kernel.git/arch/s390/kernel/smp.c, branch linux-5.13.y

s390/topology: fix topology information when calling cpu hotplug notifiers

2021-09-18T11:42:21Z

commit a052096bdd6809eeab809202726634d1ac975aa1 upstream. The cpu hotplug notifiers are called without updating the core/thread masks when a new CPU is added. This causes problems with code setting up data structures in a cpu hotplug notifier, and relying on that later in normal code. This caused a crash in the new core scheduling code (SCHED_CORE), where rq->core was set up in a notifier depending on cpu masks. To fix this, add a cpu_setup_mask which is used in update_cpu_masks() instead of the cpu_online_mask to determine whether the cpu masks should be set for a certain cpu. Also move update_cpu_masks() to update the masks before calling notify_cpu_starting() so that the notifiers are seeing the updated masks. Signed-off-by: Sven Schnelle Cc: [hca@linux.ibm.com: get rid of cpu_online_mask handling] Signed-off-by: Heiko Carstens Signed-off-by: Greg Kroah-Hartman

s390: preempt: Fix preempt_count initialization

2021-07-14T15:07:51Z

commit 6a942f5780545ebd11aca8b3ac4b163397962322 upstream. S390's init_idle_preempt_count(p, cpu) doesn't actually let us initialize the preempt_count of the requested CPU's idle task: it unconditionally writes to the current CPU's. This clearly conflicts with idle_threads_init(), which intends to initialize *all* the idle tasks, including their preempt_count (or their CPU's, if the arch uses a per-CPU preempt_count). Unfortunately, it seems the way s390 does things doesn't let us initialize every possible CPU's preempt_count early on, as the pages where this resides are only allocated when a CPU is brought up and are freed when it is brought down. Let the arch-specific code set a CPU's preempt_count when its lowcore is allocated, and turn init_idle_preempt_count() into an empty stub. Fixes: f1a0a376ca0c ("sched/core: Initialize the idle task with preemption disabled") Reported-by: Guenter Roeck Signed-off-by: Valentin Schneider Tested-by: Guenter Roeck Reviewed-by: Heiko Carstens Link: https://lore.kernel.org/r/20210707163338.1623014-1-valentin.schneider@arm.com Signed-off-by: Vasily Gorbik Signed-off-by: Greg Kroah-Hartman

sched/core: Initialize the idle task with preemption disabled

2021-07-14T15:06:21Z

[ Upstream commit f1a0a376ca0c4ef1fc3d24e3e502acbb5b795674 ] As pointed out by commit de9b8f5dcbd9 ("sched: Fix crash trying to dequeue/enqueue the idle thread") init_idle() can and will be invoked more than once on the same idle task. At boot time, it is invoked for the boot CPU thread by sched_init(). Then smp_init() creates the threads for all the secondary CPUs and invokes init_idle() on them. As the hotplug machinery brings the secondaries to life, it will issue calls to idle_thread_get(), which itself invokes init_idle() yet again. In this case it's invoked twice more per secondary: at _cpu_up(), and at bringup_cpu(). Given smp_init() already initializes the idle tasks for all *possible* CPUs, no further initialization should be required. Now, removing init_idle() from idle_thread_get() exposes some interesting expectations with regards to the idle task's preempt_count: the secondary startup always issues a preempt_disable(), requiring some reset of the preempt count to 0 between hot-unplug and hotplug, which is currently served by idle_thread_get() -> idle_init(). Given the idle task is supposed to have preemption disabled once and never see it re-enabled, it seems that what we actually want is to initialize its preempt_count to PREEMPT_DISABLED and leave it there. Do that, and remove init_idle() from idle_thread_get(). Secondary startups were patched via coccinelle: @begone@ @@ -preempt_disable(); ... cpu_startup_entry(CPUHP_AP_ONLINE_IDLE); Signed-off-by: Valentin Schneider Signed-off-by: Ingo Molnar Acked-by: Peter Zijlstra Link: https://lore.kernel.org/r/20210512094636.2958515-1-valentin.schneider@arm.com Signed-off-by: Sasha Levin

KVM: s390: diag9c (directed yield) forwarding

2021-03-09T09:16:26Z

When we intercept a DIAG_9C from the guest we verify that the target real CPU associated with the virtual CPU designated by the guest is running and if not we forward the DIAG_9C to the target real CPU. To avoid a diag9c storm we allow a maximal rate of diag9c forwarding. The rate is calculated as a count per second defined as a new parameter of the s390 kvm module: diag9c_forwarding_hz . The default value of 0 is to not forward diag9c. Signed-off-by: Pierre Morel Link: https://lore.kernel.org/r/1613997661-22525-2-git-send-email-pmorel@linux.ibm.com Reviewed-by: Cornelia Huck Signed-off-by: Christian Borntraeger

s390/smp: implement arch_irq_work_raise()

