From 432c6bacbd0c16ec210c43da411ccc3855c4c010 Mon Sep 17 00:00:00 2001 From: Paul Burton Date: Fri, 8 Jul 2016 11:06:19 +0100 Subject: MIPS: Use per-mm page to execute branch delay slot instructions In some cases the kernel needs to execute an instruction from the delay slot of an emulated branch instruction. These cases include: - Emulated floating point branch instructions (bc1[ft]l?) for systems which don't include an FPU, or upon which the kernel is run with the "nofpu" parameter. - MIPSr6 systems running binaries targeting older revisions of the architecture, which may include branch instructions whose encodings are no longer valid in MIPSr6. Executing instructions from such delay slots is done by writing the instruction to memory followed by a trap, as part of an "emuframe", and executing it. This avoids the requirement of an emulator for the entire MIPS instruction set. Prior to this patch such emuframes are written to the user stack and executed from there. This patch moves FP branch delay emuframes off of the user stack and into a per-mm page. Allocating a page per-mm leaves userland with access to only what it had access to previously, and compared to other solutions is relatively simple. When a thread requires a delay slot emulation, it is allocated a frame. A thread may only have one frame allocated at any one time, since it may only ever be executing one instruction at any one time. In order to ensure that we can free up allocated frame later, its index is recorded in struct thread_struct. In the typical case, after executing the delay slot instruction we'll execute a break instruction with the BRK_MEMU code. This traps back to the kernel & leads to a call to do_dsemulret which frees the allocated frame & moves the user PC back to the instruction that would have executed following the emulated branch. In some cases the delay slot instruction may be invalid, such as a branch, or may trigger an exception. In these cases the BRK_MEMU break instruction will not be hit. In order to ensure that frames are freed this patch introduces dsemul_thread_cleanup() and calls it to free any allocated frame upon thread exit. If the instruction generated an exception & leads to a signal being delivered to the thread, or indeed if a signal simply happens to be delivered to the thread whilst it is executing from the struct emuframe, then we need to take care to exit the frame appropriately. This is done by either rolling back the user PC to the branch or advancing it to the continuation PC prior to signal delivery, using dsemul_thread_rollback(). If this were not done then a sigreturn would return to the struct emuframe, and if that frame had meanwhile been used in response to an emulated branch instruction within the signal handler then we would execute the wrong user code. Whilst a user could theoretically place something like a compact branch to self in a delay slot and cause their thread to become stuck in an infinite loop with the frame never being deallocated, this would: - Only affect the users single process. - Be architecturally invalid since there would be a branch in the delay slot, which is forbidden. - Be extremely unlikely to happen by mistake, and provide a program with no more ability to harm the system than a simple infinite loop would. If a thread requires a delay slot emulation & no frame is available to it (ie. the process has enough other threads that all frames are currently in use) then the thread joins a waitqueue. It will sleep until a frame is freed by another thread in the process. Since we now know whether a thread has an allocated frame due to our tracking of its index, the cookie field of struct emuframe is removed as we can be more certain whether we have a valid frame. Since a thread may only ever have a single frame at any given time, the epc field of struct emuframe is also removed & the PC to continue from is instead stored in struct thread_struct. Together these changes simplify & shrink struct emuframe somewhat, allowing twice as many frames to fit into the page allocated for them. The primary benefit of this patch is that we are now free to mark the user stack non-executable where that is possible. Signed-off-by: Paul Burton Cc: Leonid Yegoshin Cc: Maciej Rozycki Cc: Faraz Shahbazker Cc: Raghu Gandham Cc: Matthew Fortune Cc: linux-mips@linux-mips.org Patchwork: https://patchwork.linux-mips.org/patch/13764/ Signed-off-by: Ralf Baechle --- arch/mips/math-emu/cp1emu.c | 8 +- arch/mips/math-emu/dsemul.c | 