lkp reported a randconfig failure:
In file included from arch/powerpc/include/asm/book3s/64/pkeys.h:6,
from arch/powerpc/kvm/book3s_64_mmu_host.c:15:
arch/powerpc/include/asm/book3s/64/hash-pkey.h: In function 'hash__vmflag_to_pte_pkey_bits':
>> arch/powerpc/include/asm/book3s/64/hash-pkey.h:10:23: error: 'VM_PKEY_BIT0' undeclared
10 | return (((vm_flags & VM_PKEY_BIT0) ? H_PTE_PKEY_BIT0 : 0x0UL) |
| ^~~~~~~~~~~~
We added the include of book3s/64/pkeys.h for pte_to_hpte_pkey_bits(),
but that header on its own should only be included when PPC_MEM_KEYS=y.
Instead include linux/pkeys.h, which brings in the right definitions
when PPC_MEM_KEYS=y and also provides empty stubs when PPC_MEM_KEYS=n.
Fixes: e4e8bc1df6 ("powerpc/kvm: Fix PR KVM with KUAP/MEM_KEYS enabled")
Cc: stable@vger.kernel.org # v5.11+
Reported-by: kernel test robot <lkp@intel.com>
Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
Link: https://lore.kernel.org/r/20210425115831.2818434-1-mpe@ellerman.id.au
As of today, doing iommu_range_alloc() only for !largealloc (npages <= 15)
will only be able to use 3/4 of the available pages, given pages on
largepool not being available for !largealloc.
This could mean some drivers not being able to fully use all the available
pages for the DMA window.
Add pages on largepool as a last resort for !largealloc, making all pages
of the DMA window available.
Signed-off-by: Leonardo Bras <leobras.c@gmail.com>
Reviewed-by: Alexey Kardashevskiy <aik@ozlabs.ru>
Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
Link: https://lore.kernel.org/r/20210318174414.684630-2-leobras.c@gmail.com
Currently both iommu_alloc_coherent() and iommu_free_coherent() align the
desired allocation size to PAGE_SIZE, and gets system pages and IOMMU
mappings (TCEs) for that value.
When IOMMU_PAGE_SIZE < PAGE_SIZE, this behavior may cause unnecessary
TCEs to be created for mapping the whole system page.
Example:
- PAGE_SIZE = 64k, IOMMU_PAGE_SIZE() = 4k
- iommu_alloc_coherent() is called for 128 bytes
- 1 system page (64k) is allocated
- 16 IOMMU pages (16 x 4k) are allocated (16 TCEs used)
It would be enough to use a single TCE for this, so 15 TCEs are
wasted in the process.
Update iommu_*_coherent() to make sure the size alignment happens only
for IOMMU_PAGE_SIZE() before calling iommu_alloc() and iommu_free().
Also, on iommu_range_alloc(), replace ALIGN(n, 1 << tbl->it_page_shift)
with IOMMU_PAGE_ALIGN(n, tbl), which is easier to read and does the
same.
Signed-off-by: Leonardo Bras <leobras.c@gmail.com>
Reviewed-by: Alexey Kardashevskiy <aik@ozlabs.ru>
Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
Link: https://lore.kernel.org/r/20210318174414.684630-1-leobras.c@gmail.com
The IOMMU table is divided into pools for concurrent mappings and each
pool has a separate spinlock. When taking the ownership of an IOMMU group
to pass through a device to a VM, we lock these spinlocks which triggers
a false negative warning in lockdep (below).
This fixes it by annotating the large pool's spinlock as a nest lock
which makes lockdep not complaining when locking nested locks if
the nest lock is locked already.
