commit 2925c2fdf1 upstream.
Currently the first thing we do in cxl_probe is to grab a reference
on the pci device. Later on, we call device_register on our adapter.
In our remove path, we call device_unregister, but we never call
pci_dev_put. We therefore leak the device every time we do a
reflash.
device_register/unregister is sufficient to hold the reference.
Therefore, drop the call to pci_dev_get.
Here's why this is safe.
The proposed cxl_probe(pdev) calls cxl_adapter_init:
a) init calls cxl_adapter_alloc, which creates a struct cxl,
conventionally called adapter. This struct contains a
device entry, adapter->dev.
b) init calls cxl_configure_adapter, where we set
adapter->dev.parent = &dev->dev (here dev is the pci dev)
So at this point, the cxl adapter's device's parent is the PCI
device that I want to be refcounted properly.
c) init calls cxl_register_adapter
*) cxl_register_adapter calls device_register(&adapter->dev)
So now we're in device_register, where dev is the adapter device, and
we want to know if the PCI device is safe after we return.
device_register(&adapter->dev) calls device_initialize() and then
device_add().
device_add() does a get_device(). device_add() also explicitly grabs
the device's parent, and calls get_device() on it:
parent = get_device(dev->parent);
So therefore, device_register() takes a lock on the parent PCI dev,
which is what pci_dev_get() was guarding. pci_dev_get() can therefore
be safely removed.
Fixes: f204e0b8ce ("cxl: Driver code for powernv PCIe based cards for userspace access")
Signed-off-by: Daniel Axtens <dja@axtens.net>
Acked-by: Ian Munsie <imunsie@au1.ibm.com>
Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
commit 9d8e27673c upstream.
cxl_reset currently PERSTs the slot, and then repeatedly tries to
read MMIO space in order to kick off EEH.
There are 2 problems with this: it's unnecessary, and it's racy.
It's unnecessary because the PERST will bring down the PHB link.
That will be picked up by the CAPP, which will send out an HMI.
Skiboot, noticing an HMI from the CAPP, will send an OPAL
notification to the kernel, which will trigger EEH recovery.
It's also racy: the EEH recovery triggered by the CAPP will
eventually cause the MMIO space to have its mapping invalidated
and the pointer NULLed out. This races with our attempt to read
the MMIO space. This is causing OOPSes in testing.
Simply drop all the attempts to force EEH detection, and trust
that Skiboot will send the notification and that we'll act on it.
The Skiboot code to send the EEH notification has been in Skiboot
for as long as CAPP recovery has been supported, so we don't need
to worry about breaking obscure setups with ancient firmware.
Cc: Ryan Grimm <grimm@linux.vnet.ibm.com>
Fixes: 62fa19d4b4 ("cxl: Add ability to reset the card")
Signed-off-by: Daniel Axtens <dja@axtens.net>
Acked-by: Ian Munsie <imunsie@au1.ibm.com>
Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
If an AFU claims to have a configuration record but doesn't actually
contain a vendor and device ID, fail the AFU initialisation. Right now
this is just a way of politely letting AFU developers know that they
need to fix their config space, but later on we may expose the AFUs as
actual PCI devices in their own right and don't want to inadvertendly
expose an AFU with a bad config space.
Signed-off-by: Ian Munsie <imunsie@au1.ibm.com>
Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
An AFU may optionally contain one or more PCIe like configuration
records, which can be used to identify the AFU.
This patch adds support for exposing the raw config space and the
vendor, device and class code under sysfs. These will appear in a
subdirectory of the AFU device corresponding with the configuration
record number, e.g.
cat /sys/class/cxl/afu0.0/cr0/vendor
0x1014
cat /sys/class/cxl/afu0.0/cr0/device
0x4350
cat /sys/class/cxl/afu0.0/cr0/class
0x120000
hexdump -C /sys/class/cxl/afu0.0/cr0/config
00000000 14 10 50 43 00 00 00 00 06 00 00 12 00 00 00 00 |..PC............|
00000010 00 00 00 00 00 00 00 00 00 00 00 00 00 00 00 00 |................|
*
00000100
These files behave in much the same way as the equivalent files for PCI
devices, with one exception being that the config file is currently
read-only and restricted to the root user. It is not necessarily
required to be this strict, but we currently do not have a compelling
use-case to make it writable and/or world-readable, so I erred on the
side of being restrictive.
