RISC-V/bpf: Enable bpf_probe_read{, str}()
Subsystems (4)
PATCH 1/1 RISC-V/bpf: Enable bpf_probe_read{, str}()
Enable this option to fix a bcc error in RISC-V platform
And, the error shows as follows:
~ # runqlen
WARNING: This target JIT is not designed for the host you are running. \
If bad things happen, please choose a different -march switch.
bpf: Failed to load program: Invalid argument
0: R1=ctx(off=0,imm=0) R10=fp0
0: (85) call bpf_get_current_task#35 ; R0_w=scalar()
1: (b7) r6 = 0 ; R6_w=0 2: (7b) *(u64 *)(r10 -8) = r6 ; R6_w=P0 R10=fp0 fp-8_w=00000000 3: (07) r0 += 312 ; R0_w=scalar() 4: (bf) r1 = r10 ; R1_w=fp0 R10=fp0 5: (07) r1 += -8 ; R1_w=fp-8 6: (b7) r2 = 8 ; R2_w=8 7: (bf) r3 = r0 ; R0_w=scalar(id=1) R3_w=scalar(id=1)
8: (85) call bpf_probe_read#4
unknown func bpf_probe_read#4
processed 9 insns (limit 1000000) max_states_per_insn 0 total_states 0 peak_states 0 mark_read 0
Traceback (most recent call last):
File "/usr/lib/python-exec/python3.9/runqlen", line 187, in <module>
b.attach_perf_event(ev_type=PerfType.SOFTWARE,
File "/usr/lib/python3.9/site-packages/bcc/__init__.py", line 1228, in attach_perf_event
fn = self.load_func(fn_name, BPF.PERF_EVENT)
File "/usr/lib/python3.9/site-packages/bcc/__init__.py", line 522, in load_func
raise Exception("Failed to load BPF program %s: %s" %
Exception: Failed to load BPF program b'do_perf_event': Invalid argument
--- arch/riscv/Kconfig | 1 + 1 file changed, 1 insertion(+)
arch/riscv/Kconfig +1 -0
@@ -25,6 +25,7 @@ config RISCV select ARCH_HAS_GIGANTIC_PAGE select ARCH_HAS_KCOV select ARCH_HAS_MMIOWB + select ARCH_HAS_NON_OVERLAPPING_ADDRESS_SPACE select ARCH_HAS_PTE_SPECIAL select ARCH_HAS_SET_DIRECT_MAP if MMU select ARCH_HAS_SET_MEMORY if MMU
Signature · 1 line
2.35.1
Quoted 30 lines · Yixun Lan
Enable this option to fix a bcc error in RISC-V platform
And, the error shows as follows:
~ # runqlen
WARNING: This target JIT is not designed for the host you are running. \
If bad things happen, please choose a different -march switch.
bpf: Failed to load program: Invalid argument
0: R1=ctx(off=0,imm=0) R10=fp0
0: (85) call bpf_get_current_task#35 ; R0_w=scalar()1: (b7) r6 = 0 ; R6_w=0 2: (7b) *(u64 *)(r10 -8) = r6 ; R6_w=P0 R10=fp0 fp-8_w=00000000 3: (07) r0 += 312 ; R0_w=scalar() 4: (bf) r1 = r10 ; R1_w=fp0 R10=fp0 5: (07) r1 += -8 ; R1_w=fp-8 6: (b7) r2 = 8 ; R2_w=8 7: (bf) r3 = r0 ; R0_w=scalar(id=1) R3_w=scalar(id=1)8: (85) call bpf_probe_read#4
unknown func bpf_probe_read#4
processed 9 insns (limit 1000000)max_states_per_insn0total_states0peak_states0mark_read0Traceback (most recent call last):
File "/usr/lib/python-exec/python3.9/runqlen", line 187, in <module>
b.attach_perf_event(ev_type=PerfType.SOFTWARE,
File "/usr/lib/python3.9/site-packages/bcc/__init__.py", line 1228, inattach_perf_event
fn =self.load_func(fn_name, BPF.PERF_EVENT)
File "/usr/lib/python3.9/site-packages/bcc/__init__.py", line 522, inload_func
raise Exception("Failed to load BPF program %s: %s" %
Exception: Failed to load BPF program b'do_perf_event': Invalid argument
Quoted 1 line
Signed-off-by Yixun Lan <[email protected]>
Do you know what commit this fixes?
