Lazily and dynamically create i386-linux target descriptions

Instead of using pre-generated target descriptions, this patch
changes GDB to lazily and dynamically create target descriptions
according to the target hardware capability (xcr0 in i386).
This support any combination of target features.

Some reg in target description has "regnum" attribute, so its register
number is got from the attribute value instead from sequential allocation.

  <reg name="xmm0" bitsize="128" type="vec128" regnum="32"/>

when target description is created, it should match the regnum, so this
patch adds a new field m_next_regnum to track it, if attribute number is
greater than the m_next_regnum, print the code to set register number
explicitly.

gdb:

2017-07-26  Yao Qi  <yao.qi@linaro.org>

	* i386-linux-tdep.c: Don't include features/i386/i386-*linux.c.
	Include features/i386/32bit-*.c.
	(i386_linux_read_description): Generate target description if it
	doesn't exist.
	(_initialize_i386_linux_tdep): Don't call _initialize_tdesc_i386
	functions.
	* features/i386/32bit-linux.c: Re-generated.
	* features/i386/32bit-sse.c: Likewise.
	* target-descriptions.c (print_c_feature::visit): Print code to
	set register number if needed.
	(print_c_feature) <m_next_regnum>: New field.

1 files changed, 47 insertions, 0 deletions

diff --git a/gdb/target-descriptions.c b/gdb/target-descriptions.c
index 5fd07ef0451..333cbd89ab7 100644
--- a/gdb/target-descriptions.c
+++ b/gdb/target-descriptions.c
@@ -2104,6 +2104,48 @@ public:
 
   void visit (const tdesc_reg *reg) override
   {
+    /* Most "reg" in XML target descriptions don't have "regnum"
+       attribute, so the register number is allocated sequentially.
+       In case that reg has "regnum" attribute, register number
+       should be set by that explicitly.  */
+
+    if (reg->target_regnum < m_next_regnum)
+      {
+	/* The integrity check, it can catch some errors on register
+	   number collision, like this,
+
+	  <reg name="x0" bitsize="32"/>
+	  <reg name="x1" bitsize="32"/>
+	  <reg name="x2" bitsize="32"/>
+	  <reg name="x3" bitsize="32"/>
+	  <reg name="ps" bitsize="32" regnum="3"/>
+
+	  but it also has false negatives.  The target description
+	  below is correct,
+
+	  <reg name="x1" bitsize="32" regnum="1"/>
+	  <reg name="x3" bitsize="32" regnum="3"/>
+	  <reg name="x2" bitsize="32" regnum="2"/>
+	  <reg name="x4" bitsize="32" regnum="4"/>
+
+	  but it is not a good practice, so still error on this,
+	  and also print the message so that it can be saved in the
+	  generated c file.  */
+
+	printf_unfiltered ("ERROR: \"regnum\" attribute %ld ",
+			   reg->target_regnum);
+	printf_unfiltered ("is not the largest number (%d).\n",
+			   m_next_regnum);
+	error (_("\"regnum\" attribute %ld is not the largest number (%d)."),
+	       reg->target_regnum, m_next_regnum);
+      }
+
+    if (reg->target_regnum > m_next_regnum)
+      {
+	printf_unfiltered ("  regnum = %ld;\n", reg->target_regnum);
+	m_next_regnum = reg->target_regnum;
+      }
+
     printf_unfiltered ("  tdesc_create_reg (feature, \"%s\", regnum++, %d, ",
 		       reg->name, reg->save_restore);
     if (reg->group)
@@ -2111,8 +2153,13 @@ public:
     else
       printf_unfiltered ("NULL, ");
     printf_unfiltered ("%d, \"%s\");\n", reg->bitsize, reg->type);
+
+    m_next_regnum++;
   }
 
+private:
+  /* The register number to use for the next register we see.  */
+  int m_next_regnum = 0;
 };
 
 static void