PR record/18691: Fix fails in solib-precsave.exp
authorYao Qi <yao.qi@linaro.org>
Wed, 29 Jul 2015 11:43:10 +0000 (12:43 +0100)
committerYao Qi <yao.qi@linaro.org>
Wed, 29 Jul 2015 11:43:10 +0000 (12:43 +0100)
commitcc9f16aa882eb22cb2128c5eb8237fd453ab2988
tree1731fb259bc51d6759f4710fc0081dcd6a3d938f
parenteb1a79028c63ca565850c5d64ae60193f90cd841
PR record/18691: Fix fails in solib-precsave.exp

We see the following regressions in testing on x86_64-linux,

 reverse-step^M
 Cannot access memory at address 0x2aaaaaed26c0^M
 (gdb) FAIL: gdb.reverse/solib-precsave.exp: reverse-step into solib function one

when GDB reverse step into a function, GDB wants to skip prologue so
it requests TARGET_OBJECT_CODE_MEMORY to read some code memory in
memory_xfer_partial_1.  However in dcache_read_memory_partial, the object
becomes TARGET_OBJECT_MEMORY

      return ops->to_xfer_partial (ops, TARGET_OBJECT_MEMORY, NULL,
                                   myaddr, NULL, memaddr, len,
                                   xfered_len);

in reverse debugging, ops->to_xfer_partial is record_full_core_xfer_partial
and it will return TARGET_XFER_E_IO because it can't find any records.
The test fails.

At this moment, the delegate relationship is like

  dcache -> record-core -> core -> exec

and we want to GDB read memory across targets, which means if the
requested memory isn't found in record-core, GDB can read memory from
core, and exec even further if needed.  I find raw_memory_xfer_partial
is exactly what I want.

gdb:

2015-07-29  Yao Qi  <yao.qi@linaro.org>

PR record/18691
* dcache.c (dcache_read_memory_partial): Call
raw_memory_xfer_partial.
* target.c (raw_memory_xfer_partial): Make it non-static.
* target.h (raw_memory_xfer_partial): Declare.
gdb/ChangeLog
gdb/dcache.c
gdb/target.c
gdb/target.h