]> git.karo-electronics.de Git - karo-tx-linux.git/commitdiff
tilegx: support KGDB
authorChris Metcalf <cmetcalf@tilera.com>
Mon, 12 Aug 2013 16:49:24 +0000 (12:49 -0400)
committerChris Metcalf <cmetcalf@tilera.com>
Tue, 13 Aug 2013 20:28:09 +0000 (16:28 -0400)
Enter kernel debugger at boot with:
  --hvd UART_1=1 --hvx kgdbwait --hvx kgdboc=ttyS1,115200
or at runtime with:
  echo ttyS1,115200 > /sys/module/kgdboc/parameters/kgdboc
  echo g > /proc/sysrq-trigger

Signed-off-by: Chris Metcalf <cmetcalf@tilera.com>
arch/tile/Kconfig
arch/tile/include/asm/kgdb.h [new file with mode: 0644]
arch/tile/kernel/Makefile
arch/tile/kernel/kgdb.c [new file with mode: 0644]
arch/tile/kernel/traps.c

index c7137b368c7da834e185f9cf231e065c6bb0073d..e2390130613d85cc07f0228ce153e76ef9b1d5f6 100644 (file)
@@ -128,6 +128,7 @@ config TILEGX
        select HAVE_KPROBES
        select HAVE_KRETPROBES
        select HAVE_KVM if !KVM_GUEST
+       select HAVE_ARCH_KGDB
 
 config TILEPRO
        def_bool !TILEGX
diff --git a/arch/tile/include/asm/kgdb.h b/arch/tile/include/asm/kgdb.h
new file mode 100644 (file)
index 0000000..280c181
--- /dev/null
@@ -0,0 +1,71 @@
+/*
+ * Copyright 2013 Tilera Corporation. All Rights Reserved.
+ *
+ *   This program is free software; you can redistribute it and/or
+ *   modify it under the terms of the GNU General Public License
+ *   as published by the Free Software Foundation, version 2.
+ *
+ *   This program is distributed in the hope that it will be useful, but
+ *   WITHOUT ANY WARRANTY; without even the implied warranty of
+ *   MERCHANTABILITY OR FITNESS FOR A PARTICULAR PURPOSE, GOOD TITLE or
+ *   NON INFRINGEMENT.  See the GNU General Public License for
+ *   more details.
+ *
+ * TILE-Gx KGDB support.
+ */
+
+#ifndef __TILE_KGDB_H__
+#define __TILE_KGDB_H__
+
+#include <linux/kdebug.h>
+#include <arch/opcode.h>
+
+#define GDB_SIZEOF_REG         sizeof(unsigned long)
+
+/*
+ * TILE-Gx gdb is expecting the following register layout:
+ * 56 GPRs(R0 - R52, TP, SP, LR), 8 special GPRs(networks and ZERO),
+ * plus the PC and the faultnum.
+ *
+ * Even though kernel not use the 8 special GPRs, they need to be present
+ * in the registers sent for correct processing in the host-side gdb.
+ *
+ */
+#define DBG_MAX_REG_NUM                (56+8+2)
+#define NUMREGBYTES            (DBG_MAX_REG_NUM * GDB_SIZEOF_REG)
+
+/*
+ * BUFMAX defines the maximum number of characters in inbound/outbound
+ * buffers at least NUMREGBYTES*2 are needed for register packets,
+ * Longer buffer is needed to list all threads.
+ */
+#define BUFMAX                 2048
+
+#define BREAK_INSTR_SIZE       TILEGX_BUNDLE_SIZE_IN_BYTES
+
+/*
+ * Require cache flush for set/clear a software breakpoint or write memory.
+ */
+#define CACHE_FLUSH_IS_SAFE    1
+
+/*
+ * The compiled-in breakpoint instruction can be used to "break" into
+ * the debugger via magic system request key (sysrq-G).
+ */
+static tile_bundle_bits compiled_bpt = TILEGX_BPT_BUNDLE | DIE_COMPILED_BPT;
+
+enum tilegx_regnum {
+       TILEGX_PC_REGNUM = TREG_LAST_GPR + 9,
+       TILEGX_FAULTNUM_REGNUM,
+};
+
+/*
+ * Generate a breakpoint exception to "break" into the debugger.
