KVM: PPC: Book3s: Retire H_PUT_TCE/etc real mode handlers

LoPAPR defines guest visible IOMMU with hypercalls to use it -
H_PUT_TCE/etc. Implemented first on POWER7 where hypercalls would trap
in the KVM in the real mode (with MMU off). The problem with the real mode
is some memory is not available and some API usage crashed the host but
enabling MMU was an expensive operation.

The problems with the real mode handlers are:
1. Occasionally these cannot complete the request so the code is
copied+modified to work in the virtual mode, very little is shared;
2. The real mode handlers have to be linked into vmlinux to work;
3. An exception in real mode immediately reboots the machine.

If the small DMA window is used, the real mode handlers bring better
performance. However since POWER8, there has always been a bigger DMA
window which VMs use to map the entire VM memory to avoid calling
H_PUT_TCE. Such 1:1 mapping happens once and uses H_PUT_TCE_INDIRECT
(a bulk version of H_PUT_TCE) which virtual mode handler is even closer
to its real mode version.

On POWER9 hypercalls trap straight to the virtual mode so the real mode
handlers never execute on POWER9 and later CPUs.

So with the current use of the DMA windows and MMU improvements in
POWER9 and later, there is no point in duplicating the code.
The 32bit passed through devices may slow down but we do not have many
of these in practice. For example, with this applied, a 1Gbit ethernet
adapter still demostrates above 800Mbit/s of actual throughput.

This removes the real mode handlers from KVM and related code from
the powernv platform.

This updates the list of implemented hcalls in KVM-HV as the realmode
handlers are removed.

This changes ABI - kvmppc_h_get_tce() moves to the KVM module and
kvmppc_find_table() is static now.

Signed-off-by: Alexey Kardashevskiy <aik@ozlabs.ru>
Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
Link: https://lore.kernel.org/r/20220506053755.3820702-1-aik@ozlabs.ru

powerpc/powernv/pci: fix a RCU-list lock

It is unsafe to traverse tbl->it_group_list without the RCU read lock.

WARNING: suspicious RCU usage
5.7.0-rc4-next-20200508 #1 Not tainted
-----------------------------
arch/powerpc/platforms/powernv/pci-ioda-tce.c:355 RCU-list traversed in non-reader section!!

other info that might help us debug this:

rcu_scheduler_active = 2, debug_locks = 1
3 locks held by qemu-kvm/4305:
#0: c000000bc3fe6988 (&container->group_lock){++++}-{3:3}, at: vfio_fops_unl_ioctl+0x108/0x410 [vfio]
#1: c00800000fcc7400 (&vfio.iommu_drivers_lock){+.+.}-{3:3}, at: vfio_fops_unl_ioctl+0x148/0x410 [vfio]
#2: c000000bc3fe4d68 (&container->lock){+.+.}-{3:3}, at: tce_iommu_attach_group+0x3c/0x4f0 [vfio_iommu_spapr_tce]

stack backtrace:
CPU: 4 PID: 4305 Comm: qemu-kvm Not tainted 5.7.0-rc4-next-20200508 #1
Call Trace:
[c0000010f29afa60] [c0000000007154c8] dump_stack+0xfc/0x174 (unreliable)
[c0000010f29afab0] [c0000000001d8ff0] lockdep_rcu_suspicious+0x140/0x164
[c0000010f29afb30] [c0000000000dae2c] pnv_pci_unlink_table_and_group+0x11c/0x200
[c0000010f29afb70] [c0000000000d4a34] pnv_pci_ioda2_unset_window+0xc4/0x190
[c0000010f29afbf0] [c0000000000d4b4c] pnv_ioda2_take_ownership+0x4c/0xd0
[c0000010f29afc30] [c00800000fd60ee0] tce_iommu_attach_group+0x2c8/0x4f0 [vfio_iommu_spapr_tce]
[c0000010f29afcd0] [c00800000fcc11a0] vfio_fops_unl_ioctl+0x238/0x410 [vfio]
[c0000010f29afd50] [c0000000005430a8] ksys_ioctl+0xd8/0x130
[c0000010f29afda0] [c000000000543128] sys_ioctl+0x28/0x40
[c0000010f29afdc0] [c000000000038af4] system_call_exception+0x114/0x1e0
[c0000010f29afe20] [c00000000000c8f0] system_call_common+0xf0/0x278

