At the moment updates in a TCE table are made by iommu_table_ops::exchange
which update one TCE and invalidates an entry in the PHB/NPU TCE cache
via set of registers called "TCE Kill" (hence the naming).
Writing a TCE is a simple xchg() but invalidating the TCE cache is
a relatively expensive OPAL call. Mapping a 100GB guest with PCI+NPU
passed through devices takes about 20s.
Thankfully we can do better. Since such big mappings happen at the boot
time and when memory is plugged/onlined (i.e. not often), these requests
come in 512 pages so we call call OPAL 512 times less which brings 20s
from the above to less than 10s. Also, since TCE caches can be flushed
entirely, calling OPAL for 512 TCEs helps skiboot [1] to decide whether
to flush the entire cache or not.
This implements 2 new iommu_table_ops callbacks:
- xchg_no_kill() to update a single TCE with no TCE invalidation;
- tce_kill() to invalidate multiple TCEs.
This uses the same xchg_no_kill() callback for IODA1/2.
This implements 2 new wrappers on top of the new callbacks similar to
the existing iommu_tce_xchg().
This does not use the new callbacks yet, the next patches will;
so this should not cause any behavioral change.
Signed-off-by: Alexey Kardashevskiy <aik@ozlabs.ru>
Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
Link: https://lore.kernel.org/r/20190829085252.72370-2-aik@ozlabs.ru
The existing code uses bunch of hardcoded values from the PCI Bus
Binding to IEEE Std 1275 spec; and it does so in quite non-obvious
way.
This defines fields from the cell#0 of the "reg" property of a PCI
device and uses them for parsing.
This should cause no behavioral change.
Signed-off-by: Alexey Kardashevskiy <aik@ozlabs.ru>
[mpe: Unsplit some 80/81 char lines, space the code with some newlines]
Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
Link: https://lore.kernel.org/r/20190829084417.71873-1-aik@ozlabs.ru
Prior to commit 1bd98d7fbaf5 ("ppc64: Update BUG handling based on
ppc32"), BUG() family was using BUG_ILLEGAL_INSTRUCTION which
was an invalid instruction opcode to trap into program check
exception.
That commit converted them to using standard trap instructions,
but prom/prom_init and their PROM_BUG() macro were left over.
head_64.S and exception-64s.S were left aside as well.
Convert them to using the standard BUG infrastructure.
Signed-off-by: Christophe Leroy <christophe.leroy@c-s.fr>
Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
Link: https://lore.kernel.org/r/cdaf4bbbb64c288a077845846f04b12683f8875a.1566817807.git.christophe.leroy@c-s.fr
Today LOAD_REG_IMMEDIATE() is a basic #define which loads all
parts on a value into a register, including the parts that are NUL.
This means always 2 instructions on PPC32 and always 5 instructions
on PPC64. And those instructions cannot run in parallele as they are
updating the same register.
Ex: LOAD_REG_IMMEDIATE(r1,THREAD_SIZE) in head_64.S results in:
3c 20 00 00 lis r1,0
60 21 00 00 ori r1,r1,0
78 21 07 c6 rldicr r1,r1,32,31
64 21 00 00 oris r1,r1,0
60 21 40 00 ori r1,r1,16384
Rewrite LOAD_REG_IMMEDIATE() with GAS macro in order to skip
the parts that are NUL.
Rename existing LOAD_REG_IMMEDIATE() as LOAD_REG_IMMEDIATE_SYM()
and use that one for loading value of symbols which are not known
at compile time.
Now LOAD_REG_IMMEDIATE(r1,THREAD_SIZE) in head_64.S results in:
38 20 40 00 li r1,16384
Signed-off-by: Christophe Leroy <christophe.leroy@c-s.fr>
Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
Link: https://lore.kernel.org/r/d60ce8dd3a383c7adbfc322bf1d53d81724a6000.1566311636.git.christophe.leroy@c-s.fr
ppc_md.ioremap() is only used for I/O workaround on CELL platform,
so indirect function call can be avoided.
This patch reworks the io-workaround and ioremap() functions to
use the global 'io_workaround_inited' flag for the activation
of io-workaround.
When CONFIG_PPC_IO_WORKAROUNDS or CONFIG_PPC_INDIRECT_MMIO are not
selected, the I/O workaround ioremap() voids and the global flag is
not used.
Signed-off-by: Christophe Leroy <christophe.leroy@c-s.fr>
Reviewed-by: Christoph Hellwig <hch@lst.de>
Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
Link: https://lore.kernel.org/r/5fa3ef069fbd0f152512afaae19e7a60161454cf.1566309262.git.christophe.leroy@c-s.fr
We still treat devices without a DMA mask as defaulting to 32-bits for
both mask, but a few releases ago we've started warning about such
cases, as they require special cases to work around this sloppyness.
