Add the rt linux 4.1.3-rt3 as base
[kvmfornfv.git] / kernel / arch / x86 / kernel / step.c
diff --git a/kernel/arch/x86/kernel/step.c b/kernel/arch/x86/kernel/step.c
new file mode 100644 (file)
index 0000000..9b4d51d
--- /dev/null
@@ -0,0 +1,230 @@
+/*
+ * x86 single-step support code, common to 32-bit and 64-bit.
+ */
+#include <linux/sched.h>
+#include <linux/mm.h>
+#include <linux/ptrace.h>
+#include <asm/desc.h>
+
+unsigned long convert_ip_to_linear(struct task_struct *child, struct pt_regs *regs)
+{
+       unsigned long addr, seg;
+
+       addr = regs->ip;
+       seg = regs->cs & 0xffff;
+       if (v8086_mode(regs)) {
+               addr = (addr & 0xffff) + (seg << 4);
+               return addr;
+       }
+
+       /*
+        * We'll assume that the code segments in the GDT
+        * are all zero-based. That is largely true: the
+        * TLS segments are used for data, and the PNPBIOS
+        * and APM bios ones we just ignore here.
+        */
+       if ((seg & SEGMENT_TI_MASK) == SEGMENT_LDT) {
+               struct desc_struct *desc;
+               unsigned long base;
+
+               seg &= ~7UL;
+
+               mutex_lock(&child->mm->context.lock);
+               if (unlikely((seg >> 3) >= child->mm->context.size))
+                       addr = -1L; /* bogus selector, access would fault */
+               else {
+                       desc = child->mm->context.ldt + seg;
+                       base = get_desc_base(desc);
+
+                       /* 16-bit code segment? */
+                       if (!desc->d)
+                               addr &= 0xffff;
+                       addr += base;
+               }
+               mutex_unlock(&child->mm->context.lock);
+       }
+
+       return addr;
+}
+
+static int is_setting_trap_flag(struct task_struct *child, struct pt_regs *regs)
+{
+       int i, copied;
+       unsigned char opcode[15];
+       unsigned long addr = convert_ip_to_linear(child, regs);
+
+       copied = access_process_vm(child, addr, opcode, sizeof(opcode), 0);
+       for (i = 0; i < copied; i++) {
+               switch (opcode[i]) {
+               /* popf and iret */
+               case 0x9d: case 0xcf:
+                       return 1;
+
+                       /* CHECKME: 64 65 */
+
+               /* opcode and address size prefixes */
+               case 0x66: case 0x67:
+                       continue;
+               /* irrelevant prefixes (segment overrides and repeats) */
+               case 0x26: case 0x2e:
+               case 0x36: case 0x3e:
+               case 0x64: case 0x65:
+               case 0xf0: case 0xf2: case 0xf3:
+                       continue;
+
+#ifdef CONFIG_X86_64
+               case 0x40 ... 0x4f:
+                       if (!user_64bit_mode(regs))
+                               /* 32-bit mode: register increment */
+                               return 0;
+                       /* 64-bit mode: REX prefix */
+                       continue;
+#endif
+
+                       /* CHECKME: f2, f3 */
+
+               /*
+                * pushf: NOTE! We should probably not let
+                * the user see the TF bit being set. But
+                * it's more pain than it's worth to avoid
+                * it, and a debugger could emulate this
+                * all in user space if it _really_ cares.
+                */
+               case 0x9c:
+               default:
+                       return 0;
+               }
+       }
+       return 0;
+}
+
+/*
+ * Enable single-stepping.  Return nonzero if user mode is not using TF itself.
+ */
+static int enable_single_step(struct task_struct *child)
+{
+       struct pt_regs *regs = task_pt_regs(child);
+       unsigned long oflags;
+
+       /*
+        * If we stepped into a sysenter/syscall insn, it trapped in
+        * kernel mode; do_debug() cleared TF and set TIF_SINGLESTEP.
+        * If user-mode had set TF itself, then it's still clear from
+        * do_debug() and we need to set it again to restore the user
+        * state so we don't wrongly set TIF_FORCED_TF below.
