linux/virt/kvm/irqchip.c

240 lines
5.6 KiB
C
Raw Normal View History

// SPDX-License-Identifier: GPL-2.0-only
/*
* irqchip.c: Common API for in kernel interrupt controllers
* Copyright (c) 2007, Intel Corporation.
* Copyright 2010 Red Hat, Inc. and/or its affiliates.
* Copyright (c) 2013, Alexander Graf <agraf@suse.de>
*
* This file is derived from virt/kvm/irq_comm.c.
*
* Authors:
* Yaozu (Eddie) Dong <Eddie.dong@intel.com>
* Alexander Graf <agraf@suse.de>
*/
#include <linux/kvm_host.h>
#include <linux/slab.h>
kvm/irqchip: Speed up KVM_SET_GSI_ROUTING When starting lots of dataplane devices the bootup takes very long on Christian's s390 with irqfd patches. With larger setups he is even able to trigger some timeouts in some components. Turns out that the KVM_SET_GSI_ROUTING ioctl takes very long (strace claims up to 0.1 sec) when having multiple CPUs. This is caused by the synchronize_rcu and the HZ=100 of s390. By changing the code to use a private srcu we can speed things up. This patch reduces the boot time till mounting root from 8 to 2 seconds on my s390 guest with 100 disks. Uses of hlist_for_each_entry_rcu, hlist_add_head_rcu, hlist_del_init_rcu are fine because they do not have lockdep checks (hlist_for_each_entry_rcu uses rcu_dereference_raw rather than rcu_dereference, and write-sides do not do rcu lockdep at all). Note that we're hardly relying on the "sleepable" part of srcu. We just want SRCU's faster detection of grace periods. Testing was done by Andrew Theurer using netperf tests STREAM, MAERTS and RR. The difference between results "before" and "after" the patch has mean -0.2% and standard deviation 0.6%. Using a paired t-test on the data points says that there is a 2.5% probability that the patch is the cause of the performance difference (rather than a random fluctuation). (Restricting the t-test to RR, which is the most likely to be affected, changes the numbers to respectively -0.3% mean, 0.7% stdev, and 8% probability that the numbers actually say something about the patch. The probability increases mostly because there are fewer data points). Cc: Marcelo Tosatti <mtosatti@redhat.com> Cc: Michael S. Tsirkin <mst@redhat.com> Tested-by: Christian Borntraeger <borntraeger@de.ibm.com> # s390 Reviewed-by: Christian Borntraeger <borntraeger@de.ibm.com> Signed-off-by: Christian Borntraeger <borntraeger@de.ibm.com> Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
2014-01-16 20:44:20 +08:00
#include <linux/srcu.h>
#include <linux/export.h>
#include <trace/events/kvm.h>
int kvm_irq_map_gsi(struct kvm *kvm,
struct kvm_kernel_irq_routing_entry *entries, int gsi)
{
struct kvm_irq_routing_table *irq_rt;
struct kvm_kernel_irq_routing_entry *e;
int n = 0;
irq_rt = srcu_dereference_check(kvm->irq_routing, &kvm->irq_srcu,
lockdep_is_held(&kvm->irq_lock));
KVM: irqfd: fix NULL pointer dereference in kvm_irq_map_gsi Found by syzkaller: BUG: unable to handle kernel NULL pointer dereference at 0000000000000120 IP: [<ffffffffa0797202>] kvm_irq_map_gsi+0x12/0x90 [kvm] PGD 6f80b067 PUD b6535067 PMD 0 Oops: 0000 [#1] SMP CPU: 3 PID: 4988 Comm: a.out Not tainted 4.4.9-300.fc23.x86_64 #1 [...] Call Trace: [<ffffffffa0795f62>] irqfd_update+0x32/0xc0 [kvm] [<ffffffffa0796c7c>] kvm_irqfd+0x3dc/0x5b0 [kvm] [<ffffffffa07943f4>] kvm_vm_ioctl+0x164/0x6f0 [kvm] [<ffffffff81241648>] do_vfs_ioctl+0x298/0x480 [<ffffffff812418a9>] SyS_ioctl+0x79/0x90 [<ffffffff817a1062>] tracesys_phase2+0x84/0x89 Code: b5 71 a7 e0 5b 41 5c 41 5d 5d f3 c3 66 66 66 66 2e 0f 1f 84 00 00 00 00 00 0f 1f 44 00 00 55 48 8b 8f 10 2e 00 00 31 c0 48 89 e5 <39> 91 20 01 00 00 76 6a 48 63 