Merge branch 'x86/pebs' into x86/unify-cpu-detect
Conflicts:
arch/x86/Kconfig.cpu
include/asm-x86/ds.h
Signed-off-by: Ingo Molnar <mingo@elte.hu>
diff --git a/arch/x86/Kconfig.cpu b/arch/x86/Kconfig.cpu
index ab77d40..1b29d6a 100644
--- a/arch/x86/Kconfig.cpu
+++ b/arch/x86/Kconfig.cpu
@@ -477,3 +477,21 @@
depends on !64BIT
help
This enables extended support for UMC processors
+
+config X86_DS
+ bool "Debug Store support"
+ default y
+ help
+ Add support for Debug Store.
+ This allows the kernel to provide a memory buffer to the hardware
+ to store various profiling and tracing events.
+
+config X86_PTRACE_BTS
+ bool "ptrace interface to Branch Trace Store"
+ default y
+ depends on (X86_DS && X86_DEBUGCTLMSR)
+ help
+ Add a ptrace interface to allow collecting an execution trace
+ of the traced task.
+ This collects control flow changes in a (cyclic) buffer and allows
+ debuggers to fill in the gaps and show an execution trace of the debuggee.
diff --git a/arch/x86/kernel/cpu/intel.c b/arch/x86/kernel/cpu/intel.c
index 4a8ac9d..a669895 100644
--- a/arch/x86/kernel/cpu/intel.c
+++ b/arch/x86/kernel/cpu/intel.c
@@ -221,10 +221,11 @@
set_cpu_cap(c, X86_FEATURE_BTS);
if (!(l1 & (1<<12)))
set_cpu_cap(c, X86_FEATURE_PEBS);
+ ds_init_intel(c);
}
if (cpu_has_bts)
- ds_init_intel(c);
+ ptrace_bts_init_intel(c);
/*
* See if we have a good local APIC by checking for buggy Pentia,
diff --git a/arch/x86/kernel/ds.c b/arch/x86/kernel/ds.c
index 11c11b8..ab21c27 100644
--- a/arch/x86/kernel/ds.c
+++ b/arch/x86/kernel/ds.c
@@ -2,26 +2,49 @@
* Debug Store support
*
* This provides a low-level interface to the hardware's Debug Store
- * feature that is used for last branch recording (LBR) and
+ * feature that is used for branch trace store (BTS) and
* precise-event based sampling (PEBS).
*
- * Different architectures use a different DS layout/pointer size.
- * The below functions therefore work on a void*.
+ * It manages:
+ * - per-thread and per-cpu allocation of BTS and PEBS
+ * - buffer memory allocation (optional)
+ * - buffer overflow handling
+ * - buffer access
+ *
+ * It assumes:
+ * - get_task_struct on all parameter tasks
+ * - current is allowed to trace parameter tasks
*
*
- * Since there is no user for PEBS, yet, only LBR (or branch
- * trace store, BTS) is supported.
- *
- *
- * Copyright (C) 2007 Intel Corporation.
- * Markus Metzger <markus.t.metzger@intel.com>, Dec 2007
+ * Copyright (C) 2007-2008 Intel Corporation.
+ * Markus Metzger <markus.t.metzger@intel.com>, 2007-2008
*/
+
+#ifdef CONFIG_X86_DS
+
#include <asm/ds.h>
#include <linux/errno.h>
#include <linux/string.h>
#include <linux/slab.h>
+#include <linux/sched.h>
+#include <linux/mm.h>
+
+
+/*
+ * The configuration for a particular DS hardware implementation.
+ */
+struct ds_configuration {
+ /* the size of the DS structure in bytes */
+ unsigned char sizeof_ds;
+ /* the size of one pointer-typed field in the DS structure in bytes;
+ this covers the first 8 fields related to buffer management. */
+ unsigned char sizeof_field;
+ /* the size of a BTS/PEBS record in bytes */
+ unsigned char sizeof_rec[2];
+};
+static struct ds_configuration ds_cfg;
/*
@@ -44,378 +67,747 @@
* (interrupt occurs when write pointer passes interrupt pointer)
* - value to which counter is reset following counter overflow
*
- * On later architectures, the last branch recording hardware uses
- * 64bit pointers even in 32bit mode.
+ * Later architectures use 64bit pointers throughout, whereas earlier
+ * architectures use 32bit pointers in 32bit mode.
*
*
- * Branch Trace Store (BTS) records store information about control
- * flow changes. They at least provide the following information:
- * - source linear address
- * - destination linear address
+ * We compute the base address for the first 8 fields based on:
+ * - the field size stored in the DS configuration
+ * - the relative field position
+ * - an offset giving the start of the respective region
*
- * Netburst supported a predicated bit that had been dropped in later
- * architectures. We do not suppor it.
+ * This offset is further used to index various arrays holding
+ * information for BTS and PEBS at the respective index.
*
- *
- * In order to abstract from the actual DS and BTS layout, we describe
- * the access to the relevant fields.
- * Thanks to Andi Kleen for proposing this design.
- *
- * The implementation, however, is not as general as it might seem. In
- * order to stay somewhat simple and efficient, we assume an
- * underlying unsigned type (mostly a pointer type) and we expect the
- * field to be at least as big as that type.
+ * On later 32bit processors, we only access the lower 32bit of the
+ * 64bit pointer fields. The upper halves will be zeroed out.
*/
-/*
- * A special from_ip address to indicate that the BTS record is an
- * info record that needs to be interpreted or skipped.
- */
-#define BTS_ESCAPE_ADDRESS (-1)
-
-/*
- * A field access descriptor
- */
-struct access_desc {
- unsigned char offset;
- unsigned char size;
+enum ds_field {
+ ds_buffer_base = 0,
+ ds_index,
+ ds_absolute_maximum,
+ ds_interrupt_threshold,
};
-/*
- * The configuration for a particular DS/BTS hardware implementation.
- */
-struct ds_configuration {
- /* the DS configuration */
- unsigned char sizeof_ds;
- struct access_desc bts_buffer_base;
- struct access_desc bts_index;
- struct access_desc bts_absolute_maximum;
- struct access_desc bts_interrupt_threshold;
- /* the BTS configuration */
- unsigned char sizeof_bts;
- struct access_desc from_ip;
- struct access_desc to_ip;
- /* BTS variants used to store additional information like
- timestamps */
- struct access_desc info_type;
- struct access_desc info_data;
- unsigned long debugctl_mask;
+enum ds_qualifier {
+ ds_bts = 0,
+ ds_pebs
};
-/*
- * The global configuration used by the below accessor functions
- */
-static struct ds_configuration ds_cfg;
+static inline unsigned long ds_get(const unsigned char *base,
+ enum ds_qualifier qual, enum ds_field field)
+{
+ base += (ds_cfg.sizeof_field * (field + (4 * qual)));
+ return *(unsigned long *)base;
+}
+
+static inline void ds_set(unsigned char *base, enum ds_qualifier qual,
+ enum ds_field field, unsigned long value)
+{
+ base += (ds_cfg.sizeof_field * (field + (4 * qual)));
+ (*(unsigned long *)base) = value;
+}
+
/*
- * Accessor functions for some DS and BTS fields using the above
- * global ptrace_bts_cfg.
+ * Locking is done only for allocating BTS or PEBS resources and for
+ * guarding context and buffer memory allocation.
+ *
+ * Most functions require the current task to own the ds context part
+ * they are going to access. All the locking is done when validating
+ * access to the context.
