Re: [PATCH 13/19] accel/tcg: Fixed tsan warnings.

[Emilio G. Cota]

On Fri, May 22, 2020 at 12:07:49 -0400, Robert Foley wrote:
> For example:
> WARNING: ThreadSanitizer: data race (pid=35425)
>   Write of size 4 at 0x7bbc000000ac by main thread (mutexes: write M875):
>     #0 cpu_reset_interrupt hw/core/cpu.c:107:28 (qemu-system-aarch64+0x843790)
>     #1 arm_cpu_set_irq target/arm/cpu.c (qemu-system-aarch64+0x616265)
>     #2 qemu_set_irq hw/core/irq.c:44:5 (qemu-system-aarch64+0x8462ca)
>   Previous atomic read of size 4 at 0x7bbc000000ac by thread T6:
>     #0 __tsan_atomic32_load <null> (qemu-system-aarch64+0x394c1c)
>     #1 cpu_handle_interrupt accel/tcg/cpu-exec.c:534:9 
> (qemu-system-aarch64+0x4b7e79)
>     #2 cpu_exec accel/tcg/cpu-exec.c:720:17 (qemu-system-aarch64+0x4b7e79)
> or
> WARNING: ThreadSanitizer: data race (pid=25425)
>   Read of size 8 at 0x7f8ad8e138d0 by thread T10:
>     #0 tb_lookup_cmp accel/tcg/cpu-exec.c:307:13 
> (qemu-system-aarch64+0x4ac4d2)
>     #1 qht_do_lookup util/qht.c:502:34 (qemu-system-aarch64+0xd05264)
>   Previous write of size 8 at 0x7f8ad8e138d0 by thread T15 (mutexes: write 
> M728311726235541804):
>     #0 tb_link_page accel/tcg/translate-all.c:1625:26 
> (qemu-system-aarch64+0x4b0bf2)
>     #1 tb_gen_code accel/tcg/translate-all.c:1865:19 
> (qemu-system-aarch64+0x4b0bf2)
>     #2 tb_find accel/tcg/cpu-exec.c:407:14 (qemu-system-aarch64+0x4ad77c)

I see you're working through the warnings in this file, but I think it would
be better to forget about files and focus on the data itself.
Therefore this patch should be split in two: (1) cpu-<interrupt_request
and (2) gen_code_buf. (1) requires a lot of changes with a proper audit;
the per-cpu-lock series has a possible solution for that, so I will
ignore those hunks and just comment on (2) below.

> Cc: Richard Henderson <address@hidden>
> Cc: Paolo Bonzini <address@hidden>
> Signed-off-by: Robert Foley <address@hidden>
> ---
>  accel/tcg/tcg-all.c       | 4 ++--
>  accel/tcg/tcg-runtime.c   | 7 ++++++-
>  accel/tcg/translate-all.c | 6 +++++-
>  hw/core/cpu.c             | 2 +-
>  4 files changed, 14 insertions(+), 5 deletions(-)
> 
(snip)
> diff --git a/accel/tcg/tcg-runtime.c b/accel/tcg/tcg-runtime.c
> index 446465a09a..bd0cd77450 100644
> --- a/accel/tcg/tcg-runtime.c
> +++ b/accel/tcg/tcg-runtime.c
> @@ -31,6 +31,7 @@
>  #include "disas/disas.h"
>  #include "exec/log.h"
>  #include "tcg/tcg.h"
> +#include "qemu/tsan.h"
>  
>  /* 32-bit helpers */
>  
> @@ -151,6 +152,7 @@ void *HELPER(lookup_tb_ptr)(CPUArchState *env)
>      TranslationBlock *tb;
>      target_ulong cs_base, pc;
>      uint32_t flags;
> +    void *tc_ptr;
>  
>      tb = tb_lookup__cpu_state(cpu, &pc, &cs_base, &flags, curr_cflags());
>      if (tb == NULL) {
> @@ -161,7 +163,10 @@ void *HELPER(lookup_tb_ptr)(CPUArchState *env)
>                             TARGET_FMT_lx "/" TARGET_FMT_lx "/%#x] %s\n",
>                             cpu->cpu_index, tb->tc.ptr, cs_base, pc, flags,
>                             lookup_symbol(pc));
> -    return tb->tc.ptr;
> +    TSAN_ANNOTATE_IGNORE_READS_BEGIN();
> +    tc_ptr = tb->tc.ptr;
> +    TSAN_ANNOTATE_IGNORE_READS_END();
> +    return tc_ptr;

I'm not sure these are needed. At least after applying all other patches
in this series, I don't get a warning here.

>  }
>  
>  void HELPER(exit_atomic)(CPUArchState *env)
> diff --git a/accel/tcg/translate-all.c b/accel/tcg/translate-all.c
> index 3fb71a1503..6c0e61994c 100644
> --- a/accel/tcg/translate-all.c
> +++ b/accel/tcg/translate-all.c
> @@ -58,6 +58,7 @@
>  #include "exec/log.h"
>  #include "sysemu/cpus.h"
>  #include "sysemu/tcg.h"
> +#include "qemu/tsan.h"
>  
>  /* #define DEBUG_TB_INVALIDATE */
>  /* #define DEBUG_TB_FLUSH */
> @@ -1704,6 +1705,7 @@ TranslationBlock *tb_gen_code(CPUState *cpu,
>          max_insns = 1;
>      }
>  
> +    TSAN_ANNOTATE_IGNORE_WRITES_BEGIN();

Same here, I don't get a warning in this hunk if I remove these,
except for:
---
WARNING: ThreadSanitizer: data race (pid=445867)
  Atomic read of size 1 at 0x7f906e050158 by thread T7:
    #0 __tsan_mutex_post_lock <null> (qemu-system-aarch64+0x481721)
    #1 qemu_spin_lock /home/cota/src/qemu/include/qemu/thread.h:244:5 
(qemu-system-aarch64+0x5578e9)
    #2 tb_add_jump /home/cota/src/qemu/accel/tcg/cpu-exec.c:363:5 
(qemu-system-aarch64+0x5578e9)
    #3 tb_find /home/cota/src/qemu/accel/tcg/cpu-exec.c:423:9 
(qemu-system-aarch64+0x5578e9)

  Previous write of size 1 at 0x7f906e050158 by thread T8:
    #0 __tsan_mutex_create <null> (qemu-system-aarch64+0x481589)
    #1 qemu_spin_init /home/cota/src/qemu/include/qemu/thread.h:221:5 
(qemu-system-aarch64+0x559a71)
    #2 tb_gen_code /home/cota/src/qemu/accel/tcg/translate-all.c:1875:5 
(qemu-system-aarch64+0x559a71)

  Thread T7 'CPU 0/TCG' (tid=445875, running) created by main thread at:
    #0 pthread_create <null> (qemu-system-aarch64+0x43915b)
    #1 qemu_thread_create /home/cota/src/qemu/util/qemu-thread-posix.c:558:11 
(qemu-system-aarch64+0xaf91ff)

  Thread T8 'CPU 1/TCG' (tid=445876, running) created by main thread at:
    #0 pthread_create <null> (qemu-system-aarch64+0x43915b)
    #1 qemu_thread_create /home/cota/src/qemu/util/qemu-thread-posix.c:558:11 
(qemu-system-aarch64+0xaf91ff)

SUMMARY: ThreadSanitizer: data race 
(/home/cota/src/qemu/build/aarch64-softmmu/qemu-system-aarch64+0x481721) in 
__tsan_mutex_post_lock
---

Seems like tsan is confusing itself here.

Thanks,
                E.