On Tue, Apr 02, 2024 at 04:28:06PM +0530, Hari Bathini wrote:
> Currently, bpf jit code on powerpc assumes all the bpf functions and
> helpers to be kernel text. This is false for kfunc case, as function
> addresses can be module addresses as well. So, ensure module addresses
> are supported to enable kfunc support.
>
> Emit instructions based on whether the function address is kernel text
> address or module address to retain optimized instruction sequence for
> kernel text address case.
>
> Also, as bpf programs are always module addresses and a bpf helper can
> be within kernel address as well, using relative addressing often fails
> with "out of range of pcrel address" error. Use unoptimized instruction
> sequence for both kernel and module addresses to work around this, when
> PCREL addressing is used.
I guess we need a fixes tag for this?
Fixes: 7e3a68be42e1 ("powerpc/64: vmlinux support building with PCREL addresing")
It will be good to separate out this fix into a separate patch.
Also, I know I said we could use the generic PPC_LI64() for pcrel, but
we may be able to use a more optimized sequence when calling bpf kernel
helpers. See stub_insns[] in module_64.c for an example where we load
paca->kernelbase, then use a prefixed load instruction to populate the
lower 34-bit value. For calls out to module area, we can use the generic
PPC_LI64() macro only if it is outside range of a prefixed load
instruction.
>
> With module addresses supported, override bpf_jit_supports_kfunc_call()
> to enable kfunc support. Since module address offsets can be more than
> 32-bit long on PPC64, override bpf_jit_supports_far_kfunc_call() to
> enable 64-bit pointers.
>
> Signed-off-by: Hari Bathini <hbathini@xxxxxxxxxxxxx>
> ---
>
> * Changes in v3:
> - Retained optimized instruction sequence when function address is
> a core kernel address as suggested by Naveen.
> - Used unoptimized instruction sequence for PCREL addressing to
> avoid out of range errors for core kernel function addresses.
> - Folded patch that adds support for kfunc calls with patch that
> enables/advertises this support as suggested by Naveen.
>
>
> arch/powerpc/net/bpf_jit_comp.c | 10 +++++++
> arch/powerpc/net/bpf_jit_comp64.c | 48 ++++++++++++++++++++-----------
> 2 files changed, 42 insertions(+), 16 deletions(-)
>
> diff --git a/arch/powerpc/net/bpf_jit_comp.c b/arch/powerpc/net/bpf_jit_comp.c
> index 0f9a21783329..dc7ffafd7441 100644
> --- a/arch/powerpc/net/bpf_jit_comp.c
> +++ b/arch/powerpc/net/bpf_jit_comp.c
> @@ -359,3 +359,13 @@ void bpf_jit_free(struct bpf_prog *fp)
>
> bpf_prog_unlock_free(fp);
> }
> +
> +bool bpf_jit_supports_kfunc_call(void)
> +{
> + return true;
> +}
> +
> +bool bpf_jit_supports_far_kfunc_call(void)
> +{
> + return IS_ENABLED(CONFIG_PPC64) ? true : false;
> +}
> diff --git a/arch/powerpc/net/bpf_jit_comp64.c
> b/arch/powerpc/net/bpf_jit_comp64.c
> index 7f62ac4b4e65..ec3adf715c55 100644
> --- a/arch/powerpc/net/bpf_jit_comp64.c
> +++ b/arch/powerpc/net/bpf_jit_comp64.c
> @@ -207,24 +207,14 @@ static int bpf_jit_emit_func_call_hlp(u32 *image, struct codegen_context *ctx, u
> unsigned long func_addr = func ? ppc_function_entry((void *)func) : 0;
> long reladdr;
>
> - if (WARN_ON_ONCE(!core_kernel_text(func_addr)))
> + /*
> + * With the introduction of kfunc feature, BPF helpers can be part of kernel as
> + * well as module text address.
> + */
> + if (WARN_ON_ONCE(!kernel_text_address(func_addr)))
> return -EINVAL;
>
> - if (IS_ENABLED(CONFIG_PPC_KERNEL_PCREL)) {
> - reladdr = func_addr - CTX_NIA(ctx);
> -
> - if (reladdr >= (long)SZ_8G || reladdr < -(long)SZ_8G) {
> - pr_err("eBPF: address of %ps out of range of pcrel address.\n",
> - (void *)func);
> - return -ERANGE;
> - }
> - /* pla r12,addr */
> - EMIT(PPC_PREFIX_MLS | __PPC_PRFX_R(1) | IMM_H18(reladdr));
> - EMIT(PPC_INST_PADDI | ___PPC_RT(_R12) | IMM_L(reladdr));
> - EMIT(PPC_RAW_MTCTR(_R12));
> - EMIT(PPC_RAW_BCTR());
> -
> - } else {
> + if (core_kernel_text(func_addr) && !IS_ENABLED(CONFIG_PPC_KERNEL_PCREL)) {
> reladdr = func_addr - kernel_toc_addr();
> if (reladdr > 0x7FFFFFFF || reladdr < -(0x80000000L)) {
> pr_err("eBPF: address of %ps out of range of kernel_toc.\n", (void *)func);
> @@ -235,6 +225,32 @@ static int bpf_jit_emit_func_call_hlp(u32 *image, struct codegen_context *ctx, u
> EMIT(PPC_RAW_ADDI(_R12, _R12, PPC_LO(reladdr)));
> EMIT(PPC_RAW_MTCTR(_R12));
> EMIT(PPC_RAW_BCTRL());
> + } else {
> + if (IS_ENABLED(CONFIG_PPC64_ELF_ABI_V1)) {
> + /* func points to the function descriptor */
> + PPC_LI64(bpf_to_ppc(TMP_REG_2), func);
> + /* Load actual entry point from function descriptor */
> + EMIT(PPC_RAW_LD(bpf_to_ppc(TMP_REG_1), bpf_to_ppc(TMP_REG_2), 0));
> + /* ... and move it to CTR */
> + EMIT(PPC_RAW_MTCTR(bpf_to_ppc(TMP_REG_1)));
> + /*
> + * Load TOC from function descriptor at offset 8.
> + * We can clobber r2 since we get called through a
> + * function pointer (so caller will save/restore r2)
> + * and since we don't use a TOC ourself.
> + */
> + EMIT(PPC_RAW_LD(2, bpf_to_ppc(TMP_REG_2), 8));
> + EMIT(PPC_RAW_BCTRL());
I thought we started using TOC for ABIv1 when we moved to using an
optimized function call sequence for bpf helpers? If so, we will need
to load kernel toc here for subsequent calls to BPF helpers.
> + } else {
> + /* We can clobber r12 */
> + PPC_LI64(12, func);
> + EMIT(PPC_RAW_MTCTR(12));
> + EMIT(PPC_RAW_BCTRL());
> +#ifndef CONFIG_PPC_KERNEL_PCREL
> + /* Restore kernel TOC */
> + EMIT(PPC_RAW_LD(2, 13, offsetof(struct paca_struct, kernel_toc)));
> +#endif
> + }
> }
>
> return 0;
> --
> 2.44.0
>
- Naveen
