Re: [Qemu-devel] [PATCH v7 03/16] gdbstub: add multiprocess support to '?' packets

[Luc Michel]

On 11/25/18 10:22 PM, Philippe Mathieu-Daudé wrote:
> Hi Luc,
> 
> On 23/11/18 10:17, Luc Michel wrote:
>> The gdb_get_cpu_pid() function does the PID lookup for the given CPU. It
>> checks if the CPU is a direct child of a CPU cluster. If it is, the
>> returned PID is the cluster ID plus one (cluster IDs start at 0, GDB
>> PIDs at 1). When the CPU is not a child of such a container, the PID of
>> the first process is returned.
>>
>> The gdb_fmt_thread_id() function generates the string to be used to identify
>> a given thread, in a response packet for the peer. This function
>> supports generating thread IDs when multiprocess mode is enabled (in the
>> form `p<pid>.<tid>').
>>
>> Use them in the reply to a '?' request.
>>
>> Signed-off-by: Luc Michel <address@hidden>
>> Acked-by: Alistair Francis <address@hidden>
>> Reviewed-by: Edgar E. Iglesias <address@hidden>
>> ---
>>  gdbstub.c | 60 +++++++++++++++++++++++++++++++++++++++++++++++++++++--
>>  1 file changed, 58 insertions(+), 2 deletions(-)
>>
>> diff --git a/gdbstub.c b/gdbstub.c
>> index 26f5a7449a..4fbc05dfe3 100644
>> --- a/gdbstub.c
>> +++ b/gdbstub.c
>> @@ -638,10 +638,52 @@ static int memtox(char *buf, const char *mem, int len)
>>          }
>>      }
>>      return p - buf;
>>  }
>>  
>> +static uint32_t gdb_get_cpu_pid(const GDBState *s, CPUState *cpu)
>> +{
>> +#ifndef CONFIG_USER_ONLY
>> +    gchar *path, *name;
> 
> Setting ...
> 
>        gchar *path, *name = NULL;
> 
>> +    Object *obj;
>> +    CPUClusterState *cluster;
>> +    uint32_t ret;
>> +
>> +    path = object_get_canonical_path(OBJECT(cpu));
>> +    name = object_get_canonical_path_component(OBJECT(cpu));
> 
> ... we might move this line ...
> 
>> +
>> +    if (path == NULL) {
>> +        ret = s->processes[0].pid;
>> +        goto out;
>> +    }
> 
> ... hereOK I'll change that.
> 
>        name = object_get_canonical_path_component(OBJECT(cpu));
> 
>> +
>> +    /*
>> +     * Retrieve the CPU parent path by removing the last '/' and the CPU 
>> name
>> +     * from the CPU canonical path. */
>> +    path[strlen(path) - strlen(name) - 1] = '\0';
> 
> Can we get there with path != NULL && name == NULL?
I think the only way we could end up in this case is if cpu ==
object_get_root(), which does not make much sense. I can add an
assert(name != NULL) here to enforce that.

> 
>> +
>> +    obj = object_resolve_path_type(path, TYPE_CPU_CLUSTER, NULL);
>> +
>> +    if (obj == NULL) {
>> +        ret = s->processes[0].pid;
>> +        goto out;
>> +    }
>> +
>> +    cluster = CPU_CLUSTER(obj);
>> +    ret = cluster->cluster_id + 1;
>> +
>> +out:
>> +    g_free(name);
>> +    g_free(path);
>> +
>> +    return ret;
>> +
>> +#else
> 
> [*]
> 
>> +    return s->processes[0].pid;
>> +#endif
>> +}
>> +
>>  static const char *get_feature_xml(const char *p, const char **newp,
>>                                     CPUClass *cc)
>>  {
>>      size_t len;
>>      int i;
>> @@ -907,10 +949,23 @@ static CPUState *find_cpu(uint32_t thread_id)
>>      }
>>  
>>      return NULL;
>>  }
>>  
>> +static char *gdb_fmt_thread_id(const GDBState *s, CPUState *cpu,
>> +                           char *buf, size_t buf_size)
>> +{
>> +    if (s->multiprocess) {
>> +        snprintf(buf, buf_size, "p%02x.%02x",
>> +                 gdb_get_cpu_pid(s, cpu), cpu_gdb_index(cpu));
>> +    } else {
>> +        snprintf(buf, buf_size, "%02x", cpu_gdb_index(cpu));
>> +    }
>> +
>> +    return buf;
>> +}
>> +
>>  static int is_query_packet(const char *p, const char *query, char separator)
>>  {
>>      unsigned int query_len = strlen(query);
>>  
>>      return strncmp(p, query, query_len) == 0 &&
>> @@ -1018,22 +1073,23 @@ static int gdb_handle_packet(GDBState *s, const char 
>> *line_buf)
>>      const char *p;
>>      uint32_t thread;
>>      int ch, reg_size, type, res;
>>      uint8_t mem_buf[MAX_PACKET_LENGTH];
>>      char buf[sizeof(mem_buf) + 1 /* trailing NUL */];
>> +    char thread_id[16];
>>      uint8_t *registers;
>>      target_ulong addr, len;
>>  
>>      trace_gdbstub_io_command(line_buf);
>>  
>>      p = line_buf;
>>      ch = *p++;
>>      switch(ch) {
>>      case '?':
>>          /* TODO: Make this return the correct value for user-mode.  */
> 
> Is this comment still relevant?
> 
> If so, wouldn't it be better placed in [*]?
git blame shows that at the time when this comment was added
(1fddef4b1ba from 2005), the Stop Reply packet was like this:

+        /* TODO: Make this return the correct value for user-mode.  */
         snprintf(buf, sizeof(buf), "S%02x", SIGTRAP);

Which is the form that contains only a signal number. So this comment
must refer to this hard-coded signal, so I think it is still valid :-)

However, you are right pointing out that the PID used in user-mode
should probably be the one of the QEMU process running the guest binary
(as it is done for TIDs I believe). I'll add a comment at [*] to point
that out.

Thanks.

-- 
Luc

> 
>> -        snprintf(buf, sizeof(buf), "T%02xthread:%02x;", GDB_SIGNAL_TRAP,
>> -                 cpu_gdb_index(s->c_cpu));
>> +        snprintf(buf, sizeof(buf), "T%02xthread:%s;", GDB_SIGNAL_TRAP,
>> +                 gdb_fmt_thread_id(s, s->c_cpu, thread_id, 
>> sizeof(thread_id)));
>>          put_packet(s, buf);
>>          /* Remove all the breakpoints when this query is issued,
>>           * because gdb is doing and initial connect and the state
>>           * should be cleaned up.
>>           */
>>