This is the mail archive of the
gdb-patches@sourceware.org
mailing list for the GDB project.
[PATCH] RISC-V: Don't decrement pc after break.
- From: Jim Wilson <jimw at sifive dot com>
- To: gdb-patches at sourceware dot org
- Cc: Jim Wilson <jimw at sifive dot com>, andrew dot burgess at embecosm dot com
- Date: Mon, 16 Jul 2018 17:12:41 -0700
- Subject: [PATCH] RISC-V: Don't decrement pc after break.
This is the gdb patch for the RISC-V Linux kernel patch I just submitted.
https://patchwork.kernel.org/patch/10524925/
This removes the code that decrements the pc after a break, now that we have
a patch to stop linux from pointlessly adding to the pc after a break. It
would be nice if this goes in now, to avoid unnecessary divergence between
gdb and the linux kernel. Palmer, the RISC-V linux kernel maintainer, has
already agreed to accept that patch.
This will also be needed by the FreeBSD gdb port which may be started soon.
This also fixes a bug in the code. The fact that we have two different
mechanisms to decide breakpoint size, used_compressed_breakpoints and
has_compressed_isa, means that it is possible for gdb to emit a 4 byte
breakpoint and then subtract 2 from the pc, and vice versa. Removing the
unnecessary pc decrement fixes that problem.
OK?
Jim
gdb/
* riscv-tdep.c (riscv_has_feature): Delete comment that refers to
set_gdbarch_decr_pc_after_break. Call riscv_read_misa_reg always.
(riscv_gdbarch_init): Delete local has_compressed_isa. Delete now
unecessary braces after EF_RISCV_RVC test. Delete call to
set_gdbarch_decr_pc_after_break.
---
gdb/riscv-tdep.c | 25 ++-----------------------
1 file changed, 2 insertions(+), 23 deletions(-)
diff --git a/gdb/riscv-tdep.c b/gdb/riscv-tdep.c
index 72dab0f897..f5d1af822c 100644
--- a/gdb/riscv-tdep.c
+++ b/gdb/riscv-tdep.c
@@ -335,23 +335,7 @@ riscv_has_feature (struct gdbarch *gdbarch, char feature)
gdb_assert (feature >= 'A' && feature <= 'Z');
- /* It would be nice to always check with the real target where possible,
- however, for compressed instructions this is a bad idea.
-
- The call to `set_gdbarch_decr_pc_after_break' is made just once per
- GDBARCH and we decide at that point if we should decrement by 2 or 4
- bytes based on whether the BFD has compressed instruction support or
- not.
-
- If the BFD was not compiled with compressed instruction support, but we
- are running on a target with compressed instructions then we might
- place a 4-byte breakpoint, then decrement the $pc by 2 bytes leading to
- confusion.
-
- It's safer if we just make decisions about compressed instruction
- support based on the BFD. */
- if (feature != 'C')
- misa = riscv_read_misa_reg (&have_read_misa);
+ misa = riscv_read_misa_reg (&have_read_misa);
if (!have_read_misa || misa == 0)
misa = gdbarch_tdep (gdbarch)->core_features;
@@ -2440,7 +2424,6 @@ riscv_gdbarch_init (struct gdbarch_info info,
struct gdbarch *gdbarch;
struct gdbarch_tdep *tdep;
struct gdbarch_tdep tmp_tdep;
- bool has_compressed_isa = false;
int i;
/* Ideally, we'd like to get as much information from the target for
@@ -2472,10 +2455,7 @@ riscv_gdbarch_init (struct gdbarch_info info,
_("unknown ELF header class %d"), eclass);
if (e_flags & EF_RISCV_RVC)
- {
- has_compressed_isa = true;
- tmp_tdep.core_features |= (1 << ('C' - 'A'));
- }
+ tmp_tdep.core_features |= (1 << ('C' - 'A'));
if (e_flags & EF_RISCV_FLOAT_ABI_DOUBLE)
{
@@ -2545,7 +2525,6 @@ riscv_gdbarch_init (struct gdbarch_info info,
set_gdbarch_register_reggroup_p (gdbarch, riscv_register_reggroup_p);
/* Functions to analyze frames. */
- set_gdbarch_decr_pc_after_break (gdbarch, (has_compressed_isa ? 2 : 4));
set_gdbarch_skip_prologue (gdbarch, riscv_skip_prologue);
set_gdbarch_inner_than (gdbarch, core_addr_lessthan);
set_gdbarch_frame_align (gdbarch, riscv_frame_align);
--
2.17.1