On 7/3/20 2:48 PM, Nicholas Piggin wrote:
Excerpts from Aneesh Kumar K.V's message of June 15, 2020 4:14 pm:This prepare kernel to operate with a different value than userspace AMR. For this, AMR needs to be saved and restored on entry and return from the kernel.With KUAP we modify kernel AMR when accessing user address from the kernel via copy_to/from_user interfaces. If MMU_FTR_KEY is enabled we always use the key mechanism to implement KUAP feature. If MMU_FTR_KEY is not supported and if we support MMU_FTR_KUAP (radix translation on POWER9), we can skip restoring AMR on return to userspace. Userspace won't be using AMR in that specific config. Signed-off-by: Aneesh Kumar K.V <aneesh.ku...@linux.ibm.com> --- arch/powerpc/include/asm/book3s/64/kup.h | 141 ++++++++++++++++++----- arch/powerpc/kernel/entry_64.S | 6 +- arch/powerpc/kernel/exceptions-64s.S | 4 +- arch/powerpc/kernel/syscall_64.c | 26 ++++- 4 files changed, 144 insertions(+), 33 deletions(-) diff --git a/arch/powerpc/include/asm/book3s/64/kup.h b/arch/powerpc/include/asm/book3s/64/kup.h index e6ee1c34842f..6979cd1a0003 100644 --- a/arch/powerpc/include/asm/book3s/64/kup.h +++ b/arch/powerpc/include/asm/book3s/64/kup.h @@ -13,18 +13,47 @@#ifdef __ASSEMBLY__ -.macro kuap_restore_amr gpr1, gpr2-#ifdef CONFIG_PPC_KUAP +.macro kuap_restore_user_amr gpr1 +#if defined(CONFIG_PPC_MEM_KEYS) BEGIN_MMU_FTR_SECTION_NESTED(67) - mfspr \gpr1, SPRN_AMR + /* + * AMR is going to be different when + * returning to userspace. + */ + ld \gpr1, STACK_REGS_KUAP(r1) + isync + mtspr SPRN_AMR, \gpr1 + + /* No isync required, see kuap_restore_user_amr() */ + END_MMU_FTR_SECTION_NESTED_IFSET(MMU_FTR_PKEY , 67) +#endif +.endm + +.macro kuap_restore_kernel_amr gpr1, gpr2 +#if defined(CONFIG_PPC_MEM_KEYS) + BEGIN_MMU_FTR_SECTION_NESTED(67) + b 99f // handle_pkey_restore_amr + END_MMU_FTR_SECTION_NESTED_IFSET(MMU_FTR_PKEY , 67) + + BEGIN_MMU_FTR_SECTION_NESTED(68) + b 99f // handle_kuap_restore_amr + MMU_FTR_SECTION_ELSE_NESTED(68) + b 100f // skip_restore_amr + ALT_MMU_FTR_SECTION_END_NESTED_IFSET(MMU_FTR_KUAP, 68) + +99: + /* + * AMR is going to be mostly the same since we are + * returning to the kernel. Compare and do a mtspr. + */ ld \gpr2, STACK_REGS_KUAP(r1) + mfspr \gpr1, SPRN_AMR cmpd \gpr1, \gpr2 - beq 998f + beq 100f isync mtspr SPRN_AMR, \gpr2 /* No isync required, see kuap_restore_amr() */ -998: - END_MMU_FTR_SECTION_NESTED_IFSET(MMU_FTR_KUAP, 67) +100: // skip_restore_amrCan't you code it like this? (_IFCLR requires none of the bits to be set) BEGIN_MMU_FTR_SECTION_NESTED(67) b 99f // nothing using AMR, no need to restore END_MMU_FTR_SECTION_NESTED_IFCLR(MMU_FTR_PKEY | MMU_FTR_KUAP, 67) That saves you a branch in the common case of using AMR. Similar for others.
Yes i could switch to that. The code is taking extra 200 cycles even with KUAP/KUEP disabled and no keys being used on hash. I am yet to analyze this closely. So will rework things based on that analysis.