+ */
+static inline void arch_kgdb_breakpoint(void)
+{
+       asm volatile (".quad %0\n\t"
+                     ::""(compiled_bpt));
+}
+
+#endif /* __TILE_KGDB_H__ */
index b638d3efd2d10796aad11c076b53cb4494856889..4fb7098ff7456425a064f0c58beac0b3a23b8b60 100644 (file)
@@ -30,5 +30,6 @@ obj-$(CONFIG_TILE_HVGLUE_TRACE)       += hvglue_trace.o
 obj-$(CONFIG_FUNCTION_TRACER)  += ftrace.o mcount_64.o
 obj-$(CONFIG_KPROBES)          += kprobes.o
 obj-$(CONFIG_KVM_GUEST)                += kvm_virtio.o
+obj-$(CONFIG_KGDB)             += kgdb.o
 
 obj-y                          += vdso/
diff --git a/arch/tile/kernel/kgdb.c b/arch/tile/kernel/kgdb.c
new file mode 100644 (file)
index 0000000..4cd8838
--- /dev/null
@@ -0,0 +1,499 @@
+/*
+ * Copyright 2013 Tilera Corporation. All Rights Reserved.
+ *
+ *   This program is free software; you can redistribute it and/or
+ *   modify it under the terms of the GNU General Public License
+ *   as published by the Free Software Foundation, version 2.
+ *
+ *   This program is distributed in the hope that it will be useful, but
+ *   WITHOUT ANY WARRANTY; without even the implied warranty of
+ *   MERCHANTABILITY OR FITNESS FOR A PARTICULAR PURPOSE, GOOD TITLE or
+ *   NON INFRINGEMENT.  See the GNU General Public License for
+ *   more details.
+ *
+ * TILE-Gx KGDB support.
+ */
+
+#include <linux/ptrace.h>
+#include <linux/kgdb.h>
+#include <linux/kdebug.h>
+#include <linux/uaccess.h>
+#include <linux/module.h>
+#include <asm/cacheflush.h>
+
+static tile_bundle_bits singlestep_insn = TILEGX_BPT_BUNDLE | DIE_SSTEPBP;
+static unsigned long stepped_addr;
+static tile_bundle_bits stepped_instr;
+
+struct dbg_reg_def_t dbg_reg_def[DBG_MAX_REG_NUM] = {
+       { "r0", GDB_SIZEOF_REG, offsetof(struct pt_regs, regs[0])},
+       { "r1", GDB_SIZEOF_REG, offsetof(struct pt_regs, regs[1])},
+       { "r2", GDB_SIZEOF_REG, offsetof(struct pt_regs, regs[2])},
+       { "r3", GDB_SIZEOF_REG, offsetof(struct pt_regs, regs[3])},
+       { "r4", GDB_SIZEOF_REG, offsetof(struct pt_regs, regs[4])},
+       { "r5", GDB_SIZEOF_REG, offsetof(struct pt_regs, regs[5])},
+       { "r6", GDB_SIZEOF_REG, offsetof(struct pt_regs, regs[6])},
+       { "r7", GDB_SIZEOF_REG, offsetof(struct pt_regs, regs[7])},
+       { "r8", GDB_SIZEOF_REG, offsetof(struct pt_regs, regs[8])},
+       { "r9", GDB_SIZEOF_REG, offsetof(struct pt_regs, regs[9])},
+       { "r10", GDB_SIZEOF_REG, offsetof(struct pt_regs, regs[10])},
+       { "r11", GDB_SIZEOF_REG, offsetof(struct pt_regs, regs[11])},
+       { "r12", GDB_SIZEOF_REG, offsetof(struct pt_regs, regs[12])},
+       { "r13", GDB_SIZEOF_REG, offsetof(struct pt_regs, regs[13])},
+       { "r14", GDB_SIZEOF_REG, offsetof(struct pt_regs, regs[14])},
+       { "r15", GDB_SIZEOF_REG, offsetof(struct pt_regs, regs[15])},