===
WARNING: possible recursive locking detected
5.11.0-le_syzkaller_a+fstn1 #100 Not tainted
--------------------------------------------
qemu-system-ppc/4129 is trying to acquire lock:
c0000000119bddb0 (&(p->lock)/1){....}-{2:2}, at: iommu_take_ownership+0xac/0x1e0
but task is already holding lock:
c0000000119bdd30 (&(p->lock)/1){....}-{2:2}, at: iommu_take_ownership+0xac/0x1e0
other info that might help us debug this:
Possible unsafe locking scenario:
CPU0
----
lock(&(p->lock)/1);
lock(&(p->lock)/1);
===
Signed-off-by: Alexey Kardashevskiy <aik@ozlabs.ru>
Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
Link: https://lore.kernel.org/r/20210301063653.51003-1-aik@ozlabs.ru
Most platforms allocate IOMMU table structures (specifically it_map)
at the boot time and when this fails - it is a valid reason for panic().
However the powernv platform allocates it_map after a device is returned
to the host OS after being passed through and this happens long after
the host OS booted. It is quite possible to trigger the it_map allocation
panic() and kill the host even though it is not necessary - the host OS
can still use the DMA bypass mode (requires a tiny fraction of it_map's
memory) and even if that fails, the host OS is runnnable as it was without
the device for which allocating it_map causes the panic.
Instead of immediately crashing in a powernv/ioda2 system, this prints
an error and continues. All other platforms still call panic().
Signed-off-by: Alexey Kardashevskiy <aik@ozlabs.ru>
Reviewed-by: David Gibson <david@gibson.dropbear.id.au>
Reviewed-by: Leonardo Bras <leobras.c@gmail.com>
Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
Link: https://lore.kernel.org/r/20210216033307.69863-3-aik@ozlabs.ru
The IOMMU table uses the it_map bitmap to keep track of allocated DMA
pages. This has always been a contiguous array allocated at either
the boot time or when a passed through device is returned to the host OS.
The it_map memory is allocated by alloc_pages() which allocates
contiguous physical memory.
Such allocation method occasionally creates a problem when there is
no big chunk of memory available (no free memory or too fragmented).
On powernv/ioda2 the default DMA window requires 16MB for it_map.
This replaces alloc_pages_node() with vzalloc_node() which allocates
contiguous block but in virtual memory. This should reduce changes of
failure but should not cause other behavioral changes as it_map is only
used by the kernel's DMA hooks/api when MMU is on.
Signed-off-by: Alexey Kardashevskiy <aik@ozlabs.ru>
Reviewed-by: David Gibson <david@gibson.dropbear.id.au>
Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
Link: https://lore.kernel.org/r/20210216033307.69863-2-aik@ozlabs.ru
Eliminate the following coccicheck warning:
./tools/testing/selftests/powerpc/nx-gzip/gzfht_test.c:327:4-5: Unneeded
semicolon
Reported-by: Abaci Robot <abaci@linux.alibaba.com>
Signed-off-by: Yang Li <yang.lee@linux.alibaba.com>
Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
Link: https://lore.kernel.org/r/1612780870-95890-1-git-send-email-yang.lee@linux.alibaba.com
Eliminate the following coccicheck warning:
./arch/powerpc/platforms/powernv/setup.c:160:2-3: Unneeded semicolon
Reported-by: Abaci Robot <abaci@linux.alibaba.com>
Signed-off-by: Yang Li <yang.lee@linux.alibaba.com>
Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
Link: https://lore.kernel.org/r/1612236877-104974-1-git-send-email-yang.lee@linux.alibaba.com
Eliminate the following coccicheck warning:
./arch/powerpc/kernel/eeh.c:782:2-3: Unneeded semicolon
Reported-by: Abaci Robot <abaci@linux.alibaba.com>
Signed-off-by: Yang Li <yang.lee@linux.alibaba.com>
Reviewed-by: Oliver O'Halloran <oohall@gmail.com>
Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
Link: https://lore.kernel.org/r/1612236096-91154-1-git-send-email-yang.lee@linux.alibaba.com
ptrace and perf watchpoints can't co-exists if their address range
overlaps. See commit 29da4f91c0 ("powerpc/watchpoint: Don't allow
concurrent perf and ptrace events") for more detail. Add selftest
for the same.