Signed-off-by: Qian Cai <cai@lca.pw>
Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
Link: https://lore.kernel.org/r/20200510051347.1906-1-cai@lca.pw

powerpc/powernv/ioda: Return correct error if TCE level allocation failed

The iommu_table_ops::xchg_no_kill() callback updates TCE. It is quite
possible that not entire table is allocated if it is huge and multilevel
so xchg may also allocate subtables. If failed, it returns H_HARDWARE
for failed allocation and H_TOO_HARD if it needs it but cannot do because
the alloc parameter is "false" (set when called with MMU=off to force
retry with MMU=on).

The problem is that having separate errors only matters in real mode
(MMU=off) but the only caller with alloc="false" does not check the exact
error code and simply returns H_TOO_HARD; and for every other mode
alloc is "true". Also, the function is also called from the ioctl()
handler of the VFIO SPAPR TCE IOMMU subdriver which does not expect
hypervisor error codes (H_xxx) and will expose them to the userspace.

This converts wrong error codes to -ENOMEM.

Signed-off-by: Alexey Kardashevskiy <aik@ozlabs.ru>
Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
Link: https://lore.kernel.org/r/20200617003835.48831-1-aik@ozlabs.ru

powerpc/powernv/pci: Move tce size parsing to pci-ioda-tce.c

Move it in with the rest of the TCE wrangling rather than carting around
a static prototype in pci-ioda.c

Signed-off-by: Oliver O'Halloran <oohall@gmail.com>
Reviewed-by: Alexey Kardashevskiy <aik@ozlabs.ru>
Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
Link: https://lore.kernel.org/r/20200406030745.24595-7-oohall@gmail.com

powerpc/powernv/ioda: using kfree_rcu() to simplify the code

The callback function of call_rcu() just calls a kfree(), so we
can use kfree_rcu() instead of call_rcu() + callback function.

Signed-off-by: YueHaibing <yuehaibing@huawei.com>
Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
Link: https://lore.kernel.org/r/20190711141818.18044-1-yuehaibing@huawei.com

powerpc/powernv/ioda2: Allocate TCE table levels on demand for default DMA window

We allocate only the first level of multilevel TCE tables for KVM
already (alloc_userspace_copy==true), and the rest is allocated on demand.
This is not enabled though for bare metal.

This removes the KVM limitation (implicit, via the alloc_userspace_copy
parameter) and always allocates just the first level. The on-demand
allocation of missing levels is already implemented.

As from now on DMA map might happen with disabled interrupts, this
allocates TCEs with GFP_ATOMIC; otherwise lockdep reports errors 1].
In practice just a single page is allocated there so chances for failure
are quite low.

To save time when creating a new clean table, this skips non-allocated
indirect TCE entries in pnv_tce_free just like we already do in
the VFIO IOMMU TCE driver.

This changes the default level number from 1 to 2 to reduce the amount
of memory required for the default 32bit DMA window at the boot time.
The default window size is up to 2GB which requires 4MB of TCEs which is
unlikely to be used entirely or at all as most devices these days are
64bit capable so by switching to 2 levels by default we save 4032KB of
RAM per a device.

While at this, add __GFP_NOWARN to alloc_pages_node() as the userspace
can trigger this path via VFIO, see the failure and try creating a table
again with different parameters which might succeed.