*/
-static inline unsigned long get_bts_buffer_base(char *base)
-{
- return *(unsigned long *)(base + ds_cfg.bts_buffer_base.offset);
-}
-static inline void set_bts_buffer_base(char *base, unsigned long value)
-{
- (*(unsigned long *)(base + ds_cfg.bts_buffer_base.offset)) = value;
-}
-static inline unsigned long get_bts_index(char *base)
-{
- return *(unsigned long *)(base + ds_cfg.bts_index.offset);
-}
-static inline void set_bts_index(char *base, unsigned long value)
-{
- (*(unsigned long *)(base + ds_cfg.bts_index.offset)) = value;
-}
-static inline unsigned long get_bts_absolute_maximum(char *base)
-{
- return *(unsigned long *)(base + ds_cfg.bts_absolute_maximum.offset);
-}
-static inline void set_bts_absolute_maximum(char *base, unsigned long value)
-{
- (*(unsigned long *)(base + ds_cfg.bts_absolute_maximum.offset)) = value;
-}
-static inline unsigned long get_bts_interrupt_threshold(char *base)
-{
- return *(unsigned long *)(base + ds_cfg.bts_interrupt_threshold.offset);
-}
-static inline void set_bts_interrupt_threshold(char *base, unsigned long value)
-{
- (*(unsigned long *)(base + ds_cfg.bts_interrupt_threshold.offset)) = value;
-}
-static inline unsigned long get_from_ip(char *base)
-{
- return *(unsigned long *)(base + ds_cfg.from_ip.offset);
-}
-static inline void set_from_ip(char *base, unsigned long value)
-{
- (*(unsigned long *)(base + ds_cfg.from_ip.offset)) = value;
-}
-static inline unsigned long get_to_ip(char *base)
-{
- return *(unsigned long *)(base + ds_cfg.to_ip.offset);
-}
-static inline void set_to_ip(char *base, unsigned long value)
-{
- (*(unsigned long *)(base + ds_cfg.to_ip.offset)) = value;
-}
-static inline unsigned char get_info_type(char *base)
-{
- return *(unsigned char *)(base + ds_cfg.info_type.offset);
-}
-static inline void set_info_type(char *base, unsigned char value)
-{
- (*(unsigned char *)(base + ds_cfg.info_type.offset)) = value;
-}
-static inline unsigned long get_info_data(char *base)
-{
- return *(unsigned long *)(base + ds_cfg.info_data.offset);
-}
-static inline void set_info_data(char *base, unsigned long value)
-{
- (*(unsigned long *)(base + ds_cfg.info_data.offset)) = value;
-}
+static spinlock_t ds_lock = __SPIN_LOCK_UNLOCKED(ds_lock);
-
-int ds_allocate(void **dsp, size_t bts_size_in_bytes)
+/*
+ * Validate that the current task is allowed to access the BTS/PEBS
+ * buffer of the parameter task.
+ *
+ * Returns 0, if access is granted; -Eerrno, otherwise.
+ */
+static inline int ds_validate_access(struct ds_context *context,
+ enum ds_qualifier qual)
{
- size_t bts_size_in_records;
- unsigned long bts;
- void *ds;
+ if (!context)
+ return -EPERM;
- if (!ds_cfg.sizeof_ds || !ds_cfg.sizeof_bts)
- return -EOPNOTSUPP;
-
- if (bts_size_in_bytes < 0)
- return -EINVAL;
-
- bts_size_in_records =
- bts_size_in_bytes / ds_cfg.sizeof_bts;
- bts_size_in_bytes =
- bts_size_in_records * ds_cfg.sizeof_bts;
-
- if (bts_size_in_bytes <= 0)
- return -EINVAL;
-
- bts = (unsigned long)kzalloc(bts_size_in_bytes, GFP_KERNEL);
-
- if (!bts)
- return -ENOMEM;
-
- ds = kzalloc(ds_cfg.sizeof_ds, GFP_KERNEL);
-
- if (!ds) {
- kfree((void *)bts);
- return -ENOMEM;
- }
-
- set_bts_buffer_base(ds, bts);
- set_bts_index(ds, bts);
- set_bts_absolute_maximum(ds, bts + bts_size_in_bytes);
- set_bts_interrupt_threshold(ds, bts + bts_size_in_bytes + 1);
-
- *dsp = ds;
- return 0;
-}
-
-int ds_free(void **dsp)
-{
- if (*dsp) {
- kfree((void *)get_bts_buffer_base(*dsp));
- kfree(*dsp);
- *dsp = NULL;
- }
- return 0;
-}
-
-int ds_get_bts_size(void *ds)
-{
- int size_in_bytes;
-
- if (!ds_cfg.sizeof_ds || !ds_cfg.sizeof_bts)
- return -EOPNOTSUPP;
-
- if (!ds)
+ if (context->owner[qual] == current)
return 0;
- size_in_bytes =
- get_bts_absolute_maximum(ds) -
- get_bts_buffer_base(ds);
- return size_in_bytes;
+ return -EPERM;
}
-int ds_get_bts_end(void *ds)
+
+/*
+ * We either support (system-wide) per-cpu or per-thread allocation.
+ * We distinguish the two based on the task_struct pointer, where a
+ * NULL pointer indicates per-cpu allocation for the current cpu.
+ *
+ * Allocations are use-counted. As soon as resources are allocated,
+ * further allocations must be of the same type (per-cpu or
+ * per-thread). We model this by counting allocations (i.e. the number
+ * of tracers of a certain type) for one type negatively:
+ * =0 no tracers
+ * >0 number of per-thread tracers
+ * <0 number of per-cpu tracers
+ *
+ * The below functions to get and put tracers and to check the
+ * allocation type require the ds_lock to be held by the caller.
+ *
+ * Tracers essentially gives the number of ds contexts for a certain
+ * type of allocation.
+ */
+static long tracers;
+
+static inline void get_tracer(struct task_struct *task)
{
- int size_in_bytes = ds_get_bts_size(ds);
-
- if (size_in_bytes <= 0)
- return size_in_bytes;
-
- return size_in_bytes / ds_cfg.sizeof_bts;
+ tracers += (task ? 1 : -1);
}
-int ds_get_bts_index(void *ds)
+static inline void put_tracer(struct task_struct *task)
{
- int index_offset_in_bytes;
-
- if (!ds_cfg.sizeof_ds || !ds_cfg.sizeof_bts)
- return -EOPNOTSUPP;
-
- index_offset_in_bytes =
- get_bts_index(ds) -
- get_bts_buffer_base(ds);
-
- return index_offset_in_bytes / ds_cfg.sizeof_bts;
+ tracers -= (task ? 1 : -1);
}
-int ds_set_overflow(void *ds, int method)
+static inline int check_tracer(struct task_struct *task)
{
- switch (method) {
- case DS_O_SIGNAL:
- return -EOPNOTSUPP;
- case DS_O_WRAP:
- return 0;
- default:
- return -EINVAL;
- }
+ return (task ? (tracers >= 0) : (tracers <= 0));
}
-int ds_get_overflow(void *ds)
+
+/*
+ * The DS context is either attached to a thread or to a cpu:
+ * - in the former case, the thread_struct contains a pointer to the
+ * attached context.
+ * - in the latter case, we use a static array of per-cpu context
+ * pointers.
+ *
+ * Contexts are use-counted. They are allocated on first access and
+ * deallocated when the last user puts the context.
+ *
+ * We distinguish between an allocating and a non-allocating get of a
+ * context:
+ * - the allocating get is used for requesting BTS/PEBS resources. It
+ * requires the caller to hold the global ds_lock.
+ * - the non-allocating get is used for all other cases. A
+ * non-existing context indicates an error. It acquires and releases
+ * the ds_lock itself for obtaining the context.
+ *
+ * A context and its DS configuration are allocated and deallocated
+ * together. A context always has a DS configuration of the
+ * appropriate size.
+ */
+static DEFINE_PER_CPU(struct ds_context *, system_context);
+
+#define this_system_context per_cpu(system_context, smp_processor_id())
+
+/*
+ * Returns the pointer to the parameter task's context or to the
+ * system-wide context, if task is NULL.
+ *
+ * Increases the use count of the returned context, if not NULL.
+ */
+static inline struct ds_context *ds_get_context(struct task_struct *task)
{
- return DS_O_WRAP;
+ struct ds_context *context;
+
+ spin_lock(&ds_lock);
+
+ context = (task ? task->thread.ds_ctx : this_system_context);
+ if (context)
+ context->count++;
+
+ spin_unlock(&ds_lock);
+
+ return context;
}
-int ds_clear(void *ds)
+/*
+ * Same as ds_get_context, but allocates the context and it's DS
+ * structure, if necessary; returns NULL; if out of memory.
