6.6-stable review patch. If anyone has any objections, please let me know.
------------------
From: Ilya Leoshkevich iii@linux.ibm.com
commit 7ba4f43e16de351fe9821de80e15d88c884b2967 upstream.
Using a mask instead of an array saves a small amount of memory and allows marking multiple registers as seen with a simple "or". Another positive side-effect is that it speeds up verification with jitterbug.
Signed-off-by: Ilya Leoshkevich iii@linux.ibm.com Signed-off-by: Daniel Borkmann daniel@iogearbox.net Link: https://lore.kernel.org/bpf/20240703005047.40915-2-iii@linux.ibm.com Signed-off-by: Greg Kroah-Hartman gregkh@linuxfoundation.org --- arch/s390/net/bpf_jit_comp.c | 32 ++++++++++++++++---------------- 1 file changed, 16 insertions(+), 16 deletions(-)
--- a/arch/s390/net/bpf_jit_comp.c +++ b/arch/s390/net/bpf_jit_comp.c @@ -35,7 +35,7 @@
struct bpf_jit { u32 seen; /* Flags to remember seen eBPF instructions */ - u32 seen_reg[16]; /* Array to remember which registers are used */ + u16 seen_regs; /* Mask to remember which registers are used */ u32 *addrs; /* Array with relative instruction addresses */ u8 *prg_buf; /* Start of program */ int size; /* Size of program and literal pool */ @@ -118,8 +118,8 @@ static inline void reg_set_seen(struct b { u32 r1 = reg2hex[b1];
- if (r1 >= 6 && r1 <= 15 && !jit->seen_reg[r1]) - jit->seen_reg[r1] = 1; + if (r1 >= 6 && r1 <= 15) + jit->seen_regs |= (1 << r1); }
#define REG_SET_SEEN(b1) \ @@ -127,8 +127,6 @@ static inline void reg_set_seen(struct b reg_set_seen(jit, b1); \ })
-#define REG_SEEN(b1) jit->seen_reg[reg2hex[(b1)]] - /* * EMIT macros for code generation */ @@ -436,12 +434,12 @@ static void restore_regs(struct bpf_jit /* * Return first seen register (from start) */ -static int get_start(struct bpf_jit *jit, int start) +static int get_start(u16 seen_regs, int start) { int i;
for (i = start; i <= 15; i++) { - if (jit->seen_reg[i]) + if (seen_regs & (1 << i)) return i; } return 0; @@ -450,15 +448,15 @@ static int get_start(struct bpf_jit *jit /* * Return last seen register (from start) (gap >= 2) */ -static int get_end(struct bpf_jit *jit, int start) +static int get_end(u16 seen_regs, int start) { int i;
for (i = start; i < 15; i++) { - if (!jit->seen_reg[i] && !jit->seen_reg[i + 1]) + if (!(seen_regs & (3 << i))) return i - 1; } - return jit->seen_reg[15] ? 15 : 14; + return (seen_regs & (1 << 15)) ? 15 : 14; }
#define REGS_SAVE 1 @@ -467,8 +465,10 @@ static int get_end(struct bpf_jit *jit, * Save and restore clobbered registers (6-15) on stack. * We save/restore registers in chunks with gap >= 2 registers. */ -static void save_restore_regs(struct bpf_jit *jit, int op, u32 stack_depth) +static void save_restore_regs(struct bpf_jit *jit, int op, u32 stack_depth, + u16 extra_regs) { + u16 seen_regs = jit->seen_regs | extra_regs; const int last = 15, save_restore_size = 6; int re = 6, rs;
@@ -482,10 +482,10 @@ static void save_restore_regs(struct bpf }
do { - rs = get_start(jit, re); + rs = get_start(seen_regs, re); if (!rs) break; - re = get_end(jit, rs + 1); + re = get_end(seen_regs, rs + 1); if (op == REGS_SAVE) save_regs(jit, rs, re); else @@ -579,7 +579,7 @@ static void bpf_jit_prologue(struct bpf_ /* Tail calls have to skip above initialization */ jit->tail_call_start = jit->prg; /* Save registers */ - save_restore_regs(jit, REGS_SAVE, stack_depth); + save_restore_regs(jit, REGS_SAVE, stack_depth, 0); /* Setup literal pool */ if (is_first_pass(jit) || (jit->seen & SEEN_LITERAL)) { if (!is_first_pass(jit) && @@ -653,7 +653,7 @@ static void bpf_jit_epilogue(struct bpf_ /* Load exit code: lgr %r2,%b0 */ EMIT4(0xb9040000, REG_2, BPF_REG_0); /* Restore registers */ - save_restore_regs(jit, REGS_RESTORE, stack_depth); + save_restore_regs(jit, REGS_RESTORE, stack_depth, 0); if (nospec_uses_trampoline()) { jit->r14_thunk_ip = jit->prg; /* Generate __s390_indirect_jump_r14 thunk */ @@ -1519,7 +1519,7 @@ static noinline int bpf_jit_insn(struct /* * Restore registers before calling function */ - save_restore_regs(jit, REGS_RESTORE, stack_depth); + save_restore_regs(jit, REGS_RESTORE, stack_depth, 0);
/* * goto *(prog->bpf_func + tail_call_start);