| 1 |
// SPDX-License-Identifier: GPL-2.0 |
| 2 |
#include <linux/err.h> |
| 3 |
#include <linux/slab.h> |
| 4 |
#include <linux/mm_types.h> |
| 5 |
#include <linux/sched/task.h> |
| 6 |
|
| 7 |
#include <asm/branch.h> |
| 8 |
#include <asm/cacheflush.h> |
| 9 |
#include <asm/fpu_emulator.h> |
| 10 |
#include <asm/inst.h> |
| 11 |
#include <asm/mipsregs.h> |
| 12 |
#include <linux/uaccess.h> |
| 13 |
|
| 14 |
/** |
| 15 |
* struct emuframe - The 'emulation' frame structure |
| 16 |
* @emul: The instruction to 'emulate'. |
| 17 |
* @badinst: A break instruction to cause a return to the kernel. |
| 18 |
* |
| 19 |
* This structure defines the frames placed within the delay slot emulation |
| 20 |
* page in response to a call to mips_dsemul(). Each thread may be allocated |
| 21 |
* only one frame at any given time. The kernel stores within it the |
| 22 |
* instruction to be 'emulated' followed by a break instruction, then |
| 23 |
* executes the frame in user mode. The break causes a trap to the kernel |
| 24 |
* which leads to do_dsemulret() being called unless the instruction in |
| 25 |
* @emul causes a trap itself, is a branch, or a signal is delivered to |
| 26 |
* the thread. In these cases the allocated frame will either be reused by |
| 27 |
* a subsequent delay slot 'emulation', or be freed during signal delivery or |
| 28 |
* upon thread exit. |
| 29 |
* |
| 30 |
* This approach is used because: |
| 31 |
* |
| 32 |
* - Actually emulating all instructions isn't feasible. We would need to |
| 33 |
* be able to handle instructions from all revisions of the MIPS ISA, |
| 34 |
* all ASEs & all vendor instruction set extensions. This would be a |
| 35 |
* whole lot of work & continual maintenance burden as new instructions |
| 36 |
* are introduced, and in the case of some vendor extensions may not |
| 37 |
* even be possible. Thus we need to take the approach of actually |
| 38 |
* executing the instruction. |
| 39 |
* |
| 40 |
* - We must execute the instruction within user context. If we were to |
| 41 |
* execute the instruction in kernel mode then it would have access to |
| 42 |
* kernel resources without very careful checks, leaving us with a |
| 43 |
* high potential for security or stability issues to arise. |
| 44 |
* |
| 45 |
* - We used to place the frame on the users stack, but this requires |
| 46 |
* that the stack be executable. This is bad for security so the |
| 47 |
* per-process page is now used instead. |
| 48 |
* |
| 49 |
* - The instruction in @emul may be something entirely invalid for a |
| 50 |
* delay slot. The user may (intentionally or otherwise) place a branch |
| 51 |
* in a delay slot, or a kernel mode instruction, or something else |
| 52 |
* which generates an exception. Thus we can't rely upon the break in |
| 53 |
* @badinst always being hit. For this reason we track the index of the |
| 54 |
* frame allocated to each thread, allowing us to clean it up at later |