+ *
+ * pre: requires ds_lock to be held
+ */
+static inline struct ds_context *ds_alloc_context(struct task_struct *task)
{
- int bts_size = ds_get_bts_size(ds);
- unsigned long bts_base;
+ struct ds_context **p_context =
+ (task ? &task->thread.ds_ctx : &this_system_context);
+ struct ds_context *context = *p_context;
- if (bts_size <= 0)
- return bts_size;
+ if (!context) {
+ context = kzalloc(sizeof(*context), GFP_KERNEL);
- bts_base = get_bts_buffer_base(ds);
- memset((void *)bts_base, 0, bts_size);
+ if (!context)
+ return NULL;
- set_bts_index(ds, bts_base);
- return 0;
-}
+ context->ds = kzalloc(ds_cfg.sizeof_ds, GFP_KERNEL);
+ if (!context->ds) {
+ kfree(context);
+ return NULL;
+ }
-int ds_read_bts(void *ds, int index, struct bts_struct *out)
-{
- void *bts;
+ *p_context = context;
- if (!ds_cfg.sizeof_ds || !ds_cfg.sizeof_bts)
- return -EOPNOTSUPP;
+ context->this = p_context;
+ context->task = task;
- if (index < 0)
- return -EINVAL;
+ if (task)
+ set_tsk_thread_flag(task, TIF_DS_AREA_MSR);
- if (index >= ds_get_bts_size(ds))
- return -EINVAL;
+ if (!task || (task == current))
+ wrmsr(MSR_IA32_DS_AREA, (unsigned long)context->ds, 0);
- bts = (void *)(get_bts_buffer_base(ds) + (index * ds_cfg.sizeof_bts));
-
- memset(out, 0, sizeof(*out));
- if (get_from_ip(bts) == BTS_ESCAPE_ADDRESS) {
- out->qualifier = get_info_type(bts);
- out->variant.jiffies = get_info_data(bts);
- } else {
- out->qualifier = BTS_BRANCH;
- out->variant.lbr.from_ip = get_from_ip(bts);
- out->variant.lbr.to_ip = get_to_ip(bts);
+ get_tracer(task);
}
- return sizeof(*out);;
+ context->count++;
+
+ return context;
}
-int ds_write_bts(void *ds, const struct bts_struct *in)
+/*
+ * Decreases the use count of the parameter context, if not NULL.
+ * Deallocates the context, if the use count reaches zero.
+ */
+static inline void ds_put_context(struct ds_context *context)
{
- unsigned long bts;
+ if (!context)
+ return;
- if (!ds_cfg.sizeof_ds || !ds_cfg.sizeof_bts)
+ spin_lock(&ds_lock);
+
+ if (--context->count)
+ goto out;
+
+ *(context->this) = NULL;
+
+ if (context->task)
+ clear_tsk_thread_flag(context->task, TIF_DS_AREA_MSR);
+
+ if (!context->task || (context->task == current))
+ wrmsrl(MSR_IA32_DS_AREA, 0);
+
+ put_tracer(context->task);
+
+ /* free any leftover buffers from tracers that did not
+ * deallocate them properly. */
+ kfree(context->buffer[ds_bts]);
+ kfree(context->buffer[ds_pebs]);
+ kfree(context->ds);
+ kfree(context);
+ out:
+ spin_unlock(&ds_lock);
+}
+
+
+/*
+ * Handle a buffer overflow
+ *
+ * task: the task whose buffers are overflowing;
+ * NULL for a buffer overflow on the current cpu
+ * context: the ds context
+ * qual: the buffer type
+ */
+static void ds_overflow(struct task_struct *task, struct ds_context *context,
+ enum ds_qualifier qual)
+{
+ if (!context)
+ return;
+
+ if (context->callback[qual])
+ (*context->callback[qual])(task);
+
+ /* todo: do some more overflow handling */
+}
+
+
+/*
+ * Allocate a non-pageable buffer of the parameter size.
+ * Checks the memory and the locked memory rlimit.
+ *
+ * Returns the buffer, if successful;
+ * NULL, if out of memory or rlimit exceeded.
+ *
+ * size: the requested buffer size in bytes
+ * pages (out): if not NULL, contains the number of pages reserved
+ */
+static inline void *ds_allocate_buffer(size_t size, unsigned int *pages)
+{
+ unsigned long rlim, vm, pgsz;
+ void *buffer;
+
+ pgsz = PAGE_ALIGN(size) >> PAGE_SHIFT;
+
+ rlim = current->signal->rlim[RLIMIT_AS].rlim_cur >> PAGE_SHIFT;
+ vm = current->mm->total_vm + pgsz;
+ if (rlim < vm)
+ return NULL;
+
+ rlim = current->signal->rlim[RLIMIT_MEMLOCK].rlim_cur >> PAGE_SHIFT;
+ vm = current->mm->locked_vm + pgsz;
+ if (rlim < vm)
+ return NULL;
+
+ buffer = kzalloc(size, GFP_KERNEL);
+ if (!buffer)
+ return NULL;
+
+ current->mm->total_vm += pgsz;
+ current->mm->locked_vm += pgsz;
+
+ if (pages)
+ *pages = pgsz;
+
+ return buffer;
+}
+
+static int ds_request(struct task_struct *task, void *base, size_t size,
+ ds_ovfl_callback_t ovfl, enum ds_qualifier qual)
+{
+ struct ds_context *context;
+ unsigned long buffer, adj;
+ const unsigned long alignment = (1 << 3);
+ int error = 0;
+
+ if (!ds_cfg.sizeof_ds)
return -EOPNOTSUPP;
- if (ds_get_bts_size(ds) <= 0)
- return -ENXIO;
-
- bts = get_bts_index(ds);
-
- memset((void *)bts, 0, ds_cfg.sizeof_bts);
- switch (in->qualifier) {
- case BTS_INVALID:
- break;
-
- case BTS_BRANCH:
- set_from_ip((void *)bts, in->variant.lbr.from_ip);
- set_to_ip((void *)bts, in->variant.lbr.to_ip);
- break;
-
- case BTS_TASK_ARRIVES:
- case BTS_TASK_DEPARTS:
- set_from_ip((void *)bts, BTS_ESCAPE_ADDRESS);
- set_info_type((void *)bts, in->qualifier);
- set_info_data((void *)bts, in->variant.jiffies);
- break;
-
- default:
+ /* we require some space to do alignment adjustments below */
+ if (size < (alignment + ds_cfg.sizeof_rec[qual]))
return -EINVAL;
+
+ /* buffer overflow notification is not yet implemented */
+ if (ovfl)
+ return -EOPNOTSUPP;
+
+
+ spin_lock(&ds_lock);
+
+ if (!check_tracer(task))
+ return -EPERM;
+
+ error = -ENOMEM;
+ context = ds_alloc_context(task);
+ if (!context)
+ goto out_unlock;
+
+ error = -EALREADY;
+ if (context->owner[qual] == current)
+ goto out_unlock;
+ error = -EPERM;
+ if (context->owner[qual] != NULL)
+ goto out_unlock;
+ context->owner[qual] = current;
+
+ spin_unlock(&ds_lock);
+
+
+ error = -ENOMEM;
+ if (!base) {
+ base = ds_allocate_buffer(size, &context->pages[qual]);
+ if (!base)
+ goto out_release;
+
+ context->buffer[qual] = base;
+ }
+ error = 0;
+
+ context->callback[qual] = ovfl;
+
+ /* adjust the buffer address and size to meet alignment
+ * constraints:
+ * - buffer is double-word aligned
+ * - size is multiple of record size
+ *
+ * We checked the size at the very beginning; we have enough
+ * space to do the adjustment.