+       { "r16", GDB_SIZEOF_REG, offsetof(struct pt_regs, regs[16])},
+       { "r17", GDB_SIZEOF_REG, offsetof(struct pt_regs, regs[17])},
+       { "r18", GDB_SIZEOF_REG, offsetof(struct pt_regs, regs[18])},
+       { "r19", GDB_SIZEOF_REG, offsetof(struct pt_regs, regs[19])},
+       { "r20", GDB_SIZEOF_REG, offsetof(struct pt_regs, regs[20])},
+       { "r21", GDB_SIZEOF_REG, offsetof(struct pt_regs, regs[21])},
+       { "r22", GDB_SIZEOF_REG, offsetof(struct pt_regs, regs[22])},
+       { "r23", GDB_SIZEOF_REG, offsetof(struct pt_regs, regs[23])},
+       { "r24", GDB_SIZEOF_REG, offsetof(struct pt_regs, regs[24])},
+       { "r25", GDB_SIZEOF_REG, offsetof(struct pt_regs, regs[25])},
+       { "r26", GDB_SIZEOF_REG, offsetof(struct pt_regs, regs[26])},
+       { "r27", GDB_SIZEOF_REG, offsetof(struct pt_regs, regs[27])},
+       { "r28", GDB_SIZEOF_REG, offsetof(struct pt_regs, regs[28])},
+       { "r29", GDB_SIZEOF_REG, offsetof(struct pt_regs, regs[29])},
+       { "r30", GDB_SIZEOF_REG, offsetof(struct pt_regs, regs[30])},
+       { "r31", GDB_SIZEOF_REG, offsetof(struct pt_regs, regs[31])},
+       { "r32", GDB_SIZEOF_REG, offsetof(struct pt_regs, regs[32])},
+       { "r33", GDB_SIZEOF_REG, offsetof(struct pt_regs, regs[33])},
+       { "r34", GDB_SIZEOF_REG, offsetof(struct pt_regs, regs[34])},
+       { "r35", GDB_SIZEOF_REG, offsetof(struct pt_regs, regs[35])},
+       { "r36", GDB_SIZEOF_REG, offsetof(struct pt_regs, regs[36])},
+       { "r37", GDB_SIZEOF_REG, offsetof(struct pt_regs, regs[37])},
+       { "r38", GDB_SIZEOF_REG, offsetof(struct pt_regs, regs[38])},
+       { "r39", GDB_SIZEOF_REG, offsetof(struct pt_regs, regs[39])},
+       { "r40", GDB_SIZEOF_REG, offsetof(struct pt_regs, regs[40])},
+       { "r41", GDB_SIZEOF_REG, offsetof(struct pt_regs, regs[41])},
+       { "r42", GDB_SIZEOF_REG, offsetof(struct pt_regs, regs[42])},
+       { "r43", GDB_SIZEOF_REG, offsetof(struct pt_regs, regs[43])},
+       { "r44", GDB_SIZEOF_REG, offsetof(struct pt_regs, regs[44])},
+       { "r45", GDB_SIZEOF_REG, offsetof(struct pt_regs, regs[45])},
+       { "r46", GDB_SIZEOF_REG, offsetof(struct pt_regs, regs[46])},
+       { "r47", GDB_SIZEOF_REG, offsetof(struct pt_regs, regs[47])},
+       { "r48", GDB_SIZEOF_REG, offsetof(struct pt_regs, regs[48])},
+       { "r49", GDB_SIZEOF_REG, offsetof(struct pt_regs, regs[49])},
+       { "r50", GDB_SIZEOF_REG, offsetof(struct pt_regs, regs[50])},
+       { "r51", GDB_SIZEOF_REG, offsetof(struct pt_regs, regs[51])},
+       { "r52", GDB_SIZEOF_REG, offsetof(struct pt_regs, regs[52])},
+       { "tp", GDB_SIZEOF_REG, offsetof(struct pt_regs, tp)},
+       { "sp", GDB_SIZEOF_REG, offsetof(struct pt_regs, sp)},
+       { "lr", GDB_SIZEOF_REG, offsetof(struct pt_regs, lr)},
