trap in the fpcsr register to step in the calculation.
 116 +        * However, the UDF and IEX trap enable bit in $fpcsr also lose
 117 +        * their use.
 118 +        *
 119 +        * UDF_IEX_trap replaces the feature of UDF and IEX trap enable bit in
 120 +        * $fpcsr to control the trap of underflow and inexact. The bit filed
 121 +        * of UDF_IEX_trap is the same as $fpcsr, 10th bit is used to enable UDF
 122 +        * exception trapping and 11th bit is used to enable IEX exception
 123 +        * trapping.
 124 +        *
 125 +        * UDF_IEX_trap is only modified through fp_udfiex_crtl syscall.
 126 +        * Therefore, UDF_IEX_trap needn't be saved and restored in each
 127 +        * context switch.
 128          */
 129 -       unsigned long UDF_trap;
 130 +       unsigned long UDF_IEX_trap;
 131  };
 132
 133  struct zol_struct {
 134 diff --git a/arch/nds32/include/uapi/asm/udftrap.h b/arch/nds32/include/uapi/asm/udftrap.h
 135 deleted file mode 100644
 136 index 433f79d679c0..000000000000
 137 --- a/arch/nds32/include/uapi/asm/udftrap.h
 138 +++ /dev/null
 139 @@ -1,13 +0,0 @@
 140 -/* SPDX-License-Identifier: GPL-2.0 */
 141 -/* Copyright (C) 2005-2018 Andes Technology Corporation */
 142 -#ifndef        _ASM_SETFPUTRAP
 143 -#define        _ASM_SETFPUTRAP
 144 -
 145 -/*
 146 - * Options for setfputrap system call
 147 - */
 148 -#define        DISABLE_UDFTRAP 0       /* disable underflow exception trap */
 149 -#define        ENABLE_UDFTRAP  1       /* enable undeflos exception trap */
 150 -#define        GET_UDFTRAP     2       /* only get undeflos exception trap status */
 151 -
 152 -#endif /* _ASM_CACHECTL */
 153 diff --git a/arch/nds32/include/uapi/asm/unistd.h b/arch/nds32/include/uapi/asm/unistd.h
 154 index 4ec8f543103f..6b9ff90e3ae5 100644
 155 --- a/arch/nds32/include/uapi/asm/unistd.h
 156 +++ b/arch/nds32/include/uapi/asm/unistd.h
 157 @@ -11,6 +11,6 @@
 158
 159  /* Additional NDS32 specific syscalls. */
 160  #define __NR_cacheflush                (__NR_arch_specific_syscall)
 161 -#define __NR_udftrap           (__NR_arch_specific_syscall + 1)
 162 +#define __NR_fp_udfiex_crtl    (__NR_arch_specific_syscall + 1)
 163  __SYSCALL(__NR_cacheflush, sys_cacheflush)
 164 -__SYSCALL(__NR_udftrap, sys_udftrap)
 165 +__SYSCALL(__NR_fp_udfiex_crtl, sys_fp_udfiex_crtl)
 166 diff --git a/arch/nds32/kernel/fpu.c b/arch/nds32/kernel/fpu.c
 167 index fddd40c7a16f..cf0b8760f261 100644
 168 --- a/arch/nds32/kernel/fpu.c
 169 +++ b/arch/nds32/kernel/fpu.c
 170 @@ -14,7 +14,7 @@ const struct fpu_struct init_fpuregs = {
 171         .fd_regs = {[0 ... 31] = sNAN64},
 172         .fpcsr = FPCSR_INIT,
 173  #if IS_ENABLED(CONFIG_SUPPORT_DENORMAL_ARITHMETIC)
 174 -       .UDF_trap = 0
 175 +       .UDF_IEX_trap = 0
 176  #endif
 177  };
 178
 179 @@ -178,7 +178,7 @@ inline void do_fpu_context_switch(struct pt_regs *regs)
 180                 /* First time FPU user.  */
 181                 load_fpu(&init_fpuregs);
 182  #if IS_ENABLED(CONFIG_SUPPORT_DENORMAL_ARITHMETIC)
 183 -               current->thread.fpu.UDF_trap = init_fpuregs.UDF_trap;
 184 +               current->thread.fpu.UDF_IEX_trap = init_fpuregs.UDF_IEX_trap;
 185  #endif
 186                 set_used_math();
 187         }
 188 @@ -206,7 +206,7 @@ inline void handle_fpu_exception(struct pt_regs *regs)
 189         unsigned int fpcsr;
 190         int si_code = 0, si_signo = SIGFPE;
 191  #if IS_ENABLED(CONFIG_SUPPORT_DENORMAL_ARITHMETIC)