[1]:
===
BUG: sleeping function called from invalid context at mm/page_alloc.c:4596
in_atomic(): 1, irqs_disabled(): 1, pid: 1038, name: scsi_eh_1
2 locks held by scsi_eh_1/1038:
#0: 000000005efd659a (&host->eh_mutex){+.+.}, at: ata_eh_acquire+0x34/0x80
#1: 0000000006cf56a6 (&(&host->lock)->rlock){....}, at: ata_exec_internal_sg+0xb0/0x5c0
irq event stamp: 500
hardirqs last enabled at (499): [<c000000000cb8a74>] _raw_spin_unlock_irqrestore+0x94/0xd0
hardirqs last disabled at (500): [<c000000000cb85c4>] _raw_spin_lock_irqsave+0x44/0x120
softirqs last enabled at (0): [<c000000000101120>] copy_process.isra.4.part.5+0x640/0x1a80
softirqs last disabled at (0): [<0000000000000000>] 0x0
CPU: 73 PID: 1038 Comm: scsi_eh_1 Not tainted 5.2.0-rc6-le_nv2_aikATfstn1-p1 #634
Call Trace:
[c000003d064cef50] [c000000000c8e6c4] dump_stack+0xe8/0x164 (unreliable)
[c000003d064cefa0] [c00000000014ed78] ___might_sleep+0x2f8/0x310
[c000003d064cf020] [c0000000003ca084] __alloc_pages_nodemask+0x2a4/0x1560
[c000003d064cf220] [c0000000000c2530] pnv_alloc_tce_level.isra.0+0x90/0x130
[c000003d064cf290] [c0000000000c2888] pnv_tce+0x128/0x3b0
[c000003d064cf360] [c0000000000c2c00] pnv_tce_build+0xb0/0xf0
[c000003d064cf3c0] [c0000000000bbd9c] pnv_ioda2_tce_build+0x3c/0xb0
[c000003d064cf400] [c00000000004cfe0] ppc_iommu_map_sg+0x210/0x550
[c000003d064cf510] [c00000000004b7a4] dma_iommu_map_sg+0x74/0xb0
[c000003d064cf530] [c000000000863944] ata_qc_issue+0x134/0x470
[c000003d064cf5b0] [c000000000863ec4] ata_exec_internal_sg+0x244/0x5c0
[c000003d064cf700] [c0000000008642d0] ata_exec_internal+0x90/0xe0
[c000003d064cf780] [c0000000008650ac] ata_dev_read_id+0x2ec/0x640
[c000003d064cf8d0] [c000000000878e28] ata_eh_recover+0x948/0x16d0
[c000003d064cfa10] [c00000000087d760] sata_pmp_error_handler+0x480/0xbf0
[c000003d064cfbc0] [c000000000884624] ahci_error_handler+0x74/0xe0
[c000003d064cfbf0] [c000000000879fa8] ata_scsi_port_error_handler+0x2d8/0x7c0
[c000003d064cfca0] [c00000000087a544] ata_scsi_error+0xb4/0x100
[c000003d064cfd00] [c000000000802450] scsi_error_handler+0x120/0x510
[c000003d064cfdb0] [c000000000140c48] kthread+0x1b8/0x1c0
[c000003d064cfe20] [c00000000000bd8c] ret_from_kernel_thread+0x5c/0x70
ata1: SATA link up 6.0 Gbps (SStatus 133 SControl 300)
irq event stamp: 2305

========================================================
hardirqs last enabled at (2305): [<c00000000000e4c8>] fast_exc_return_irq+0x28/0x34
hardirqs last disabled at (2303): [<c000000000cb9fd0>] __do_softirq+0x4a0/0x654
WARNING: possible irq lock inversion dependency detected
5.2.0-rc6-le_nv2_aikATfstn1-p1 #634 Tainted: G W
softirqs last enabled at (2304): [<c000000000cba054>] __do_softirq+0x524/0x654
softirqs last disabled at (2297): [<c00000000010f278>] irq_exit+0x128/0x180
--------------------------------------------------------
swapper/0/0 just changed the state of lock:
0000000006cf56a6 (&(&host->lock)->rlock){-...}, at: ahci_single_level_irq_intr+0xac/0x120
but this lock took another, HARDIRQ-unsafe lock in the past:
(fs_reclaim){+.+.}

and interrupts could create inverse lock ordering between them.