+       { "sn", GDB_SIZEOF_REG, -1},
+       { "idn0", GDB_SIZEOF_REG, -1},
+       { "idn1", GDB_SIZEOF_REG, -1},
+       { "udn0", GDB_SIZEOF_REG, -1},
+       { "udn1", GDB_SIZEOF_REG, -1},
+       { "udn2", GDB_SIZEOF_REG, -1},
+       { "udn3", GDB_SIZEOF_REG, -1},
+       { "zero", GDB_SIZEOF_REG, -1},
+       { "pc", GDB_SIZEOF_REG, offsetof(struct pt_regs, pc)},
+       { "faultnum", GDB_SIZEOF_REG, offsetof(struct pt_regs, faultnum)},
+};
+
+char *dbg_get_reg(int regno, void *mem, struct pt_regs *regs)
+{
+       if (regno >= DBG_MAX_REG_NUM || regno < 0)
+               return NULL;
+
+       if (dbg_reg_def[regno].offset != -1)
+               memcpy(mem, (void *)regs + dbg_reg_def[regno].offset,
+                      dbg_reg_def[regno].size);
+       else
+               memset(mem, 0, dbg_reg_def[regno].size);
+       return dbg_reg_def[regno].name;
+}
+
+int dbg_set_reg(int regno, void *mem, struct pt_regs *regs)
+{
+       if (regno >= DBG_MAX_REG_NUM || regno < 0)
+               return -EINVAL;
+
+       if (dbg_reg_def[regno].offset != -1)
+               memcpy((void *)regs + dbg_reg_def[regno].offset, mem,
+                      dbg_reg_def[regno].size);
+       return 0;
+}
+
+/*
+ * Similar to pt_regs_to_gdb_regs() except that process is sleeping and so
+ * we may not be able to get all the info.
+ */
+void
+sleeping_thread_to_gdb_regs(unsigned long *gdb_regs, struct task_struct *task)
+{
+       int reg;
+       struct pt_regs *thread_regs;
+       unsigned long *ptr = gdb_regs;
+
+       if (task == NULL)
+               return;
+
+       /* Initialize to zero. */
+       memset(gdb_regs, 0, NUMREGBYTES);
+
+       thread_regs = task_pt_regs(task);
+       for (reg = 0; reg <= TREG_LAST_GPR; reg++)
+               *(ptr++) = thread_regs->regs[reg];
+
+       gdb_regs[TILEGX_PC_REGNUM] = thread_regs->pc;
+       gdb_regs[TILEGX_FAULTNUM_REGNUM] = thread_regs->faultnum;
+}
+
+void kgdb_arch_set_pc(struct pt_regs *regs, unsigned long pc)
+{
+       regs->pc = pc;
+}
+
+static void kgdb_call_nmi_hook(void *ignored)
+{
+       kgdb_nmicallback(raw_smp_processor_id(), NULL);
+}
+
+void kgdb_roundup_cpus(unsigned long flags)
+{
+       local_irq_enable();
+       smp_call_function(kgdb_call_nmi_hook, NULL, 0);
+       local_irq_disable();
+}
+
+/*
+ * Convert a kernel address to the writable kernel text mapping.
+ */
+static unsigned long writable_address(unsigned long addr)
+{
+       unsigned long ret = 0;
+
+       if (core_kernel_text(addr))
+               ret = addr - MEM_SV_START + PAGE_OFFSET;
+       else if (is_module_text_address(addr))
+               ret = addr;
+       else
+               pr_err("Unknown virtual address 0x%lx\n", addr);
+
+       return ret;
+}
+
+/*
+ * Calculate the new address for after a step.