d2 48 8b 94 d1 28 01 00 00 48 85 RIP [<ffffffffa0797202>] kvm_irq_map_gsi+0x12/0x90 [kvm] RSP <ffff8800926cbca8> CR2: 0000000000000120 Testcase: #include <unistd.h> #include <sys/syscall.h> #include <string.h> #include <stdint.h> #include <linux/kvm.h> #include <fcntl.h> #include <sys/ioctl.h> long r[26]; int main() { memset(r, -1, sizeof(r)); r[2] = open("/dev/kvm", 0); r[3] = ioctl(r[2], KVM_CREATE_VM, 0); struct kvm_irqfd ifd; ifd.fd = syscall(SYS_eventfd2, 5, 0); ifd.gsi = 3; ifd.flags = 2; ifd.resamplefd = ifd.fd; r[25] = ioctl(r[3], KVM_IRQFD, &ifd); return 0; } Reported-by: Dmitry Vyukov <dvyukov@google.com> Cc: stable@vger.kernel.org Signed-off-by: Paolo Bonzini <pbonzini@redhat.com> Signed-off-by: Radim Krčmář <rkrcmar@redhat.com>
2016-06-01 20:09:21 +08:00
if (irq_rt && gsi < irq_rt->nr_rt_entries) {
hlist_for_each_entry(e, &irq_rt->map[gsi], link) {
entries[n] = *e;
++n;
}
}
return n;
}
int kvm_irq_map_chip_pin(struct kvm *kvm, unsigned irqchip, unsigned pin)
{
struct kvm_irq_routing_table *irq_rt;
irq_rt = srcu_dereference(kvm->irq_routing, &kvm->irq_srcu);
return irq_rt->chip[irqchip][pin];
}
int kvm_send_userspace_msi(struct kvm *kvm, struct kvm_msi *msi)
{
struct kvm_kernel_irq_routing_entry route;
if (!kvm_arch_irqchip_in_kernel(kvm) || (msi->flags & ~KVM_MSI_VALID_DEVID))
return -EINVAL;
route.msi.address_lo = msi->address_lo;
route.msi.address_hi = msi->address_hi;
route.msi.data = msi->data;
route.msi.flags = msi->flags;
route.msi.devid = msi->devid;
return kvm_set_msi(&route, kvm, KVM_USERSPACE_IRQ_SOURCE_ID, 1, false);
}
/*
* Return value:
* < 0 Interrupt was ignored (masked or not delivered for other reasons)
* = 0 Interrupt was coalesced (previous irq is still pending)
* > 0 Number of CPUs interrupt was delivered to
*/
int kvm_set_irq(struct kvm *kvm, int irq_source_id, u32 irq, int level,
bool line_status)
{
struct kvm_kernel_irq_routing_entry irq_set[KVM_NR_IRQCHIPS];
int ret = -1, i, idx;
trace_kvm_set_irq(irq, level, irq_source_id);
/* Not possible to detect if the guest uses the PIC or the
* IOAPIC. So set the bit in both. The guest will ignore
* writes to the unused one.
*/
kvm/irqchip: Speed up KVM_SET_GSI_ROUTING When starting lots of dataplane devices the bootup takes very long on Christian's s390 with irqfd patches. With larger setups he is even able to trigger some timeouts in some components. Turns out that the KVM_SET_GSI_ROUTING ioctl takes very long (strace claims up to 0.1 sec) when having multiple CPUs. This is caused by the synchronize_rcu and the HZ=100 of s390. By changing the code to use a private srcu we can speed things up. This patch reduces the boot time till mounting root from 8 to 2 seconds on my s390 guest with 100 disks. Uses of hlist_for_each_entry_rcu, hlist_add_head_rcu, hlist_del_init_rcu are fine because they do not have lockdep checks (hlist_for_each_entry_rcu uses rcu_dereference_raw rather than rcu_dereference, and write-sides do not do rcu lockdep at all). Note that we're hardly relying on the "sleepable" part of srcu. We just want SRCU's faster detection of grace periods. Testing was done by Andrew Theurer using netperf tests STREAM, MAERTS and RR. The difference between results "before" and "after" the patch has mean -0.2% and standard deviation 0.6%. Using a paired t-test on the data points says that there is a 2.5% probability that the patch is the cause of the performance difference (rather than a random fluctuation). (Restricting the t-test to RR, which is the most likely to be affected, changes the numbers to respectively -0.3% mean, 0.7% stdev, and 8% probability that the numbers actually say something about the patch. The probability increases mostly because there are fewer data points). Cc: Marcelo Tosatti <mtosatti@redhat.com> Cc: Michael S. Tsirkin <mst@redhat.com> Tested-by: Christian Borntraeger <borntraeger@de.ibm.com> # s390 Reviewed-by: Christian Borntraeger <borntraeger@de.ibm.com> Signed-off-by: Christian Borntraeger <borntraeger@de.ibm.com> Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