333 +++++++++++++++++++++++++++++--------------- 2 files changed, 227 insertions(+), 114 deletions(-) (limited to 'arch/mips/math-emu') diff --git a/arch/mips/math-emu/cp1emu.c b/arch/mips/math-emu/cp1emu.c index d96e912b9d44..8afa090f9536 100644 --- a/arch/mips/math-emu/cp1emu.c +++ b/arch/mips/math-emu/cp1emu.c @@ -434,8 +434,8 @@ static int microMIPS32_to_MIPS32(union mips_instruction *insn_ptr) * a single subroutine should be used across both * modules. */ -static int isBranchInstr(struct pt_regs *regs, struct mm_decoded_insn dec_insn, - unsigned long *contpc) +int isBranchInstr(struct pt_regs *regs, struct mm_decoded_insn dec_insn, + unsigned long *contpc) { union mips_instruction insn = (union mips_instruction)dec_insn.insn; unsigned int fcr31; @@ -1268,7 +1268,7 @@ branch_common: * instruction in the dslot. */ sig = mips_dsemul(xcp, ir, - contpc); + bcpc, contpc); if (sig < 0) break; if (sig) @@ -1323,7 +1323,7 @@ branch_common: * Single step the non-cp1 * instruction in the dslot */ - sig = mips_dsemul(xcp, ir, contpc); + sig = mips_dsemul(xcp, ir, bcpc, contpc); if (sig < 0) break; if (sig) diff --git a/arch/mips/math-emu/dsemul.c b/arch/mips/math-emu/dsemul.c index 47074887e64c..72a4642eee2c 100644 --- a/arch/mips/math-emu/dsemul.c +++ b/arch/mips/math-emu/dsemul.c @@ -1,3 +1,6 @@ +#include +#include + #include #include #include @@ -5,43 +8,211 @@ #include #include -#include "ieee754.h" - -/* - * Emulate the arbitrary instruction ir at xcp->cp0_epc. Required when - * we have to emulate the instruction in a COP1 branch delay slot. Do - * not change cp0_epc due to the instruction +/** + * struct emuframe - The 'emulation' frame structure + * @emul: The instruction to 'emulate'. + * @badinst: A break instruction to cause a return to the kernel. * - * According to the spec: - * 1) it shouldn't be a branch :-) - * 2) it can be a COP instruction :-( - * 3) if we are tring to run a protected memory space we must take - * special care on memory access instructions :-( - */ - -/* - * "Trampoline" return routine to catch exception following - * execution of delay-slot instruction execution. + * This structure defines the frames placed within the delay slot emulation + * page in response to a call to mips_dsemul(). Each thread may be allocated + * only one frame at any given time. The kernel stores within it the + * instruction to be 'emulated' followed by a break instruction, then + * executes the frame in user mode. The break causes a trap to the kernel + * which leads to do_dsemulret() being called unless the instruction in + * @emul causes a trap itself, is a branch, or a signal is delivered to + * the thread. In these cases the allocated frame will either be reused by + * a subsequent delay slot 'emulation', or be freed during signal delivery or + * upon thread exit. + * + * This approach is used because: + * + * - Actually emulating all instructions isn't feasible. We would need to + * be able to handle instructions from all revisions of the MIPS ISA, + * all ASEs & all vendor instruction set extensions. This would be a + * whole lot of work & continual maintenance burden as new instructions + * are introduced, and in the case of some vendor extensions may not + * even be possible. Thus we need to take the approach of actually + * executing the instruction. + * + * - We must execute the instruction within user context. If we were to + * execute the instruction in kernel mode then it would have access to + * kernel resources without very careful checks, leaving us with a + * high potential for security or stability issues to arise. + * + * - We used to place the frame on the users stack, but this requires + * that the stack be executable. This is bad for security so the + * per-process page is now used instead. + * + * - The instruction in @emul may be something entirely invalid for a + * delay slot. The user may (intentionally or otherwise) place a branch + * in a delay slot, or a kernel mode instruction, or something else + * which generates an exception. Thus we can't rely upon the break in + * @badinst always being hit. For this reason we track the index of the + * frame allocated to each thread, allowing us to clean it up at later + * points such as signal delivery or thread exit. + * + * - The user may generate a fake struct emuframe if they wish, invoking + * the BRK_MEMU break instruction themselves. We must therefore not + * trust that BRK_MEMU means there's actually a valid frame allocated + * to the thread, and must