+ */
+static unsigned long get_step_address(struct pt_regs *regs)
+{
+       int src_reg;
+       int jump_off;
+       int br_off;
+       unsigned long addr;
+       unsigned int opcode;
+       tile_bundle_bits bundle;
+
+       /* Move to the next instruction by default. */
+       addr = regs->pc + TILEGX_BUNDLE_SIZE_IN_BYTES;
+       bundle = *(unsigned long *)instruction_pointer(regs);
+
+       /* 0: X mode, Otherwise: Y mode. */
+       if (bundle & TILEGX_BUNDLE_MODE_MASK) {
+               if (get_Opcode_Y1(bundle) == RRR_1_OPCODE_Y1 &&
+                   get_RRROpcodeExtension_Y1(bundle) ==
+                   UNARY_RRR_1_OPCODE_Y1) {
+                       opcode = get_UnaryOpcodeExtension_Y1(bundle);
+
+                       switch (opcode) {
+                       case JALR_UNARY_OPCODE_Y1:
+                       case JALRP_UNARY_OPCODE_Y1:
+                       case JR_UNARY_OPCODE_Y1:
+                       case JRP_UNARY_OPCODE_Y1:
+                               src_reg = get_SrcA_Y1(bundle);
+                               dbg_get_reg(src_reg, &addr, regs);
+                               break;
+                       }
+               }
+       } else if (get_Opcode_X1(bundle) == RRR_0_OPCODE_X1) {
+               if (get_RRROpcodeExtension_X1(bundle) ==
+                   UNARY_RRR_0_OPCODE_X1) {
+                       opcode = get_UnaryOpcodeExtension_X1(bundle);
+
+                       switch (opcode) {
+                       case JALR_UNARY_OPCODE_X1:
+                       case JALRP_UNARY_OPCODE_X1:
+                       case JR_UNARY_OPCODE_X1:
+                       case JRP_UNARY_OPCODE_X1:
+                               src_reg = get_SrcA_X1(bundle);
+                               dbg_get_reg(src_reg, &addr, regs);
+                               break;
+                       }
+               }
+       } else if (get_Opcode_X1(bundle) == JUMP_OPCODE_X1) {
+               opcode = get_JumpOpcodeExtension_X1(bundle);
+
+               switch (opcode) {
+               case JAL_JUMP_OPCODE_X1:
+               case J_JUMP_OPCODE_X1:
+                       jump_off = sign_extend(get_JumpOff_X1(bundle), 27);
+                       addr = regs->pc +
+                               (jump_off << TILEGX_LOG2_BUNDLE_SIZE_IN_BYTES);
+                       break;
+               }
+       } else if (get_Opcode_X1(bundle) == BRANCH_OPCODE_X1) {
+               br_off = 0;
+               opcode = get_BrType_X1(bundle);
+
+               switch (opcode) {
+               case BEQZT_BRANCH_OPCODE_X1:
+               case BEQZ_BRANCH_OPCODE_X1:
+                       if (get_SrcA_X1(bundle) == 0)
+                               br_off = get_BrOff_X1(bundle);
+                       break;
+               case BGEZT_BRANCH_OPCODE_X1:
+               case BGEZ_BRANCH_OPCODE_X1:
+                       if (get_SrcA_X1(bundle) >= 0)
+                               br_off = get_BrOff_X1(bundle);
+                       break;
+               case BGTZT_BRANCH_OPCODE_X1:
+               case BGTZ_BRANCH_OPCODE_X1:
+                       if (get_SrcA_X1(bundle) > 0)
+                               br_off = get_BrOff_X1(bundle);
+                       break;
+               case BLBCT_BRANCH_OPCODE_X1:
+               case BLBC_BRANCH_OPCODE_X1:
+                       if (!(get_SrcA_X1(bundle) & 1))
+                               br_off = get_BrOff_X1(bundle);
+                       break;
+               case BLBST_BRANCH_OPCODE_X1:
+               case BLBS_BRANCH_OPCODE_X1:
+                       if (get_SrcA_X1(bundle) & 1)
+                               br_off = get_BrOff_X1(bundle);
+                       break;
+               case BLEZT_BRANCH_OPCODE_X1:
+               case BLEZ_BRANCH_OPCODE_X1:
+                       if (get_SrcA_X1(bundle) <= 0)
+                               br_off = get_BrOff_X1(bundle);
+                       break;
+               case BLTZT_BRANCH_OPCODE_X1:
+               case BLTZ_BRANCH_OPCODE_X1:
+                       if (get_SrcA_X1(bundle) < 0)
+                               br_off = get_BrOff_X1(bundle);
+                       break;
+               case BNEZT_BRANCH_OPCODE_X1:
+               case BNEZ_BRANCH_OPCODE_X1:
+                       if (get_SrcA_X1(bundle) != 0)
+                               br_off = get_BrOff_X1(bundle);
+                       break;
+               }
+
+               if (br_off != 0) {
+                       br_off = sign_extend(br_off, 17);
+                       addr = regs->pc +
+                               (br_off << TILEGX_LOG2_BUNDLE_SIZE_IN_BYTES);
+               }
+       }
+
+       return addr;
+}
+
+/*
+ * Replace the next instruction after the current instruction with a
+ * breakpoint instruction.