| 55 |
* points such as signal delivery or thread exit. |
| 56 |
* |
| 57 |
* - The user may generate a fake struct emuframe if they wish, invoking |
| 58 |
* the BRK_MEMU break instruction themselves. We must therefore not |
| 59 |
* trust that BRK_MEMU means there's actually a valid frame allocated |
| 60 |
* to the thread, and must not allow the user to do anything they |
| 61 |
* couldn't already. |
| 62 |
*/ |
| 63 |
struct emuframe { |
| 64 |
mips_instruction emul; |
| 65 |
mips_instruction badinst; |
| 66 |
}; |
| 67 |
|
| 68 |
static const int emupage_frame_count = PAGE_SIZE / sizeof(struct emuframe); |
| 69 |
|
| 70 |
static inline __user struct emuframe *dsemul_page(void) |
| 71 |
{ |
| 72 |
return (__user struct emuframe *)STACK_TOP; |
| 73 |
} |
| 74 |
|
| 75 |
static int alloc_emuframe(void) |
| 76 |
{ |
| 77 |
mm_context_t *mm_ctx = ¤t->mm->context; |
| 78 |
int idx; |
| 79 |
|
| 80 |
retry: |
| 81 |
spin_lock(&mm_ctx->bd_emupage_lock); |
| 82 |
|
| 83 |
/* Ensure we have an allocation bitmap */ |
| 84 |
if (!mm_ctx->bd_emupage_allocmap) { |
| 85 |
mm_ctx->bd_emupage_allocmap = bitmap_zalloc(emupage_frame_count, |
| 86 |
GFP_ATOMIC); |
| 87 |
if (!mm_ctx->bd_emupage_allocmap) { |
| 88 |
idx = BD_EMUFRAME_NONE; |
| 89 |
goto out_unlock; |
| 90 |
} |
| 91 |
} |
| 92 |
|
| 93 |
/* Attempt to allocate a single bit/frame */ |
| 94 |
idx = bitmap_find_free_region(mm_ctx->bd_emupage_allocmap, |
| 95 |
emupage_frame_count, 0); |
| 96 |
if (idx < 0) { |
| 97 |
/* |
| 98 |
* Failed to allocate a frame. We'll wait until one becomes |
| 99 |
* available. We unlock the page so that other threads actually |
| 100 |
* get the opportunity to free their frames, which means |
| 101 |
* technically the result of bitmap_full may be incorrect. |
| 102 |
* However the worst case is that we repeat all this and end up |
| 103 |
* back here again. |
| 104 |
*/ |
| 105 |
spin_unlock(&mm_ctx->bd_emupage_lock); |
| 106 |
if (!wait_event_killable(mm_ctx->bd_emupage_queue, |
| 107 |
!bitmap_full(mm_ctx->bd_emupage_allocmap, |
| 108 |
emupage_frame_count))) |
| 109 |
goto retry; |
| 110 |
|
| 111 |
/* Received a fatal signal - just give in */ |
| 112 |
return BD_EMUFRAME_NONE; |
| 113 |
} |
| 114 |
|
| 115 |
/* Success! */ |
| 116 |
pr_debug("allocate emuframe %d to %d\n", idx, current->pid); |
| 117 |
out_unlock: |
| 118 |
spin_unlock(&mm_ctx->bd_emupage_lock); |
| 119 |
return idx; |
| 120 |
} |
| 121 |
|
| 122 |
static void free_emuframe(int idx, struct mm_struct *mm) |
| 123 |
{ |
| 124 |
mm_context_t *mm_ctx = &mm->context; |
| 125 |
|
| 126 |
spin_lock(&mm_ctx->bd_emupage_lock); |
| 127 |
|
| 128 |
pr_debug("free emuframe %d from %d\n", idx, current->pid); |
| 129 |
bitmap_clear(mm_ctx->bd_emupage_allocmap, idx, 1); |
| 130 |
|
| 131 |
/* If some thread is waiting for a frame, now's its chance */ |
| 132 |
wake_up(&mm_ctx->bd_emupage_queue); |
| 133 |
|
| 134 |
