- 19 May, 2022 21 commits
-
-
Christophe Leroy authored
Instead of returning -EPERM when patch_instruction() fails, just return what patch_instruction returns. That simplifies ftrace_modify_code(): 0: 94 21 ff c0 stwu r1,-64(r1) 4: 93 e1 00 3c stw r31,60(r1) 8: 7c 7f 1b 79 mr. r31,r3 c: 40 80 00 30 bge 3c <ftrace_modify_code+0x3c> 10: 93 c1 00 38 stw r30,56(r1) 14: 7c 9e 23 78 mr r30,r4 18: 7c a4 2b 78 mr r4,r5 1c: 80 bf 00 00 lwz r5,0(r31) 20: 7c 1e 28 40 cmplw r30,r5 24: 40 82 00 34 bne 58 <ftrace_modify_code+0x58> 28: 83 c1 00 38 lwz r30,56(r1) 2c: 7f e3 fb 78 mr r3,r31 30: 83 e1 00 3c lwz r31,60(r1) 34: 38 21 00 40 addi r1,r1,64 38: 48 00 00 00 b 38 <ftrace_modify_code+0x38> 38: R_PPC_REL24 patch_instruction Before: 0: 94 21 ff c0 stwu r1,-64(r1) 4: 93 e1 00 3c stw r31,60(r1) 8: 7c 7f 1b 79 mr. r31,r3 c: 40 80 00 4c bge 58 <ftrace_modify_code+0x58> 10: 93 c1 00 38 stw r30,56(r1) 14: 7c 9e 23 78 mr r30,r4 18: 7c a4 2b 78 mr r4,r5 1c: 80 bf 00 00 lwz r5,0(r31) 20: 7c 08 02 a6 mflr r0 24: 90 01 00 44 stw r0,68(r1) 28: 7c 1e 28 40 cmplw r30,r5 2c: 40 82 00 48 bne 74 <ftrace_modify_code+0x74> 30: 7f e3 fb 78 mr r3,r31 34: 48 00 00 01 bl 34 <ftrace_modify_code+0x34> 34: R_PPC_REL24 patch_instruction 38: 80 01 00 44 lwz r0,68(r1) 3c: 20 63 00 00 subfic r3,r3,0 40: 83 c1 00 38 lwz r30,56(r1) 44: 7c 63 19 10 subfe r3,r3,r3 48: 7c 08 03 a6 mtlr r0 4c: 83 e1 00 3c lwz r31,60(r1) 50: 38 21 00 40 addi r1,r1,64 54: 4e 80 00 20 blr It improves ftrace activation/deactivation duration by about 3%. Modify patch_instruction() return on failure to -EPERM in order to match with ftrace expectations. Other users of patch_instruction() do not care about the exact error value returned. Signed-off-by:
Christophe Leroy <christophe.leroy@csgroup.eu> Signed-off-by:
Michael Ellerman <mpe@ellerman.id.au> Link: https://lore.kernel.org/r/49a8597230713e2633e7d9d7b56140787c4a7e20.1652074503.git.christophe.leroy@csgroup.eu
-
Christophe Leroy authored
Inlining ftrace_modify_code(), it increases a bit the size of ftrace code but brings 5% improvment on ftrace activation. Usually in C files we let gcc decide what to do but here it really help to 'help' gcc to decide to inline, thought we don't want to force it with an __always_inline that would be too much for CONFIG_CC_OPTIMIZE_FOR_SIZE. Signed-off-by:
-
Christophe Leroy authored
create_branch() is a good candidate for inlining because: - Flags can be folded in. - Range tests are likely to be already done. Hence reducing the create_branch() to only a set of instructions. So inline it. It improves ftrace activation by 10%. Signed-off-by:
-
Christophe Leroy authored
Use is_offset_in_branch_range() instead of create_branch() to check if a target is within branch range. This patch together with the previous one improves ftrace activation time by 7% Signed-off-by:
-
Christophe Leroy authored
Test in is_offset_in_branch_range() and is_offset_in_cond_branch_range() are simple tests that are worth inlining. Signed-off-by:
-
Christophe Leroy authored
Since commit d5937db1 ("powerpc/code-patching: Fix patch_branch() return on out-of-range failure") patch_branch() fails with -ERANGE when trying to branch out of range. No need to perform the test twice. Remove redundant create_branch() calls. Signed-off-by:
-
Christophe Leroy authored
When we have CONFIG_DYNAMIC_FTRACE_WITH_ARGS, prepare_ftrace_return() is called by ftrace_graph_func() otherwise prepare_ftrace_return() is called from assembly. Refactor prepare_ftrace_return() into a static __prepare_ftrace_return() that will be called by both prepare_ftrace_return() and ftrace_graph_func(). It will allow GCC to fold __prepare_ftrace_return() inside ftrace_graph_func(). Signed-off-by:
-
Nicholas Piggin authored
rtas_call must not be called with the MMU disabled because in case of rtas error, log_error is called which requires MMU enabled. Add a test and warning for this. Signed-off-by:
Nicholas Piggin <npiggin@gmail.com> Reviewed-by:
Laurent Dufour <ldufour@linux.ibm.com> Signed-off-by:
-
Nicholas Piggin authored
PAPR specifies that RTAS may be called with MSR[RI] enabled if the calling context is recoverable, and RTAS will manage RI as necessary. Call the rtas entry point with RI enabled, and add a check to ensure the caller has RI enabled. Signed-off-by:
-
Nicholas Piggin authored
On 64-bit, PACA is saved in a SPRG so it does not need to be saved on stack. We also don't need to mask off the top bits for real mode addresses because the architecture does this for us. Signed-off-by:
-
Nicholas Piggin authored
Disable MSR[EE] in C code rather than asm. Signed-off-by:
-
Nicholas Piggin authored
The code was moved verbatim including whitespace cruft. Fix that. Signed-off-by:
-
Nicholas Piggin authored
This symbol is marked nokprobe on 32-bit but not 64-bit, add it. Signed-off-by:
-
Nicholas Piggin authored
This makes working on the code a bit easier. Signed-off-by:
-
Nicholas Piggin authored
Implement the AT_MINSIGSTKSZ AUXV entry, allowing userspace to dynamically size stack allocations in a manner forward-compatible with new processor state saved in the signal frame For now these statically find the maximum signal frame size rather than doing any runtime testing of features to minimise the size. glibc 2.34 will take advantage of this, as will applications that use use _SC_MINSIGSTKSZ and _SC_SIGSTKSZ. Reviewed-by:
Tulio Magno Quites Machado Filho <tuliom@linux.ibm.com> Signed-off-by:
-
Nicholas Piggin authored
The sad tale of SIGSTKSZ and MINSIGSTKSZ is documented in glibc.git commit f7c399cff5bd ("PowerPC SIGSTKSZ"), which explains why glibc does not use the kernel defines for these constants. Since then in fact there has been a further expansion of the signal stack frame size on little-endian with linux commit 573ebfa6 ("powerpc: Increase stack redzone for 64-bit userspace to 512 bytes"), which has caused it to exceed even the glibc defines. See kernel commit 63dee5df ("powerpc: Allow 4224 bytes of stack expansion for the signal frame") for more details of the history of the expansion. Increase MINSIGSTKSZ to 8192 which is double the current glibc value and fits the current stack frame with room to grow. SIGSTKSZ is set to 4x the minimum as convention. glibc will have to be updated as well. Reviewed-by: -
Nathan Chancellor authored
The PowerPC vDSO uses $(CC) to link, which differs from the rest of the kernel, which uses $(LD) directly. As a result, the default linker of the compiler is used, which may differ from the linker requested by the builder. For example: $ make ARCH=powerpc LLVM=1 mrproper defconfig arch/powerpc/kernel/vdso/ ... $ llvm-readelf -p .comment arch/powerpc/kernel/vdso/vdso{32,64}.so.dbg File: arch/powerpc/kernel/vdso/vdso32.so.dbg String dump of section '.comment': [ 0] clang version 14.0.0 (Fedora 14.0.0-1.fc37) File: arch/powerpc/kernel/vdso/vdso64.so.dbg String dump of section '.comment': [ 0] clang version 14.0.0 (Fedora 14.0.0-1.fc37) LLVM=1 sets LD=ld.lld but ld.lld is not used to link the vDSO; GNU ld is because "ld" is the default linker for clang on most Linux platforms. This is a problem for Clang's Link Time Optimization as implemented in the kernel because use of GNU ld with LTO requires the LLVMgold plugin, which is not technically supported for ld.bfd per https://llvm.org/docs/GoldPlugin.html. Furthermore, if LLVMgold.so is missing from a user's system, the build will fail, even though LTO as it is implemented in the kernel requires ld.lld to avoid this dependency in the first place. Ultimately, the PowerPC vDSO should be converted to compiling and linking with $(CC) and $(LD) respectively but there were issues last time this was tried, potentially due to older but supported tool versions. To avoid regressing GCC + binutils, use the compiler option '-fuse-ld', which tells the compiler which linker to use when it is invoked as both the compiler and linker. Use '-fuse-ld=lld' when LD=ld.lld has been specified (CONFIG_LD_IS_LLD) so that the vDSO is linked with the same linker as the rest of the kernel. $ llvm-readelf -p .comment arch/powerpc/kernel/vdso/vdso{32,64}.so.dbg File: arch/powerpc/kernel/vdso/vdso32.so.dbg String dump of section '.comment': [ 0] Linker: LLD 14.0.0 [ 14] clang version 14.0.0 (Fedora 14.0.0-1.fc37) File: arch/powerpc/kernel/vdso/vdso64.so.dbg String dump of section '.comment': [ 0] Linker: LLD 14.0.0 [ 14] clang version 14.0.0 (Fedora 14.0.0-1.fc37) LD can be a full path to ld.lld, which will not be handled properly by '-fuse-ld=lld' if the full path to ld.lld is outside of the compiler's search path. '-fuse-ld' can take a path to the linker but it is deprecated in clang 12.0.0; '--ld-path' is preferred for this scenario. Use '--ld-path' if it is supported, as it will handle a full path or just 'ld.lld' properly. See the LLVM commit below for the full details of '--ld-path'. Signed-off-by:Nathan Chancellor <nathan@kernel.org> Tested-by:
Alexey Kardashevskiy <aik@ozlabs.ru> Reviewed-by:
Nick Desaulniers <ndesaulniers@google.com> Reviewed-by:
-
Nathan Chancellor authored
When linking vdso{32,64}.so.dbg with ld.lld, there is a warning about not finding _start for the starting address: ld.lld: warning: cannot find entry symbol _start; not setting start address ld.lld: warning: cannot find entry symbol _start; not setting start address Looking at GCC + GNU ld, the entry point address is 0x0: $ llvm-readelf -h vdso{32,64}.so.dbg &| rg "(File|Entry point address):" File: vdso32.so.dbg Entry point address: 0x0 File: vdso64.so.dbg Entry point address: 0x0 This matches what ld.lld emits: $ powerpc64le-linux-gnu-readelf -p .comment vdso{32,64}.so.dbg File: vdso32.so.dbg String dump of section '.comment': [ 0] Linker: LLD 14.0.0 [ 14] clang version 14.0.0 (Fedora 14.0.0-1.fc37) File: vdso64.so.dbg String dump of section '.comment': [ 0] Linker: LLD 14.0.0 [ 14] clang version 14.0.0 (Fedora 14.0.0-1.fc37) $ llvm-readelf -h vdso{32,64}.so.dbg &| rg "(File|Entry point address):" File: vdso32.so.dbg Entry point address: 0x0 File: vdso64.so.dbg Entry point address: 0x0 Remove ENTRY to remove the warning, as it is unnecessary for the vDSO to function correctly. Signed-off-by: -
Kevin Hao authored
When the mmu_feature_keys[] was introduced in the commit c12e6f24 ("powerpc: Add option to use jump label for mmu_has_feature()"), it is unlikely that it would be used either directly or indirectly in the out of tree modules. So we exported it as GPL only. But with the evolution of the codes, especially the PPC_KUAP support, it may be indirectly referenced by some primitive macro or inline functions such as get_user() or __copy_from_user_inatomic(), this will make it impossible to build many non GPL modules (such as ZFS) on ppc architecture. Fix this by exposing the mmu_feature_keys[] to the non-GPL modules too. Fixes: 7613f5a6 ("powerpc/64s/kuap: Use mmu_has_feature()") Reported-by:
Nathaniel Filardo <nwfilardo@gmail.com> Signed-off-by:
Kevin Hao <haokexin@gmail.com> Signed-off-by:
-
Guilherme G. Piccoli authored
The panic notifiers infrastructure is a bit limited in the scope of the callbacks - basically every kind of functionality is dropped in a list that runs in the same point during the kernel panic path. This is not really on par with the complexities and particularities of architecture / hypervisors' needs, and a refactor is ongoing. As part of this refactor, it was observed that powerpc has 2 notifiers, with mixed goals: one is just a KASLR offset dumper, whereas the other aims to hard-disable IRQs (necessary on panic path), warn firmware of the panic event (fadump) and run low-level platform-specific machinery that might stop kernel execution and never come back. Clearly, the 2nd notifier has opposed goals: disable IRQs / fadump should run earlier while low-level platform actions should run late since it might not even return. Hence, this patch decouples the notifiers splitting them in three: - First one is responsible for hard-disable IRQs and fadump, should run early; - The kernel KASLR offset dumper is really an informative notifier, harmless and may run at any moment in the panic path; - The last notifier should run last, since it aims to perform low-level actions for specific platforms, and might never return. It is also only registered for 2 platforms, pseries and ps3. The patch better documents the notifiers and clears the code too, also removing a useless header. Currently no functionality change should be observed, but after the planned panic refactor we should expect more panic reliability with this patch. Signed-off-by:
Guilherme G. Piccoli <gpiccoli@igalia.com> Reviewed-by:
Hari Bathini <hbathini@linux.ibm.com> Signed-off-by:
-
Michael Ellerman authored
Merge our KVM topic branch.