Sample o/p:
# ./ptrace-perf-hwbreak
test: ptrace-perf-hwbreak
tags: git_version:powerpc-5.8-7-118-g937fa174a15d-dirty
perf cpu event -> ptrace thread event (Overlapping): Ok
perf cpu event -> ptrace thread event (Non-overlapping): Ok
perf thread event -> ptrace same thread event (Overlapping): Ok
perf thread event -> ptrace same thread event (Non-overlapping): Ok
perf thread event -> ptrace other thread event: Ok
ptrace thread event -> perf kernel event: Ok
ptrace thread event -> perf same thread event (Overlapping): Ok
ptrace thread event -> perf same thread event (Non-overlapping): Ok
ptrace thread event -> perf other thread event: Ok
ptrace thread event -> perf cpu event (Overlapping): Ok
ptrace thread event -> perf cpu event (Non-overlapping): Ok
ptrace thread event -> perf same thread & cpu event (Overlapping): Ok
ptrace thread event -> perf same thread & cpu event (Non-overlapping): Ok
ptrace thread event -> perf other thread & cpu event: Ok
success: ptrace-perf-hwbreak
Signed-off-by: Ravi Bangoria <ravi.bangoria@linux.ibm.com>
Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
Link: https://lore.kernel.org/r/20210412112218.128183-5-ravi.bangoria@linux.ibm.com
Extend perf-hwbreak.c selftest to test multiple DAWRs. Also add
testcase for testing 512 byte boundary removal.
Sample o/p:
# ./perf-hwbreak
...
TESTED: Process specific, Two events, diff addr
TESTED: Process specific, Two events, same addr
TESTED: Process specific, Two events, diff addr, one is RO, other is WO
TESTED: Process specific, Two events, same addr, one is RO, other is WO
TESTED: Systemwide, Two events, diff addr
TESTED: Systemwide, Two events, same addr
TESTED: Systemwide, Two events, diff addr, one is RO, other is WO
TESTED: Systemwide, Two events, same addr, one is RO, other is WO
TESTED: Process specific, 512 bytes, unaligned
success: perf_hwbreak
Signed-off-by: Ravi Bangoria <ravi.bangoria@linux.ibm.com>
Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
Link: https://lore.kernel.org/r/20210412112218.128183-4-ravi.bangoria@linux.ibm.com
perf-hwbreak selftest opens hw-breakpoint event at multiple places for
which it has same code repeated. Coalesce that code into a function.
Signed-off-by: Ravi Bangoria <ravi.bangoria@linux.ibm.com>
Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
Link: https://lore.kernel.org/r/20210412112218.128183-3-ravi.bangoria@linux.ibm.com
Also based on the RFI and entry flush tests, it counts the L1D misses
by doing a syscall that does user access: uname, in this case.
Signed-off-by: Thadeu Lima de Souza Cascardo <cascardo@canonical.com>
[dja: forward port, rename function]
Signed-off-by: Daniel Axtens <dja@axtens.net>
Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
Link: https://lore.kernel.org/r/20210225061949.1213404-1-dja@axtens.net
The memory ordering comment no longer applies, because mm_ctx_id is
no longer used anywhere. At best always been difficult to follow.
It's better to consider the load on which the slbmte depends on, which
the MMU depends on before it can start loading TLBs, rather than a
store which may or may not have a subsequent dependency chain to the
slbmte.
So update the comment and we use the load of the mm's user context ID.
This is much more analogous the radix ordering too, which is good.
Signed-off-by: Nicholas Piggin <npiggin@gmail.com>
Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
Link: https://lore.kernel.org/r/20210421151733.212858-1-npiggin@gmail.com
Memory events (mem-loads and mem-stores) currently use the threshold
event selection as issue to finish. Power10 supports issue to complete
as part of thresholding which is more appropriate for mem-loads and
mem-stores. Hence fix the event code for memory events to use issue
to complete.