Add a dma_mask field to struct platform_device so that we can initialize
the dma_mask pointer in struct device and initialize both masks to
32-bits by default, replacing similar functionality in m68k and
powerpc. The arch_setup_pdev_archdata hooks is now unused and removed.
Note that the code looks a little odd with the various conditionals
because we have to support platform_device structures that are
statically allocated.
Signed-off-by: Christoph Hellwig <hch@lst.de>
Acked-by: Geert Uytterhoeven <geert@linux-m68k.org>
Link: https://lore.kernel.org/r/20190816062435.881-7-hch@lst.de
Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
If a PCI device is removed during eeh_pe_report_edev(), between the
calls to device_lock() and device_unlock(), edev->pdev will change and
cause a crash as the wrong mutex is released.
To correct this, hold the PCI rescan/remove lock while taking a copy
of edev->pdev and performing a get_device() on it. Use this value to
release the mutex, but also pass it through to the device driver's EEH
handlers so that they always see the same device.
Signed-off-by: Sam Bobroff <sbobroff@linux.ibm.com>
Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
Link: https://lore.kernel.org/r/3c590579a0faa24d20c826dcd26c739eb4d454e6.1565930772.git.sbobroff@linux.ibm.com
Convert existing messages, where appropriate, to use the eeh_edev_*
logging macros.
The only effect should be minor adjustments to the log messages, apart
from:
- A new message in pseries_eeh_probe() "Probing device" to match the
powernv case.
- The "Probing device" message in pnv_eeh_probe() is now generated
slightly later, which will mean that it is no longer emitted for
devices that aren't probed due to the initial checks.
Signed-off-by: Sam Bobroff <sbobroff@linux.ibm.com>
Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
Link: https://lore.kernel.org/r/ce505a0a7a4a5b0367f0f40f8b26e7c0a9cf4cb7.1565930772.git.sbobroff@linux.ibm.com
Now that struct eeh_dev includes the BDFN of it's PCI device, make use
of it to replace eeh_edev_info() with a set of dev_dbg()-style macros
that only need a struct edev.
With the BDFN available without the struct pci_dev, eeh_pci_name() is
now unnecessary, so remove it.
While only the "info" level function is used here, the others will be
used in followup work.
Signed-off-by: Sam Bobroff <sbobroff@linux.ibm.com>
Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
Link: https://lore.kernel.org/r/f90ae9a53d762be7b0ccbad79e62b5a1b4f4996e.1565930772.git.sbobroff@linux.ibm.com
Now that EEH support for all devices (on PowerNV and pSeries) is
provided by the pcibios bus add device hooks, eeh_probe_devices() and
eeh_addr_cache_build() are redundant and can be removed.
Move the EEH enabled message into it's own function so that it can be
called from multiple places.
Note that previously on pSeries, useless EEH sysfs files were created
for some devices that did not have EEH support and this change
prevents them from being created.
Signed-off-by: Sam Bobroff <sbobroff@linux.ibm.com>
Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
Link: https://lore.kernel.org/r/33b0a6339d5ac88693de092d6fba984f2a5add66.1565930772.git.sbobroff@linux.ibm.com
On PowerNV and pSeries, devices currently acquire EEH support from
several different places: Boot-time devices from eeh_probe_devices()
and eeh_addr_cache_build(), Virtual Function devices from the pcibios
bus add device hooks and hot plugged devices from pci_hp_add_devices()
(with other platforms using other methods as well). Unfortunately,
pSeries machines currently discover hot plugged devices using
pci_rescan_bus(), not pci_hp_add_devices(), and so those devices do
not receive EEH support.
Rather than adding another case for pci_rescan_bus(), this change
widens the scope of the pcibios bus add device hooks so that they can
handle all devices. As a side effect this also supports devices
discovered after manually rescanning via /sys/bus/pci/rescan.
Note that on PowerNV, this change allows the EEH subsystem to become
enabled after boot as long as it has not been forced off, which was
not previously possible (it was already possible on pSeries).
Signed-off-by: Sam Bobroff <sbobroff@linux.ibm.com>
Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
Link: https://lore.kernel.org/r/72ae8ae9c54097158894a52de23690448de38ea9.1565930772.git.sbobroff@linux.ibm.com
The EEH address cache is currently initialized and populated by a
single function: eeh_addr_cache_build(). While the initial population
of the cache can only be done once resources are allocated,
initialization (just setting up a spinlock) could be done much
earlier.
So move the initialization step into a separate function and call it
from a core_initcall (rather than a subsys initcall).
This will allow future work to make use of the cache during boot time
PCI scanning.