+ */
+static void do_single_step(struct pt_regs *regs)
+{
+       unsigned long addr_wr;
+
+       /* Determine where the target instruction will send us to. */
+       stepped_addr = get_step_address(regs);
+       probe_kernel_read((char *)&stepped_instr, (char *)stepped_addr,
+                         BREAK_INSTR_SIZE);
+
+       addr_wr = writable_address(stepped_addr);
+       probe_kernel_write((char *)addr_wr, (char *)&singlestep_insn,
+                          BREAK_INSTR_SIZE);
+       smp_wmb();
+       flush_icache_range(stepped_addr, stepped_addr + BREAK_INSTR_SIZE);
+}
+
+static void undo_single_step(struct pt_regs *regs)
+{
+       unsigned long addr_wr;
+
+       if (stepped_instr == 0)
+               return;
+
+       addr_wr = writable_address(stepped_addr);
+       probe_kernel_write((char *)addr_wr, (char *)&stepped_instr,
+                          BREAK_INSTR_SIZE);
+       stepped_instr = 0;
+       smp_wmb();
+       flush_icache_range(stepped_addr, stepped_addr + BREAK_INSTR_SIZE);
+}
+
+/*
+ * Calls linux_debug_hook before the kernel dies. If KGDB is enabled,
+ * then try to fall into the debugger.
+ */
+static int
+kgdb_notify(struct notifier_block *self, unsigned long cmd, void *ptr)
+{
+       int ret;
+       unsigned long flags;
+       struct die_args *args = (struct die_args *)ptr;
+       struct pt_regs *regs = args->regs;
+
+#ifdef CONFIG_KPROBES
+       /*
+        * Return immediately if the kprobes fault notifier has set
+        * DIE_PAGE_FAULT.
+        */
+       if (cmd == DIE_PAGE_FAULT)
+               return NOTIFY_DONE;
+#endif /* CONFIG_KPROBES */
+
+       switch (cmd) {
+       case DIE_BREAK:
+       case DIE_COMPILED_BPT:
+               break;
+       case DIE_SSTEPBP:
+               local_irq_save(flags);
+               kgdb_handle_exception(0, SIGTRAP, 0, regs);
+               local_irq_restore(flags);
+               return NOTIFY_STOP;
+       default:
+               /* Userspace events, ignore. */
+               if (user_mode(regs))
+                       return NOTIFY_DONE;
+       }
+
+       local_irq_save(flags);
+       ret = kgdb_handle_exception(args->trapnr, args->signr, args->err, regs);
+       local_irq_restore(flags);
+       if (ret)
+               return NOTIFY_DONE;
+
+       return NOTIFY_STOP;
+}
+
+static struct notifier_block kgdb_notifier = {
+       .notifier_call = kgdb_notify,
+};
+
+/*
+ * kgdb_arch_handle_exception - Handle architecture specific GDB packets.
+ * @vector: The error vector of the exception that happened.
+ * @signo: The signal number of the exception that happened.
+ * @err_code: The error code of the exception that happened.
+ * @remcom_in_buffer: The buffer of the packet we have read.
+ * @remcom_out_buffer: The buffer of %BUFMAX bytes to write a packet into.
+ * @regs: The &struct pt_regs of the current process.
+ *
+ * This function MUST handle the 'c' and 's' command packets,
+ * as well packets to set / remove a hardware breakpoint, if used.
+ * If there are additional packets which the hardware needs to handle,
+ * they are handled here. The code should return -1 if it wants to
+ * process more packets, and a %0 or %1 if it wants to exit from the
+ * kgdb callback.