Thanks,
Conor.
Quoted 16 lines
--- arch/riscv/Kconfig | 1 + 1 file changed, 1 insertion(+)arch/riscv/Kconfig +1
-0@@ -25,6 +25,7 @@ config RISCV select ARCH_HAS_GIGANTIC_PAGE select ARCH_HAS_KCOV select ARCH_HAS_MMIOWB + select ARCH_HAS_NON_OVERLAPPING_ADDRESS_SPACE select ARCH_HAS_PTE_SPECIAL select ARCH_HAS_SET_DIRECT_MAP if MMU select ARCH_HAS_SET_MEMORY if MMU
Hi Conor Dooley:
Quoted 31 lines · Conor Dooley
Quoted 30 lines · Yixun Lan
Enable this option to fix a bcc error in RISC-V platform
And, the error shows as follows:
~ # runqlen
WARNING: This target JIT is not designed for the host you are running. \
If bad things happen, please choose a different -march switch.
bpf: Failed to load program: Invalid argument
0: R1=ctx(off=0,imm=0) R10=fp0
0: (85) call bpf_get_current_task#35 ; R0_w=scalar()1: (b7) r6 = 0 ; R6_w=0 2: (7b) *(u64 *)(r10 -8) = r6 ; R6_w=P0 R10=fp0 fp-8_w=00000000 3: (07) r0 += 312 ; R0_w=scalar() 4: (bf) r1 = r10 ; R1_w=fp0 R10=fp0 5: (07) r1 += -8 ; R1_w=fp-8 6: (b7) r2 = 8 ; R2_w=8 7: (bf) r3 = r0 ; R0_w=scalar(id=1) R3_w=scalar(id=1)8: (85) call bpf_probe_read#4
unknown func bpf_probe_read#4
processed 9 insns (limit 1000000)max_states_per_insn0total_states0peak_states0mark_read0Traceback (most recent call last):
File "/usr/lib/python-exec/python3.9/runqlen", line 187, in <module>
b.attach_perf_event(ev_type=PerfType.SOFTWARE,
File "/usr/lib/python3.9/site-packages/bcc/__init__.py", line 1228, inattach_perf_event
fn =self.load_func(fn_name, BPF.PERF_EVENT)
File "/usr/lib/python3.9/site-packages/bcc/__init__.py", line 522, inload_func
raise Exception("Failed to load BPF program %s: %s" %
Exception: Failed to load BPF program b'do_perf_event': Invalid argument
Quoted 6 lines
Quoted 1 line
Signed-off-by Yixun Lan <[email protected]>Do you know what commit this fixes?
Thanks,
Conor.
I think this is effectively broken for RISC-V 64 at the commit:
0ebeea8ca8a4: bpf: Restrict bpf_probe_read{, str}() only to archs where they work
However, the bcc tools haven't got BPF support for RISC-V at that time, so no one noticed it
I can add a Fixes tag if you think it's a proper way
Quoted 16 lines
Quoted 16 lines
--- arch/riscv/Kconfig | 1 + 1 file changed, 1 insertion(+)arch/riscv/Kconfig +1
-0@@ -25,6 +25,7 @@ config RISCV select ARCH_HAS_GIGANTIC_PAGE select ARCH_HAS_KCOV select ARCH_HAS_MMIOWB + select ARCH_HAS_NON_OVERLAPPING_ADDRESS_SPACE select ARCH_HAS_PTE_SPECIAL select ARCH_HAS_SET_DIRECT_MAP if MMU select ARCH_HAS_SET_MEMORY if MMU
Signature · 3 lines
Yixun Lan (dlan) Gentoo Linux Developer GPG Key ID AABEFD55
Quoted 45 lines · Yixun Lan
[You don't often get email from [email protected]. Learn why this is important at https://aka.ms/LearnAboutSenderIdentification ]
EXTERNAL EMAIL: Do not click links or open attachments unless you know the content is safe
Hi Conor Dooley:
Quoted 31 lines · Conor Dooley
Quoted 30 lines · Yixun Lan
Enable this option to fix a bcc error in RISC-V platform
And, the error shows as follows:
~ # runqlen
WARNING: This target JIT is not designed for the host you are running. \
If bad things happen, please choose a different -march switch.