Signed-off-by: Sam Bobroff <sbobroff@linux.ibm.com>
Reviewed-by: Alexey Kardashevskiy <aik@ozlabs.ru>
Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
Link: https://lore.kernel.org/r/0557206741bffee76cdfff042f65321f6f7a5b41.1565930772.git.sbobroff@linux.ibm.com
The EEH_DEV_NO_HANDLER flag is used by the EEH system to prevent the
use of driver callbacks in drivers that have been bound part way
through the recovery process. This is necessary to prevent later stage
handlers from being called when the earlier stage handlers haven't,
which can be confusing for drivers.
However, the flag is set for all devices that are added after boot
time and only cleared at the end of the EEH recovery process. This
results in hot plugged devices erroneously having the flag set during
the first recovery after they are added (causing their driver's
handlers to be incorrectly ignored).
To remedy this, clear the flag at the beginning of recovery
processing. The flag is still cleared at the end of recovery
processing, although it is no longer really necessary.
Also clear the flag during eeh_handle_special_event(), for the same
reasons.
Signed-off-by: Sam Bobroff <sbobroff@linux.ibm.com>
Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
Link: https://lore.kernel.org/r/b8ca5629d27de74c957d4f4b250177d1b6fc4bbd.1565930772.git.sbobroff@linux.ibm.com
The pcibios_init() function for PowerPC 64 currently calls
pci_bus_add_devices() before pcibios_resource_survey(). This means
that at boot time, when the pcibios_bus_add_device() hooks are called
by pci_bus_add_devices(), device resources have not been allocated and
they are unable to perform EEH setup, so a separate pass is needed.
This patch adjusts that order so that it will become possible to
consolidate the EEH setup work into a single location.
The only functional change is to execute pcibios_resource_survey()
(excepting ppc_md.pcibios_fixup(), see below) before
pci_bus_add_devices() instead of after it.
Because pcibios_scan_phb() and pci_bus_add_devices() are called
together in a loop, this must be broken into one loop for each call.
Then the call to pcibios_resource_survey() is moved up in between
them. This changes the ordering but because pcibios_resource_survey()
also calls ppc_md.pcibios_fixup(), that call is extracted out into
pcibios_init() to where pcibios_resource_survey() was, so that it is
not moved.
The only other caller of pcibios_resource_survey() is the PowerPC 32
version of pcibios_init(), and therefore, that is modified to call
ppc_md.pcibios_fixup() right after pcibios_resource_survey() so that
there is no functional change there at all.
The re-arrangement will cause very few side-effects because at this
stage in the boot, pci_bus_add_devices() does very little:
- pci_create_sysfs_dev_files() does nothing (no sysfs yet)
- pci_proc_attach_device() does nothing (no proc yet)
- device_attach() does nothing (no drivers yet)
This leaves only the pci_final_fixup calls, D3 support, and marking
the device as added. Of those, only the pci_final_fixup calls have the
potential to be affected by resource allocation.
The only pci_final_fixup handlers that touch resources seem to be one
for x86 (pci_amd_enable_64bit_bar()), and a PowerPC 32 platform driver
(quirk_final_uli1575()), neither of which use this pcibios_init()
function. Even if they did, it would almost certainly be a bug, under
the current ordering, to rely on or make changes to resources before
they were allocated.
Signed-off-by: Sam Bobroff <sbobroff@linux.ibm.com>
Reviewed-by: Alexey Kardashevskiy <aik@ozlabs.ru>
Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
Link: https://lore.kernel.org/r/4506b0489eabd0921a3587d90bd44c7683f3472d.1565930772.git.sbobroff@linux.ibm.com
The current code would fail on huge pages addresses, since the shift would
be incorrect. Use the correct page shift value returned by
__find_linux_pte() to get the correct physical address. The code is more
generic and can handle both regular and compound pages.
Fixes: ba41e1e1cc ("powerpc/mce: Hookup derror (load/store) UE errors")
Signed-off-by: Balbir Singh <bsingharora@gmail.com>
[arbab@linux.ibm.com: Fixup pseries_do_memory_failure()]
Signed-off-by: Reza Arbab <arbab@linux.ibm.com>
Tested-by: Mahesh Salgaonkar <mahesh@linux.vnet.ibm.com>
Signed-off-by: Santosh Sivaraj <santosh@fossix.org>
Cc: stable@vger.kernel.org # v4.15+
Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
Link: https://lore.kernel.org/r/20190820081352.8641-3-santosh@fossix.org
The LPAR migration implementation and userspace-initiated cpu hotplug
can interleave their executions like so:
1. Set cpu 7 offline via sysfs.
2. Begin a partition migration, whose implementation requires the OS
to ensure all present cpus are online; cpu 7 is onlined:
rtas_ibm_suspend_me -> rtas_online_cpus_mask -> cpu_up
This sets cpu 7 online in all respects except for the cpu's
corresponding struct device; dev->offline remains true.
3. Set cpu 7 online via sysfs. _cpu_up() determines that cpu 7 is
already online and returns success. The driver core (device_online)
sets dev->offline = false.