2014-01-16 20:44:20 +08:00
idx = srcu_read_lock(&kvm->irq_srcu);
i = kvm_irq_map_gsi(kvm, irq_set, irq);
kvm/irqchip: Speed up KVM_SET_GSI_ROUTING When starting lots of dataplane devices the bootup takes very long on Christian's s390 with irqfd patches. With larger setups he is even able to trigger some timeouts in some components. Turns out that the KVM_SET_GSI_ROUTING ioctl takes very long (strace claims up to 0.1 sec) when having multiple CPUs. This is caused by the synchronize_rcu and the HZ=100 of s390. By changing the code to use a private srcu we can speed things up. This patch reduces the boot time till mounting root from 8 to 2 seconds on my s390 guest with 100 disks. Uses of hlist_for_each_entry_rcu, hlist_add_head_rcu, hlist_del_init_rcu are fine because they do not have lockdep checks (hlist_for_each_entry_rcu uses rcu_dereference_raw rather than rcu_dereference, and write-sides do not do rcu lockdep at all). Note that we're hardly relying on the "sleepable" part of srcu. We just want SRCU's faster detection of grace periods. Testing was done by Andrew Theurer using netperf tests STREAM, MAERTS and RR. The difference between results "before" and "after" the patch has mean -0.2% and standard deviation 0.6%. Using a paired t-test on the data points says that there is a 2.5% probability that the patch is the cause of the performance difference (rather than a random fluctuation). (Restricting the t-test to RR, which is the most likely to be affected, changes the numbers to respectively -0.3% mean, 0.7% stdev, and 8% probability that the numbers actually say something about the patch. The probability increases mostly because there are fewer data points). Cc: Marcelo Tosatti <mtosatti@redhat.com> Cc: Michael S. Tsirkin <mst@redhat.com> Tested-by: Christian Borntraeger <borntraeger@de.ibm.com> # s390 Reviewed-by: Christian Borntraeger <borntraeger@de.ibm.com> Signed-off-by: Christian Borntraeger <borntraeger@de.ibm.com> Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
2014-01-16 20:44:20 +08:00
srcu_read_unlock(&kvm->irq_srcu, idx);
while (i--) {
int r;
r = irq_set[i].set(&irq_set[i], kvm, irq_source_id, level,
line_status);
if (r < 0)
continue;
ret = r + ((ret < 0) ? 0 : ret);
}
return ret;
}
static void free_irq_routing_table(struct kvm_irq_routing_table *rt)
{
int i;
if (!rt)
return;
for (i = 0; i < rt->nr_rt_entries; ++i) {
struct kvm_kernel_irq_routing_entry *e;
struct hlist_node *n;
hlist_for_each_entry_safe(e, n, &rt->map[i], link) {
hlist_del(&e->link);
kfree(e);
}
}
kfree(rt);
}
void kvm_free_irq_routing(struct kvm *kvm)
{
/* Called only during vm destruction. Nobody can use the pointer
at this stage */
struct kvm_irq_routing_table *rt = rcu_access_pointer(kvm->irq_routing);
free_irq_routing_table(rt);
}
static int setup_routing_entry(struct kvm *kvm,
struct kvm_irq_routing_table *rt,
struct kvm_kernel_irq_routing_entry *e,
const struct kvm_irq_routing_entry *ue)
{
struct kvm_kernel_irq_routing_entry *ei;
int r;
u32 gsi = array_index_nospec(ue->gsi, KVM_MAX_IRQ_ROUTES);
/*
* Do not allow GSI to be mapped to the same irqchip more than once.
* Allow only one to one mapping between GSI and non-irqchip routing.