-
- 18 May, 2022 10 commits
-
-
Fabiano Rosas authored
We removed most of the vcore logic from the P9 path but there's still a tracepoint that tried to dereference vc->runner. Fixes: ecb6a720 ("KVM: PPC: Book3S HV P9: Remove most of the vcore logic") Signed-off-by:
Fabiano Rosas <farosas@linux.ibm.com> Reviewed-by:
-
Alexey Kardashevskiy authored
Currently we have 2 sets of interrupt controller hypercalls handlers for real and virtual modes, this is from POWER8 times when switching MMU on was considered an expensive operation. POWER9 however does not have dependent threads and MMU is enabled for handling hcalls so the XIVE native or XICS-on-XIVE real mode handlers never execute on real P9 and later CPUs. This untemplate the handlers and only keeps the real mode handlers for XICS native (up to POWER8) and remove the rest of dead code. Changes in functions are mechanical except few missing empty lines to make checkpatch.pl happy. The default implemented hcalls list already contains XICS hcalls so no change there. This should not cause any behavioral change. Reported-by:
kernel test robot <lkp@intel.com> Signed-off-by:
Cédric Le Goater <clg@kaod.org> Signed-off-by:
-
Alexey Kardashevskiy authored
When KVM_CAP_PPC_ENABLE_HCALL was introduced, H_GET_TCE and H_PUT_TCE were already implemented and enabled by default; however H_GET_TCE was missed out on PR KVM (probably because the handler was in the real mode code at the time). This enables H_GET_TCE by default. While at this, this wraps the checks in ifdef CONFIG_SPAPR_TCE_IOMMU just like HV KVM. Signed-off-by:
-
Alexey Kardashevskiy authored
LoPAPR defines guest visible IOMMU with hypercalls to use it - H_PUT_TCE/etc. Implemented first on POWER7 where hypercalls would trap in the KVM in the real mode (with MMU off). The problem with the real mode is some memory is not available and some API usage crashed the host but enabling MMU was an expensive operation. The problems with the real mode handlers are: 1. Occasionally these cannot complete the request so the code is copied+modified to work in the virtual mode, very little is shared; 2. The real mode handlers have to be linked into vmlinux to work; 3. An exception in real mode immediately reboots the machine. If the small DMA window is used, the real mode handlers bring better performance. However since POWER8, there has always been a bigger DMA window which VMs use to map the entire VM memory to avoid calling H_PUT_TCE. Such 1:1 mapping happens once and uses H_PUT_TCE_INDIRECT (a bulk version of H_PUT_TCE) which virtual mode handler is even closer to its real mode version. On POWER9 hypercalls trap straight to the virtual mode so the real mode handlers never execute on POWER9 and later CPUs. So with the current use of the DMA windows and MMU improvements in POWER9 and later, there is no point in duplicating the code. The 32bit passed through devices may slow down but we do not have many of these in practice. For example, with this applied, a 1Gbit ethernet adapter still demostrates above 800Mbit/s of actual throughput. This removes the real mode handlers from KVM and related code from the powernv platform. This updates the list of implemented hcalls in KVM-HV as the realmode handlers are removed. This changes ABI - kvmppc_h_get_tce() moves to the KVM module and kvmppc_find_table() is static now. Signed-off-by:
-
Michael Ellerman authored
Merge our fixes branch. In parciular this brings in the KVM TCE handling fix, which is a prerequisite for a subsequent patch.
-
Fabiano Rosas authored
The hypervisor always sets AMOR to ~0, but let's ensure we're not passing stale values around. Signed-off-by:
-
Michael Ellerman authored
Merge our fixes branch from this cycle. In particular this brings in a papr_scm.c change which a subsequent patch has a dependency on.