+ */
+ buffer = (unsigned long)base;
+
+ adj = ALIGN(buffer, alignment) - buffer;
+ buffer += adj;
+ size -= adj;
+
+ size /= ds_cfg.sizeof_rec[qual];
+ size *= ds_cfg.sizeof_rec[qual];
+
+ ds_set(context->ds, qual, ds_buffer_base, buffer);
+ ds_set(context->ds, qual, ds_index, buffer);
+ ds_set(context->ds, qual, ds_absolute_maximum, buffer + size);
+
+ if (ovfl) {
+ /* todo: select a suitable interrupt threshold */
+ } else
+ ds_set(context->ds, qual,
+ ds_interrupt_threshold, buffer + size + 1);
+
+ /* we keep the context until ds_release */
+ return error;
+
+ out_release:
+ context->owner[qual] = NULL;
+ ds_put_context(context);
+ return error;
+
+ out_unlock:
+ spin_unlock(&ds_lock);
+ ds_put_context(context);
+ return error;
+}
+
+int ds_request_bts(struct task_struct *task, void *base, size_t size,
+ ds_ovfl_callback_t ovfl)
+{
+ return ds_request(task, base, size, ovfl, ds_bts);
+}
+
+int ds_request_pebs(struct task_struct *task, void *base, size_t size,
+ ds_ovfl_callback_t ovfl)
+{
+ return ds_request(task, base, size, ovfl, ds_pebs);
+}
+
+static int ds_release(struct task_struct *task, enum ds_qualifier qual)
+{
+ struct ds_context *context;
+ int error;
+
+ context = ds_get_context(task);
+ error = ds_validate_access(context, qual);
+ if (error < 0)
+ goto out;
+
+ kfree(context->buffer[qual]);
+ context->buffer[qual] = 0;
+
+ current->mm->total_vm -= context->pages[qual];
+ current->mm->locked_vm -= context->pages[qual];
+ context->pages[qual] = 0;
+ context->owner[qual] = 0;
+
+ /*
+ * we put the context twice:
+ * once for the ds_get_context
+ * once for the corresponding ds_request
+ */
+ ds_put_context(context);
+ out:
+ ds_put_context(context);
+ return error;
+}
+
+int ds_release_bts(struct task_struct *task)
+{
+ return ds_release(task, ds_bts);
+}
+
+int ds_release_pebs(struct task_struct *task)
+{
+ return ds_release(task, ds_pebs);
+}
+
+static int ds_get_index(struct task_struct *task, size_t *pos,
+ enum ds_qualifier qual)
+{
+ struct ds_context *context;
+ unsigned long base, index;
+ int error;
+
+ context = ds_get_context(task);
+ error = ds_validate_access(context, qual);
+ if (error < 0)
+ goto out;
+
+ base = ds_get(context->ds, qual, ds_buffer_base);
+ index = ds_get(context->ds, qual, ds_index);
+
+ error = ((index - base) / ds_cfg.sizeof_rec[qual]);
+ if (pos)
+ *pos = error;
+ out:
+ ds_put_context(context);
+ return error;
+}
+
+int ds_get_bts_index(struct task_struct *task, size_t *pos)
+{
+ return ds_get_index(task, pos, ds_bts);
+}
+
+int ds_get_pebs_index(struct task_struct *task, size_t *pos)
+{
+ return ds_get_index(task, pos, ds_pebs);
+}
+
+static int ds_get_end(struct task_struct *task, size_t *pos,
+ enum ds_qualifier qual)
+{
+ struct ds_context *context;
+ unsigned long base, end;
+ int error;
+
+ context = ds_get_context(task);
+ error = ds_validate_access(context, qual);
+ if (error < 0)
+ goto out;
+
+ base = ds_get(context->ds, qual, ds_buffer_base);
+ end = ds_get(context->ds, qual, ds_absolute_maximum);
+
+ error = ((end - base) / ds_cfg.sizeof_rec[qual]);
+ if (pos)
+ *pos = error;
+ out:
+ ds_put_context(context);
+ return error;
+}
+
+int ds_get_bts_end(struct task_struct *task, size_t *pos)
+{
+ return ds_get_end(task, pos, ds_bts);
+}
+
+int ds_get_pebs_end(struct task_struct *task, size_t *pos)
+{
+ return ds_get_end(task, pos, ds_pebs);
+}
+
+static int ds_access(struct task_struct *task, size_t index,
+ const void **record, enum ds_qualifier qual)
+{
+ struct ds_context *context;
+ unsigned long base, idx;
+ int error;
+
+ if (!record)
+ return -EINVAL;
+
+ context = ds_get_context(task);
+ error = ds_validate_access(context, qual);
+ if (error < 0)
+ goto out;
+
+ base = ds_get(context->ds, qual, ds_buffer_base);
+ idx = base + (index * ds_cfg.sizeof_rec[qual]);
+
+ error = -EINVAL;
+ if (idx > ds_get(context->ds, qual, ds_absolute_maximum))
+ goto out;
+
+ *record = (const void *)idx;
+ error = ds_cfg.sizeof_rec[qual];
+ out:
+ ds_put_context(context);
+ return error;
+}
+
+int ds_access_bts(struct task_struct *task, size_t index, const void **record)
+{
+ return ds_access(task, index, record, ds_bts);
+}
+
+int ds_access_pebs(struct task_struct *task, size_t index, const void **record)
+{
+ return ds_access(task, index, record, ds_pebs);
+}
+
+static int ds_write(struct task_struct *task, const void *record, size_t size,
+ enum ds_qualifier qual, int force)
+{
+ struct ds_context *context;
+ int error;
+
+ if (!record)
+ return -EINVAL;
+
+ error = -EPERM;
+ context = ds_get_context(task);
+ if (!context)
+ goto out;
+
+ if (!force) {
+ error = ds_validate_access(context, qual);
+ if (error < 0)
+ goto out;
}
- bts = bts + ds_cfg.sizeof_bts;
- if (bts >= get_bts_absolute_maximum(ds))
- bts = get_bts_buffer_base(ds);
- set_bts_index(ds, bts);
+ error = 0;
+ while (size) {
+ unsigned long base, index, end, write_end, int_th;
+ unsigned long write_size, adj_write_size;
- return ds_cfg.sizeof_bts;
+ /*
+ * write as much as possible without producing an
+ * overflow interrupt.
+ *
+ * interrupt_threshold must either be
+ * - bigger than absolute_maximum or
+ * - point to a record between buffer_base and absolute_maximum
+ *
+ * index points to a valid record.