Fixes: a64e697cef ("powerpc/perf: power10 Performance Monitoring support")
Signed-off-by: Athira Rajeev <atrajeev@linux.vnet.ibm.com>
Reviewed-by: Madhavan Srinivasan <maddy@linux.ibm.com>
Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
Link: https://lore.kernel.org/r/1614840015-1535-1-git-send-email-atrajeev@linux.vnet.ibm.com
Sampled Instruction Event Register (SIER) field [46:48] identifies the
sampled instruction type. ISA v3.1 says value of 0b111 for this field as
reserved, but in POWER10 it denotes LARX/STCX type which will hopefully
be fixed in ISA v3.1 update.
Patch fixes the functions to handle type value 7 for CPU_FTR_ARCH_31.
Fixes: a64e697cef ("powerpc/perf: power10 Performance Monitoring support")
Signed-off-by: Athira Rajeev <atrajeev@linux.vnet.ibm.com>
Reviewed-by: Madhavan Srinivasan <maddy@linux.ibm.com>
[mpe: Avoid reading mmcra until necessary, use early return to deindent if block]
Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
Link: https://lore.kernel.org/r/1614858937-1485-1-git-send-email-atrajeev@linux.vnet.ibm.com
[ 0.000000] ioremap() called early from find_legacy_serial_ports+0x3cc/0x474. Use early_ioremap() instead
find_legacy_serial_ports() is called early from setup_arch(), before
paging_init(). vmalloc is not available yet, ioremap shouldn't be
used that early.
Use early_ioremap() and switch to a regular ioremap() later.
Signed-off-by: Christophe Leroy <christophe.leroy@c-s.fr>
Signed-off-by: Christophe Leroy <christophe.leroy@csgroup.eu>
Tested-by: Chris Packham <chris.packham@alliedtelesis.co.nz>
Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
Link: https://lore.kernel.org/r/103ed8ee9e5973c958ec1da2d0b0764f69395d01.1618925560.git.christophe.leroy@csgroup.eu
At the time being, the fixmap area is defined at the top of
the address space or just below KASAN.
This definition is not valid for PPC64.
For PPC64, use the top of the I/O space.
Because of circular dependencies, it is not possible to include
asm/fixmap.h in asm/book3s/64/pgtable.h , so define a fixed size
AREA at the top of the I/O space for fixmap and ensure during
build that the size is big enough.
Fixes: 265c3491c4 ("powerpc: Add support for GENERIC_EARLY_IOREMAP")
Signed-off-by: Christophe Leroy <christophe.leroy@csgroup.eu>
Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
Link: https://lore.kernel.org/r/0d51620eacf036d683d1a3c41328f69adb601dc0.1618925560.git.christophe.leroy@csgroup.eu
On a kernel config with ALTIVEC=y and PPC_FPU not set/enabled,
there are build errors:
drivers/cpufreq/pmac32-cpufreq.c:262:2: error: implicit declaration of function 'enable_kernel_fp' [-Werror,-Wimplicit-function-declaration]
enable_kernel_fp();
../arch/powerpc/lib/sstep.c: In function 'do_vec_load':
../arch/powerpc/lib/sstep.c:637:3: error: implicit declaration of function 'put_vr' [-Werror=implicit-function-declaration]
637 | put_vr(rn, &u.v);
| ^~~~~~
../arch/powerpc/lib/sstep.c: In function 'do_vec_store':
../arch/powerpc/lib/sstep.c:660:3: error: implicit declaration of function 'get_vr'; did you mean 'get_oc'? [-Werror=implicit-function-declaration]
660 | get_vr(rn, &u.v);
| ^~~~~~
In theory ALTIVEC is independent of PPC_FPU but in practice nobody
is going to build such a machine, so make ALTIVEC require PPC_FPU
by selecting it.
Reported-by: kernel test robot <lkp@intel.com>
Signed-off-by: Randy Dunlap <rdunlap@infradead.org>
Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
Link: https://lore.kernel.org/r/20210421210647.20836-1-rdunlap@infradead.org
opal_mpipl_query_tag() takes a pointer to a 64-bit value, which firmware
writes a value to. As OPAL is traditionally big endian this value will
be big endian.