other info that might help us debug this:
Possible interrupt unsafe locking scenario:

CPU0 CPU1
---- ----
lock(fs_reclaim);
local_irq_disable();
lock(&(&host->lock)->rlock);
lock(fs_reclaim);
<Interrupt>
lock(&(&host->lock)->rlock);

*** DEADLOCK ***

no locks held by swapper/0/0.

the shortest dependencies between 2nd lock and 1st lock:
-> (fs_reclaim){+.+.} ops: 167579 {
HARDIRQ-ON-W at:
lock_acquire+0xf8/0x2a0
fs_reclaim_acquire.part.23+0x44/0x60
kmem_cache_alloc_node_trace+0x80/0x590
alloc_desc+0x64/0x270
__irq_alloc_descs+0x2e4/0x3a0
irq_domain_alloc_descs+0xb0/0x150
irq_create_mapping+0x168/0x2c0
xics_smp_probe+0x2c/0x98
pnv_smp_probe+0x40/0x9c
smp_prepare_cpus+0x524/0x6c4
kernel_init_freeable+0x1b4/0x650
kernel_init+0x2c/0x148
ret_from_kernel_thread+0x5c/0x70
SOFTIRQ-ON-W at:
lock_acquire+0xf8/0x2a0
fs_reclaim_acquire.part.23+0x44/0x60
kmem_cache_alloc_node_trace+0x80/0x590
alloc_desc+0x64/0x270
__irq_alloc_descs+0x2e4/0x3a0
irq_domain_alloc_descs+0xb0/0x150
irq_create_mapping+0x168/0x2c0
xics_smp_probe+0x2c/0x98
pnv_smp_probe+0x40/0x9c
smp_prepare_cpus+0x524/0x6c4
kernel_init_freeable+0x1b4/0x650
kernel_init+0x2c/0x148
ret_from_kernel_thread+0x5c/0x70
INITIAL USE at:
lock_acquire+0xf8/0x2a0
fs_reclaim_acquire.part.23+0x44/0x60
kmem_cache_alloc_node_trace+0x80/0x590
alloc_desc+0x64/0x270
__irq_alloc_descs+0x2e4/0x3a0
irq_domain_alloc_descs+0xb0/0x150
irq_create_mapping+0x168/0x2c0
xics_smp_probe+0x2c/0x98
pnv_smp_probe+0x40/0x9c
smp_prepare_cpus+0x524/0x6c4
kernel_init_freeable+0x1b4/0x650
kernel_init+0x2c/0x148
ret_from_kernel_thread+0x5c/0x70
}
===

Signed-off-by: Alexey Kardashevskiy <aik@ozlabs.ru>
Reviewed-by: Alistair Popple <alistair@popple.id.au>
Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
Link: https://lore.kernel.org/r/20190718051139.74787-4-aik@ozlabs.ru

powerpc/powernv/ioda: Fix race in TCE level allocation

pnv_tce() returns a pointer to a TCE entry and originally a TCE table
would be pre-allocated. For the default case of 2GB window the table
needs only a single level and that is fine. However if more levels are
requested, it is possible to get a race when 2 threads want a pointer
to a TCE entry from the same page of TCEs.

This adds cmpxchg to handle the race. Note that once TCE is non-zero,
it cannot become zero again.