+        * If enable_single_step() was used last and that is what
+        * set TIF_SINGLESTEP, then both TF and TIF_FORCED_TF are
+        * already set and our bookkeeping is fine.
+        */
+       if (unlikely(test_tsk_thread_flag(child, TIF_SINGLESTEP)))
+               regs->flags |= X86_EFLAGS_TF;
+
+       /*
+        * Always set TIF_SINGLESTEP - this guarantees that
+        * we single-step system calls etc..  This will also
+        * cause us to set TF when returning to user mode.
+        */
+       set_tsk_thread_flag(child, TIF_SINGLESTEP);
+
+       oflags = regs->flags;
+
+       /* Set TF on the kernel stack.. */
+       regs->flags |= X86_EFLAGS_TF;
+
+       /*
+        * ..but if TF is changed by the instruction we will trace,
+        * don't mark it as being "us" that set it, so that we
+        * won't clear it by hand later.
+        *
+        * Note that if we don't actually execute the popf because
+        * of a signal arriving right now or suchlike, we will lose
+        * track of the fact that it really was "us" that set it.
+        */
+       if (is_setting_trap_flag(child, regs)) {
+               clear_tsk_thread_flag(child, TIF_FORCED_TF);
+               return 0;
+       }
+
+       /*
+        * If TF was already set, check whether it was us who set it.
+        * If not, we should never attempt a block step.
+        */
+       if (oflags & X86_EFLAGS_TF)
+               return test_tsk_thread_flag(child, TIF_FORCED_TF);
+
+       set_tsk_thread_flag(child, TIF_FORCED_TF);
+
+       return 1;
+}
+
+void set_task_blockstep(struct task_struct *task, bool on)
+{
+       unsigned long debugctl;
+
+       /*
+        * Ensure irq/preemption can't change debugctl in between.
+        * Note also that both TIF_BLOCKSTEP and debugctl should
+        * be changed atomically wrt preemption.
+        *
+        * NOTE: this means that set/clear TIF_BLOCKSTEP is only safe if
+        * task is current or it can't be running, otherwise we can race
+        * with __switch_to_xtra(). We rely on ptrace_freeze_traced() but
+        * PTRACE_KILL is not safe.
+        */
+       local_irq_disable();
+       debugctl = get_debugctlmsr();
+       if (on) {
+               debugctl |= DEBUGCTLMSR_BTF;
+               set_tsk_thread_flag(task, TIF_BLOCKSTEP);
+       } else {
+               debugctl &= ~DEBUGCTLMSR_BTF;
+               clear_tsk_thread_flag(task, TIF_BLOCKSTEP);
+       }
+       if (task == current)
+               update_debugctlmsr(debugctl);
+       local_irq_enable();
+}
+
+/*
+ * Enable single or block step.
+ */
+static void enable_step(struct task_struct *child, bool block)
+{
+       /*
+        * Make sure block stepping (BTF) is not enabled unless it should be.
+        * Note that we don't try to worry about any is_setting_trap_flag()
+        * instructions after the first when using block stepping.
+        * So no one should try to use debugger block stepping in a program
+        * that uses user-mode single stepping itself.
+        */
+       if (enable_single_step(child) && block)
+               set_task_blockstep(child, true);
+       else if (test_tsk_thread_flag(child, TIF_BLOCKSTEP))
+               set_task_blockstep(child, false);
+}
+
+void user_enable_single_step(struct task_struct *child)
+{
+       enable_step(child, 0);
+}
+
+void user_enable_block_step(struct task_struct *child)
+{
+       enable_step(child, 1);
+}
+
+void user_disable_single_step(struct task_struct *child)
+{
+       /*
+        * Make sure block stepping (BTF) is disabled.
+        */
+       if (test_tsk_thread_flag(child, TIF_BLOCKSTEP))
+               set_task_blockstep(child, false);
+
+       /* Always clear TIF_SINGLESTEP... */
+       clear_tsk_thread_flag(child, TIF_SINGLESTEP);
+
+       /* But touch TF only if it was set by us.. */
+       if (test_and_clear_tsk_thread_flag(child, TIF_FORCED_TF))
+               task_pt_regs(child)->flags &= ~X86_EFLAGS_TF;
+}