This can be confirmed by looking at the implementation in skiboot:
static uint64_t opal_mpipl_query_tag(enum opal_mpipl_tags tag, __be64 *tag_val)
{
...
*tag_val = cpu_to_be64(opal_mpipl_tags[tag]);
return OPAL_SUCCESS;
}
Fix the declaration to annotate that the value is big endian.
Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
Link: https://lore.kernel.org/r/20210421125402.1955013-2-mpe@ellerman.id.au
Sparse says:
arch/powerpc/kernel/fadump.c:48:16: warning: symbol 'fadump_kobj' was not declared. Should it be static?
arch/powerpc/kernel/fadump.c:55:27: warning: symbol 'crash_mrange_info' was not declared. Should it be static?
arch/powerpc/kernel/fadump.c:61:27: warning: symbol 'reserved_mrange_info' was not declared. Should it be static?
arch/powerpc/kernel/fadump.c:83:12: warning: symbol 'fadump_cma_init' was not declared. Should it be static?
And indeed none of them are used outside this file, they can all be made
static. Also fadump_kobj needs to be moved inside the ifdef where it's
used.
Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
Link: https://lore.kernel.org/r/20210421125402.1955013-1-mpe@ellerman.id.au
When probe_kernel_read_inst() was created, there was no good place to
put it, so a file called lib/inst.c was dedicated for it.
Since then, probe_kernel_read_inst() has been renamed
copy_inst_from_kernel_nofault(). And mm/maccess.h didn't exist at that
time. Today, mm/maccess.h is related to copy_from_kernel_nofault().
Move copy_inst_from_kernel_nofault() into mm/maccess.c
Signed-off-by: Christophe Leroy <christophe.leroy@csgroup.eu>
Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
Link: https://lore.kernel.org/r/9655d8957313906b77b8db5700a0e33ce06f45e5.1618405715.git.christophe.leroy@csgroup.eu
When probe_kernel_read_inst() was created, it was to mimic
probe_kernel_read() function.
Since then, probe_kernel_read() has been renamed
copy_from_kernel_nofault().
Rename probe_kernel_read_inst() into copy_inst_from_kernel_nofault().
Signed-off-by: Christophe Leroy <christophe.leroy@csgroup.eu>
Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
Link: https://lore.kernel.org/r/b783d1f7cdb8914992384a669a2af57051b6bdcf.1618405715.git.christophe.leroy@csgroup.eu
We have two independant versions of probe_kernel_read_inst(), one for
PPC32 and one for PPC64.
The PPC32 is identical to the first part of the PPC64 version.
The remaining part of PPC64 version is not relevant for PPC32, but
not contradictory, so we can easily have a common function with
the PPC64 part opted out via a IS_ENABLED(CONFIG_PPC64).
The only need is to add a version of ppc_inst_prefix() for PPC32.
Signed-off-by: Christophe Leroy <christophe.leroy@csgroup.eu>
Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
Link: https://lore.kernel.org/r/f7b9dfddef3b3760182c7e5466356c121a293dc9.1618405715.git.christophe.leroy@csgroup.eu
Its name comes from former probe_user_read() function.
That function is now called copy_from_user_nofault().
probe_user_read_inst() uses copy_from_user_nofault() to read only
a few bytes. It is suboptimal.
It does the same as get_user_inst() but in addition disables
page faults.
But on the other hand, it is not used for the time being. So remove it
for now. If one day it is really needed, we can give it a new name
more in line with today's naming, and implement it using get_user_inst()
Signed-off-by: Christophe Leroy <christophe.leroy@csgroup.eu>
Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
Link: https://lore.kernel.org/r/5f6f82572242a59bfee1e19a71194d8f7ef5fca4.1618405715.git.christophe.leroy@csgroup.eu
If the target of a function call is within 32 Mbytes distance, use a
standard function call with 'bl' instead of the 'lis/ori/mtlr/blrl'
sequence.
