- 10 Feb, 2023 21 commits
-
-
Nicholas Piggin authored
The early paca and boot cpuid dance is complicated and currently does not quite work as expected for boot cpuid != 0 cases. early_init_devtree() currently allocates the paca_ptrs and boot cpuid paca, but until that returns and early_setup() calls setup_paca(), this thread is currently still executing with smp_processor_id() == 0. One problem this causes is the paca_ptrs[smp_processor_id()] pointer is poisoned, so valid_emergency_stack() (any backtrace) and any similar users will crash. Another is that the hardware id which is set here will not be returned by get_hard_smp_processor_id(smp_processor_id()), but it would work correctly for boot_cpuid == 0, which could lead to difficult to reproduce or find bugs. The hard id does not seem to be used by the rest of early_init_devtree(), it just looks like all this code might have been put here to allocate somewhere to store boot CPU hardware id while scanning the devtree. Rearrange things so the hwid is put in a global variable like boot_cpuid, and do all the paca allocation and boot paca setup in the 64-bit early_setup() after we have everything ready to go. The paca_ptrs[0] re-poisoning code in early_setup does not seem to have ever worked, because paca_ptrs[0] was never not-poisoned when boot_cpuid is not 0. Signed-off-by:
Nicholas Piggin <npiggin@gmail.com> [mpe: Fix build error on 32-bit] Signed-off-by:
Michael Ellerman <mpe@ellerman.id.au> Link: https://lore.kernel.org/r/20221216115930.2667772-4-npiggin@gmail.com
-
Nicholas Piggin authored
powerpc/64 can boot on a non-zero SMP processor id. Initially, the boot CPU is said to be "assumed to be 0" until early_init_devtree() discovers the id from the device tree. That is not a good description because the assumption can be wrong and that has to be handled, the better description is that 0 is used as a placeholder, and things are fixed after the real id is discovered. smp_processor_id() is set to the boot cpuid, but task_cpu(current) is not, which causes the smp_processor_id() == task_cpu(current) invariant to be broken until init_idle() in sched_init(). This is quite fragile and could lead to subtle bugs in future. One bug is that validate_sp_size uses task_cpu() to get the process stack, so any stack trace from the booting CPU between early_init_devtree() and sched_init() will have problems. Early on paca_ptrs[0] will be poisoned, so that can cause machine checks dereferencing that memory in real mode. Later, validating the current stack pointer against the idle task of a different secondary will probably cause no stack trace to be printed. Fix this by setting thread_info->cpu right after smp_processor_id() is set to the boot cpuid. Signed-off-by:
-
Nicholas Piggin authored
The stress_hpt memblock allocation did not pass in an alignment, which causes a stack dump in early boot (that I missed, oops). Fixes: 6b34a099 ("powerpc/64s/hash: add stress_hpt kernel boot option to increase hash faults") Signed-off-by:
-
Nicholas Piggin authored
As the earlier comment explains, __secondary_hold_spinloop does not have to be accessed at its virtual address, slightly simplifying code. Signed-off-by:
-
Nicholas Piggin authored
Move some basic Book3S initialisation after prom to a function similar to what Book3E looks like. Book3E returns from this function at the virtual address mapping, and Book3S will do the same in a later change, so making them look similar helps with that. Signed-off-by:
-
Nicholas Piggin authored
Replace open-coded toc-relative address calculation with helper macros, commit dab3b8f4 ("powerpc/64: asm use consistent global variable declaration and access") made similar conversions already but missed this one. This allows data addressing model to be changed more easily. Signed-off-by:
-
Frederic Barrat authored