bpf: Failed to load program: Invalid argument
0: R1=ctx(off=0,imm=0) R10=fp0
0: (85) call bpf_get_current_task#35 ; R0_w=scalar()1: (b7) r6 = 0 ; R6_w=0 2: (7b) *(u64 *)(r10 -8) = r6 ; R6_w=P0 R10=fp0 fp-8_w=00000000 3: (07) r0 += 312 ; R0_w=scalar() 4: (bf) r1 = r10 ; R1_w=fp0 R10=fp0 5: (07) r1 += -8 ; R1_w=fp-8 6: (b7) r2 = 8 ; R2_w=8 7: (bf) r3 = r0 ; R0_w=scalar(id=1) R3_w=scalar(id=1)8: (85) call bpf_probe_read#4
unknown func bpf_probe_read#4
processed 9 insns (limit 1000000)max_states_per_insn0total_states0peak_states0mark_read0Traceback (most recent call last):
File "/usr/lib/python-exec/python3.9/runqlen", line 187, in <module> b.attach_perf_event(ev_type=PerfType.SOFTWARE, File "/usr/lib/python3.9/site-packages/bcc/__init__.py", line 1228, in attach_perf_event fn = self.load_func(fn_name, BPF.PERF_EVENT) File "/usr/lib/python3.9/site-packages/bcc/__init__.py", line 522, in load_func raise Exception("Failed to load BPF program %s: %s" %Exception: Failed to load BPF program b'do_perf_event': Invalid argument
Quoted 6 lines
Quoted 1 line
Signed-off-by Yixun Lan <[email protected]>Do you know what commit this fixes?
Thanks,
Conor.
Quoted 7 lines
I think this is effectively broken for RISC-V 64 at the commit:
0ebeea8ca8a4: bpf: Restrict bpf_probe_read{, str}() only to archs where they workHowever, the bcc tools haven't got BPF support for RISC-V at that time, so no one noticed it
I can add a Fixes tag if you think it's a proper way
Hmm, I had a read of that commit and the breakage sounded intentional. I ran a git log --grep "0ebeea8ca8a4" & it seems like a mixed bag of fixes tags. Whether or not it deserves the explicit tag, mentioning the commit would be useful. The tag would be:
Thanks,
Conor.
Quoted 27 lines
Quoted 5 lines
Quoted 5 lines
--- arch/riscv/Kconfig | 1 + 1 file changed, 1 insertion(+)Quoted 12 lines
arch/riscv/Kconfig config RISCV · line 31
Quoted 12 lines
arch/riscv/Kconfig +1
-0@@ -25,6 +25,7 @@ config RISCV select ARCH_HAS_GIGANTIC_PAGE select ARCH_HAS_KCOV select ARCH_HAS_MMIOWB + select ARCH_HAS_NON_OVERLAPPING_ADDRESS_SPACE select ARCH_HAS_PTE_SPECIAL select ARCH_HAS_SET_DIRECT_MAP if MMU select ARCH_HAS_SET_MEMORY if MMU--
Yixun Lan (dlan)
Gentoo Linux Developer
GPG Key ID AABEFD55_______________________________________________
linux-riscv mailing list
[email protected]
http://lists.infradead.org/mailman/listinfo/linux-riscv
Quoted 3 lines · Yixun Lan
Enable this option to fix a bcc error in RISC-V platform
And, the error shows as follows:
These should not be enabled on new platforms. Use the proper helpers to probe kernel vs user pointers instead.
_______________________________________________
linux-riscv mailing list
[email protected]
http://lists.infradead.org/mailman/listinfo/linux-riscv
Quoted 8 lines · Christoph Hellwig
Quoted 3 lines · Yixun Lan
Enable this option to fix a bcc error in RISC-V platform
And, the error shows as follows:
These should not be enabled on new platforms. Use the proper helpers to probe kernel vs user pointers instead.
riscv existed as of [0], so I'd argue it is a proper bug fix, as corresponding select ARCH_HAS_NON_OVERLAPPING_ADDRESS_SPACE should have been added back then.