@@ -69,22 +133,40 @@extern u64 default_uamor; -static inline void kuap_restore_amr(struct pt_regs *regs, unsigned long amr)+static inline void kuap_restore_user_amr(struct pt_regs *regs) { - if (mmu_has_feature(MMU_FTR_KUAP) && unlikely(regs->kuap != amr)) { - isync(); - mtspr(SPRN_AMR, regs->kuap); - /* - * No isync required here because we are about to RFI back to - * previous context before any user accesses would be made, - * which is a CSI. - */ + if (!mmu_has_feature(MMU_FTR_PKEY)) + return;If you have PKEY but not KUAP, do you still have to restore?
Yes, because user space pkey is now set on the exit path. This is needed to handle things like exec/fork().
+ + isync(); + mtspr(SPRN_AMR, regs->kuap); + /* + * No isync required here because we are about to rfi + * back to previous context before any user accesses + * would be made, which is a CSI. + */ +} + +static inline void kuap_restore_kernel_amr(struct pt_regs *regs, + unsigned long amr) +{ + if (mmu_has_feature(MMU_FTR_KUAP) || mmu_has_feature(MMU_FTR_PKEY)) { + + if (unlikely(regs->kuap != amr)) { + isync(); + mtspr(SPRN_AMR, regs->kuap); + /* + * No isync required here because we are about to rfi + * back to previous context before any user accesses + * would be made, which is a CSI. + */ + } } }static inline unsigned long kuap_get_and_check_amr(void){ - if (mmu_has_feature(MMU_FTR_KUAP)) { + if (mmu_has_feature(MMU_FTR_KUAP) || mmu_has_feature(MMU_FTR_PKEY)) { unsigned long amr = mfspr(SPRN_AMR); if (IS_ENABLED(CONFIG_PPC_KUAP_DEBUG)) /* kuap_check_amr() */ WARN_ON_ONCE(amr != AMR_KUAP_BLOCKED);We could do a static key that's based on this condition, but that can wait for another day.static inline void kuap_check_amr(void){ - if (IS_ENABLED(CONFIG_PPC_KUAP_DEBUG) && mmu_has_feature(MMU_FTR_KUAP)) + if (IS_ENABLED(CONFIG_PPC_KUAP_DEBUG) && + (mmu_has_feature(MMU_FTR_KUAP) || mmu_has_feature(MMU_FTR_PKEY))) WARN_ON_ONCE(mfspr(SPRN_AMR) != AMR_KUAP_BLOCKED); }#else /* CONFIG_PPC_MEM_KEYS */ -static inline void kuap_restore_amr(struct pt_regs *regs, unsigned long amr)+static inline void kuap_restore_user_amr(struct pt_regs *regs) +{ +} + +static inline void kuap_restore_kernel_amr(struct pt_regs *regs, unsigned long amr) { }@@ -113,6 +200,7 @@ static inline unsigned long kuap_get_and_check_amr(void){ return 0; } + #endif /* CONFIG_PPC_MEM_KEYS */@@ -187,7 +275,6 @@ bad_kuap_fault(struct pt_regs *regs, unsigned long address, bool is_write)"Bug: %s fault blocked by AMR!", is_write ? "Write" : "Read"); } #endif /* CONFIG_PPC_KUAP */ - #endif /* __ASSEMBLY__ */#endif /* _ASM_POWERPC_BOOK3S_64_KUP_H */Unnecessary hunks.
