Inspired by[1], modify the code to remove the code of modifying ra to avoid imbalance RAS (return address stack) which may lead to incorret predictions on return. Link: https://lore.kernel.org/linux-riscv/20240607061335.2197383-1-cyrilbur@tenstorrent.com/ [1] Signed-off-by: Jisheng Zhang <jszhang@kernel.org> Reviewed-by: Cyril Bur <cyrilbur@tenstorrent.com> Link: https://lore.kernel.org/r/20240720170659.1522-1-jszhang@kernel.org Signed-off-by: Palmer Dabbelt <palmer@rivosinc.com>
Alexandre Ghiti <alexghiti@rivosinc.com> says: In RISC-V, after a new mapping is established, a sfence.vma needs to be emitted for different reasons: - if the uarch caches invalid entries, we need to invalidate it otherwise we would trap on this invalid entry, - if the uarch does not cache invalid entries, a reordered access could fail to see the new mapping and then trap (sfence.vma acts as a fence). We can actually avoid emitting those (mostly) useless and costly sfence.vma by handling the traps instead: - for new kernel mappings: only vmalloc mappings need to be taken care of, other new mapping are rare and already emit the required sfence.vma if needed. That must be achieved very early in the exception path as explained in patch 3, and this also fixes our fragile way of dealing with vmalloc faults. - for new user mappings: Svvptc makes update_mmu_cache() a no-op but we can take some gratuitous page faults (which are very unlikely though). Patch 1 and 2 introduce Svvptc extension probing. On our uarch that does not cache invalid entries and a 6.5 kernel, the gains are measurable: * Kernel boot: 6% * ltp - mmapstress01: 8% * lmbench - lat_pagefault: 20% * lmbench - lat_mmap: 5% Here are the corresponding numbers of sfence.vma emitted: * Ubuntu boot to login: Before: ~630k sfence.vma After: ~200k sfence.vma * ltp - mmapstress01 Before: ~45k After: ~6.3k * lmbench - lat_pagefault Before: ~665k After: 832 (!) * lmbench - lat_mmap Before: ~546k After: 718 (!) Thanks to Ved and Matt Evans for triggering the discussion that led to this patchset! * b4-shazam-merge: riscv: Stop emitting preventive sfence.vma for new userspace mappings with Svvptc riscv: Stop emitting preventive sfence.vma for new vmalloc mappings dt-bindings: riscv: Add Svvptc ISA extension description riscv: Add ISA extension parsing for Svvptc Link: https://lore.kernel.org/r/20240717060125.139416-1-alexghiti@rivosinc.com Signed-off-by: Palmer Dabbelt <palmer@rivosinc.com>
In 6.5, we removed the vmalloc fault path because that can't work (see [1] [2]). Then in order to make sure that new page table entries were seen by the page table walker, we had to preventively emit a sfence.vma on all harts [3] but this solution is very costly since it relies on IPI. And even there, we could end up in a loop of vmalloc faults if a vmalloc allocation is done in the IPI path (for example if it is traced, see [4]), which could result in a kernel stack overflow. Those preventive sfence.vma needed to be emitted because: - if the uarch caches invalid entries, the new mapping may not be observed by the page table walker and an invalidation may be needed. - if the uarch does not cache invalid entries, a reordered access could "miss" the new mapping and traps: in that case, we would actually only need to retry the access, no sfence.vma is required. So this patch removes those preventive sfence.vma and actually handles the possible (and unlikely) exceptions. And since the kernel stacks mappings lie in the vmalloc area, this handling must be done very early when the trap is taken, at the very beginning of handle_exception: this also rules out the vmalloc allocations in the fault path. Link: https://lore.kernel.org/linux-riscv/20230531093817.665799-1-bjorn@kernel.org/ [1] Link: https://lore.kernel.org/linux-riscv/20230801090927.2018653-1-dylan@andestech.com [2] Link: https://lore.kernel.org/linux-riscv/20230725132246.817726-1-alexghiti@rivosinc.com/ [3] Link: https://lore.kernel.org/lkml/20200508144043.13893-1-joro@8bytes.org/ [4] Signed-off-by: Alexandre Ghiti <alexghiti@rivosinc.com> Reviewed-by: Yunhui Cui <cuiyunhui@bytedance.com> Link: https://lore.kernel.org/r/20240717060125.139416-4-alexghiti@rivosinc.com Signed-off-by: Palmer Dabbelt <palmer@rivosinc.com>