not allow the user to do anything they + * couldn't already. */ - struct emuframe { mips_instruction emul; mips_instruction badinst; - mips_instruction cookie; - unsigned long epc; }; -/* - * Set up an emulation frame for instruction IR, from a delay slot of - * a branch jumping to CPC. Return 0 if successful, -1 if no emulation - * required, otherwise a signal number causing a frame setup failure. - */ -int mips_dsemul(struct pt_regs *regs, mips_instruction ir, unsigned long cpc) +static const int emupage_frame_count = PAGE_SIZE / sizeof(struct emuframe); + +static inline __user struct emuframe *dsemul_page(void) +{ + return (__user struct emuframe *)STACK_TOP; +} + +static int alloc_emuframe(void) +{ + mm_context_t *mm_ctx = ¤t->mm->context; + int idx; + +retry: + spin_lock(&mm_ctx->bd_emupage_lock); + + /* Ensure we have an allocation bitmap */ + if (!mm_ctx->bd_emupage_allocmap) { + mm_ctx->bd_emupage_allocmap = + kcalloc(BITS_TO_LONGS(emupage_frame_count), + sizeof(unsigned long), + GFP_ATOMIC); + + if (!mm_ctx->bd_emupage_allocmap) { + idx = BD_EMUFRAME_NONE; + goto out_unlock; + } + } + + /* Attempt to allocate a single bit/frame */ + idx = bitmap_find_free_region(mm_ctx->bd_emupage_allocmap, + emupage_frame_count, 0); + if (idx < 0) { + /* + * Failed to allocate a frame. We'll wait until one becomes + * available. We unlock the page so that other threads actually + * get the opportunity to free their frames, which means + * technically the result of bitmap_full may be incorrect. + * However the worst case is that we repeat all this and end up + * back here again. + */ + spin_unlock(&mm_ctx->bd_emupage_lock); + if (!wait_event_killable(mm_ctx->bd_emupage_queue, + !bitmap_full(mm_ctx->bd_emupage_allocmap, + emupage_frame_count))) + goto retry; + + /* Received a fatal signal - just give in */ + return BD_EMUFRAME_NONE; + } + + /* Success! */ + pr_debug("allocate emuframe %d to %d\n", idx, current->pid); +out_unlock: + spin_unlock(&mm_ctx->bd_emupage_lock); + return idx; +} + +static void free_emuframe(int idx, struct mm_struct *mm) +{ + mm_context_t *mm_ctx = &mm->context; + + spin_lock(&mm_ctx->bd_emupage_lock); + + pr_debug("free emuframe %d from %d\n", idx, current->pid); + bitmap_clear(mm_ctx->bd_emupage_allocmap, idx, 1); + + /* If some thread is waiting for a frame, now's its chance */ + wake_up(&mm_ctx->bd_emupage_queue); + + spin_unlock(&mm_ctx->bd_emupage_lock); +} + +static bool within_emuframe(struct pt_regs *regs) +{ + unsigned long base = (unsigned long)dsemul_page(); + + if (regs->cp0_epc < base) + return false; + if (regs->cp0_epc >= (base + PAGE_SIZE)) + return false; + + return true; +} + +bool dsemul_thread_cleanup(struct task_struct *tsk) +{ + int fr_idx; + + /* Clear any allocated frame, retrieving its index */ + fr_idx = atomic_xchg(&tsk->thread.bd_emu_frame, BD_EMUFRAME_NONE); + + /* If no frame was allocated, we're done */ + if (fr_idx == BD_EMUFRAME_NONE) + return false; + + task_lock(tsk); + + /* Free the frame that this thread had allocated */ + if (tsk->mm) + free_emuframe(fr_idx, tsk->mm); + + task_unlock(tsk); + return true; +} + +bool dsemul_thread_rollback(struct pt_regs *regs) +{ + struct emuframe __user *fr; + int fr_idx; + + /* Do nothing if we're not executing from a frame */ + if (!within_emuframe(regs)) + return false; + + /* Find the frame being executed */ + fr_idx = atomic_read(¤t->thread.bd_emu_frame); + if (fr_idx == BD_EMUFRAME_NONE) + return false; + fr = &dsemul_page()[fr_idx]; + + /* + * If the PC is at the emul instruction, roll back to the branch. If + * PC is at the badinst (break) instruction, we've already emulated the + * instruction so progress to the continue PC. If it's anything else + * then something is amiss & the user has branched into some other area + * of the emupage - we'll free the allocated frame anyway. + */ + if (msk_isa16_mode(regs->cp0_epc) == (unsigned long)&fr->emul) + regs->cp0_epc = current->thread.bd_emu_branch_pc; + else if (msk_isa16_mode(regs->cp0_epc) == (unsigned long)&fr->badinst) + regs->cp0_epc = current->thread.bd_emu_cont_pc; + + atomic_set(¤t->thread.bd_emu_frame, BD_EMUFRAME_NONE); + free_emuframe(fr_idx, current->mm); + return true; +} + +void dsemul_mm_cleanup(struct mm_struct *mm) +{ + mm_context_t *mm_ctx = &mm->context; + + kfree(mm_ctx->bd_emupage_allocmap); +} + +int mips_dsemul(struct pt_regs *regs, mips_instruction ir, + unsigned long branch_pc, unsigned long cont_pc) { int isa16 = get_isa16_mode(regs->cp0_epc); mips_instruction