spin_unlock(&mm_ctx->bd_emupage_lock); |
| 135 |
} |
| 136 |
|
| 137 |
static bool within_emuframe(struct pt_regs *regs) |
| 138 |
{ |
| 139 |
unsigned long base = (unsigned long)dsemul_page(); |
| 140 |
|
| 141 |
if (regs->cp0_epc < base) |
| 142 |
return false; |
| 143 |
if (regs->cp0_epc >= (base + PAGE_SIZE)) |
| 144 |
return false; |
| 145 |
|
| 146 |
return true; |
| 147 |
} |
| 148 |
|
| 149 |
bool dsemul_thread_cleanup(struct task_struct *tsk) |
| 150 |
{ |
| 151 |
int fr_idx; |
| 152 |
|
| 153 |
/* Clear any allocated frame, retrieving its index */ |
| 154 |
fr_idx = atomic_xchg(&tsk->thread.bd_emu_frame, BD_EMUFRAME_NONE); |
| 155 |
|
| 156 |
/* If no frame was allocated, we're done */ |
| 157 |
if (fr_idx == BD_EMUFRAME_NONE) |
| 158 |
return false; |
| 159 |
|
| 160 |
task_lock(tsk); |
| 161 |
|
| 162 |
/* Free the frame that this thread had allocated */ |
| 163 |
if (tsk->mm) |
| 164 |
free_emuframe(fr_idx, tsk->mm); |
| 165 |
|
| 166 |
task_unlock(tsk); |
| 167 |
return true; |
| 168 |
} |
| 169 |
|
| 170 |
bool dsemul_thread_rollback(struct pt_regs *regs) |
| 171 |
{ |
| 172 |
struct emuframe __user *fr; |
| 173 |
int fr_idx; |
| 174 |
|
| 175 |
/* Do nothing if we're not executing from a frame */ |
| 176 |
if (!within_emuframe(regs)) |
| 177 |
return false; |
| 178 |
|
| 179 |
/* Find the frame being executed */ |
| 180 |
fr_idx = atomic_read(¤t->thread.bd_emu_frame); |
| 181 |
if (fr_idx == BD_EMUFRAME_NONE) |
| 182 |
return false; |
| 183 |
fr = &dsemul_page()[fr_idx]; |
| 184 |
|
| 185 |
/* |
| 186 |
* If the PC is at the emul instruction, roll back to the branch. If |
| 187 |
* PC is at the badinst (break) instruction, we've already emulated the |
| 188 |
* instruction so progress to the continue PC. If it's anything else |
| 189 |
* then something is amiss & the user has branched into some other area |
| 190 |
* of the emupage - we'll free the allocated frame anyway. |
| 191 |
*/ |
| 192 |
if (msk_isa16_mode(regs->cp0_epc) == (unsigned long)&fr->emul) |
| 193 |
regs->cp0_epc = current->thread.bd_emu_branch_pc; |
| 194 |
else if (msk_isa16_mode(regs->cp0_epc) == (unsigned long)&fr->badinst) |
| 195 |
regs->cp0_epc = current->thread.bd_emu_cont_pc; |
| 196 |
|
| 197 |
atomic_set(¤t->thread.bd_emu_frame, BD_EMUFRAME_NONE); |
| 198 |
free_emuframe(fr_idx, current->mm); |
| 199 |
return true; |
| 200 |
} |
| 201 |
|
| 202 |
void dsemul_mm_cleanup(struct mm_struct *mm) |
| 203 |
{ |
| 204 |
mm_context_t *mm_ctx = &mm->context; |
| 205 |
|
| 206 |
bitmap_free(mm_ctx->bd_emupage_allocmap); |
| 207 |
} |
| 208 |
|
| 209 |
int mips_dsemul(struct pt_regs *regs, mips_instruction ir, |
| 210 |
unsigned long branch_pc, unsigned long cont_pc) |
| 211 |
{ |
| 212 |
int isa16 = get_isa16_mode(regs->cp0_epc); |
| 213 |
mips_instruction break_math; |
| 214 |
unsigned long fr_uaddr; |
| 215 |
struct emuframe fr; |
| 216 |
int fr_idx, ret; |
| 217 |
|
| 218 |
/* NOP is easy */ |
| 219 |