But I also agree that BCC tools should be updated to use proper bpf_probe_read_{kernel,user}[_str()] helpers, please contribute such fixes to BCC tools and BCC itself as well. Cc'ed Alan as his ksnoop in libbpf-tools seems to be using bpf_probe_read() as well and needs to be fixed.
[0] 0ebeea8ca8a4 ("bpf: Restrict bpf_probe_read{, str}() only to
archs where they work")
Quoted 14 lines · Andrii Nakryiko
Quoted 8 lines · Christoph Hellwig
Quoted 3 lines · Yixun Lan
Enable this option to fix a bcc error in RISC-V platform
And, the error shows as follows:
These should not be enabled on new platforms. Use the proper helpers to probe kernel vs user pointers instead.
riscv existed as of [0], so I'd argue it is a proper bug fix, as corresponding select
ARCH_HAS_NON_OVERLAPPING_ADDRESS_SPACEshould have been added back then.
Quoted 5 lines
But I also agree that BCC tools should be updated to use proper bpf_probe_read_{kernel,user}[_str()] helpers, please contribute such fixes to BCC tools and BCC itself as well. Cc'ed Alan as his ksnoop in libbpf-tools seems to be using
bpf_probe_read()as well and needs to be fixed.
Yixun, the bcc change looks like below:
src/cc/frontends/clang/b_frontend_action.cc +0 -0
@@ -132,7 +132,7 @@ static std::string check_bpf_probe_read_user(llvm::StringRef probe, /* For arch with overlapping address space, dont use
bpf_probe_read for
* user read. Just error out */
-#if defined(__s390x__)
+#if defined(__s390x__) || defined(__riscv_)
overlap_addr = true;
return "";#endif
Basically, prevent using bpf_probe_read() helper, so it will force user to use bpf_probe_read_user() or bpf_probe_read_kernel(). and this should make it work for old kernels.
Quoted 3 lines
[0] 0ebeea8ca8a4 ("bpf: Restrict bpf_probe_read{, str}() only toarchs where they work")
_______________________________________________
linux-riscv mailing list
[email protected]
http://lists.infradead.org/mailman/listinfo/linux-riscv
Quoted 17 lines · Yonghong Song
Quoted 14 lines · Andrii Nakryiko
Quoted 8 lines · Christoph Hellwig
Quoted 3 lines · Yixun Lan
Enable this option to fix a bcc error in RISC-V platform
And, the error shows as follows:
These should not be enabled on new platforms. Use the proper helpers to probe kernel vs user pointers instead.
riscv existed as of [0], so I'd argue it is a proper bug fix, as corresponding select
ARCH_HAS_NON_OVERLAPPING_ADDRESS_SPACEshould have been added back then.
Quoted 7 lines
Quoted 5 lines
But I also agree that BCC tools should be updated to use proper bpf_probe_read_{kernel,user}[_str()] helpers, please contribute such fixes to BCC tools and BCC itself as well. Cc'ed Alan as his ksnoop in libbpf-tools seems to be using
bpf_probe_read()as well and needs to be fixed.Yixun, the bcc change looks like below:
No, this is broken. bcc needs to stop using bpf_probe_read entirely for user addresses and unconditionally use bpf_probe_read_user first and only fall back to bpf_probe_read if not supported.
Hi Christoph, YongHong
Quoted 24 lines · Christoph Hellwig
Quoted 23 lines · Yonghong Song
Quoted 19 lines · Andrii Nakryiko
Quoted 8 lines · Christoph Hellwig
Quoted 3 lines · Yixun Lan
Enable this option to fix a bcc error in RISC-V platform
And, the error shows as follows:
These should not be enabled on new platforms. Use the proper helpers to probe kernel vs user pointers instead.
riscv existed as of [0], so I'd argue it is a proper bug fix, as corresponding select
ARCH_HAS_NON_OVERLAPPING_ADDRESS_SPACEshould have been added back then.But I also agree that BCC tools should be updated to use proper bpf_probe_read_{kernel,user}[_str()] helpers, please contribute such fixes to BCC tools and BCC itself as well. Cc'ed Alan as his ksnoop in libbpf-tools seems to be using
bpf_probe_read()as well and needs to be fixed.