Fixes: a68bd1267b72 ("powerpc/powernv/ioda: Allocate indirect TCE levels on demand")
CC: stable@vger.kernel.org # v4.19+
Signed-off-by: Alexey Kardashevskiy <aik@ozlabs.ru>
Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
Link: https://lore.kernel.org/r/20190718051139.74787-2-aik@ozlabs.ru

powerpc/powernv/ioda: Fix locked_vm counting for memory used by IOMMU tables

We store 2 multilevel tables in iommu_table - one for the hardware and
one with the corresponding userspace addresses. Before allocating
the tables, the iommu_table_group_ops::get_table_size() hook returns
the combined size of the two and VFIO SPAPR TCE IOMMU driver adjusts
the locked_vm counter correctly. When the table is actually allocated,
the amount of allocated memory is stored in iommu_table::it_allocated_size
and used to decrement the locked_vm counter when we release the memory
used by the table; .get_table_size() and .create_table() calculate it
independently but the result is expected to be the same.

However the allocator does not add the userspace table size to
.it_allocated_size so when we destroy the table because of VFIO PCI
unplug (i.e. VFIO container is gone but the userspace keeps running),
we decrement locked_vm by just a half of size of memory we are
releasing.

To make things worse, since we enabled on-demand allocation of
indirect levels, it_allocated_size contains only the amount of memory
actually allocated at the table creation time which can just be a
fraction. It is not a problem with incrementing locked_vm (as
get_table_size() value is used) but it is with decrementing.

As the result, we leak locked_vm and may not be able to allocate more
IOMMU tables after few iterations of hotplug/unplug.

This sets it_allocated_size in the pnv_pci_ioda2_ops::create_table()
hook to what pnv_pci_ioda2_get_table_size() returns so from now on we
have a single place which calculates the maximum memory a table can
occupy. The original meaning of it_allocated_size is somewhat lost now
though.

We do not ditch it_allocated_size whatsoever here and we do not call
get_table_size() from vfio_iommu_spapr_tce.c when decrementing
locked_vm as we may have multiple IOMMU groups per container and even
though they all are supposed to have the same get_table_size()
implementation, there is a small chance for failure or confusion.

Fixes: 090bad39b237 ("powerpc/powernv: Add indirect levels to it_userspace")
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>

powerpc/powernv: Reference iommu_table while it is linked to a group

The iommu_table pointer stored in iommu_table_group may get stale
by accident, this adds referencing and removes a redundant comment
about this.

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>

powerpc/powernv/ioda: Allocate indirect TCE levels of cached userspace addresses on demand

The powernv platform maintains 2 TCE tables for VFIO - a hardware TCE
table and a table with userspace addresses; the latter is used for
marking pages dirty when corresponging TCEs are unmapped from
the hardware table.

a68bd1267b72 ("powerpc/powernv/ioda: Allocate indirect TCE levels
on demand") enabled on-demand allocation of the hardware table,
however it missed the other table so it has still been fully allocated
at the boot time. This fixes the issue by allocating a single level,
just like we do for the hardware table.

Fixes: a68bd1267b72 ("powerpc/powernv/ioda: Allocate indirect TCE levels on demand")
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>

powerpc/powernv/ioda2: Reduce upper limit for DMA window size (again)

mpe: This was fixed originally in commit d3d4ffaae439
("powerpc/powernv/ioda2: Reduce upper limit for DMA window size"), but
contrary to what the merge commit says was inadvertently lost by me in
commit ce57c6610cc2 ("Merge branch 'topic/ppc-kvm' into next") which
brought in changes that moved the code to a new file. So reapply it to
the new file.

Original commit message follows:

We use PHB in mode1 which uses bit 59 to select a correct DMA window.
However there is mode2 which uses bits 59:55 and allows up to 32 DMA
windows per a PE.

Even though documentation does not clearly specify that, it seems that
the actual hardware does not support bits 59:55 even in mode1, in
other words we can create a window as big as 1<<58 but DMA simply
won't work.

This reduces the upper limit from 59 to 55 bits to let the userspace
know about the hardware limits.