will remove
diff --git a/arch/powerpc/kernel/entry_64.S b/arch/powerpc/kernel/entry_64.S index 9d49338e0c85..a087cbe0b17d 100644 --- a/arch/powerpc/kernel/entry_64.S +++ b/arch/powerpc/kernel/entry_64.S @@ -481,8 +481,8 @@ _ASM_NOKPROBE_SYMBOL(fast_interrupt_return) kuap_check_amr r3, r4 ld r5,_MSR(r1) andi. r0,r5,MSR_PR - bne .Lfast_user_interrupt_return - kuap_restore_amr r3, r4 + bne .Lfast_user_interrupt_return_amr + kuap_restore_kernel_amr r3, r4 andi. r0,r5,MSR_RI li r3,0 /* 0 return value, no EMULATE_STACK_STORE */ bne+ .Lfast_kernel_interrupt_return @@ -502,6 +502,8 @@ _ASM_NOKPROBE_SYMBOL(interrupt_return) cmpdi r3,0 bne- .Lrestore_nvgprs+.Lfast_user_interrupt_return_amr:+ kuap_restore_user_amr r3 .Lfast_user_interrupt_return: ld r11,_NIP(r1) ld r12,_MSR(r1) diff --git a/arch/powerpc/kernel/exceptions-64s.S b/arch/powerpc/kernel/exceptions-64s.S index e70ebb5c318c..8226af444d77 100644 --- a/arch/powerpc/kernel/exceptions-64s.S +++ b/arch/powerpc/kernel/exceptions-64s.S @@ -971,7 +971,7 @@ EXC_COMMON_BEGIN(system_reset_common) ld r10,SOFTE(r1) stb r10,PACAIRQSOFTMASK(r13)- kuap_restore_amr r9, r10+ kuap_restore_kernel_amr r9, r10 EXCEPTION_RESTORE_REGS RFI_TO_USER_OR_KERNEL@@ -2784,7 +2784,7 @@ EXC_COMMON_BEGIN(soft_nmi_common)ld r10,SOFTE(r1) stb r10,PACAIRQSOFTMASK(r13)- kuap_restore_amr r9, r10+ kuap_restore_kernel_amr r9, r10 EXCEPTION_RESTORE_REGS hsrr=0 RFI_TO_KERNELdiff --git a/arch/powerpc/kernel/syscall_64.c b/arch/powerpc/kernel/syscall_64.cindex 7e560a01afa4..fded67982fbe 100644 --- a/arch/powerpc/kernel/syscall_64.c +++ b/arch/powerpc/kernel/syscall_64.c @@ -35,7 +35,21 @@ notrace long system_call_exception(long r3, long r4, long r5, BUG_ON(!FULL_REGS(regs)); BUG_ON(regs->softe != IRQS_ENABLED);- kuap_check_amr();+#ifdef CONFIG_PPC_MEM_KEYS + if (mmu_has_feature(MMU_FTR_PKEY)) { + unsigned long amr; + /* + * When entering from userspace we mostly have the AMR + * different from kernel default values. Hence don't compare. + */ + amr = mfspr(SPRN_AMR); + regs->kuap = amr; + if (mmu_has_feature(MMU_FTR_KUAP)) + mtspr(SPRN_AMR, AMR_KUAP_BLOCKED); + isync();isync should be inside the if(). Again do pkeys need to save this if KUAP is not being used? I haven't really looked at how all that works, but what's changing for the PKEY && !KUAP case?
There is no SPR switch in context switch now and all the AMR/IAMR handling is now in the exit to userspace.
This would be nice if it could all go into a wrapper function rather than ifdef.
will do
+ } else +#endif + kuap_check_amr();account_cpu_user_entry(); @@ -222,6 +236,10 @@ notrace unsigned long syscall_exit_prepare(unsigned long r3, account_cpu_user_exit(); + /*+ * We do this at the end so that we do context switch with KERNEL AMR + */ + kuap_restore_user_amr(regs); return ret;Comment doesn't make sense, newline required before return.
Ok the detail there was we need to make sure we restore AMR towrads the end and make sure all the kernel code continue to run with KERNEL AMR value. There is a schedule() call in there with _TIF_NEED_RESCHED. But those details are not really relevant. That was me tracking down some issues and writing comment around that part of the code. The only real detail is switch to userspace AMR in the end.
}@@ -306,6 +324,10 @@ notrace unsigned long interrupt_exit_user_prepare(struct pt_regs *regs, unsigned account_cpu_user_exit(); + /*+ * We do this at the end so that we do context switch with KERNEL AMR + */ + kuap_restore_user_amr(regs);Duplicated comments I prefer to just have like this instead of trying to keep them in sync. Can complete the circular reference by having a * similarly in interrupt_exit_user_prepare in the main comment, but if they come close to one another in the same file it's not so important to keep them together. + kuap_restore_user_amr(regs); /* see syscall_exit_prepare */return ret; }@@ -376,7 +398,7 @@ notrace unsigned long interrupt_exit_kernel_prepare(struct pt_regs *regs, unsign* which would cause Read-After-Write stalls. Hence, we take the AMR * value from the check above. */ - kuap_restore_amr(regs, amr); + kuap_restore_kernel_amr(regs, amr);return ret;} -- 2.26.2
-aneesh