+ */
+ base = ds_get(context->ds, qual, ds_buffer_base);
+ index = ds_get(context->ds, qual, ds_index);
+ end = ds_get(context->ds, qual, ds_absolute_maximum);
+ int_th = ds_get(context->ds, qual, ds_interrupt_threshold);
+
+ write_end = min(end, int_th);
+
+ /* if we are already beyond the interrupt threshold,
+ * we fill the entire buffer */
+ if (write_end <= index)
+ write_end = end;
+
+ if (write_end <= index)
+ goto out;
+
+ write_size = min((unsigned long) size, write_end - index);
+ memcpy((void *)index, record, write_size);
+
+ record = (const char *)record + write_size;
+ size -= write_size;
+ error += write_size;
+
+ adj_write_size = write_size / ds_cfg.sizeof_rec[qual];
+ adj_write_size *= ds_cfg.sizeof_rec[qual];
+
+ /* zero out trailing bytes */
+ memset((char *)index + write_size, 0,
+ adj_write_size - write_size);
+ index += adj_write_size;
+
+ if (index >= end)
+ index = base;
+ ds_set(context->ds, qual, ds_index, index);
+
+ if (index >= int_th)
+ ds_overflow(task, context, qual);
+ }
+
+ out:
+ ds_put_context(context);
+ return error;
}
-unsigned long ds_debugctl_mask(void)
+int ds_write_bts(struct task_struct *task, const void *record, size_t size)
{
- return ds_cfg.debugctl_mask;
+ return ds_write(task, record, size, ds_bts, /* force = */ 0);
}
-#ifdef __i386__
-static const struct ds_configuration ds_cfg_netburst = {
- .sizeof_ds = 9 * 4,
- .bts_buffer_base = { 0, 4 },
- .bts_index = { 4, 4 },
- .bts_absolute_maximum = { 8, 4 },
- .bts_interrupt_threshold = { 12, 4 },
- .sizeof_bts = 3 * 4,
- .from_ip = { 0, 4 },
- .to_ip = { 4, 4 },
- .info_type = { 4, 1 },
- .info_data = { 8, 4 },
- .debugctl_mask = (1<<2)|(1<<3)
+int ds_write_pebs(struct task_struct *task, const void *record, size_t size)
+{
+ return ds_write(task, record, size, ds_pebs, /* force = */ 0);
+}
+
+int ds_unchecked_write_bts(struct task_struct *task,
+ const void *record, size_t size)
+{
+ return ds_write(task, record, size, ds_bts, /* force = */ 1);
+}
+
+int ds_unchecked_write_pebs(struct task_struct *task,
+ const void *record, size_t size)
+{
+ return ds_write(task, record, size, ds_pebs, /* force = */ 1);
+}
+
+static int ds_reset_or_clear(struct task_struct *task,
+ enum ds_qualifier qual, int clear)
+{
+ struct ds_context *context;
+ unsigned long base, end;
+ int error;
+
+ context = ds_get_context(task);
+ error = ds_validate_access(context, qual);
+ if (error < 0)
+ goto out;
+
+ base = ds_get(context->ds, qual, ds_buffer_base);
+ end = ds_get(context->ds, qual, ds_absolute_maximum);
+
+ if (clear)
+ memset((void *)base, 0, end - base);
+
+ ds_set(context->ds, qual, ds_index, base);
+
+ error = 0;
+ out:
+ ds_put_context(context);
+ return error;
+}
+
+int ds_reset_bts(struct task_struct *task)
+{
+ return ds_reset_or_clear(task, ds_bts, /* clear = */ 0);
+}
+
+int ds_reset_pebs(struct task_struct *task)
+{
+ return ds_reset_or_clear(task, ds_pebs, /* clear = */ 0);
+}
+
+int ds_clear_bts(struct task_struct *task)
+{
+ return ds_reset_or_clear(task, ds_bts, /* clear = */ 1);
+}
+
+int ds_clear_pebs(struct task_struct *task)
+{
+ return ds_reset_or_clear(task, ds_pebs, /* clear = */ 1);
+}
+
+int ds_get_pebs_reset(struct task_struct *task, u64 *value)
+{
+ struct ds_context *context;
+ int error;
+
+ if (!value)
+ return -EINVAL;
+
+ context = ds_get_context(task);
+ error = ds_validate_access(context, ds_pebs);
+ if (error < 0)
+ goto out;
+
+ *value = *(u64 *)(context->ds + (ds_cfg.sizeof_field * 8));
+
+ error = 0;
+ out:
+ ds_put_context(context);
+ return error;
+}
+
+int ds_set_pebs_reset(struct task_struct *task, u64 value)
+{
+ struct ds_context *context;
+ int error;
+
+ context = ds_get_context(task);
+ error = ds_validate_access(context, ds_pebs);
+ if (error < 0)
+ goto out;
+
+ *(u64 *)(context->ds + (ds_cfg.sizeof_field * 8)) = value;
+
+ error = 0;
+ out:
+ ds_put_context(context);
+ return error;
+}
+
+static const struct ds_configuration ds_cfg_var = {
+ .sizeof_ds = sizeof(long) * 12,
+ .sizeof_field = sizeof(long),
+ .sizeof_rec[ds_bts] = sizeof(long) * 3,
+ .sizeof_rec[ds_pebs] = sizeof(long) * 10
};
-
-static const struct ds_configuration ds_cfg_pentium_m = {
- .sizeof_ds = 9 * 4,
- .bts_buffer_base = { 0, 4 },
- .bts_index = { 4, 4 },
- .bts_absolute_maximum = { 8, 4 },
- .bts_interrupt_threshold = { 12, 4 },
- .sizeof_bts = 3 * 4,
- .from_ip = { 0, 4 },
- .to_ip = { 4, 4 },
- .info_type = { 4, 1 },
- .info_data = { 8, 4 },
- .debugctl_mask = (1<<6)|(1<<7)
-};
-#endif /* _i386_ */
-
-static const struct ds_configuration ds_cfg_core2 = {
- .sizeof_ds = 9 * 8,
- .bts_buffer_base = { 0, 8 },
- .bts_index = { 8, 8 },
- .bts_absolute_maximum = { 16, 8 },
- .bts_interrupt_threshold = { 24, 8 },
- .sizeof_bts = 3 * 8,
- .from_ip = { 0, 8 },
- .to_ip = { 8, 8 },
- .info_type = { 8, 1 },
- .info_data = { 16, 8 },
- .debugctl_mask = (1<<6)|(1<<7)|(1<<9)
+static const struct ds_configuration ds_cfg_64 = {
+ .sizeof_ds = 8 * 12,
+ .sizeof_field = 8,
+ .sizeof_rec[ds_bts] = 8 * 3,
+ .sizeof_rec[ds_pebs] = 8 * 10
};
static inline void
@@ -429,14 +821,13 @@
switch (c->x86) {
case 0x6:
switch (c->x86_model) {
-#ifdef __i386__
case 0xD:
case 0xE: /* Pentium M */
- ds_configure(&ds_cfg_pentium_m);
+ ds_configure(&ds_cfg_var);
break;
-#endif /* _i386_ */
case 0xF: /* Core2 */
- ds_configure(&ds_cfg_core2);
+ case 0x1C: /* Atom */
+ ds_configure(&ds_cfg_64);
break;
default:
/* sorry, don't know about them */
@@ -445,13 +836,11 @@
break;
case 0xF:
switch (c->x86_model) {
-#ifdef __i386__
case 0x0:
case 0x1:
case 0x2: /* Netburst */
- ds_configure(&ds_cfg_netburst);
+ ds_configure(&ds_cfg_var);
break;
-#endif /* _i386_ */
default:
/* sorry, don't know about them */
break;
@@ -462,3 +851,14 @@
break;
}
}
+
+void ds_free(struct ds_context *context)
+{