break_math; struct emuframe __user *fr; - int err; + int err, fr_idx; /* NOP is easy */ if (ir == 0) @@ -68,30 +239,20 @@ int mips_dsemul(struct pt_regs *regs, mips_instruction ir, unsigned long cpc) } } - pr_debug("dsemul %lx %lx\n", regs->cp0_epc, cpc); + pr_debug("dsemul 0x%08lx cont at 0x%08lx\n", regs->cp0_epc, cont_pc); - /* - * The strategy is to push the instruction onto the user stack - * and put a trap after it which we can catch and jump to - * the required address any alternative apart from full - * instruction emulation!!. - * - * Algorithmics used a system call instruction, and - * borrowed that vector. MIPS/Linux version is a bit - * more heavyweight in the interests of portability and - * multiprocessor support. For Linux we use a BREAK 514 - * instruction causing a breakpoint exception. - */ - break_math = BREAK_MATH(isa16); - - /* Ensure that the two instructions are in the same cache line */ - fr = (struct emuframe __user *) - ((regs->regs[29] - sizeof(struct emuframe)) & ~0x7); - - /* Verify that the stack pointer is not completely insane */ - if (unlikely(!access_ok(VERIFY_WRITE, fr, sizeof(struct emuframe)))) + /* Allocate a frame if we don't already have one */ + fr_idx = atomic_read(¤t->thread.bd_emu_frame); + if (fr_idx == BD_EMUFRAME_NONE) + fr_idx = alloc_emuframe(); + if (fr_idx == BD_EMUFRAME_NONE) return SIGBUS; + fr = &dsemul_page()[fr_idx]; + + /* Retrieve the appropriately encoded break instruction */ + break_math = BREAK_MATH(isa16); + /* Write the instructions to the frame */ if (isa16) { err = __put_user(ir >> 16, (u16 __user *)(&fr->emul)); @@ -106,84 +267,36 @@ int mips_dsemul(struct pt_regs *regs, mips_instruction ir, unsigned long cpc) err |= __put_user(break_math, &fr->badinst); } - err |= __put_user((mips_instruction)BD_COOKIE, &fr->cookie); - err |= __put_user(cpc, &fr->epc); - if (unlikely(err)) { MIPS_FPU_EMU_INC_STATS(errors); + free_emuframe(fr_idx, current->mm); return SIGBUS; } + /* Record the PC of the branch, PC to continue from & frame index */ + current->thread.bd_emu_branch_pc = branch_pc; + current->thread.bd_emu_cont_pc = cont_pc; + atomic_set(¤t->thread.bd_emu_frame, fr_idx); + + /* Change user register context to execute the frame */ regs->cp0_epc = (unsigned long)&fr->emul | isa16; + /* Ensure the icache observes our newly written frame */ flush_cache_sigtramp((unsigned long)&fr->emul); return 0; } -int do_dsemulret(struct pt_regs *xcp) +bool do_dsemulret(struct pt_regs *xcp) { - int isa16 = get_isa16_mode(xcp->cp0_epc); - struct emuframe __user *fr; - unsigned long epc; - u32 insn, cookie; - int err = 0; - u16 instr[2]; - - fr = (struct emuframe __user *) - (msk_isa16_mode(xcp->cp0_epc) - sizeof(mips_instruction)); - - /* - * If we can't even access the area, something is very wrong, but we'll - * leave that to the default handling - */ - if (!access_ok(VERIFY_READ, fr, sizeof(struct emuframe))) - return 0; - - /* - * Do some sanity checking on the stackframe: - * - * - Is the instruction pointed to by the EPC an BREAK_MATH? - * - Is the following memory word the BD_COOKIE? - */ - if (isa16) { - err = __get_user(instr[0], - (u16 __user *)(&fr->badinst)); - err |= __get_user(instr[1], - (u16 __user *)((long)(&fr->badinst) + 2)); - insn = (instr[0] << 16) | instr[1]; - } else { - err = __get_user(insn, &fr->badinst); - } - err |= __get_user(cookie, &fr->cookie); - - if (unlikely(err || - insn != BREAK_MATH(isa16) || cookie != BD_COOKIE)) { + /* Cleanup the allocated frame, returning if there wasn't one */ + if (!dsemul_thread_cleanup(current)) { MIPS_FPU_EMU_INC_STATS(errors); - return 0; - } - - /* - * At this point, we are satisfied that it's a BD emulation trap. Yes, - * a user might have deliberately put two malformed and useless - * instructions in a row in his program, in which case he's in for a - * nasty surprise - the next instruction will be treated as a - * continuation address! Alas, this seems to be the only way that we - * can handle signals, recursion, and longjmps() in the context of - * emulating the branch delay instruction. - */ - - pr_debug("dsemulret\n"); - - if (__get_user(epc, &fr->epc)) { /* Saved EPC */ - /* This is not a good situation to be in */ - force_sig(SIGBUS, current); - - return 0; + return false; } /* Set EPC to return to post-branch instruction */ - xcp->cp0_epc = epc; - MIPS_FPU_EMU_INC_STATS(ds_emul); - return 1; + xcp->cp0_epc = current->thread.bd_emu_cont_pc; + pr_debug("dsemulret to 0x%08lx\n", xcp->cp0_epc); + return true; } -- cgit v1.2.3