Re: [PATCH v2 bpf-next 2/7] bpf: use bytes instead of pages for bpf_jit_[charge|uncharge]_modmem

On Dec 15, 2021, at 12:56 AM, Peter Zijlstra [off-list ref] wrote:

On Tue, Dec 14, 2021 at 10:00:57PM -0800, Song Liu wrote:

quoted

From: Song Liu <redacted>

This enables sub-page memory charge and allocation.

Signed-off-by: Song Liu <redacted>
---
include/linux/bpf.h     |  4 ++--
kernel/bpf/core.c       | 19 +++++++++----------
kernel/bpf/trampoline.c |  6 +++---
3 files changed, 14 insertions(+), 15 deletions(-)

diff --git a/include/linux/bpf.h b/include/linux/bpf.h
index 965fffaf0308..adcdda0019f1 100644
--- a/include/linux/bpf.h
+++ b/include/linux/bpf.h

@@ -775,8 +775,8 @@ void bpf_image_ksym_add(void *data, struct bpf_ksym *ksym);

void bpf_image_ksym_del(struct bpf_ksym *ksym);
void bpf_ksym_add(struct bpf_ksym *ksym);
void bpf_ksym_del(struct bpf_ksym *ksym);
-int bpf_jit_charge_modmem(u32 pages);
-void bpf_jit_uncharge_modmem(u32 pages);
+int bpf_jit_charge_modmem(u32 size);
+void bpf_jit_uncharge_modmem(u32 size);
bool bpf_prog_has_trampoline(const struct bpf_prog *prog);
#else
static inline int bpf_trampoline_link_prog(struct bpf_prog *prog,

diff --git a/kernel/bpf/core.c b/kernel/bpf/core.c
index de3e5bc6781f..495e3b2c36ff 100644
--- a/kernel/bpf/core.c
+++ b/kernel/bpf/core.c

@@ -808,7 +808,7 @@ int bpf_jit_add_poke_descriptor(struct bpf_prog *prog,

	return slot;
}

-static atomic_long_t bpf_jit_current;
+static atomic64_t bpf_jit_current;

atomic64_t is atrocious crap on much of 32bit. I suppose it doesn't
matter since this is slow path accounting?

Yeah, speed shouldn't matter for bpf_jit_charge|uncharge(). 

Thanks,
Song