+ /* This is called when the task owning the parameter context
+ * is dying. There should not be any user of that context left
+ * to disturb us, anymore. */
+ unsigned long leftovers = context->count;
+ while (leftovers--)
+ ds_put_context(context);
+}
+#endif /* CONFIG_X86_DS */
diff --git a/arch/x86/kernel/process_32.c b/arch/x86/kernel/process_32.c
index 2c9abc9..1962d27 100644
--- a/arch/x86/kernel/process_32.c
+++ b/arch/x86/kernel/process_32.c
@@ -277,6 +277,14 @@
tss->x86_tss.io_bitmap_base = INVALID_IO_BITMAP_OFFSET;
put_cpu();
}
+#ifdef CONFIG_X86_DS
+ /* Free any DS contexts that have not been properly released. */
+ if (unlikely(current->thread.ds_ctx)) {
+ /* we clear debugctl to make sure DS is not used. */
+ update_debugctlmsr(0);
+ ds_free(current->thread.ds_ctx);
+ }
+#endif /* CONFIG_X86_DS */
}
void flush_thread(void)
@@ -438,6 +446,35 @@
return 0;
}
+#ifdef CONFIG_X86_DS
+static int update_debugctl(struct thread_struct *prev,
+ struct thread_struct *next, unsigned long debugctl)
+{
+ unsigned long ds_prev = 0;
+ unsigned long ds_next = 0;
+
+ if (prev->ds_ctx)
+ ds_prev = (unsigned long)prev->ds_ctx->ds;
+ if (next->ds_ctx)
+ ds_next = (unsigned long)next->ds_ctx->ds;
+
+ if (ds_next != ds_prev) {
+ /* we clear debugctl to make sure DS
+ * is not in use when we change it */
+ debugctl = 0;
+ update_debugctlmsr(0);
+ wrmsr(MSR_IA32_DS_AREA, ds_next, 0);
+ }
+ return debugctl;
+}
+#else
+static int update_debugctl(struct thread_struct *prev,
+ struct thread_struct *next, unsigned long debugctl)
+{
+ return debugctl;
+}
+#endif /* CONFIG_X86_DS */
+
static noinline void
__switch_to_xtra(struct task_struct *prev_p, struct task_struct *next_p,
struct tss_struct *tss)
@@ -448,14 +485,7 @@
prev = &prev_p->thread;
next = &next_p->thread;
- debugctl = prev->debugctlmsr;
- if (next->ds_area_msr != prev->ds_area_msr) {
- /* we clear debugctl to make sure DS
- * is not in use when we change it */
- debugctl = 0;
- update_debugctlmsr(0);
- wrmsr(MSR_IA32_DS_AREA, next->ds_area_msr, 0);
- }
+ debugctl = update_debugctl(prev, next, prev->debugctlmsr);
if (next->debugctlmsr != debugctl)
update_debugctlmsr(next->debugctlmsr);
@@ -479,13 +509,13 @@
hard_enable_TSC();
}
-#ifdef X86_BTS
+#ifdef CONFIG_X86_PTRACE_BTS
if (test_tsk_thread_flag(prev_p, TIF_BTS_TRACE_TS))
ptrace_bts_take_timestamp(prev_p, BTS_TASK_DEPARTS);
if (test_tsk_thread_flag(next_p, TIF_BTS_TRACE_TS))
ptrace_bts_take_timestamp(next_p, BTS_TASK_ARRIVES);
-#endif
+#endif /* CONFIG_X86_PTRACE_BTS */
if (!test_tsk_thread_flag(next_p, TIF_IO_BITMAP)) {
diff --git a/arch/x86/kernel/process_64.c b/arch/x86/kernel/process_64.c
index 00263c9..08a9df0 100644
--- a/arch/x86/kernel/process_64.c
+++ b/arch/x86/kernel/process_64.c
@@ -239,6 +239,14 @@
t->io_bitmap_max = 0;
put_cpu();
}
+#ifdef CONFIG_X86_DS
+ /* Free any DS contexts that have not been properly released. */
+ if (unlikely(t->ds_ctx)) {
+ /* we clear debugctl to make sure DS is not used. */
+ update_debugctlmsr(0);
+ ds_free(t->ds_ctx);
+ }
+#endif /* CONFIG_X86_DS */
}
void flush_thread(void)
@@ -472,13 +480,27 @@
next = &next_p->thread;
debugctl = prev->debugctlmsr;
- if (next->ds_area_msr != prev->ds_area_msr) {
- /* we clear debugctl to make sure DS
- * is not in use when we change it */
- debugctl = 0;
- update_debugctlmsr(0);
- wrmsrl(MSR_IA32_DS_AREA, next->ds_area_msr);
+
+#ifdef CONFIG_X86_DS
+ {
+ unsigned long ds_prev = 0, ds_next = 0;
+
+ if (prev->ds_ctx)
+ ds_prev = (unsigned long)prev->ds_ctx->ds;
+ if (next->ds_ctx)
+ ds_next = (unsigned long)next->ds_ctx->ds;
+
+ if (ds_next != ds_prev) {
+ /*
+ * We clear debugctl to make sure DS
+ * is not in use when we change it:
+ */
+ debugctl = 0;
+ update_debugctlmsr(0);
+ wrmsrl(MSR_IA32_DS_AREA, ds_next);
+ }
}
+#endif /* CONFIG_X86_DS */
if (next->debugctlmsr != debugctl)
update_debugctlmsr(next->debugctlmsr);
@@ -516,13 +538,13 @@
memset(tss->io_bitmap, 0xff, prev->io_bitmap_max);
}
-#ifdef X86_BTS
+#ifdef CONFIG_X86_PTRACE_BTS
if (test_tsk_thread_flag(prev_p, TIF_BTS_TRACE_TS))
ptrace_bts_take_timestamp(prev_p, BTS_TASK_DEPARTS);
if (test_tsk_thread_flag(next_p, TIF_BTS_TRACE_TS))
ptrace_bts_take_timestamp(next_p, BTS_TASK_ARRIVES);
-#endif
+#endif /* CONFIG_X86_PTRACE_BTS */
}
/*
diff --git a/arch/x86/kernel/ptrace.c b/arch/x86/kernel/ptrace.c
index fc3e8dc..58ce4b5 100644
--- a/arch/x86/kernel/ptrace.c
+++ b/arch/x86/kernel/ptrace.c
@@ -554,45 +554,115 @@
return 0;
}
-#ifdef X86_BTS
+#ifdef CONFIG_X86_PTRACE_BTS
+/*
+ * The configuration for a particular BTS hardware implementation.
+ */
+struct bts_configuration {
+ /* the size of a BTS record in bytes; at most BTS_MAX_RECORD_SIZE */
+ unsigned char sizeof_bts;
+ /* the size of a field in the BTS record in bytes */
+ unsigned char sizeof_field;
+ /* a bitmask to enable/disable BTS in DEBUGCTL MSR */
+ unsigned long debugctl_mask;
+};
+static struct bts_configuration bts_cfg;
-static int ptrace_bts_get_size(struct task_struct *child)
+#define BTS_MAX_RECORD_SIZE (8 * 3)
+
+
+/*
+ * Branch Trace Store (BTS) uses the following format. Different
+ * architectures vary in the size of those fields.
+ * - source linear address
+ * - destination linear address
+ * - flags
+ *
+ * Later architectures use 64bit pointers throughout, whereas earlier
+ * architectures use 32bit pointers in 32bit mode.
+ *
+ * We compute the base address for the first 8 fields based on:
+ * - the field size stored in the DS configuration
+ * - the relative field position
+ *
+ * In order to store additional information in the BTS buffer, we use
+ * a special source address to indicate that the record requires
+ * special interpretation.
+ *
+ * Netburst indicated via a bit in the flags field whether the branch
+ * was predicted; this is ignored.