Quoted 5 lines
Quoted 1 line
Yixun, the bcc change looks like below:
No, this is broken. bcc needs to stop using
bpf_probe_readentirely for user addresses and unconditionally usebpf_probe_read_userfirst and only fall back tobpf_probe_readif not supported.
I agree with Christoph, there is something in the bcc tools that need to adjust in order to use new bpf_probe_read_{kernel,user}
Please check the ongoing discussion [0] in the bcc tools if you're interested in, advice and comments are welcome
[0] https://github.com/iovisor/bcc/pull/4085#issuecomment-1179446738
Signature · 3 lines
Yixun Lan (dlan) Gentoo Linux Developer GPG Key ID AABEFD55
Quoted 4 lines · Yixun Lan
Please check the ongoing discussion [0] in the bcc tools if you're interested in, advice and comments are welcome
[0] https://github.com/iovisor/bcc/pull/4085#issuecomment-1179446738
I can't find a way to post there, as replying eems to require a login. Is there a mailing list discussion somewhere that is broadly accessible?
_______________________________________________
linux-riscv mailing list
[email protected]
http://lists.infradead.org/mailman/listinfo/linux-riscv
Hi Christoph:
Quoted 8 lines · Christoph Hellwig
Quoted 4 lines · Yixun Lan
Please check the ongoing discussion [0] in the bcc tools if you're interested in, advice and comments are welcome
[0] https://github.com/iovisor/bcc/pull/4085#issuecomment-1179446738
I can't find a way to post there, as replying eems to require a login. Is there a mailing list discussion somewhere that is broadly accessible?
never mind, I think the logic is quite clear, we can do something in bcc:
1) adopt new _{kernel,user} interface whenever possible, this will work fine for all arch with new kernel versions
2) for old kernel versions which lack the _{kernel,user} support, fall back to old bpf_probe_read(), but take care of the Archs which have overlaping address space - like s390, and just error out for it
Signature · 3 lines
Yixun Lan (dlan) Gentoo Linux Developer GPG Key ID AABEFD55
Quoted 8 lines · Yixun Lan
never mind, I think the logic is quite clear, we can do something in bcc:
1) adopt new _{kernel,user} interface whenever possible, this will work fine for all arch with new kernel versions
2) for old kernel versions which lack the _{kernel,user} support, fall back to old
bpf_probe_read(), but take care of the Archs which have overlaping address space - like s390, and just error out for it
Yes, that is the right thing to do.
_______________________________________________
linux-riscv mailing list
[email protected]
http://lists.infradead.org/mailman/listinfo/linux-riscv
Quoted 3 lines · Andrii Nakryiko
riscv existed as of [0], so I'd argue it is a proper bug fix, as corresponding select
ARCH_HAS_NON_OVERLAPPING_ADDRESS_SPACEshould have been added back then.
How much of an eBPF ecosystem was there on RISC-V at the point?
_______________________________________________
linux-riscv mailing list
[email protected]
http://lists.infradead.org/mailman/listinfo/linux-riscv
Quoted 7 lines · Christoph Hellwig
Quoted 3 lines · Andrii Nakryiko
riscv existed as of [0], so I'd argue it is a proper bug fix, as corresponding select
ARCH_HAS_NON_OVERLAPPING_ADDRESS_SPACEshould have been added back then.How much of an eBPF ecosystem was there on RISC-V at the point?
No idea, never used RISC-V and didn't pay much attention. But why does it matter?
Quoted 11 lines · Andrii Nakryiko
Quoted 7 lines · Christoph Hellwig
Quoted 3 lines · Andrii Nakryiko
riscv existed as of [0], so I'd argue it is a proper bug fix, as corresponding select
ARCH_HAS_NON_OVERLAPPING_ADDRESS_SPACEshould have been added back then.How much of an eBPF ecosystem was there on RISC-V at the point?
No idea, never used RISC-V and didn't pay much attention. But why does it matter?
It matters because we should not spread broken legacy interfaces.
_______________________________________________
linux-riscv mailing list
[email protected]
http://lists.infradead.org/mailman/listinfo/linux-riscv