In the first pass, no memory has been allocated yet and the code
position is not known yet (image pointer is NULL). This pass is there
to calculate the amount of memory to allocate for the EBPF code, so
assume the 4 instructions sequence is required, so that enough memory
is allocated.
Signed-off-by: Christophe Leroy <christophe.leroy@csgroup.eu>
Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
Link: https://lore.kernel.org/r/74944a1e3e5cfecc141e440a6ccd37920e186b70.1618227846.git.christophe.leroy@csgroup.eu
wrtspr() is a function to write an arbitrary value in a special
register. It is used on 8xx to write to SPRN_NRI, SPRN_EID and
SPRN_EIE. Writing any value to one of those will play with MSR EE
and MSR RI regardless of that value.
r0 is used many places in the generated code and using r0 for
that creates an unnecessary dependency of this instruction with
preceding ones using r0 in a few places in vmlinux.
r2 is most likely the most stable register as it contains the
pointer to 'current'.
Using r2 instead of r0 avoids that unnecessary dependency.
Signed-off-by: Christophe Leroy <christophe.leroy@csgroup.eu>
Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
Link: https://lore.kernel.org/r/69f9968f4b592fefda55227f0f7430ea612cc950.1611299687.git.christophe.leroy@csgroup.eu
When we hit an UE while using machine check safe copy routines,
ignore_event flag is set and the event is ignored by mce handler,
And the flag is also saved for defered handling and printing of
mce event information, But as of now saving of this flag is done
on checking if the effective address is provided or physical address
is calculated, which is not right.
Save ignore_event flag regardless of whether the effective address is
provided or physical address is calculated.
Without this change following log is seen, when the event is to be
ignored.
[ 512.971365] MCE: CPU1: machine check (Severe) UE Load/Store [Recovered]
[ 512.971509] MCE: CPU1: NIP: [c0000000000b67c0] memcpy+0x40/0x90
[ 512.971655] MCE: CPU1: Initiator CPU
[ 512.971739] MCE: CPU1: Unknown
[ 512.972209] MCE: CPU1: machine check (Severe) UE Load/Store [Recovered]
[ 512.972334] MCE: CPU1: NIP: [c0000000000b6808] memcpy+0x88/0x90
[ 512.972456] MCE: CPU1: Initiator CPU
[ 512.972534] MCE: CPU1: Unknown
Signed-off-by: Ganesh Goudar <ganeshgr@linux.ibm.com>
Reviewed-by: Santosh Sivaraj <santosh@fossix.org>
Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
Link: https://lore.kernel.org/r/20210407045816.352276-1-ganeshgr@linux.ibm.com
For that, create a 32 bits version of patch_imm64_load_insns()
and create a patch_imm_load_insns() which calls
patch_imm32_load_insns() on PPC32 and patch_imm64_load_insns()
on PPC64.
Adapt optprobes_head.S for PPC32. Use PPC_LL/PPC_STL macros instead
of raw ld/std, opt out things linked to paca and use stmw/lmw to
save/restore registers.
Signed-off-by: Christophe Leroy <christophe.leroy@csgroup.eu>
Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
Link: https://lore.kernel.org/r/bad58c66859b2a475c0ad516b53164ae3b4853cd.1618927318.git.christophe.leroy@csgroup.eu
As of today, if the DDW is big enough to fit (1 << MAX_PHYSMEM_BITS)
it's possible to use direct DMA mapping even with pmem region.
But, if that happens, the window size (len) is set to (MAX_PHYSMEM_BITS
- page_shift) instead of MAX_PHYSMEM_BITS, causing a pagesize times
smaller DDW to be created, being insufficient for correct usage.
Fix this so the correct window size is used in this case.