+ */
+
+enum bts_field {
+ bts_from = 0,
+ bts_to,
+ bts_flags,
+
+ bts_escape = (unsigned long)-1,
+ bts_qual = bts_to,
+ bts_jiffies = bts_flags
+};
+
+static inline unsigned long bts_get(const char *base, enum bts_field field)
{
- if (!child->thread.ds_area_msr)
- return -ENXIO;
-
- return ds_get_bts_index((void *)child->thread.ds_area_msr);
+ base += (bts_cfg.sizeof_field * field);
+ return *(unsigned long *)base;
}
-static int ptrace_bts_read_record(struct task_struct *child,
- long index,
+static inline void bts_set(char *base, enum bts_field field, unsigned long val)
+{
+ base += (bts_cfg.sizeof_field * field);;
+ (*(unsigned long *)base) = val;
+}
+
+/*
+ * Translate a BTS record from the raw format into the bts_struct format
+ *
+ * out (out): bts_struct interpretation
+ * raw: raw BTS record
+ */
+static void ptrace_bts_translate_record(struct bts_struct *out, const void *raw)
+{
+ memset(out, 0, sizeof(*out));
+ if (bts_get(raw, bts_from) == bts_escape) {
+ out->qualifier = bts_get(raw, bts_qual);
+ out->variant.jiffies = bts_get(raw, bts_jiffies);
+ } else {
+ out->qualifier = BTS_BRANCH;
+ out->variant.lbr.from_ip = bts_get(raw, bts_from);
+ out->variant.lbr.to_ip = bts_get(raw, bts_to);
+ }
+}
+
+static int ptrace_bts_read_record(struct task_struct *child, size_t index,
struct bts_struct __user *out)
{
struct bts_struct ret;
- int retval;
- int bts_end;
- int bts_index;
+ const void *bts_record;
+ size_t bts_index, bts_end;
+ int error;
- if (!child->thread.ds_area_msr)
- return -ENXIO;
+ error = ds_get_bts_end(child, &bts_end);
+ if (error < 0)
+ return error;
- if (index < 0)
- return -EINVAL;
-
- bts_end = ds_get_bts_end((void *)child->thread.ds_area_msr);
if (bts_end <= index)
return -EINVAL;
- /* translate the ptrace bts index into the ds bts index */
- bts_index = ds_get_bts_index((void *)child->thread.ds_area_msr);
- bts_index -= (index + 1);
- if (bts_index < 0)
- bts_index += bts_end;
+ error = ds_get_bts_index(child, &bts_index);
+ if (error < 0)
+ return error;
- retval = ds_read_bts((void *)child->thread.ds_area_msr,
- bts_index, &ret);
- if (retval < 0)
- return retval;
+ /* translate the ptrace bts index into the ds bts index */
+ bts_index += bts_end - (index + 1);
+ if (bts_end <= bts_index)
+ bts_index -= bts_end;
+
+ error = ds_access_bts(child, bts_index, &bts_record);
+ if (error < 0)
+ return error;
+
+ ptrace_bts_translate_record(&ret, bts_record);
if (copy_to_user(out, &ret, sizeof(ret)))
return -EFAULT;
@@ -600,101 +670,106 @@
return sizeof(ret);
}
-static int ptrace_bts_clear(struct task_struct *child)
-{
- if (!child->thread.ds_area_msr)
- return -ENXIO;
-
- return ds_clear((void *)child->thread.ds_area_msr);
-}
-
static int ptrace_bts_drain(struct task_struct *child,
long size,
struct bts_struct __user *out)
{
- int end, i;
- void *ds = (void *)child->thread.ds_area_msr;
+ struct bts_struct ret;
+ const unsigned char *raw;
+ size_t end, i;
+ int error;
- if (!ds)
- return -ENXIO;
-
- end = ds_get_bts_index(ds);
- if (end <= 0)
- return end;
+ error = ds_get_bts_index(child, &end);
+ if (error < 0)
+ return error;
if (size < (end * sizeof(struct bts_struct)))
return -EIO;
- for (i = 0; i < end; i++, out++) {
- struct bts_struct ret;
- int retval;
+ error = ds_access_bts(child, 0, (const void **)&raw);
+ if (error < 0)
+ return error;
- retval = ds_read_bts(ds, i, &ret);
- if (retval < 0)
- return retval;
+ for (i = 0; i < end; i++, out++, raw += bts_cfg.sizeof_bts) {
+ ptrace_bts_translate_record(&ret, raw);
if (copy_to_user(out, &ret, sizeof(ret)))
return -EFAULT;
}
- ds_clear(ds);
+ error = ds_clear_bts(child);
+ if (error < 0)
+ return error;
return end;
}
+static void ptrace_bts_ovfl(struct task_struct *child)
+{
+ send_sig(child->thread.bts_ovfl_signal, child, 0);
+}
+
static int ptrace_bts_config(struct task_struct *child,
long cfg_size,
const struct ptrace_bts_config __user *ucfg)
{
struct ptrace_bts_config cfg;
- int bts_size, ret = 0;
- void *ds;
+ int error = 0;
+ error = -EOPNOTSUPP;
+ if (!bts_cfg.sizeof_bts)
+ goto errout;
+
+ error = -EIO;
if (cfg_size < sizeof(cfg))
- return -EIO;
+ goto errout;
+ error = -EFAULT;
if (copy_from_user(&cfg, ucfg, sizeof(cfg)))
- return -EFAULT;
+ goto errout;
- if ((int)cfg.size < 0)
- return -EINVAL;
+ error = -EINVAL;
+ if ((cfg.flags & PTRACE_BTS_O_SIGNAL) &&
+ !(cfg.flags & PTRACE_BTS_O_ALLOC))
+ goto errout;
- bts_size = 0;
- ds = (void *)child->thread.ds_area_msr;
- if (ds) {
- bts_size = ds_get_bts_size(ds);
- if (bts_size < 0)
- return bts_size;
- }
- cfg.size = PAGE_ALIGN(cfg.size);
+ if (cfg.flags & PTRACE_BTS_O_ALLOC) {
+ ds_ovfl_callback_t ovfl = 0;
+ unsigned int sig = 0;
- if (bts_size != cfg.size) {
- ret = ptrace_bts_realloc(child, cfg.size,
- cfg.flags & PTRACE_BTS_O_CUT_SIZE);
- if (ret < 0)
+ /* we ignore the error in case we were not tracing child */
+ (void)ds_release_bts(child);
+
+ if (cfg.flags & PTRACE_BTS_O_SIGNAL) {
+ if (!cfg.signal)
+ goto errout;
+
+ sig = cfg.signal;
+ ovfl = ptrace_bts_ovfl;
+ }
+
+ error = ds_request_bts(child, /* base = */ 0, cfg.size, ovfl);
+ if (error < 0)
goto errout;
- ds = (void *)child->thread.ds_area_msr;
+ child->thread.bts_ovfl_signal = sig;
}
- if (cfg.flags & PTRACE_BTS_O_SIGNAL)
- ret = ds_set_overflow(ds, DS_O_SIGNAL);
- else
- ret = ds_set_overflow(ds, DS_O_WRAP);
- if (ret < 0)
+ error = -EINVAL;
+ if (!child->thread.ds_ctx && cfg.flags)
goto errout;
if (cfg.flags & PTRACE_BTS_O_TRACE)
- child->thread.debugctlmsr |= ds_debugctl_mask();
+ child->thread.debugctlmsr |= bts_cfg.debugctl_mask;
else
- child->thread.debugctlmsr &= ~ds_debugctl_mask();
+ child->thread.debugctlmsr &= ~bts_cfg.debugctl_mask;
if (cfg.flags & PTRACE_BTS_O_SCHED)
set_tsk_thread_flag(child, TIF_BTS_TRACE_TS);
else
clear_tsk_thread_flag(child, TIF_BTS_TRACE_TS);
- ret = sizeof(cfg);
+ error = sizeof(cfg);
out:
if (child->thread.debugctlmsr)
@@ -702,10 +777,10 @@
else
clear_tsk_thread_flag(child, TIF_DEBUGCTLMSR);
- return ret;
+ return error;
errout:
- child->thread.debugctlmsr &= ~ds_debugctl_mask();
+ child->thread.debugctlmsr &= ~bts_cfg.debugctl_mask;
clear_tsk_thread_flag(child, TIF_BTS_TRACE_TS);
goto out;
}
@@ -714,119 +789,79 @@
long cfg_size,
struct ptrace_bts_config __user *ucfg)
{
- void *ds = (void *)child->thread.ds_area_msr;
struct ptrace_bts_config cfg;
+ size_t end;
+ const void *base, *max;
+ int error;
if (cfg_size < sizeof(cfg))
return -EIO;
+ error = ds_get_bts_end(child, &end);
