On Mon, Nov 4, 2024 at 12:05 PM Mathias Krause <minipli@xxxxxxxxxxxxxx> wrote:
>
> The "staggered jumps" tests currently fail with constant blinding
> enabled as the increased program size makes jump offsets overflow.
>
> Fix that by decreasing the number of jumps depending on the expected
> size increase caused by blinding the program.
>
> As the test for JIT blinding makes use of bpf_jit_blinding_enabled(NULL)
> and test_bpf.ko is a kernel modules, 'bpf_token_capable' and
> 'bpf_jit_harden' need to be exported.
>
> Fixes: a7d2e752e520 ("bpf/tests: Add staggered JMP and JMP32 tests")
> Cc: Johan Almbladh <johan.almbladh@xxxxxxxxxxxxxxxxx>
> Signed-off-by: Mathias Krause <minipli@xxxxxxxxxxxxxx>
> ---
> kernel/bpf/core.c | 3 +++
> kernel/bpf/token.c | 3 +++
> lib/test_bpf.c | 19 +++++++++++++++++--
> 3 files changed, 23 insertions(+), 2 deletions(-)
>
> diff --git a/kernel/bpf/core.c b/kernel/bpf/core.c
> index 233ea78f8f1b..fe7eada54d4b 100644
> --- a/kernel/bpf/core.c
> +++ b/kernel/bpf/core.c
> @@ -570,6 +570,9 @@ int bpf_jit_kallsyms __read_mostly = IS_BUILTIN(CONFIG_BPF_JIT_DEFAULT_ON);
> int bpf_jit_harden __read_mostly;
> long bpf_jit_limit __read_mostly;
> long bpf_jit_limit_max __read_mostly;
> +#if IS_MODULE(CONFIG_TEST_BPF)
> +EXPORT_SYMBOL_GPL(bpf_jit_harden);
> +#endif
>
> static void
> bpf_prog_ksym_set_addr(struct bpf_prog *prog)
> diff --git a/kernel/bpf/token.c b/kernel/bpf/token.c
> index dcbec1a0dfb3..aed98a958c73 100644
> --- a/kernel/bpf/token.c
> +++ b/kernel/bpf/token.c
> @@ -26,6 +26,9 @@ bool bpf_token_capable(const struct bpf_token *token, int cap)
> return false;
> return true;
> }
> +#if IS_MODULE(CONFIG_TEST_BPF)
> +EXPORT_SYMBOL_GPL(bpf_token_capable);
> +#endif
I don't like the extra exports and hack in patch 2
just to test jit blinding.
In general lib/test_bpf is there for testing JITs
path that are impossible to do with normal asm/c from selftests.
I don't think jit blinding false into this category.
The current code coverage is good enough.
pw-bot: cr
> void bpf_token_inc(struct bpf_token *token)
> {
> diff --git a/lib/test_bpf.c b/lib/test_bpf.c
> index c1140bab280d..3469631c0aba 100644
> --- a/lib/test_bpf.c
> +++ b/lib/test_bpf.c
> @@ -2700,10 +2700,25 @@ static int __bpf_fill_staggered_jumps(struct bpf_test *self,
> u64 r1, u64 r2)
> {
> int size = self->test[0].result - 1;
> - int len = 4 + 3 * (size + 1);
> struct bpf_insn *insns;
> - int off, ind;
> + int len, off, ind;
>
> + /* Constant blinding triples the size of each instruction making use
> + * of immediate values. Tweak the test to not overflow jump offsets.
> + */
> + if (bpf_jit_blinding_enabled(NULL)) {
> + int bloat_factor = 2 * 3;
> +
> + if (BPF_SRC(jmp->code) == BPF_K)
> + bloat_factor += 3;
> +
> + size /= bloat_factor;
> + size &= ~1;
> +
> + self->test[0].result = size + 1;
> + }
> +
> + len = 4 + 3 * (size + 1);
> insns = kmalloc_array(len, sizeof(*insns), GFP_KERNEL);
> if (!insns)
> return -ENOMEM;
> --
> 2.30.2
>