Fixes: bf6e2d562b ("powerpc/dma: Fallback to dma_ops when persistent memory present")
Signed-off-by: Leonardo Bras <leobras.c@gmail.com>
Reviewed-by: Alexey Kardashevskiy <aik@ozlabs.ru>
Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
Link: https://lore.kernel.org/r/20210420045404.438735-1-leobras.c@gmail.com
The changes to add KUAP support with the hash MMU broke booting of KVM
PR guests. The symptom is no visible progress of the guest, or possibly
just "SLOF" being printed to the qemu console.
Host code is still executing, but breaking into xmon might show a stack
trace such as:
__might_fault+0x84/0xe0 (unreliable)
kvm_read_guest+0x1c8/0x2f0 [kvm]
kvmppc_ld+0x1b8/0x2d0 [kvm]
kvmppc_load_last_inst+0x50/0xa0 [kvm]
kvmppc_exit_pr_progint+0x178/0x220 [kvm_pr]
kvmppc_handle_exit_pr+0x31c/0xe30 [kvm_pr]
after_sprg3_load+0x80/0x90 [kvm_pr]
kvmppc_vcpu_run_pr+0x104/0x260 [kvm_pr]
kvmppc_vcpu_run+0x34/0x48 [kvm]
kvm_arch_vcpu_ioctl_run+0x340/0x450 [kvm]
kvm_vcpu_ioctl+0x2ac/0x8c0 [kvm]
sys_ioctl+0x320/0x1060
system_call_exception+0x160/0x270
system_call_common+0xf0/0x27c
Bisect points to commit b2ff33a10c ("powerpc/book3s64/hash/kuap:
Enable kuap on hash"), but that's just the commit that enabled KUAP with
hash and made the bug visible.
The root cause seems to be that KVM PR is creating kernel mappings that
don't use the correct key, since we switched to using key 3.
We have a helper for adding the right key value, however it's designed
to take a pteflags variable, which the KVM code doesn't have. But we can
make it work by passing 0 for the pteflags, and tell it explicitly that
it should use the kernel key.
With that changed guests boot successfully.
Fixes: d94b827e89 ("powerpc/book3s64/kuap: Use Key 3 for kernel mapping with hash translation")
Cc: stable@vger.kernel.org # v5.11+
Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
Link: https://lore.kernel.org/r/20210419120139.1455937-1-mpe@ellerman.id.au
RCU complains about us calling printk() from an offline CPU:
=============================
WARNING: suspicious RCU usage
5.12.0-rc7-02874-g7cf90e481cb8 #1 Not tainted
-----------------------------
kernel/locking/lockdep.c:3568 RCU-list traversed in non-reader section!!
other info that might help us debug this:
RCU used illegally from offline CPU!
rcu_scheduler_active = 2, debug_locks = 1
no locks held by swapper/0/0.
stack backtrace:
CPU: 0 PID: 0 Comm: swapper/0 Not tainted 5.12.0-rc7-02874-g7cf90e481cb8 #1
Call Trace:
dump_stack+0xec/0x144 (unreliable)
lockdep_rcu_suspicious+0x124/0x144
__lock_acquire+0x1098/0x28b0
lock_acquire+0x128/0x600
_raw_spin_lock_irqsave+0x6c/0xc0
down_trylock+0x2c/0x70
__down_trylock_console_sem+0x60/0x140
vprintk_emit+0x1a8/0x4b0
vprintk_func+0xcc/0x200
printk+0x40/0x54
pseries_cpu_offline_self+0xc0/0x120
arch_cpu_idle_dead+0x54/0x70
do_idle+0x174/0x4a0
cpu_startup_entry+0x38/0x40
rest_init+0x268/0x388
start_kernel+0x748/0x790
start_here_common+0x1c/0x614
Which happens because by the time we get to rtas_stop_self() we are
already offline. In addition the message can be spammy, and is not that
helpful for users, so remove it.
Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
Link: https://lore.kernel.org/r/20210418135413.1204031-1-mpe@ellerman.id.au
We have some interesting code in our Makefile to define _TASK_CPU, based
on awk'ing the value out of asm-offsets.h. It exists to circumvent some
circular header dependencies that prevent us from referring to
task_struct in the relevant code. See the comment around _TASK_CPU in
smp.h for more detail.