+ if (error < 0)
+ return error;
+
+ error = ds_access_bts(child, /* index = */ 0, &base);
+ if (error < 0)
+ return error;
+
+ error = ds_access_bts(child, /* index = */ end, &max);
+ if (error < 0)
+ return error;
+
memset(&cfg, 0, sizeof(cfg));
-
- if (ds) {
- cfg.size = ds_get_bts_size(ds);
-
- if (ds_get_overflow(ds) == DS_O_SIGNAL)
- cfg.flags |= PTRACE_BTS_O_SIGNAL;
-
- if (test_tsk_thread_flag(child, TIF_DEBUGCTLMSR) &&
- child->thread.debugctlmsr & ds_debugctl_mask())
- cfg.flags |= PTRACE_BTS_O_TRACE;
-
- if (test_tsk_thread_flag(child, TIF_BTS_TRACE_TS))
- cfg.flags |= PTRACE_BTS_O_SCHED;
- }
-
+ cfg.size = (max - base);
+ cfg.signal = child->thread.bts_ovfl_signal;
cfg.bts_size = sizeof(struct bts_struct);
+ if (cfg.signal)
+ cfg.flags |= PTRACE_BTS_O_SIGNAL;
+
+ if (test_tsk_thread_flag(child, TIF_DEBUGCTLMSR) &&
+ child->thread.debugctlmsr & bts_cfg.debugctl_mask)
+ cfg.flags |= PTRACE_BTS_O_TRACE;
+
+ if (test_tsk_thread_flag(child, TIF_BTS_TRACE_TS))
+ cfg.flags |= PTRACE_BTS_O_SCHED;
+
if (copy_to_user(ucfg, &cfg, sizeof(cfg)))
return -EFAULT;
return sizeof(cfg);
}
-
static int ptrace_bts_write_record(struct task_struct *child,
const struct bts_struct *in)
{
- int retval;
+ unsigned char bts_record[BTS_MAX_RECORD_SIZE];
- if (!child->thread.ds_area_msr)
- return -ENXIO;
+ BUG_ON(BTS_MAX_RECORD_SIZE < bts_cfg.sizeof_bts);
- retval = ds_write_bts((void *)child->thread.ds_area_msr, in);
- if (retval)
- return retval;
+ memset(bts_record, 0, bts_cfg.sizeof_bts);
+ switch (in->qualifier) {
+ case BTS_INVALID:
+ break;
- return sizeof(*in);
-}
+ case BTS_BRANCH:
+ bts_set(bts_record, bts_from, in->variant.lbr.from_ip);
+ bts_set(bts_record, bts_to, in->variant.lbr.to_ip);
+ break;
-static int ptrace_bts_realloc(struct task_struct *child,
- int size, int reduce_size)
-{
- unsigned long rlim, vm;
- int ret, old_size;
+ case BTS_TASK_ARRIVES:
+ case BTS_TASK_DEPARTS:
+ bts_set(bts_record, bts_from, bts_escape);
+ bts_set(bts_record, bts_qual, in->qualifier);
+ bts_set(bts_record, bts_jiffies, in->variant.jiffies);
+ break;
- if (size < 0)
+ default:
return -EINVAL;
-
- old_size = ds_get_bts_size((void *)child->thread.ds_area_msr);
- if (old_size < 0)
- return old_size;
-
- ret = ds_free((void **)&child->thread.ds_area_msr);
- if (ret < 0)
- goto out;
-
- size >>= PAGE_SHIFT;
- old_size >>= PAGE_SHIFT;
-
- current->mm->total_vm -= old_size;
- current->mm->locked_vm -= old_size;
-
- if (size == 0)
- goto out;
-
- rlim = current->signal->rlim[RLIMIT_AS].rlim_cur >> PAGE_SHIFT;
- vm = current->mm->total_vm + size;
- if (rlim < vm) {
- ret = -ENOMEM;
-
- if (!reduce_size)
- goto out;
-
- size = rlim - current->mm->total_vm;
- if (size <= 0)
- goto out;
}
- rlim = current->signal->rlim[RLIMIT_MEMLOCK].rlim_cur >> PAGE_SHIFT;
- vm = current->mm->locked_vm + size;
- if (rlim < vm) {
- ret = -ENOMEM;
-
- if (!reduce_size)
- goto out;
-
- size = rlim - current->mm->locked_vm;
- if (size <= 0)
- goto out;
- }
-
- ret = ds_allocate((void **)&child->thread.ds_area_msr,
- size << PAGE_SHIFT);
- if (ret < 0)
- goto out;
-
- current->mm->total_vm += size;
- current->mm->locked_vm += size;
-
-out:
- if (child->thread.ds_area_msr)
- set_tsk_thread_flag(child, TIF_DS_AREA_MSR);
- else
- clear_tsk_thread_flag(child, TIF_DS_AREA_MSR);
-
- return ret;
+ /* The writing task will be the switched-to task on a context
+ * switch. It needs to write into the switched-from task's BTS
+ * buffer. */
+ return ds_unchecked_write_bts(child, bts_record, bts_cfg.sizeof_bts);
}
void ptrace_bts_take_timestamp(struct task_struct *tsk,
@@ -839,7 +874,66 @@
ptrace_bts_write_record(tsk, &rec);
}
-#endif /* X86_BTS */
+
+static const struct bts_configuration bts_cfg_netburst = {
+ .sizeof_bts = sizeof(long) * 3,
+ .sizeof_field = sizeof(long),
+ .debugctl_mask = (1<<2)|(1<<3)|(1<<5)
+};
+
+static const struct bts_configuration bts_cfg_pentium_m = {
+ .sizeof_bts = sizeof(long) * 3,
+ .sizeof_field = sizeof(long),
+ .debugctl_mask = (1<<6)|(1<<7)
+};
+
+static const struct bts_configuration bts_cfg_core2 = {
+ .sizeof_bts = 8 * 3,
+ .sizeof_field = 8,
+ .debugctl_mask = (1<<6)|(1<<7)|(1<<9)
+};
+
+static inline void bts_configure(const struct bts_configuration *cfg)
+{
+ bts_cfg = *cfg;
+}
+
+void __cpuinit ptrace_bts_init_intel(struct cpuinfo_x86 *c)
+{
+ switch (c->x86) {
+ case 0x6:
+ switch (c->x86_model) {
+ case 0xD:
+ case 0xE: /* Pentium M */
+ bts_configure(&bts_cfg_pentium_m);
+ break;
+ case 0xF: /* Core2 */
+ case 0x1C: /* Atom */
+ bts_configure(&bts_cfg_core2);
+ break;
+ default:
+ /* sorry, don't know about them */
+ break;
+ }
+ break;
+ case 0xF:
+ switch (c->x86_model) {
+ case 0x0:
+ case 0x1:
+ case 0x2: /* Netburst */
+ bts_configure(&bts_cfg_netburst);
+ break;
+ default:
+ /* sorry, don't know about them */
+ break;
+ }
+ break;
+ default:
+ /* sorry, don't know about them */
+ break;
+ }
+}
+#endif /* CONFIG_X86_PTRACE_BTS */
/*
* Called by kernel/ptrace.c when detaching..
@@ -852,15 +946,15 @@
#ifdef TIF_SYSCALL_EMU
clear_tsk_thread_flag(child, TIF_SYSCALL_EMU);
#endif
- if (child->thread.ds_area_msr) {
-#ifdef X86_BTS
- ptrace_bts_realloc(child, 0, 0);
-#endif
- child->thread.debugctlmsr &= ~ds_debugctl_mask();
- if (!child->thread.debugctlmsr)
- clear_tsk_thread_flag(child, TIF_DEBUGCTLMSR);
- clear_tsk_thread_flag(child, TIF_BTS_TRACE_TS);
- }
+#ifdef CONFIG_X86_PTRACE_BTS
+ (void)ds_release_bts(child);
+
+ child->thread.debugctlmsr &= ~bts_cfg.debugctl_mask;
+ if (!child->thread.debugctlmsr)
+ clear_tsk_thread_flag(child, TIF_DEBUGCTLMSR);
+
+ clear_tsk_thread_flag(child, TIF_BTS_TRACE_TS);
+#endif /* CONFIG_X86_PTRACE_BTS */
}
#if defined CONFIG_X86_32 || defined CONFIG_IA32_EMULATION
@@ -980,7 +1074,7 @@
/*
* These bits need more cooking - not enabled yet:
*/
-#ifdef X86_BTS
+#ifdef CONFIG_X86_PTRACE_BTS
case PTRACE_BTS_CONFIG:
ret = ptrace_bts_config
(child, data, (struct ptrace_bts_config __user *)addr);
@@ -992,7 +1086,7 @@
break;
case PTRACE_BTS_SIZE:
- ret = ptrace_bts_get_size(child);
+ ret = ds_get_bts_index(child, /* pos = */ 0);
break;
case PTRACE_BTS_GET:
@@ -1001,14 +1095,14 @@
break;
case PTRACE_BTS_CLEAR:
- ret = ptrace_bts_clear(child);
+ ret = ds_clear_bts(child);
break;
case PTRACE_BTS_DRAIN:
ret = ptrace_bts_drain
(child, data, (struct bts_struct __user *) addr);
break;
-#endif
+#endif /* CONFIG_X86_PTRACE_BTS */
default:
ret = ptrace_request(child, request, addr, data);