pnv_ioda_setup_pe_res() calls opal to map a resource with a PE. However, the code assumes the resource is allocated and it uses the resource address to find out the segment(s) which need to be mapped to the PE. In the unlikely case where the resource hasn't been allocated, the computation for the segment number is garbage, which can lead to invalid memory access and potentially a kernel crash, such as: [ ] pci_bus 0002:02: Configuring PE for bus [ ] pci 0002:02 : [PE# fc] Secondary bus 0x0000000000000002..0x0000000000000002 associated with PE#fc [ ] BUG: Kernel NULL pointer dereference on write at 0x00000000 [ ] Faulting instruction address: 0xc00000000005eac4 [ ] Oops: Kernel access of bad area, sig: 7 [#1] [ ] LE PAGE_SIZE=64K MMU=Radix SMP NR_CPUS=2048 NUMA PowerNV [ ] Modules linked in: [ ] CPU: 12 PID: 1 Comm: swapper/20 Not tainted 5.10.50-openpower1 #2 [ ] NIP: c00000000005eac4 LR: c00000000005ea44 CTR: 0000000030061b9c [ ] REGS: c000200007383650 TRAP: 0300 Not tainted (5.10.50-openpower1) [ ] MSR: 9000000000009033 <SF,HV,EE,ME,IR,DR,RI,LE> CR: 44000224 XER: 20040000 [ ] CFAR: c00000000005eaa0 DAR: 0000000000000000 DSISR: 02080000 IRQMASK: 0 [ ] GPR00: c00000000005dd98 c0002000073838e0 c00000000185de00 c000200fff018960 [ ] GPR04: 00000000000000fc 0000000000000003 0000000000000000 0000000000000000 [ ] GPR08: 0000000000000000 0000000000000000 0000000000000000 9000000000001033 [ ] GPR12: 0000000031cb0000 c000000ffffe6a80 c000000000010a58 0000000000000000 [ ] GPR16: 0000000000000000 0000000000000000 0000000000000000 0000000000000000 [ ] GPR20: 0000000000000000 0000000000000000 0000000000000000 c00000000711e200 [ ] GPR24: 0000000000000100 c000200009501120 c00020000cee2800 00000000000003ff [ ] GPR28: c000200fff018960 0000000000000000 c000200ffcb7fd00 0000000000000000 [ ] NIP [c00000000005eac4] pnv_ioda_setup_pe_res+0x94/0x1a0 [ ] LR [c00000000005ea44] pnv_ioda_setup_pe_res+0x14/0x1a0 [ ] Call Trace: [ ] [c0002000073838e0] [c00000000005eb98] pnv_ioda_setup_pe_res+0x168/0x1a0 (unreliable) [ ] [c000200007383970] [c00000000005dd98] pnv_pci_ioda_dma_dev_setup+0x43c/0x970 [ ] [c000200007383a60] [c000000000032cdc] pcibios_bus_add_device+0x78/0x18c [ ] [c000200007383aa0] [c00000000028f2bc] pci_bus_add_device+0x28/0xbc [ ] [c000200007383b10] [c00000000028f3a0] pci_bus_add_devices+0x50/0x7c [ ] [c000200007383b50] [c00000000028f3c4] pci_bus_add_devices+0x74/0x7c [ ] [c000200007383b90] [c00000000028f3c4] pci_bus_add_devices+0x74/0x7c [ ] [c000200007383bd0] [c00000000069ad0c] pcibios_init+0xf0/0x104 [ ] [c000200007383c50] [c0000000000106d8] do_one_initcall+0x84/0x1c4 [ ] [c000200007383d20] [c0000000006910b8] kernel_init_freeable+0x264/0x268 [ ] [c000200007383dc0] [c000000000010a68] kernel_init+0x18/0x138 [ ] [c000200007383e20] [c00000000000cbfc] ret_from_kernel_thread+0x5c/0x80 [ ] Instruction dump: [ ] 7f89e840 409d000c 7fbbf840 409c000c 38210090 4848f448 809c002c e95e0120 [ ] 7ba91764 38a00003 57a7043e 38c00000 <7c8a492e> 5484043e e87e0018 4bff23bd Hitting the problem is not that easy. It was seen with a (semi-bogus) PCI device with a class code of 0. The generic PCI framework doesn't allocate resources in such a case. The patch is simply skipping resources which are still flagged with IORESOURCE_UNSET. We don't have the problem with 64-bit mem resources, as the address of the resource is checked to be within the range of the 64-bit mmio window. See pnv_ioda_reserve_dev_m64_pe() and pnv_pci_is_m64(). Reported-by:
Andrew Jeffery <andrew@aj.id.au> Fixes: 23e79425 ("powerpc/powernv: Simplify pnv_ioda_setup_pe_seg()") Signed-off-by:
Frederic Barrat <fbarrat@linux.ibm.com> Signed-off-by:
-
Nicholas Piggin authored
cputime_t is no longer a type, so VIRT_CPU_ACCOUNTING_GEN does not have any affect on the type for 32-bit architectures, so there is no reason it can't be supported. Signed-off-by:
-
Nicholas Piggin authored
Context tracking involves tracking user, kernel, guest switches. 32-bit shares interrupt and syscall entry and exit code (and context tracking calls) with 64-bit, and KVM can not be selected if CONTEXT_TRACKING_USER is enabled, so context tracking can be enabled for 32-bit. Signed-off-by:
-
Nicholas Piggin authored
There are two separeate implementations for 32-bit and 64-bit which mostly do the same thing. Consolidating on one implementation ends up being smaller and simpler, there is just irq soft-mask reconcile that is specific to 64-bit. There should be no real functional change with this patch, but it does make the context tracking calls necessary for 32-bit to support context tracking. Signed-off-by:
-
Kajol Jain authored
Commit ec3eb9d9 ("powerpc/perf: Use PVR rather than oprofile field to determine CPU version") added usage of pvr value instead of oprofile field to determine the platform. In hv-24x7 pmu driver code, pvr check uses PVR_POWER8 when assigning the interface version for power8 platform. But power8 can also have other pvr values like PVR_POWER8E and PVR_POWER8NVL. Hence the interface version won't be set properly incase of PVR_POWER8E and PVR_POWER8NVL. Fix this issue by adding the checks for PVR_POWER8E and PVR_POWER8NVL as well. Fixes: ec3eb9d9 ("powerpc/perf: Use PVR rather than oprofile field to determine CPU version") Reported-by:
Sachin Sant <sachinp@linux.ibm.com> Signed-off-by:
Kajol Jain <kjain@linux.ibm.com> Tested-by:
-
Christophe Leroy authored
When an ALU instruction is preceded by a MOV instruction that just moves a source register into the destination register of the ALU, replace that MOV+ALU instructions by an ALU operation taking the source of the MOV as second source instead of using its destination. Before the change, code could look like the following, with superfluous separate register move (mr) instructions. 70: 7f c6 f3 78 mr r6,r30 74: 7f a5 eb 78 mr r5,r29 78: 30 c6 ff f4 addic r6,r6,-12 7c: 7c a5 01 d4 addme r5,r5 With this commit, addition instructions take r30 and r29 directly. 70: 30 de ff f4 addic r6,r30,-12 74: 7c bd 01 d4 addme r5,r29 Signed-off-by:
Christophe Leroy <christophe.leroy@csgroup.eu> Signed-off-by:
-
Christophe Leroy authored
At the time being, all ALU operation are performed with same L-source and destination, requiring the L-source to be moved into destination via a separate register move, like: 70: 7f c6 f3 78 mr r6,r30 74: 7f a5 eb 78 mr r5,r29 78: 30 c6 ff f4 addic r6,r6,-12 7c: 7c a5 01 d4 addme r5,r5 Introduce a second source register to all ALU operations. For the time being that second source register is made equal to the destination register. That change will allow, via following patch, to optimise the generated code as: 70: 30 de ff f4 addic r6,r30,-12 74: 7c bd 01 d4 addme r5,r29 Signed-off-by:
-
Christophe Leroy authored
Simplify multiplications and divisions with constants when the constant is 1 or -1. When the constant is a power of 2, replace them by bit shits. Signed-off-by:
-
Christophe Leroy authored
Now that two real additional passes are performed in case of extra pass requested by BPF core, padding is not needed anymore except during initial pass done before memory allocation to count maximum possible program size. So, only do the padding when 'image' is NULL. Signed-off-by:
-
Christophe Leroy authored
BPF core calls the jit compiler again for an extra pass in order to properly set subprog addresses. Unlike other architectures, powerpc only updates the addresses during that extra pass. It means that holes must have been left in the code in order to enable the maximum possible instruction size. In order to avoid waste of space, and waste of CPU time on powerpc processors on which the NOP instruction is not 0-cycle, perform two real additional passes. Signed-off-by:
-
Christophe Leroy authored
BPF progs are never called with more than one argument, plus the tail call count as a second argument when needed. So, no need to retrieve 9th and 10th argument (5th 64 bits argument) from the stack in prologue. Signed-off-by:
-
Christophe Leroy authored