Maybe one day we can come up with a better solution, but for now we can
at least limit that logic to 32-bit, because it's not needed for 64-bit.
Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
Link: https://lore.kernel.org/r/20210418131641.1186227-1-mpe@ellerman.id.au
Currently, neither the vio_bus or vio_driver structures provide support
for a shutdown() routine.
Add support for shutdown() by allowing drivers to provide a
implementation via function pointer in their vio_driver struct and
provide a proper implementation in the driver template for the vio_bus
that calls a vio drivers shutdown() if defined.
In the case that no shutdown() is defined by a vio driver and a kexec is
in progress we implement a big hammer that calls remove() to ensure no
further DMA for the devices is possible.
Signed-off-by: Tyrel Datwyler <tyreld@linux.ibm.com>
Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
Link: https://lore.kernel.org/r/20210402001325.939668-1-tyreld@linux.ibm.com
Performance Monitoring Unit (PMU) registers in powerpc provides
information on cycles elapsed between different stages in the
pipeline. This can be used for application tuning. On ISA v3.1
platform, this information is exposed by sampling registers.
Patch adds kernel support to capture two of the cycle counters
as part of perf sample using the sample type:
PERF_SAMPLE_WEIGHT_STRUCT.
The power PMU function 'get_mem_weight' currently uses 64 bit weight
field of perf_sample_data to capture memory latency. But following the
introduction of PERF_SAMPLE_WEIGHT_TYPE, weight field could contain
64-bit or 32-bit value depending on the architexture support for
PERF_SAMPLE_WEIGHT_STRUCT. Patches uses WEIGHT_STRUCT to expose the
pipeline stage cycles info. Hence update the ppmu functions to work for
64-bit and 32-bit weight values.
If the sample type is PERF_SAMPLE_WEIGHT, use the 64-bit weight field.
if the sample type is PERF_SAMPLE_WEIGHT_STRUCT, memory subsystem
latency is stored in the low 32bits of perf_sample_weight structure.
Also for CPU_FTR_ARCH_31, capture the two cycle counter information in
two 16 bit fields of perf_sample_weight structure.
Signed-off-by: Athira Rajeev <atrajeev@linux.vnet.ibm.com>
Reviewed-by: Madhavan Srinivasan <maddy@linux.ibm.com>
Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
Link: https://lore.kernel.org/r/1616425047-1666-2-git-send-email-atrajeev@linux.vnet.ibm.com
The links that are mentioned in this document are no longer
valid. So changed the proper links for NXGZIP user manual and
test cases.
Reported-by: Bulent Abali <abali@us.ibm.com>
Signed-off-by: Haren Myneni <haren@linux.ibm.com>
Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
Link: https://lore.kernel.org/r/08511c1e92ac239f20ac88c73c59d1f8cf02e6ad.camel@linux.ibm.com
The RTAS set-indicator call, when attempting to UNISOLATE a DRC that is
already UNISOLATED or CONFIGURED, returns RTAS_OK and does nothing else
for both QEMU and phyp. This gives us an opportunity to use this
behavior to signal the hypervisor layer when an error during device
removal happens, allowing it to do a proper error handling, while not
breaking QEMU/phyp implementations that don't have this support.
This patch introduces this idea by unisolating all CPU DRCs that failed
to be removed by dlpar_cpu_remove_by_index(), when handling the
PSERIES_HP_ELOG_ID_DRC_INDEX event. This is being done for this event
only because its the only CPU removal event QEMU uses, and there's no
need at this moment to add this mechanism for phyp only code.
Signed-off-by: Daniel Henrique Barboza <danielhb413@gmail.com>
Reviewed-by: David Gibson <david@gibson.dropbear.id.au>
Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
Link: https://lore.kernel.org/r/20210416210216.380291-3-danielhb413@gmail.com
Michael Ellerman