-
Bo Liu authored
The return value type defined in the function kvm_age_rmapp() is "bool", but the return value type defined in the implementation of the function kvm_age_rmapp() is "int". Change the return value type to "bool". Signed-off-by:
Bo Liu <liubo03@inspur.com> Signed-off-by:
-
Xiaomeng Tong authored
The bug is here: if (!p) return ret; The list iterator value 'p' will *always* be set and non-NULL by list_for_each_entry(), so it is incorrect to assume that the iterator value will be NULL if the list is empty or no element is found. To fix the bug, Use a new value 'iter' as the list iterator, while use the old value 'p' as a dedicated variable to point to the found element. Fixes: dfaa973a ("KVM: PPC: Book3S HV: In H_SVM_INIT_DONE, migrate remaining normal-GFNs to secure-GFNs") Cc: stable@vger.kernel.org # v5.9+ Signed-off-by:Xiaomeng Tong <xiam0nd.tong@gmail.com> Signed-off-by:
-
Bagas Sanjaya authored
kernel test robot reported kernel-doc warning for rm_host_ipi_action(): arch/powerpc/kvm/book3s_hv_rm_xics.c:887: warning: This comment starts with '/**', but isn't a kernel-doc comment. * Host Operations poked by RM KVM Since the function is static, remove the extraneous (second) asterisk at the head of function comment. Fixes: 0c2a6606 ("KVM: PPC: Book3S HV: Host side kick VCPU when poked by real-mode KVM") Reported-by:Bagas Sanjaya <bagasdotme@gmail.com> Signed-off-by:
-
- 13 May, 2022 9 commits
-
-
Nicholas Piggin authored
The L1 should not be able to adjust LPES mode for the L2. Setting LPES if the L0 needs it clear would cause external interrupts to be sent to L2 and missed by the L0. Clearing LPES when it may be set, as typically happens with XIVE enabled could cause a performance issue despite having no native XIVE support in the guest, because it will cause mediated interrupts for the L2 to be taken in HV mode, which then have to be injected. Signed-off-by:
-
Nicholas Piggin authored
The PowerNV L0 currently pushes the OS xive context when running a vCPU, regardless of whether it is running a nested guest. The problem is that xive OS ring interrupts will be delivered while the L2 is running. At the moment, by default, the L2 guest runs with LPCR[LPES]=0, which actually makes external interrupts go to the L0. That causes the L2 to exit and the interrupt taken or injected into the L1, so in some respects this behaves like an escalation. It's not clear if this was deliberate or not, there's no comment about it and the L1 is actually allowed to clear LPES in the L2, so it's confusing at best. When the L2 is running, the L1 is essentially in a ceded state with respect to external interrupts (it can't respond to them directly and won't get scheduled again absent some additional event). So the natural way to solve this is when the L0 handles a H_ENTER_NESTED hypercall to run the L2, have it arm the escalation interrupt and don't push the L1 context while running the L2. Signed-off-by:
-
Nicholas Piggin authored
The differences between nested and !nested will become larger in later changes so split them out for readability. Signed-off-by:
-
Nicholas Piggin authored
Move the cede abort logic out of xive escalation rearming and into the caller to prepare for handling a similar case with nested guest entry. Signed-off-by:
-
Nicholas Piggin authored
If there is a pending xive interrupt, inject it at guest entry (if MSR[EE] is enabled) rather than take another interrupt when the guest is entered. If xive is enabled then LPCR[LPES] is set so this behaviour should be expected. Signed-off-by:
-
Nicholas Piggin authored
KVMPPC_NR_LPIDS no longer represents any size restriction on the LPID space and can be removed. A CPU with more than 12 LPID bits implemented will now be able to create more than 4095 guests. Signed-off-by:
-
Nicholas Piggin authored
Rather than tie this to KVMPPC_NR_LPIDS which is becoming more dynamic, fix it to 4096 (12-bits) explicitly for now. kvmhv_get_nested() does not have to check against KVM_MAX_NESTED_GUESTS because the L1 partition table registration hcall already did that, and it checks against the partition table size. This patch also puts all the partition table size calculations into the same form, using 12 for the architected size field shift and 4 for the shift corresponding to the partition table entry size. Reviewed-by:
-
Nicholas Piggin authored
This removes the fixed sized kvm->arch.nested_guests array. Signed-off-by:
-
Nicholas Piggin authored
This removes the fixed-size lpid_inuse array. Signed-off-by:
-