Add support for the stackleak feature. Whenever the kernel returns to user space the kernel stack is filled with a poison value. At the same time, disables the plugin in EFI stub code because EFI stub is out of scope for the protection. Tested on qemu and milkv duo: / # echo STACKLEAK_ERASING > /sys/kernel/debug/provoke-crash/DIRECT [ 38.675575] lkdtm: Performing direct entry STACKLEAK_ERASING [ 38.678448] lkdtm: stackleak stack usage: [ 38.678448] high offset: 288 bytes [ 38.678448] current: 496 bytes [ 38.678448] lowest: 1328 bytes [ 38.678448] tracked: 1328 bytes [ 38.678448] untracked: 448 bytes [ 38.678448] poisoned: 14312 bytes [ 38.678448] low offset: 8 bytes [ 38.689887] lkdtm: OK: the rest of the thread stack is properly erased Signed-off-by: Jisheng Zhang <jszhang@kernel.org> Reviewed-by: Charlie Jenkins <charlie@rivosinc.com> Link: https://lore.kernel.org/r/20240623235316.2010-1-jszhang@kernel.org Signed-off-by: Palmer Dabbelt <palmer@rivosinc.com>
Many CPUs implement return address branch prediction as a stack. The RISCV architecture refers to this as a return address stack (RAS). If this gets corrupted then the CPU will mispredict at least one but potentally many function returns. There are two issues with the current RISCV exception code: - We are using the alternate link stack (x5/t0) for the indirect branch which makes the hardware think this is a function return. This will corrupt the RAS. - We modify the return address of handle_exception to point to ret_from_exception. This will also corrupt the RAS. Testing the null system call latency before and after the patch: Visionfive2 (StarFive JH7110 / U74) baseline: 189.87 ns patched: 176.76 ns Lichee pi 4a (T-Head TH1520 / C910) baseline: 666.58 ns patched: 636.90 ns Just over 7% on the U74 and just over 4% on the C910. Signed-off-by: Anton Blanchard <antonb@tenstorrent.com> Signed-off-by: Cyril Bur <cyrilbur@tenstorrent.com> Tested-by: Jisheng Zhang <jszhang@kernel.org> Reviewed-by: Jisheng Zhang <jszhang@kernel.org> Link: https://lore.kernel.org/r/20240607061335.2197383-1-cyrilbur@tenstorrent.com Signed-off-by: Palmer Dabbelt <palmer@rivosinc.com>
Adding kprobes on some assembly functions (mainly exception handling) will result in crashes (either recursive trap or panic). To avoid such errors, add ASM_NOKPROBE() macro which allow adding specific symbols into the __kprobe_blacklist section and use to blacklist the following symbols that showed to be problematic: - handle_exception() - ret_from_exception() - handle_kernel_stack_overflow() Signed-off-by: Clément Léger <cleger@rivosinc.com> Reviewed-by: Charlie Jenkins <charlie@rivosinc.com> Link: https://lore.kernel.org/r/20231004131009.409193-1-cleger@rivosinc.com Signed-off-by: Palmer Dabbelt <palmer@rivosinc.com>
Add kernel_vstate to keep track of kernel-mode Vector registers when trap introduced context switch happens. Also, provide riscv_v_flags to let context save/restore routine track context status. Context tracking happens whenever the core starts its in-kernel Vector executions. An active (dirty) kernel task's V contexts will be saved to memory whenever a trap-introduced context switch happens. Or, when a softirq, which happens to nest on top of it, uses Vector. Context retoring happens when the execution transfer back to the original Kernel context where it first enable preempt_v. Also, provide a config CONFIG_RISCV_ISA_V_PREEMPTIVE