exec.c: Fix breakpoint invalidation race

A bug (1647683) was reported showing a crash when removing
breakpoints.  The reproducer was bisected to 3359baad when tb_flush
was finally made thread safe.  While in MTTCG the locking in
breakpoint_invalidate would have prevented any problems, but
currently tb_lock() is a NOP for system emulation.

The race is between a tb_flush from the gdbstub and the
tb_invalidate_phys_addr() in breakpoint_invalidate().

Ideally we'd have actual locking here; for the moment the
simple fix is to do a full tb_flush() for a bp invalidate,
since that is thread-safe even if no lock is taken.

Reported-by: Julian Brown <julian@codesourcery.com>
Signed-off-by: Peter Maydell <peter.maydell@linaro.org>
Message-id: 1481047629-7763-1-git-send-email-peter.maydell@linaro.org
Signed-off-by: Stefan Hajnoczi <stefanha@redhat.com>
This commit is contained in:
Peter Maydell 2016-12-06 18:07:09 +00:00 committed by Stefan Hajnoczi
parent d750c3a966
commit a9353fe897
1 changed files with 6 additions and 19 deletions

25
exec.c
View File

@ -684,28 +684,15 @@ void cpu_exec_realizefn(CPUState *cpu, Error **errp)
#endif
}
#if defined(CONFIG_USER_ONLY)
static void breakpoint_invalidate(CPUState *cpu, target_ulong pc)
{
mmap_lock();
tb_lock();
tb_invalidate_phys_page_range(pc, pc + 1, 0);
tb_unlock();
mmap_unlock();
/* Flush the whole TB as this will not have race conditions
* even if we don't have proper locking yet.
* Ideally we would just invalidate the TBs for the
* specified PC.
*/
tb_flush(cpu);
}
#else
static void breakpoint_invalidate(CPUState *cpu, target_ulong pc)
{
MemTxAttrs attrs;
hwaddr phys = cpu_get_phys_page_attrs_debug(cpu, pc, &attrs);
int asidx = cpu_asidx_from_attrs(cpu, attrs);
if (phys != -1) {
/* Locks grabbed by tb_invalidate_phys_addr */
tb_invalidate_phys_addr(cpu->cpu_ases[asidx].as,
phys | (pc & ~TARGET_PAGE_MASK));
}
}
#endif
#if defined(CONFIG_USER_ONLY)
void cpu_watchpoint_remove_all(CPUState *cpu, int mask)