When the controller supports less queues than requested, we
should make sure that queue mapping does the right thing and
not assume that all queues are available. This fixes a crash
when the controller supports less queues than requested.
The rules are:
1. if no write queues are requested, we assign the available queues
to the default queue map. The default and read queue maps share the
existing queues.
2. if write queues are requested:
- first make sure that read queue map gets the requested
nr_io_queues count
- then grant the default queue map the minimum between the requested
nr_write_queues and the remaining queues. If there are no available
queues to dedicate to the default queue map, fallback to (1) and
share all the queues in the existing queue map.
Also, provide a log indication on how we constructed the different
queue maps.
Reported-by: Harris, James R <email address hidden>
Tested-by: Jim Harris <email address hidden>
Cc: <email address hidden> # v5.0+
Suggested-by: Roy Shterman <email address hidden>
Signed-off-by: Sagi Grimberg <email address hidden>
Signed-off-by: Greg Kroah-Hartman <email address hidden>
Signed-off-by: Kamal Mostafa <email address hidden>
Signed-off-by: Khalid Elmously <email address hidden>
ab6479e...
by
Sagi Grimberg <email address hidden>
nvme-tcp: fix possible null deref on a timed out io queue connect
A few new fields were added to mmu_gather to make TLB flush smarter for
huge page by telling what level of page table is changed.
__tlb_reset_range() is used to reset all these page table state to
unchanged, which is called by TLB flush for parallel mapping changes for
the same range under non-exclusive lock (i.e. read mmap_sem).
Before commit dd2283f2605e ("mm: mmap: zap pages with read mmap_sem in
munmap"), the syscalls (e.g. MADV_DONTNEED, MADV_FREE) which may update
PTEs in parallel don't remove page tables. But, the forementioned
commit may do munmap() under read mmap_sem and free page tables. This
may result in program hang on aarch64 reported by Jan Stancek. The
problem could be reproduced by his test program with slightly modified
below.
---8<---
static int map_size = 4096;
static int num_iter = 500;
static long threads_total;
static void *distant_area;
void *map_write_unmap(void *ptr)
{
int *fd = ptr;
unsigned char *map_address;
int i, j = 0;
for (i = 0; i < num_iter; i++) {
map_address = mmap(distant_area, (size_t) map_size, PROT_WRITE | PROT_READ,
MAP_SHARED | MAP_ANONYMOUS, -1, 0);
if (map_address == MAP_FAILED) {
perror("mmap");
exit(1);
}
tlb_finish_mmu()
if (mm_tlb_flush_nested(tlb->mm))
__tlb_reset_range()
__tlb_reset_range() would reset freed_tables and cleared_* bits, but this
may cause inconsistency for munmap() which do free page tables. Then it
may result in some architectures, e.g. aarch64, may not flush TLB
completely as expected to have stale TLB entries remained.
Use fullmm flush since it yields much better performance on aarch64 and
non-fullmm doesn't yields significant difference on x86.
The original proposed fix came from Jan Stancek who mainly debugged this
issue, I just wrapped up everything together.
Jan's testing results:
v5.2-rc2-24-gbec7550cca10
--------------------------
mean stddev
real 37.382 2.780
user 1.420 0.078
sys 54.658 1.855
v5.2-rc2-24-gbec7550cca10 + "mm: mmu_gather: remove __tlb_reset_range() for force flush"
---------------------------------------------------------------------------------------_
mean stddev
real 37.119 2.105
user 1.548 0.087
sys 55.698 1.357
[<email address hidden>: coding-style fixes]
Link: http://<email address hidden>
Fixes: dd2283f2605e ("mm: mmap: zap pages with read mmap_sem in munmap")
Signed-off-by: Yang Shi <email address hidden>
Signed-off-by: Jan Stancek <email address hidden>
Reported-by: Jan Stancek <email address hidden>
Tested-by: Jan Stancek <email address hidden>
Suggested-by: Will Deacon <email address hidden>
Tested-by: Will Deacon <email address hidden>
Acked-by: Will Deacon <email address hidden>
Cc: Peter Zijlstra <email address hidden>
Cc: Nick Piggin <email address hidden>
Cc: "Aneesh Kumar K.V" <email address hidden>
Cc: Nadav Amit <email address hidden>
Cc: Minchan Kim <email address hidden>
Cc: Mel Gorman <email address hidden>
Cc: <email address hidden> [4.20+]
Signed-off-by: Andrew Morton <email address hidden>
Signed-off-by: Linus Torvalds <email address hidden>
Signed-off-by: Greg Kroah-Hartman <email address hidden>
Signed-off-by: Kamal Mostafa <email address hidden>
Currently the HV KVM code uses kvm->lock in conjunction with a flag,
kvm->arch.mmu_ready, to synchronize MMU setup and hold off vcpu
execution until the MMU-related data structures are ready. However,
this means that kvm->lock is being taken inside vcpu->mutex, which
is contrary to Documentation/virtual/kvm/locking.txt and results in
lockdep warnings.
To fix this, we add a new mutex, kvm->arch.mmu_setup_lock, which nests
inside the vcpu mutexes, and is taken in the places where kvm->lock
was taken that are related to MMU setup.
Additionally we take the new mutex in the vcpu creation code at the
point where we are creating a new vcore, in order to provide mutual
exclusion with kvmppc_update_lpcr() and ensure that an update to
kvm->arch.lpcr doesn't get missed, which could otherwise lead to a
stale vcore->lpcr value.
Make sure only the portals for the online CPUs are used.
Without this change, there are issues when someone boots with
maxcpus=n, with n < actual number of cores available as frames
either received or corresponding to the transmit confirmation
path would be offered for dequeue to the offline CPU portals,
getting lost.
Signed-off-by: Madalin Bucur <email address hidden>
Signed-off-by: David S. Miller <email address hidden>
Signed-off-by: Sasha Levin <email address hidden>
Signed-off-by: Kamal Mostafa <email address hidden>
Signed-off-by: Khalid Elmously <email address hidden>