 192 -       unsigned long redo_except = FPCSR_mskDNIT|FPCSR_mskUDFT;
 193 +       unsigned long redo_except = FPCSR_mskDNIT|FPCSR_mskUDFT|FPCSR_mskIEXT;
 194  #else
 195         unsigned long redo_except = FPCSR_mskDNIT;
 196  #endif
 197 @@ -215,21 +215,18 @@ inline void handle_fpu_exception(struct pt_regs *regs)
 198         fpcsr = current->thread.fpu.fpcsr;
 199
 200         if (fpcsr & redo_except) {
 201 -#if IS_ENABLED(CONFIG_SUPPORT_DENORMAL_ARITHMETIC)
 202 -               if (fpcsr & FPCSR_mskUDFT)
 203 -                       current->thread.fpu.fpcsr &= ~FPCSR_mskIEX;
 204 -#endif
 205                 si_signo = do_fpuemu(regs, &current->thread.fpu);
 206                 fpcsr = current->thread.fpu.fpcsr;
 207 -               if (!si_signo)
 208 +               if (!si_signo) {
 209 +                       current->thread.fpu.fpcsr &= ~(redo_except);
 210                         goto done;
 211 +               }
 212         } else if (fpcsr & FPCSR_mskRIT) {
 213                 if (!user_mode(regs))
 214                         do_exit(SIGILL);
 215                 si_signo = SIGILL;
 216         }
 217
 218 -
 219         switch (si_signo) {
 220         case SIGFPE:
 221                 fill_sigfpe_signo(fpcsr, &si_code);
 222 diff --git a/arch/nds32/kernel/sys_nds32.c b/arch/nds32/kernel/sys_nds32.c
 223 index 0835277636ce..cb2d1e219bb3 100644
 224 --- a/arch/nds32/kernel/sys_nds32.c
 225 +++ b/arch/nds32/kernel/sys_nds32.c
 226 @@ -6,8 +6,8 @@
 227
 228  #include <asm/cachectl.h>
 229  #include <asm/proc-fns.h>
 230 -#include <asm/udftrap.h>
 231  #include <asm/fpu.h>
 232 +#include <asm/fp_udfiex_crtl.h>
 233
 234  SYSCALL_DEFINE6(mmap2, unsigned long, addr, unsigned long, len,
 235                unsigned long, prot, unsigned long, flags,
 236 @@ -51,31 +51,33 @@ SYSCALL_DEFINE3(cacheflush, unsigned int, start, unsigned int, end, int, cache)
 237         return 0;
 238  }
 239
 240 -SYSCALL_DEFINE1(udftrap, int, option)
 241 +SYSCALL_DEFINE2(fp_udfiex_crtl, unsigned int, cmd, unsigned int, act)
 242  {
 243  #if IS_ENABLED(CONFIG_SUPPORT_DENORMAL_ARITHMETIC)
 244 -       int old_udftrap;
 245 +       int old_udf_iex;
 246
 247         if (!used_math()) {
 248                 load_fpu(&init_fpuregs);
 249 -               current->thread.fpu.UDF_trap = init_fpuregs.UDF_trap;
 250 +               current->thread.fpu.UDF_IEX_trap = init_fpuregs.UDF_IEX_trap;
 251                 set_used_math();
 252         }
 253
 254 -       old_udftrap = current->thread.fpu.UDF_trap;
 255 -       switch (option) {
 256 -       case DISABLE_UDFTRAP:
 257 -               current->thread.fpu.UDF_trap = 0;
 258 +       old_udf_iex = current->thread.fpu.UDF_IEX_trap;
 259 +       act &= (FPCSR_mskUDFE | FPCSR_mskIEXE);
 260 +
 261 +       switch (cmd) {
 262 +       case DISABLE_UDF_IEX_TRAP:
 263 +               current->thread.fpu.UDF_IEX_trap &= ~act;
 264                 break;
 265 -       case ENABLE_UDFTRAP:
 266 -               current->thread.fpu.UDF_trap = FPCSR_mskUDFE;
 267 +       case ENABLE_UDF_IEX_TRAP:
 268 +               current->thread.fpu.UDF_IEX_trap |= act;
 269                 break;
 270 -       case GET_UDFTRAP:
 271 +       case GET_UDF_IEX_TRAP:
 272                 break;
 273         default:
 274                 return -EINVAL;
 275         }
 276 -       return old_udftrap;
 277 +       return old_udf_iex;
 278  #else
 279         return -ENOTSUPP;
 280  #endif
 281 --
 282 2.20.1
 283