Like other JITs, sparc64 maintains an array of instruction offsets but
stores the entries off by one. This is done because jumps to the
exit block are indexed to one past the last BPF instruction.
So if we size the array by the program length, we need to record
the previous instruction in order to stay within the array bounds.
This is explained in ARM JIT commit 8eee539dde ("arm64: bpf: fix
out-of-bounds read in bpf2a64_offset()").
But this scheme requires a little bit of careful handling when
the instruction before the branch destination is a 64-bit load
immediate. It takes up 2 BPF instruction slots.
Therefore, we have to fill in the array entry for the second
half of the 64-bit load immediate instruction rather than for
the one for the beginning of that instruction.
Fixes: 7a12b5031c ("sparc64: Add eBPF JIT.")
Signed-off-by: David S. Miller <davem@davemloft.net>
Doing a full 64-bit decomposition is really stupid especially for
simple values like 0 and -1.
But if we are going to optimize this, go all the way and try for all 2
and 3 instruction sequences not requiring a temporary register as
well.
First we do the easy cases where it's a zero or sign extended 32-bit
number (sethi+or, sethi+xor, respectively).
Then we try to find a range of set bits we can load simply then shift
up into place, in various ways.
Then we try negating the constant and see if we can do a simple
sequence using that with a xor at the end. (f.e. the range of set
bits can't be loaded simply, but for the negated value it can)
The final optimized strategy involves 4 instructions sequences not
needing a temporary register.
Otherwise we sadly fully decompose using a temp..
Example, from ALU64_XOR_K: 0x0000ffffffff0000 ^ 0x0 = 0x0000ffffffff0000:
0000000000000000 <foo>:
0: 9d e3 bf 50 save %sp, -176, %sp
4: 01 00 00 00 nop
8: 90 10 00 18 mov %i0, %o0
c: 13 3f ff ff sethi %hi(0xfffffc00), %o1
10: 92 12 63 ff or %o1, 0x3ff, %o1 ! ffffffff <foo+0xffffffff>
14: 93 2a 70 10 sllx %o1, 0x10, %o1
18: 15 3f ff ff sethi %hi(0xfffffc00), %o2
1c: 94 12 a3 ff or %o2, 0x3ff, %o2 ! ffffffff <foo+0xffffffff>
20: 95 2a b0 10 sllx %o2, 0x10, %o2
24: 92 1a 60 00 xor %o1, 0, %o1
28: 12 e2 40 8a cxbe %o1, %o2, 38 <foo+0x38>
2c: 9a 10 20 02 mov 2, %o5
30: 10 60 00 03 b,pn %xcc, 3c <foo+0x3c>
34: 01 00 00 00 nop
38: 9a 10 20 01 mov 1, %o5 ! 1 <foo+0x1>
3c: 81 c7 e0 08 ret
40: 91 eb 40 00 restore %o5, %g0, %o0
Signed-off-by: David S. Miller <davem@davemloft.net>
cbcond combines a compare with a branch into a single instruction.
The limitations are:
1) Only newer chips support it
2) For immediate compares we are limited to 5-bit signed immediate
values
3) The branch displacement is limited to 10-bit signed
4) We cannot use it for JSET
Also, cbcond (unlike all other sparc control transfers) lacks a delay
slot.
Currently we don't have a useful instruction we can push into the
delay slot of normal branches. So using cbcond pretty much always
increases code density, and is therefore a win.
Signed-off-by: David S. Miller <davem@davemloft.net>
This is an eBPF JIT for sparc64. All major features are supported.
All tests under tools/testing/selftests/bpf/ pass.
Signed-off-by: David S. Miller <davem@davemloft.net>
David S. Miller