if (ir == 0) |
| 220 |
return -1; |
| 221 |
|
| 222 |
/* microMIPS instructions */ |
| 223 |
if (isa16) { |
| 224 |
union mips_instruction insn = { .word = ir }; |
| 225 |
|
| 226 |
/* NOP16 aka MOVE16 $0, $0 */ |
| 227 |
if ((ir >> 16) == MM_NOP16) |
| 228 |
return -1; |
| 229 |
|
| 230 |
/* ADDIUPC */ |
| 231 |
if (insn.mm_a_format.opcode == mm_addiupc_op) { |
| 232 |
unsigned int rs; |
| 233 |
s32 v; |
| 234 |
|
| 235 |
rs = (((insn.mm_a_format.rs + 0xe) & 0xf) + 2); |
| 236 |
v = regs->cp0_epc & ~3; |
| 237 |
v += insn.mm_a_format.simmediate << 2; |
| 238 |
regs->regs[rs] = (long)v; |
| 239 |
return -1; |
| 240 |
} |
| 241 |
} |
| 242 |
|
| 243 |
pr_debug("dsemul 0x%08lx cont at 0x%08lx\n", regs->cp0_epc, cont_pc); |
| 244 |
|
| 245 |
/* Allocate a frame if we don't already have one */ |
| 246 |
fr_idx = atomic_read(¤t->thread.bd_emu_frame); |
| 247 |
if (fr_idx == BD_EMUFRAME_NONE) |
| 248 |
fr_idx = alloc_emuframe(); |
| 249 |
if (fr_idx == BD_EMUFRAME_NONE) |
| 250 |
return SIGBUS; |
| 251 |
|
| 252 |
/* Retrieve the appropriately encoded break instruction */ |
| 253 |
break_math = BREAK_MATH(isa16); |
| 254 |
|
| 255 |
/* Write the instructions to the frame */ |
| 256 |
if (isa16) { |
| 257 |
union mips_instruction _emul = { |
| 258 |
.halfword = { ir >> 16, ir } |
| 259 |
}; |
| 260 |
union mips_instruction _badinst = { |
| 261 |
.halfword = { break_math >> 16, break_math } |
| 262 |
}; |
| 263 |
|
| 264 |
fr.emul = _emul.word; |
| 265 |
fr.badinst = _badinst.word; |
| 266 |
} else { |
| 267 |
fr.emul = ir; |
| 268 |
fr.badinst = break_math; |
| 269 |
} |
| 270 |
|
| 271 |
/* Write the frame to user memory */ |
| 272 |
fr_uaddr = (unsigned long)&dsemul_page()[fr_idx]; |
| 273 |
ret = access_process_vm(current, fr_uaddr, &fr, sizeof(fr), |
| 274 |
FOLL_FORCE | FOLL_WRITE); |
| 275 |
if (unlikely(ret != sizeof(fr))) { |
| 276 |
MIPS_FPU_EMU_INC_STATS(errors); |
| 277 |
free_emuframe(fr_idx, current->mm); |
| 278 |
return SIGBUS; |
| 279 |
} |
| 280 |
|
| 281 |
/* Record the PC of the branch, PC to continue from & frame index */ |
| 282 |
current->thread.bd_emu_branch_pc = branch_pc; |
| 283 |
current->thread.bd_emu_cont_pc = cont_pc; |
| 284 |
atomic_set(¤t->thread.bd_emu_frame, fr_idx); |
| 285 |
|
| 286 |
/* Change user register context to execute the frame */ |
| 287 |
regs->cp0_epc = fr_uaddr | isa16; |
| 288 |
|
| 289 |
return 0; |
| 290 |
} |
| 291 |
|
| 292 |
bool do_dsemulret(struct pt_regs *xcp) |
| 293 |
{ |
| 294 |
/* Cleanup the allocated frame, returning if there wasn't one */ |
| 295 |
if (!dsemul_thread_cleanup(current)) { |
| 296 |
MIPS_FPU_EMU_INC_STATS(errors); |
| 297 |
return false; |
| 298 |
} |
| 299 |
|
| 300 |
/* Set EPC to return to post-branch instruction */ |
| 301 |
xcp->cp0_epc = current->thread.bd_emu_cont_pc; |
| 302 |
pr_debug("dsemulret to 0x%08lx\n", xcp->cp0_epc); |
| 303 |
MIPS_FPU_EMU_INC_STATS(ds_emul); |
| 304 |
return true; |
| 305 |
} |