*/
hlist_for_each_entry(ei, &rt->map[gsi], link)
if (ei->type != KVM_IRQ_ROUTING_IRQCHIP ||
ue->type != KVM_IRQ_ROUTING_IRQCHIP ||
ue->u.irqchip.irqchip == ei->irqchip.irqchip)
return -EINVAL;
e->gsi = gsi;
e->type = ue->type;
r = kvm_set_routing_entry(kvm, e, ue);
if (r)
return r;
if (e->type == KVM_IRQ_ROUTING_IRQCHIP)
rt->chip[e->irqchip.irqchip][e->irqchip.pin] = e->gsi;
hlist_add_head(&e->link, &rt->map[e->gsi]);
return 0;
}
void __attribute__((weak)) kvm_arch_irq_routing_update(struct kvm *kvm)
{
}
bool __weak kvm_arch_can_set_irq_routing(struct kvm *kvm)
{
return true;
}
int kvm_set_irq_routing(struct kvm *kvm,
const struct kvm_irq_routing_entry *ue,
unsigned nr,
unsigned flags)
{
struct kvm_irq_routing_table *new, *old;
struct kvm_kernel_irq_routing_entry *e;
u32 i, j, nr_rt_entries = 0;
int r;
for (i = 0; i < nr; ++i) {
if (ue[i].gsi >= KVM_MAX_IRQ_ROUTES)
return -EINVAL;
nr_rt_entries = max(nr_rt_entries, ue[i].gsi);
}
nr_rt_entries += 1;
new = kzalloc(struct_size(new, map, nr_rt_entries), GFP_KERNEL_ACCOUNT);
if (!new)
return -ENOMEM;
new->nr_rt_entries = nr_rt_entries;
for (i = 0; i < KVM_NR_IRQCHIPS; i++)
for (j = 0; j < KVM_IRQCHIP_NUM_PINS; j++)
new->chip[i][j] = -1;
for (i = 0; i < nr; ++i) {
r = -ENOMEM;
e = kzalloc(sizeof(*e), GFP_KERNEL_ACCOUNT);
if (!e)
goto out;
r = -EINVAL;
switch (ue->type) {
case KVM_IRQ_ROUTING_MSI:
if (ue->flags & ~KVM_MSI_VALID_DEVID)
goto free_entry;
break;
default:
if (ue->flags)
goto free_entry;
break;
}
r = setup_routing_entry(kvm, new, e, ue);
if (r)
goto free_entry;
++ue;
}
mutex_lock(&kvm->irq_lock);
old = rcu_dereference_protected(kvm->irq_routing, 1);
rcu_assign_pointer(kvm->irq_routing, new);
kvm_irq_routing_update(kvm);
kvm_arch_irq_routing_update(kvm);
mutex_unlock(&kvm->irq_lock);
kvm_arch_post_irq_routing_update(kvm);
kvm/irqchip: Speed up KVM_SET_GSI_ROUTING When starting lots of dataplane devices the bootup takes very long on Christian's s390 with irqfd patches. With larger setups he is even able to trigger some timeouts in some components. Turns out that the KVM_SET_GSI_ROUTING ioctl takes very long (strace claims up to 0.1 sec) when having multiple CPUs. This is caused by the synchronize_rcu and the HZ=100 of s390. By changing the code to use a private srcu we can speed things up. This patch reduces the boot time till mounting root from 8 to 2 seconds on my s390 guest with 100 disks. Uses of hlist_for_each_entry_rcu, hlist_add_head_rcu, hlist_del_init_rcu are fine because they do not have lockdep checks (hlist_for_each_entry_rcu uses rcu_dereference_raw rather than rcu_dereference, and write-sides do not do rcu lockdep at all). Note that we're hardly relying on the "sleepable" part of srcu. We just want SRCU's faster detection of grace periods. Testing was done by Andrew Theurer using netperf tests STREAM, MAERTS and RR. The difference between results "before" and "after" the patch has mean -0.2% and standard deviation 0.6%. Using a paired t-test on the data points says that there is a 2.5% probability that the patch is the cause of the performance difference (rather than a random fluctuation). (Restricting the t-test to RR, which is the most likely to be affected, changes the numbers to respectively -0.3% mean, 0.7% stdev, and 8% probability that the numbers actually say something about the patch. The probability increases mostly because there are fewer data points). Cc: Marcelo Tosatti <mtosatti@redhat.com> Cc: Michael S. Tsirkin <mst@redhat.com> Tested-by: Christian Borntraeger <borntraeger@de.ibm.com> # s390 Reviewed-by: Christian Borntraeger <borntraeger@de.ibm.com> Signed-off-by: Christian Borntraeger <borntraeger@de.ibm.com> Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
2014-01-16 20:44:20 +08:00
synchronize_srcu_expedited(&kvm->irq_srcu);
new = old;
r = 0;
goto out;
free_entry:
kfree(e);
out:
free_irq_routing_table(new);
return r;
}