to give users an option to disable preemptible kernel-mode Vector at build time. Users with constraint memory may want to disable this config as preemptible kernel-mode Vector needs extra space for tracking of per thread's kernel-mode V context. Or, users might as well want to disable it if all kernel-mode Vector code is time sensitive and cannot tolerate context switch overhead. Signed-off-by: Andy Chiu <andy.chiu@sifive.com> Tested-by: Björn Töpel <bjorn@rivosinc.com> Tested-by: Lad Prabhakar <prabhakar.mahadev-lad.rj@bp.renesas.com> Link: https://lore.kernel.org/r/20240115055929.4736-11-andy.chiu@sifive.com Signed-off-by: Palmer Dabbelt <palmer@rivosinc.com>
Some data were incorrectly annotated with SYM_FUNC_*() instead of SYM_DATA_*() ones. Use the correct ones. Signed-off-by: Clément Léger <cleger@rivosinc.com> Reviewed-by: Andrew Jones <ajones@ventanamicro.com> Link: https://lore.kernel.org/r/20231024132655.730417-4-cleger@rivosinc.com Signed-off-by: Palmer Dabbelt <palmer@rivosinc.com>
For the sake of coherency, use local labels in assembly when applicable. This also avoid kprobes being confused when applying a kprobe since the size of function is computed by checking where the next visible symbol is located. This might end up in computing some function size to be way shorter than expected and thus failing to apply kprobes to the specified offset. Signed-off-by: Clément Léger <cleger@rivosinc.com> Reviewed-by: Andrew Jones <ajones@ventanamicro.com> Link: https://lore.kernel.org/r/20231024132655.730417-2-cleger@rivosinc.com Signed-off-by: Palmer Dabbelt <palmer@rivosinc.com>
Sami Tolvanen <samitolvanen@google.com> says: This series adds Shadow Call Stack (SCS) support for RISC-V. SCS uses compiler instrumentation to store return addresses in a separate shadow stack to protect them against accidental or malicious overwrites. More information about SCS can be found here: https://clang.llvm.org/docs/ShadowCallStack.html Patch 1 is from Deepak, and it simplifies VMAP_STACK overflow handling by adding support for accessing per-CPU variables directly in assembly. The patch is included in this series to make IRQ stack switching cleaner with SCS, and I've simply rebased it and fixed a couple of minor issues. Patch 2 uses this functionality to clean up the stack switching by moving duplicate code into a single function. On RISC-V, the compiler uses the gp register for storing the current shadow call stack pointer, which is incompatible with global pointer relaxation. Patch 3 moves global pointer loading into a macro that can be easily disabled with SCS. Patch 4 implements SCS register loading and switching, and allows the feature to be enabled, and patch 5 adds separate per-CPU IRQ shadow call stacks when CONFIG_IRQ_STACKS is enabled. Patch 6 fixes the backward-edge CFI test in lkdtm for RISC-V. Note that this series requires Clang 17. Earlier Clang versions support SCS on RISC-V, but use the x18 register instead of gp, which isn't ideal. gcc has SCS support for arm64, but I'm not aware of plans to support RISC-V. Once the Zicfiss extension is ratified, it's probably preferable to use hardware-backed shadow stacks instead of SCS on hardware that supports the extension, and we may want to consider implementing CONFIG_DYNAMIC_SCS to patch between the implementation at runtime (similarly to the arm64 implementation, which switches to SCS when hardware PAC support isn't available). * b4-shazam-merge: lkdtm: Fix CFI_BACKWARD on RISC-V riscv: Use separate IRQ shadow call stacks riscv: Implement Shadow Call Stack riscv: Move global pointer loading to a macro riscv: Deduplicate IRQ stack switching riscv: VMAP_STACK overflow detection thread-safe Link: https://lore.kernel.org/r/20230927224757.1154247-8-samitolvanen@google.com Signed-off-by: Palmer Dabbelt <palmer@rivosinc.com>