+ using the pattern seen below.
+
+ SHADOW_CONTENTS has byte-based addressing (PLACED_ADDRESS and SHADOW_LEN)
+ while we need bit-based addressing as the instructions length is 41 bits and
+ we must not modify/corrupt the adjacent slots in the same bundle.
+ Fortunately we may store larger memory incl. the adjacent bits with the
+ original memory content (not the possibly already stored breakpoints there).
+ We need to be careful in ia64_memory_remove_breakpoint to always restore
+ only the specific bits of this instruction ignoring any adjacent stored
+ bits.
+
+ We use the original addressing with the low nibble in the range <0..2> which
+ gets incorrectly interpreted by generic non-ia64 breakpoint_restore_shadows
+ as the direct byte offset of SHADOW_CONTENTS. We store whole BUNDLE_LEN
+ bytes just without these two possibly skipped bytes to not to exceed to the
+ next bundle.
+
+ If we would like to store the whole bundle to SHADOW_CONTENTS we would have
+ to store already the base address (`address & ~0x0f') into PLACED_ADDRESS.
+ In such case there is no other place where to store
+ SLOTNUM (`adress & 0x0f', value in the range <0..2>). We need to know
+ SLOTNUM in ia64_memory_remove_breakpoint.
+
+ There is one special case where we need to be extra careful:
+ L-X instructions, which are instructions that occupy 2 slots
+ (The L part is always in slot 1, and the X part is always in
+ slot 2). We must refuse to insert breakpoints for an address
+ that points at slot 2 of a bundle where an L-X instruction is
+ present, since there is logically no instruction at that address.
+ However, to make things more interesting, the opcode of L-X
+ instructions is located in slot 2. This means that, to insert
+ a breakpoint at an address that points to slot 1, we actually
+ need to write the breakpoint in slot 2! Slot 1 is actually
+ the extended operand, so writing the breakpoint there would not
+ have the desired effect. Another side-effect of this issue
+ is that we need to make sure that the shadow contents buffer
+ does save byte 15 of our instruction bundle (this is the tail
+ end of slot 2, which wouldn't be saved if we were to insert
+ the breakpoint in slot 1).
+
+ ia64 16-byte bundle layout:
+ | 5 bits | slot 0 with 41 bits | slot 1 with 41 bits | slot 2 with 41 bits |
+
+ The current addressing used by the code below:
+ original PC placed_address placed_size required covered
+ == bp_tgt->shadow_len reqd \subset covered
+ 0xABCDE0 0xABCDE0 0x10 <0x0...0x5> <0x0..0xF>
+ 0xABCDE1 0xABCDE1 0xF <0x5...0xA> <0x1..0xF>
+ 0xABCDE2 0xABCDE2 0xE <0xA...0xF> <0x2..0xF>
+
+ L-X instructions are treated a little specially, as explained above:
+ 0xABCDE1 0xABCDE1 0xF <0xA...0xF> <0x1..0xF>
+
+ `objdump -d' and some other tools show a bit unjustified offsets:
+ original PC byte where starts the instruction objdump offset
+ 0xABCDE0 0xABCDE0 0xABCDE0
+ 0xABCDE1 0xABCDE5 0xABCDE6
+ 0xABCDE2 0xABCDEA 0xABCDEC
+ */