Until now a stack frame was set at all time due to the need to keep tail call counter in the stack. But since commit 89d21e25 ("powerpc/bpf/32: Fix Oops on tail call tests") the tail call counter is passed via register r4. It is therefore not necessary anymore to have a stack frame for that. Just like PPC64, implement bpf_has_stack_frame() and only sets the frame when needed. The difference with PPC64 is that PPC32 doesn't have a redzone, so the stack is required as soon as non volatile registers are used or when tail call count is set up. Signed-off-by:
-
Christophe Leroy authored
r4 is cleared at function entry and used as tail call count. But when the function does not perform tail call, r4 is ignored, so no need to clear it. Replace it by a NOP in that case. Signed-off-by:
-
Christophe Leroy authored
That test was introducted in 2006 by commit 00ae36de ("[POWERPC] Better check in show_instructions"). At that time, there was no BPF progs. As seen in message of commit 89d21e25 ("powerpc/bpf/32: Fix Oops on tail call tests"), when a page fault occurs in test_bpf.ko for instance, the code is dumped as XXXXXXXXs. Allthough __kernel_text_address() checks is_bpf_text_address(), it seems it is not enough. Today, show_instructions() uses get_kernel_nofault() to read the code, so there is no real need for additional verifications. ARM64 and x86 don't do any additional check before dumping instructions. Do the same and remove __kernel_text_address() in show_instructions(). Signed-off-by:
-
Ganesh Goudar authored
For all unrecoverable errors we are missing to log the error, Since machine_check_log_err() is not getting called for unrecoverable errors. machine_check_log_err() is called from deferred handler, To run deferred handlers we have to do irq work raise from the exception handler. For recoverable errors exception vector code takes care of running deferred handlers. For unrecoverable errors raise irq work in save_mce_event(), So that we log the error from MCE deferred handler. Log without this change MCE: CPU27: machine check (Severe) Real address Load/Store (foreign/control memory) [Not recovered] MCE: CPU27: PID: 10580 Comm: inject-ra-err NIP: [0000000010000df4] MCE: CPU27: Initiator CPU MCE: CPU27: Unknown Log with this change MCE: CPU24: machine check (Severe) Real address Load/Store (foreign/control memory) [Not recovered] MCE: CPU24: PID: 1589811 Comm: inject-ra-err NIP: [0000000010000e48] MCE: CPU24: Initiator CPU MCE: CPU24: Unknown RTAS: event: 5, Type: Platform Error (224), Severity: 3 Signed-off-by:
Ganesh Goudar <ganeshgr@linux.ibm.com> Reviewed-by:
Mahesh Salgaonkar <mahesh@linux.ibm.com> Signed-off-by:
-
- 09 Feb, 2023 5 commits
-
-
Benjamin Gray authored
A couple of tests roll their own auto-allocating file read logic. Add a generic implementation and convert them to use it. Signed-off-by:
Benjamin Gray <bgray@linux.ibm.com> Signed-off-by:
-
Benjamin Gray authored
Add helper functions to read and write (unsigned) long values directly from/to files. One of the kernel interfaces uses hex strings, so we need to allow passing a base too. Signed-off-by:
-
Benjamin Gray authored
Often a file is expected to hold an integral value. Existing functions will use a C stdlib function like atoi or strtol to parse the file. These operations are error prone, with complicated error conditions (atoi returns 0 if not a number, and is undefined behaviour if not in range. strtol returns 0 if not a number, and LONG_MIN/MAX if not in range + sets errno to ERANGE). Signed-off-by:
-
Benjamin Gray authored
Debugfs files are not always integers, so make *_file return/write a byte buffer, and *_int deal with int values specifically. This increases consistency with the other file read/write helpers. Signed-off-by:
Andrew Donnellan <ajd@linux.ibm.com> Signed-off-by:
-
Benjamin Gray authored
File read/write is reimplemented in about 5 different ways in the various PowerPC selftests. This indicates it should be a common util. Add a common read_file / write_file implementation and convert users to it where (easily) possible. Signed-off-by:
-
- 08 Feb, 2023 5 commits
-
-
Greg Kroah-Hartman authored
When calling debugfs_lookup() the result must have dput() called on it, otherwise the memory will leak over time. To make things simpler, just call debugfs_lookup_and_remove() instead which handles all of the logic at once. Signed-off-by:
Greg Kroah-Hartman <gregkh@linuxfoundation.org> Signed-off-by:
-
Nicholas Piggin authored
This looks like it came across from x86, but x86 uses TLB_FLUSH_ALL as a parameter to internal functions. Powerpc never sets it anywhere. Remove the associated logic and leave a warning for now. Signed-off-by:
-
Nicholas Piggin authored
The MMU_NO_CONTEXT checks are an unnecessary complication. Make these warn to prepare to remove them in future. Signed-off-by:
-
Nicholas Piggin authored
need_flush_all is only set by arch code to instruct generic tlb_flush to flush all. It is never set by powerpc, so it can be removed. Signed-off-by:
-
Christophe Leroy authored
CLANG only knows the following CPUs: generic, 440, 450, 601, 602, 603, 603e, 603ev, 604, 604e, 620, 630, g3, 7400, g4, 7450, g4+, 750, 8548, 970, g5, a2, e500, e500mc, e5500, power3, pwr3, power4, pwr4, power5, pwr5, power5x, pwr5x, power6, pwr6, power6x, pwr6x, power7, pwr7, power8, pwr8, power9, pwr9, power10, pwr10, powerpc, ppc, ppc32, powerpc64, ppc64, powerpc64le, ppc64le, futur Disable other ones when CC_IS_CLANG. Signed-off-by:
-
- 07 Feb, 2023 1 commit
-
-
Pali Rohár authored
The "pci-OF-bus-map" property was declared deprecated in 2006 [1] and to the best of everyone's knowledge is not used by anything anymore [2]. The creation of the property was disabled on powermac (arch/powerpc) in 2005 by commit 35499c01 ("powerpc: Merge in 64-bit powermac support."). But it is still created by default on CHRP. On powermac the actual map (pci_to_OF_bus_map) is still used by default, even though the device tree property is not created. Add an option to enable/disable use of the pci_to_OF_bus_map, and creation of the property (on CHRP). Disabling the option allows enabling CONFIG_PPC_PCI_BUS_NUM_DOMAIN_DEPENDENT which allows "normal" bus numbering and more than 256 buses, like 64-bit and other architectures. Mark the new option as default n, the intention is that the option and the code will be removed in a future release. [1]: https://lore.kernel.org/linuxppc-dev/1148016268.13249.14.camel@localhost.localdomain/ [2]: https://lore.kernel.org/all/575f239205e8635add81c9f902b7d9db7beb83ea.camel@kernel.crashing.org/Signed-off-by:
Pali Rohár <pali@kernel.org> [mpe: Reword commit & help text, shrink option name, rework to fix build errors] Signed-off-by:
-
- 30 Jan, 2023 8 commits
-
-
Pali Rohár authored
It makes sense to enable CONFIG_PPC_PCI_BUS_NUM_DOMAIN_DEPENDENT by default (when possible by dependencies) to take advantages of all 256 PCI buses on each PCI domain, like it is already on all other kernel architectures. Fixes: 56635681 ("powerpc/pci: Add config option for using all 256 PCI buses") Signed-off-by:
-
Nicholas Piggin authored
Interrupt handlers called by soft-pending irq replay code can run softirqs, softirq replay enables and disables local irqs, which allows interrupts to come in including soft-masked interrupts, and it can cause pending irqs to be replayed again. That makes the soft irq replay state machine and possible races more complicated and fragile than it needs to be. Use irq_enter/irq_exit around irq replay to prevent softirqs running while interrupts are being replayed. Softirqs will now be run at the irq_exit() call after all the irq replaying is done. This prevents irqs being replayed while irqs are being replayed, and should hopefully make things simpler and easier to think about and debug. A new PACA_IRQ_REPLAYING is added to prevent asynchronous interrupt handlers hard-enabling EE while pending irqs are being replayed, because that causes new pending irqs to arrive which is also a complexity. This means pending irqs won't be profiled quite so well because perf irqs can't be taken. Signed-off-by:
-
Christophe Leroy authored
NO_IRQ is a relic from the old days. It is not used anymore in core functions. By the way, function irq_of_parse_and_map() returns value 0 on error. In some drivers, NO_IRQ is erroneously used to check the return of irq_of_parse_and_map(). It is not a real bug today because the only architectures using the drivers being fixed by this patch define NO_IRQ as 0, but there are architectures which define NO_IRQ as -1. If one day those architectures start using the non fixed drivers, there will be a problem. Long time ago Linus advocated for not using NO_IRQ, see https://lore.kernel.org/all/Pine.LNX.4.64.0511211150040.13959@g5.osdl.org He re-iterated the same view recently in https://lore.kernel.org/all/CAHk-=wg2Pkb9kbfbstbB91AJA2SF6cySbsgHG-iQMq56j3VTcA@mail.gmail.com So test !irq instead of tesing irq == NO_IRQ. All other usage of NO_IRQ for powerpc were removed in previous cycles so the time has come to remove NO_IRQ completely for powerpc. Signed-off-by:
-
Nathan Lynch authored
At the time commit f97bb36f ("powerpc/rtas: Turn rtas lock into a raw spinlock") was written, the spinlock lockup detection code called __delay(), which will not make progress if the timebase is not advancing. Since the interprocessor timebase synchronization sequence for chrp, cell, and some now-unsupported Power models can temporarily freeze the timebase through an RTAS function (freeze-time-base), the lock that serializes most RTAS calls was converted to arch_spinlock_t to prevent kernel hangs in the lockup detection code. However, commit bc88c10d ("locking/spinlock/debug: Remove spinlock lockup detection code") removed that inconvenient property from the lock debug code several years ago. So now it should be safe to reintroduce generic locks into the RTAS support code, primarily to increase lockdep coverage. Making rtas_lock a spinlock_t would violate lock type nesting rules because it can be acquired while holding raw locks, e.g. pci_lock and irq_desc->lock. So convert it to raw_spinlock_t. There's no apparent reason not to upgrade timebase_lock as well. Signed-off-by:
Nathan Lynch <nathanl@linux.ibm.com> Signed-off-by:
-
Nathan Lynch authored
Only code internal to the RTAS subsystem needs access to the central lock and parameter block. Remove these from the globally visible 'rtas' struct and make them file-static in rtas.c. Some changed lines in rtas_call() lack appropriate spacing around operators and cause checkpatch errors; fix these as well. Suggested-by:
Laurent Dufour <ldufour@linux.ibm.com> Signed-off-by:
Laurent Dufour <laurent.dufour@fr.ibm.com> Signed-off-by:
-
Nathan Lynch authored
The first symbol exports of RTAS functions and data came with the (now removed) scanlog driver in 2003: https://git.kernel.org/pub/scm/linux/kernel/git/tglx/history.git/commit/?id=f92e361842d5251e50562b09664082dcbd0548bb At the time this was applied, EXPORT_SYMBOL_GPL() was very new, and the exports of rtas_call() etc have remained non-GPL. As new APIs have been added to the RTAS subsystem, their symbol exports have followed the convention set by existing code. However, the historical evidence is that RTAS function exports have been added over time only to satisfy the needs of in-kernel users, and these clients must have fairly intimate knowledge of how the APIs work to use them safely. No out of tree users are known, and future ones seem unlikely. Arguably the default for RTAS symbols should have become EXPORT_SYMBOL_GPL once it was available. Let's make it so now, and exceptions can be evaluated as needed. Signed-off-by:
-
Michael Ellerman authored
Some RTAS symbols are never used by modular code, drop their exports. Signed-off-by:
-
Nathan Lynch authored
No modular code needs access to the 'rtas' struct, so remove the symbol export. Suggested-by:
-