+ */
+int kgdb_arch_handle_exception(int vector, int signo, int err_code,
+                              char *remcom_in_buffer, char *remcom_out_buffer,
+                              struct pt_regs *regs)
+{
+       char *ptr;
+       unsigned long address;
+
+       /* Undo any stepping we may have done. */
+       undo_single_step(regs);
+
+       switch (remcom_in_buffer[0]) {
+       case 'c':
+       case 's':
+       case 'D':
+       case 'k':
+               /*
+                * Try to read optional parameter, pc unchanged if no parm.
+                * If this was a compiled-in breakpoint, we need to move
+                * to the next instruction or we will just breakpoint
+                * over and over again.
+                */
+               ptr = &remcom_in_buffer[1];
+               if (kgdb_hex2long(&ptr, &address))
+                       regs->pc = address;
+               else if (*(unsigned long *)regs->pc == compiled_bpt)
+                       regs->pc += BREAK_INSTR_SIZE;
+
+               if (remcom_in_buffer[0] == 's') {
+                       do_single_step(regs);
+                       kgdb_single_step = 1;
+                       atomic_set(&kgdb_cpu_doing_single_step,
+                                  raw_smp_processor_id());
+               } else
+                       atomic_set(&kgdb_cpu_doing_single_step, -1);
+
+               return 0;
+       }
+
+       return -1; /* this means that we do not want to exit from the handler */
+}
+
+struct kgdb_arch arch_kgdb_ops;
+
+/*
+ * kgdb_arch_init - Perform any architecture specific initalization.
+ *
+ * This function will handle the initalization of any architecture
+ * specific callbacks.
+ */
+int kgdb_arch_init(void)
+{
+       tile_bundle_bits bundle = TILEGX_BPT_BUNDLE;
+
+       memcpy(arch_kgdb_ops.gdb_bpt_instr, &bundle, BREAK_INSTR_SIZE);
+       return register_die_notifier(&kgdb_notifier);
+}
+
+/*
+ * kgdb_arch_exit - Perform any architecture specific uninitalization.
+ *
+ * This function will handle the uninitalization of any architecture
+ * specific callbacks, for dynamic registration and unregistration.
+ */
+void kgdb_arch_exit(void)
+{
+       unregister_die_notifier(&kgdb_notifier);
+}
+
+int kgdb_arch_set_breakpoint(struct kgdb_bkpt *bpt)
+{
+       int err;
+       unsigned long addr_wr = writable_address(bpt->bpt_addr);
+
+       if (addr_wr == 0)
+               return -1;
+
+       err = probe_kernel_read(bpt->saved_instr, (char *)bpt->bpt_addr,
+                               BREAK_INSTR_SIZE);
+       if (err)
+               return err;
+
+       err = probe_kernel_write((char *)addr_wr, arch_kgdb_ops.gdb_bpt_instr,
+                                BREAK_INSTR_SIZE);
+       smp_wmb();
+       flush_icache_range((unsigned long)bpt->bpt_addr,
+                          (unsigned long)bpt->bpt_addr + BREAK_INSTR_SIZE);
+       return err;
+}
+
+int kgdb_arch_remove_breakpoint(struct kgdb_bkpt *bpt)
+{
+       int err;
+       unsigned long addr_wr = writable_address(bpt->bpt_addr);
+
+       if (addr_wr == 0)
+               return -1;
+
+       err = probe_kernel_write((char *)addr_wr, (char *)bpt->saved_instr,
+                                BREAK_INSTR_SIZE);
+       smp_wmb();
+       flush_icache_range((unsigned long)bpt->bpt_addr,
+                          (unsigned long)bpt->bpt_addr + BREAK_INSTR_SIZE);
+       return err;
+}
index 19d465c3f37970eed0b12b14cab981598e33745e..ef95de28db32b5429e2053d5c2822f73cf98c2df 100644 (file)
@@ -240,6 +240,11 @@ static int do_bpt(struct pt_regs *regs)
                notify_die(DIE_BREAK, "debug", regs, bundle,
                        INT_ILL, SIGTRAP);
                break;
+       /* compiled_bpt */
+       case DIE_COMPILED_BPT:
+               notify_die(DIE_COMPILED_BPT, "debug", regs, bundle,
+                       INT_ILL, SIGTRAP);
+               break;
        /* breakpoint2_insn */
        case DIE_SSTEPBP:
                notify_die(DIE_SSTEPBP, "single_step", regs, bundle,