Signed-off-by: Ian Munsie <imunsie@au1.ibm.com>
Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
When unbinding and rebinding the driver on a system with a card in PHB0, this
error condition is reached after a few attempts:
ERROR: Bad of_node_put() on /pciex@3fffe40000000
CPU: 0 PID: 3040 Comm: bash Not tainted 3.18.0-rc3-12545-g3627ffe #152
Call Trace:
[c000000721acb5c0] [c00000000086ef94] .dump_stack+0x84/0xb0 (unreliable)
[c000000721acb640] [c00000000073a0a8] .of_node_release+0xd8/0xe0
[c000000721acb6d0] [c00000000044bc44] .kobject_release+0x74/0xe0
[c000000721acb760] [c0000000007394fc] .of_node_put+0x1c/0x30
[c000000721acb7d0] [c000000000545cd8] .cxl_probe+0x1a98/0x1d50
[c000000721acb900] [c0000000004845a0] .local_pci_probe+0x40/0xc0
[c000000721acb980] [c000000000484998] .pci_device_probe+0x128/0x170
[c000000721acba30] [c00000000052400c] .driver_probe_device+0xac/0x2a0
[c000000721acbad0] [c000000000522468] .bind_store+0x108/0x160
[c000000721acbb70] [c000000000521448] .drv_attr_store+0x38/0x60
[c000000721acbbe0] [c000000000293840] .sysfs_kf_write+0x60/0xa0
[c000000721acbc50] [c000000000292500] .kernfs_fop_write+0x140/0x1d0
[c000000721acbcf0] [c000000000208648] .vfs_write+0xd8/0x260
[c000000721acbd90] [c000000000208b18] .SyS_write+0x58/0x100
[c000000721acbe30] [c000000000009258] syscall_exit+0x0/0x98
We are missing a call to of_node_get(). pnv_pci_to_phb_node() should
call of_node_get() otherwise np's reference count isn't incremented and
it might go away. Rename pnv_pci_to_phb_node() to pnv_pci_get_phb_node()
so it's clear it calls of_node_get().
Signed-off-by: Ryan Grimm <grimm@linux.vnet.ibm.com>
Acked-by: Ian Munsie <imunsie@au1.ibm.com>
Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
Adds reset to sysfs which will PERST the card. If load_image_on_perst is set
to "user" or "factory", the PERST will cause that image to be loaded.
load_image_on_perst is set to "user" for production.
"none" could be used for debugging. The PSL trace arrays are preserved which
then can be read through debugfs.
PERST also triggers CAPP recovery. An HMI comes in, which is handled by EEH.
EEH unbinds the driver, calls into Sapphire to reinitialize the PHB, then
rebinds the driver.
Signed-off-by: Ryan Grimm <grimm@linux.vnet.ibm.com>
Acked-by: Ian Munsie <imunsie@au1.ibm.com>
Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
Turning snoops on is the last step in CAPP recovery. Sapphire is expected to
have reinitialized the PHB and done the previous recovery steps.
Add mode argument to opal call to do this. Driver can turn snoops off although
it does not currently.
Signed-off-by: Ryan Grimm <grimm@linux.vnet.ibm.com>
Acked-by: Ian Munsie <imunsie@au1.ibm.com>
Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
Select defaults such that a PERST causes flash image reload. Select which
image based on what the card is set up to load.
CXL_VSEC_PERST_LOADS_IMAGE selects whether PERST assertion causes flash image
load.
CXL_VSEC_PERST_SELECT_USER selects which image is loaded on the next PERST.
cxl_update_image_control writes these bits into the VSEC.
Signed-off-by: Ryan Grimm <grimm@linux.vnet.ibm.com>
Acked-by: Ian Munsie <imunsie@au1.ibm.com>
Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
Upon inspection of the implementation specific registers, it was
discovered that the high bit of the implementation specific RXCTL
register was enabled, which enables the DEADB00F debug feature.
The debug feature causes MMIO reads to a disabled AFU to respond with
0xDEADB00F instead of all Fs. In general this should not be visible as
the kernel will only allow MMIO access to enabled AFUs, but there may be
some circumstances where an AFU may become disabled while it is use.
One such case would be an AFU designed to only be used in the dedicated
process mode and to disable itself after it has completed it's work
(however even in that case the effects of this debug flag would be
limited as the userspace application must have completed any required
MMIO accesses before the AFU disables itself with or without the flag).
This patch removes the debug flag and replaces the magic value
programmed into this register with a preprocessor define so it is
clearer what the rest of this initialisation does.
Signed-off-by: Ian Munsie <imunsie@au1.ibm.com>
Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
We had a known sleep while atomic bug if a CXL device was forcefully
unbound while it was in use. This could occur as a result of EEH, or
manually induced with something like this while the device was in use:
echo 0000:01:00.0 > /sys/bus/pci/drivers/cxl-pci/unbind
The issue was that in this code path we iterated over each context and
forcefully detached it with the contexts_lock spin lock held, however
the detach also needed to take the spu_mutex, and call schedule.
This patch changes the contexts_lock to a mutex so that we are not in
atomic context while doing the detach, thereby avoiding the sleep while
atomic.
Also delete the related TODO comment, which suggested an alternate
solution which turned out to not be workable.
Cc: stable@vger.kernel.org
Signed-off-by: Ian Munsie <imunsie@au1.ibm.com>
Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
This is the core of the cxl driver.
It adds support for using cxl cards in the powernv environment only (ie POWER8
bare metal). It allows access to cxl accelerators by userspace using the
/dev/cxl/afuM.N char devices.
The kernel driver has no knowledge of the function implemented by the
accelerator. It provides services to userspace via the /dev/cxl/afuM.N
devices. When a program opens this device and runs the start work IOCTL, the
accelerator will have coherent access to that processes memory using the same
virtual addresses. That process may mmap the device to access any MMIO space
the accelerator provides. Also, reads on the device will allow interrupts to
be received. These services are further documented in a later patch in
Documentation/powerpc/cxl.txt.
Documentation of the cxl hardware architecture and userspace API is provided in
subsequent patches.
Signed-off-by: Ian Munsie <imunsie@au1.ibm.com>
Signed-off-by: Michael Neuling <mikey@neuling.org>
Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