Fixes: ce57c6610cc2 ("Merge branch 'topic/ppc-kvm' into next")
Signed-off-by: Alexey Kardashevskiy <aik@ozlabs.ru>
Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>

powerpc/powernv/ioda: Allocate indirect TCE levels on demand

At the moment we allocate the entire TCE table, twice (hardware part and
userspace translation cache). This normally works as we normally have
contigous memory and the guest will map entire RAM for 64bit DMA.

However if we have sparse RAM (one example is a memory device), then
we will allocate TCEs which will never be used as the guest only maps
actual memory for DMA. If it is a single level TCE table, there is nothing
we can really do but if it a multilevel table, we can skip allocating
TCEs we know we won't need.

This adds ability to allocate only first level, saving memory.

This changes iommu_table::free() to avoid allocating of an extra level;
iommu_table::set() will do this when needed.

This adds @alloc parameter to iommu_table::exchange() to tell the callback
if it can allocate an extra level; the flag is set to "false" for
the realmode KVM handlers of H_PUT_TCE hcalls and the callback returns
H_TOO_HARD.

This still requires the entire table to be counted in mm::locked_vm.

To be conservative, this only does on-demand allocation when
the usespace cache table is requested which is the case of VFIO.

The example math for a system replicating a powernv setup with NVLink2
in a guest:
16GB RAM mapped at 0x0
128GB GPU RAM window (16GB of actual RAM) mapped at 0x244000000000

the table to cover that all with 64K pages takes:
(((0x244000000000 + 0x2000000000) >> 16)*8)>>20 = 4556MB

If we allocate only necessary TCE levels, we will only need:
(((0x400000000 + 0x400000000) >> 16)*8)>>20 = 4MB (plus some for indirect
levels).

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>

powerpc/powernv: Rework TCE level allocation

This moves actual pages allocation to a separate function which is going
to be reused later in on-demand TCE allocation.

While we are at it, remove unnecessary level size round up as the caller
does this already.

Reviewed-by: David Gibson <david@gibson.dropbear.id.au>
Signed-off-by: Alexey Kardashevskiy <aik@ozlabs.ru>
Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>

powerpc/powernv: Add indirect levels to it_userspace

We want to support sparse memory and therefore huge chunks of DMA windows
do not need to be mapped. If a DMA window big enough to require 2 or more
indirect levels, and a DMA window is used to map all RAM (which is
a default case for 64bit window), we can actually save some memory by
not allocation TCE for regions which we are not going to map anyway.

The hardware tables alreary support indirect levels but we also keep
host-physical-to-userspace translation array which is allocated by
vmalloc() and is a flat array which might use quite some memory.

This converts it_userspace from vmalloc'ed array to a multi level table.

As the format becomes platform dependend, this replaces the direct access
to it_usespace with a iommu_table_ops::useraddrptr hook which returns
a pointer to the userspace copy of a TCE; future extension will return
NULL if the level was not allocated.

This should not change non-KVM handling of TCE tables and it_userspace
will not be allocated for non-KVM tables.

Reviewed-by: David Gibson <david@gibson.dropbear.id.au>
Signed-off-by: Alexey Kardashevskiy <aik@ozlabs.ru>
Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>

powerpc/powernv: Move TCE manupulation code to its own file

Right now we have allocation code in pci-ioda.c and traversing code in
pci.c, let's keep them toghether. However both files are big enough
already so let's move this business to a new file.

While we at it, move the code which links IOMMU table groups to
IOMMU tables as it is not specific to any PNV PHB model.

These puts exported symbols from the new file together.

This fixes several warnings from checkpatch.pl like this:
"WARNING: Prefer 'unsigned int' to bare use of 'unsigned'".

As this is almost cut-n-paste, there should be no behavioral change.

Reviewed-by: David Gibson <david@gibson.dropbear.id.au>
Signed-off-by: Alexey Kardashevskiy <aik@ozlabs.ru>
Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>