| Message ID | 20230925215324.2962716-1-song@kernel.org (mailing list archive) |
|---|---|
| Headers | show |
| Series | Allocate bpf trampoline on bpf_prog_pack | expand |
On Mon, 2023-09-25 at 14:53 -0700, Song Liu wrote: > This set enables allocating bpf trampoline from bpf_prog_pack on x86. > The > majority of this work, however, is the refactoring of trampoline > code. > This is needed because we need to handle 4 archs and 2 users > (trampoline > and struct_ops). > > 1/8 is a dependency that is already applied to bpf tree. > 2/8 through 7/8 refactors trampoline code. A few helpers are added. > 8/8 finally let bpf trampoline on x86 use bpf_prog_pack. > > Changes in v2: > 1. Add missing changes in net/bpf/bpf_dummy_struct_ops.c. > 2. Reduce one dry run in arch_prepare_bpf_trampoline. (Xu Kuohai) > 3. Other small fixes. > > Song Liu (8): > s390/bpf: Let arch_prepare_bpf_trampoline return program size > bpf: Let bpf_prog_pack_free handle any pointer > bpf: Adjust argument names of arch_prepare_bpf_trampoline() > bpf: Add helpers for trampoline image management > bpf, x86: Adjust arch_prepare_bpf_trampoline return value > bpf: Add arch_bpf_trampoline_size() > bpf: Use arch_bpf_trampoline_size > x86, bpf: Use bpf_prog_pack for bpf trampoline > > arch/arm64/net/bpf_jit_comp.c | 55 +++++++++----- > arch/riscv/net/bpf_jit_comp64.c | 24 ++++--- > arch/s390/net/bpf_jit_comp.c | 43 ++++++----- > arch/x86/net/bpf_jit_comp.c | 124 +++++++++++++++++++++++++----- > -- > include/linux/bpf.h | 12 +++- > include/linux/filter.h | 2 +- > kernel/bpf/bpf_struct_ops.c | 19 +++-- > kernel/bpf/core.c | 21 +++--- > kernel/bpf/dispatcher.c | 5 +- > kernel/bpf/trampoline.c | 93 ++++++++++++++++++------ > net/bpf/bpf_dummy_struct_ops.c | 7 +- > 11 files changed, 277 insertions(+), 128 deletions(-) > > -- > 2.34.1 Regarding the s390x part, arch_prepare_bpf_trampoline() needs to call __arch_prepare_bpf_trampoline() twice: the first time in order to compute various offsets, the second time to actually emit the code. So I would suggest to either keep the loop or use the following fixup: --- a/arch/s390/net/bpf_jit_comp.c +++ b/arch/s390/net/bpf_jit_comp.c @@ -2645,7 +2645,15 @@ int arch_prepare_bpf_trampoline(struct bpf_tramp_image *im, void *image, struct bpf_tramp_jit tjit; int ret; + /* Compute offsets. */ memset(&tjit, 0, sizeof(tjit)); + ret = __arch_prepare_bpf_trampoline(im, &tjit, m, flags, + tlinks, func_addr); + if (ret < 0) + return ret; + + /* Generate the code. */ + tjit.common.prg = 0; tjit.common.prg_buf = image; ret = __arch_prepare_bpf_trampoline(im, &tjit, m, flags, tlinks, func_addr); With that: Acked-by: Ilya Leoshkevich <iii@linux.ibm.com> Tested-by: Ilya Leoshkevich <iii@linux.ibm.com> # on s390x for the series.
> On Sep 26, 2023, at 5:26 AM, Ilya Leoshkevich <iii@linux.ibm.com> wrote: > > On Mon, 2023-09-25 at 14:53 -0700, Song Liu wrote: >> This set enables allocating bpf trampoline from bpf_prog_pack on x86. >> The >> majority of this work, however, is the refactoring of trampoline >> code. >> This is needed because we need to handle 4 archs and 2 users >> (trampoline >> and struct_ops). >> >> 1/8 is a dependency that is already applied to bpf tree. >> 2/8 through 7/8 refactors trampoline code. A few helpers are added. >> 8/8 finally let bpf trampoline on x86 use bpf_prog_pack. >> >> Changes in v2: >> 1. Add missing changes in net/bpf/bpf_dummy_struct_ops.c. >> 2. Reduce one dry run in arch_prepare_bpf_trampoline. (Xu Kuohai) >> 3. Other small fixes. >> >> Song Liu (8): >> s390/bpf: Let arch_prepare_bpf_trampoline return program size >> bpf: Let bpf_prog_pack_free handle any pointer >> bpf: Adjust argument names of arch_prepare_bpf_trampoline() >> bpf: Add helpers for trampoline image management >> bpf, x86: Adjust arch_prepare_bpf_trampoline return value >> bpf: Add arch_bpf_trampoline_size() >> bpf: Use arch_bpf_trampoline_size >> x86, bpf: Use bpf_prog_pack for bpf trampoline >> >> arch/arm64/net/bpf_jit_comp.c | 55 +++++++++----- >> arch/riscv/net/bpf_jit_comp64.c | 24 ++++--- >> arch/s390/net/bpf_jit_comp.c | 43 ++++++----- >> arch/x86/net/bpf_jit_comp.c | 124 +++++++++++++++++++++++++----- >> -- >> include/linux/bpf.h | 12 +++- >> include/linux/filter.h | 2 +- >> kernel/bpf/bpf_struct_ops.c | 19 +++-- >> kernel/bpf/core.c | 21 +++--- >> kernel/bpf/dispatcher.c | 5 +- >> kernel/bpf/trampoline.c | 93 ++++++++++++++++++------ >> net/bpf/bpf_dummy_struct_ops.c | 7 +- >> 11 files changed, 277 insertions(+), 128 deletions(-) >> >> -- >> 2.34.1 > > Regarding the s390x part, arch_prepare_bpf_trampoline() needs to call > __arch_prepare_bpf_trampoline() twice: the first time in order to > compute various offsets, the second time to actually emit the code. So > I would suggest to either keep the loop or use the following fixup: Thanks for the test and the fix! I will fold the fix in and send v3. Song > > --- a/arch/s390/net/bpf_jit_comp.c > +++ b/arch/s390/net/bpf_jit_comp.c > @@ -2645,7 +2645,15 @@ int arch_prepare_bpf_trampoline(struct > bpf_tramp_image *im, void *image, > struct bpf_tramp_jit tjit; > int ret; > > + /* Compute offsets. */ > memset(&tjit, 0, sizeof(tjit)); > + ret = __arch_prepare_bpf_trampoline(im, &tjit, m, flags, > + tlinks, func_addr); > + if (ret < 0) > + return ret; > + > + /* Generate the code. */ > + tjit.common.prg = 0; > tjit.common.prg_buf = image; > ret = __arch_prepare_bpf_trampoline(im, &tjit, m, flags, > tlinks, func_addr); > > With that: > > Acked-by: Ilya Leoshkevich <iii@linux.ibm.com> > Tested-by: Ilya Leoshkevich <iii@linux.ibm.com> # on s390x > > for the series.