2021-02-23T23:31:22Z

The immediate need to have this is to have bpf_send_signal() send the signal ASAP instead of during the next hrtimer interrupt. However, it should also improve irq_work_queue() latencies in general, as well as get s390 out of the lame architectures list [1]. [1] https://git.kernel.org/pub/scm/linux/kernel/git/torvalds/linux.git/tree/kernel/irq_work.c?h=v5.11#n45 Signed-off-by: Ilya Leoshkevich Reviewed-by: Heiko Carstens Signed-off-by: Vasily Gorbik

s390/smp: smp_emergency_stop() - move cpumask away from stack

2021-02-23T23:31:22Z

Make "cpumask_t cpumask" a static variable to avoid a potential large stack frame. Also protect against potential concurrent callers by introducing a local lock. Note: smp_emergency_stop() gets only called with irqs and machine checks disabled, therefore a cpu local deadlock is not possible. For concurrent callers the first cpu which enters the critical section wins and will stop all other cpus. Signed-off-by: Heiko Carstens Signed-off-by: Vasily Gorbik

s390/smp: __smp_rescan_cpus() - move cpumask away from stack

2021-02-23T23:31:22Z

Avoid a potentially large stack frame and overflow by making "cpumask_t avail" a static variable. There is no concurrent access due to the existing locking. Signed-off-by: Heiko Carstens Signed-off-by: Vasily Gorbik

s390/smp: consolidate locking for smp_rescan()

2021-02-23T23:31:22Z

Move locking to __smp_rescan() instead of duplicating it to all call sites. Reviewed-by: Alexander Gordeev Signed-off-by: Heiko Carstens Signed-off-by: Vasily Gorbik

s390: add stack for machine check handler

2021-02-13T16:17:53Z

The previous code used the normal kernel stack for machine checks. This is problematic when a machine check interrupts a system call or interrupt handler right at the beginning where registers are set up. Assume system_call is interrupted at the first instruction and a machine check is triggered. The machine check handler is called, checks the PSW to see whether it is coming from user space, notices that it is already in kernel mode but %r15 still contains the user space stack. This would lead to a kernel crash. There are basically two ways of fixing that: Either using the 'critical cleanup' approach which compares the address in the PSW to see whether it is already at a point where the stack has been set up, or use an extra stack for the machine check handler. For simplicity, we will go with the second approach and allocate an extra stack. This adds some memory overhead for large systems, but usually large system have plenty of memory so this isn't really a concern. But it keeps the mchk stack setup simple and less error prone. Fixes: 0b0ed657fe00 ("s390: remove critical section cleanup from entry.S") Signed-off-by: Sven Schnelle Cc: # v5.8+ Reviewed-by: Heiko Carstens Signed-off-by: Vasily Gorbik

s390: convert to generic entry

2021-01-19T11:29:26Z

This patch converts s390 to use the generic entry infrastructure from kernel/entry/*. There are a few special things on s390: - PIF_PER_TRAP is moved to TIF_PER_TRAP as the generic code doesn't know about our PIF flags in exit_to_user_mode_loop(). - The old code had several ways to restart syscalls: a) PIF_SYSCALL_RESTART, which was only set during execve to force a restart after upgrading a process (usually qemu-kvm) to pgste page table extensions. b) PIF_SYSCALL, which is set by do_signal() to indicate that the current syscall should be restarted. This is changed so that do_signal() now also uses PIF_SYSCALL_RESTART. Continuing to use PIF_SYSCALL doesn't work with the generic code, and changing it to PIF_SYSCALL_RESTART makes PIF_SYSCALL and PIF_SYSCALL_RESTART more unique. - On s390 calling sys_sigreturn or sys_rt_sigreturn is implemented by executing a svc instruction on the process stack which causes a fault. While handling that fault the fault code sets PIF_SYSCALL to hand over processing to the syscall code on exit to usermode. The patch introduces PIF_SYSCALL_RET_SET, which is set if ptrace sets a return value for a syscall. The s390x ptrace ABI uses r2 both for the syscall number and return value, so ptrace cannot set the syscall number + return value at the same time. The flag makes handling that a bit easier. do_syscall() will just skip executing the syscall if PIF_SYSCALL_RET_SET is set. CONFIG_DEBUG_ASCE was removd in favour of the generic CONFIG_DEBUG_ENTRY. CR1/7/13 will be checked both on kernel entry and exit to contain the correct asces. Signed-off-by: Sven Schnelle Signed-off-by: Vasily Gorbik