Re: [PATCH] net: filter: BPF 'JIT' compiler for PPC64

Le lundi 18 juillet 2011 à 17:50 +1000, Matt Evans a écrit :

An implementation of a code generator for BPF programs to speed up packet
filtering on PPC64, inspired by Eric Dumazet's x86-64 version.

Filter code is generated as an ABI-compliant function in module_alloc()'d mem
with stackframe & prologue/epilogue generated if required (simple filters don't
need anything more than an li/blr).  The filter's local variables, M[], live in
registers.  Supports all BPF opcodes, although "complicated" loads from negative
packet offsets (e.g. SKF_LL_OFF) are not yet supported.

There are a couple of further optimisations left for future work; many-pass
assembly with branch-reach reduction and a register allocator to push M[]
variables into volatile registers would improve the code quality further.

This currently supports big-endian 64-bit PowerPC only (but is fairly simple
to port to PPC32 or LE!).

Enabled in the same way as x86-64:

	echo 1 > /proc/sys/net/core/bpf_jit_enable

Or, enabled with extra debug output:

	echo 2 > /proc/sys/net/core/bpf_jit_enable

Signed-off-by: Matt Evans <redacted>
---

Since the RFC post, this has incorporated the bugfixes/tidies from review plus a
couple more found in further testing, plus some general/comment tidies.

Hi Matt

A small note about SEEN_XREG usage in PPC against x86_64 :

In x86_64, XREG is stored in EBX : I had to save/restore it in function
prologue epilogue. And set it to zero in prologue to avoid leak of
kernel information.

In PPC, you chose a scratch register, so you only have to zero it in
function prologue, if X is ever read.

So in PPC SEEN_XREG only is to be set of X is read, not if written.

So you dont have to set SEEN_XREG bit in this part :

+		case BPF_S_MISC_TAX: /* X = A */
+			ctx->seen |= SEEN_XREG;
+			PPC_MR(r_X, r_A);
+			break;

and in this part :

+		case BPF_S_LDX_IMM: /* X = K */
+			ctx->seen |= SEEN_XREG;
+			PPC_LI32(r_X, K);
+			break;

and :

+		case BPF_S_LDX_MEM: /* X = mem[K] */
+			PPC_MR(r_X, r_M + (K & 0xf));
+			ctx->seen |= SEEN_XREG | SEEN_MEM | (1<<(K & 0xf));
+			break;

and :

+		case BPF_S_LDX_W_LEN: /* X = skb->len; */
+			ctx->seen |= SEEN_XREG;
+			PPC_LWZ_OFFS(r_X, r_skb, offsetof(struct sk_buff, len));
+			break;
+