4. The migration completes and restores cpu 7 to offline state:
rtas_ibm_suspend_me -> rtas_offline_cpus_mask -> cpu_down
This leaves cpu7 in a state where the driver core considers the cpu
device online, but in all other respects it is offline and
unused. Attempts to online the cpu via sysfs appear to succeed but the
driver core actually does not pass the request to the lower-level
cpuhp support code. This makes the cpu unusable until the cpu device
is manually set offline and then online again via sysfs.
Instead of directly calling cpu_up/cpu_down, the migration code should
use the higher-level device core APIs to maintain consistent state and
serialize operations.
Fixes: 120496ac2d ("powerpc: Bring all threads online prior to migration/hibernation")
Signed-off-by: Nathan Lynch <nathanl@linux.ibm.com>
Reviewed-by: Gautham R. Shenoy <ego@linux.vnet.ibm.com>
Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
Link: https://lore.kernel.org/r/20190802192926.19277-2-nathanl@linux.ibm.com
At the moment we create a small window only for 32bit devices, the window
maps 0..2GB of the PCI space only. For other devices we either use
a sketchy bypass or hardware bypass but the former can only work if
the amount of RAM is no bigger than the device's DMA mask and the latter
requires devices to support at least 59bit DMA.
This extends the default DMA window to the maximum size possible to allow
a wider DMA mask than just 32bit. The default window size is now limited
by the the iommu_table::it_map allocation bitmap which is a contiguous
array, 1 bit per an IOMMU page.
This increases the default IOMMU page size from hard coded 4K to
the system page size to allow wider DMA masks.
This increases the level number to not exceed the max order allocation
limit per TCE level. By the same time, this keeps minimal levels number
as 2 in order to save memory.
As the extended window now overlaps the 32bit MMIO region, this adds
an area reservation to iommu_init_table().
After this change the default window size is 0x80000000000==1<<43 so
devices limited to DMA mask smaller than the amount of system RAM can
still use more than just 2GB of memory for DMA.
This is an optimization and not a bug fix for DMA API usage.
With the on-demand allocation of indirect TCE table levels enabled and
2 levels, the first TCE level size is just
1<<ceil((log2(0x7ffffffffff+1)-16)/2)=16384 TCEs or 2 system pages.
Signed-off-by: Alexey Kardashevskiy <aik@ozlabs.ru>
Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
Link: https://lore.kernel.org/r/20190718051139.74787-5-aik@ozlabs.ru
POWER8 and newer support a bypass mode which maps all host memory to
PCI buses so an IOMMU table is not always required. However if we fail to
create such a table, the DMA setup fails and the kernel does not boot.
This skips the 32bit DMA setup check if the bypass is selected.
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/20190718051139.74787-3-aik@ozlabs.ru
All the way back to introducing dma_common_mmap we've defaulted to mark
the pages as uncached. But this is wrong for DMA coherent devices.
Later on DMA_ATTR_WRITE_COMBINE also got incorrect treatment as that
flag is only treated special on the alloc side for non-coherent devices.
Introduce a new dma_pgprot helper that deals with the check for coherent
devices so that only the remapping cases ever reach arch_dma_mmap_pgprot
and we thus ensure no aliasing of page attributes happens, which makes
the powerpc version of arch_dma_mmap_pgprot obsolete and simplifies the
remaining ones.
Note that this means arch_dma_mmap_pgprot is a bit misnamed now, but
we'll phase it out soon.
Fixes: 64ccc9c033 ("common: dma-mapping: add support for generic dma_mmap_* calls")
Reported-by: Shawn Anastasio <shawn@anastas.io>
Reported-by: Gavin Li <git@thegavinli.com>
Signed-off-by: Christoph Hellwig <hch@lst.de>
Acked-by: Catalin Marinas <catalin.marinas@arm.com> # arm64
Mark switch cases where we are expecting to fall through.
Fixes errors such as below, seen with mpc85xx_defconfig:
arch/powerpc/kernel/align.c: In function 'emulate_spe':
arch/powerpc/kernel/align.c:178:8: error: this statement may fall through
ret |= __get_user_inatomic(temp.v[3], p++);
^~
Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
Link: https://lore.kernel.org/r/20190730141917.21817-1-mpe@ellerman.id.au
Wire up the new clone3 syscall added in commit 7f192e3cd3 ("fork:
add clone3").
This requires a ppc_clone3 wrapper, in order to save the non-volatile
GPRs before calling into the generic syscall code. Otherwise we hit
the BUG_ON in CHECK_FULL_REGS in copy_thread().
Lightly tested using Christian's test code on a Power8 LE VM.
Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
Acked-by: Christian Brauner <christian@brauner.io>
Link: https://lore.kernel.org/r/20190724140259.23554-1-mpe@ellerman.id.au
