Add the new API pci_enable_msi_block() to allow drivers to
request multiple MSI and reimplement pci_enable_msi in terms of
pci_enable_msi_block. Ensure that the architecture back ends don't
have to know about multiple MSI.
Signed-off-by: Matthew Wilcox <willy@linux.intel.com>
Signed-off-by: Jesse Barnes <jbarnes@virtuousgeek.org>
Since most of the callers already know whether they have an MSI or
an MSI-X capability, split msi_set_mask_bits() into msi_mask_irq()
and msix_mask_irq(). The only callers which don't (mask_msi_irq()
and unmask_msi_irq()) can share code in msi_set_mask_bit(). This then
becomes the only caller of msix_flush_writes(), so we can inline it.
The flushing read can be to any address that belongs to the device,
so we can eliminate the calculation too.
We can also get rid of maskbits_mask from struct msi_desc and simply
recalculate it on the rare occasion that we need it. The single-bit
'masked' element is replaced by a copy of the 32-bit 'masked' register,
so this patch does not affect the size of msi_desc.
Signed-off-by: Matthew Wilcox <willy@linux.intel.com>
Signed-off-by: Jesse Barnes <jbarnes@virtuousgeek.org>
MSI interrupts have a mask_pos where MSI-X have a mask_base. Use a
transparent union to get rid of some ugly casts.
Signed-off-by: Matthew Wilcox <willy@linux.intel.com>
Signed-off-by: Jesse Barnes <jbarnes@virtuousgeek.org>
By passing the pci_dev into alloc_msi_entry() we can be sure that
the ->dev entry is always assigned and so we don't need to check it.
Also, we used kzalloc() so we don't need to initialise ->irq to 0.
Signed-off-by: Matthew Wilcox <willy@linux.intel.com>
Signed-off-by: Jesse Barnes <jbarnes@virtuousgeek.org>
By changing from a 5-bit field to a 1-bit field, we free up some bits
that can be used by a later patch. Also rearrange the fields for better
packing on 64-bit platforms (reducing the size of msi_desc from 72 bytes
to 64 bytes).
Signed-off-by: Matthew Wilcox <willy@linux.intel.com>
Signed-off-by: Jesse Barnes <jbarnes@virtuousgeek.org>
There is code in msix_capability_init() which, when the requested number
of MSI-X couldn't be allocated, calculates how many MSI-X /could/ be
allocated and returns that to the driver. That allows the driver to then
make a second request, with a number of MSIs that should succeed.
The current code requires the arch code to setup as many msi_descs as it
can, and then return to the generic code. On some platforms the arch
code may already know how many MSI-X it can allocate, before it sets up
any of the msi_descs.
So change the logic such that if the arch code returns a positive error
code, that is taken to be the number of MSI-X that could be allocated.
If the error code is negative we still calculate the number available
using the old method.
Because it's a little subtle, make sure the error return code from
arch_setup_msi_irq() is always negative. That way only implementations
of arch_setup_msi_irqs() need to be careful about returning a positive
error code.
Signed-off-by: Michael Ellerman <michael@ellerman.id.au>
Signed-off-by: Jesse Barnes <jbarnes@virtuousgeek.org>
Weak functions aren't all they're cracked up to be. They lead to
incorrect binaries with some toolchains, they require us to have empty
functions we otherwise wouldn't, and the unused code is not elided
(as of gcc 4.3.2 anyway).
So replace the weak MSI arch hooks with the #define foo foo idiom. We no
longer need empty versions of arch_setup/teardown_msi_irq().
This is less source (by 1 line!), and results in smaller binaries too:
text data bss dec hex filename
9354300 1693916 678424 11726640 b2ef30 build/powerpc/vmlinux-before
9354052 1693852 678424 11726328 b2edf8 build/powerpc/vmlinux-after
Also smaller on x86_64 and arm (iop13xx).
Signed-off-by: Michael Ellerman <michael@ellerman.id.au>
Signed-off-by: Jesse Barnes <jbarnes@virtuousgeek.org>
Introduce new function pci_msix_table_size() returning the size of
the MSI-X table of given PCI device or 0 if the device doesn't
support MSI-X.
Signed-off-by: Rafael J. Wysocki <rjw@sisk.pl>
Reviewed-by: Hidetoshi Seto <seto.hidetoshi@jp.fujitsu.com>
Signed-off-by: Jesse Barnes <jbarnes@virtuousgeek.org>
Hidetoshi Seto points out that commit
bffac3c593 has wrong values in the array.
Rather than correct the array, we can just use a bounds check and
perform the calculation specified in the comment. As a bonus, this will
not run off the end of the array if the device specifies an illegal
value in the MSI capability.
Signed-off-by: Matthew Wilcox <willy@linux.intel.com>
Signed-off-by: Hidetoshi Seto <seto.hidetoshi@jp.fujitsu.com>
Signed-off-by: Jesse Barnes <jbarnes@virtuousgeek.org>
Add an msi_mask() function which returns the correct bitmask for the
number of MSI interrupts you have. This fixes an undefined bug in
msi_capability_init().
Signed-off-by: Matthew Wilcox <willy@linux.intel.com>
Signed-off-by: Jesse Barnes <jbarnes@virtuousgeek.org>
The _OSC capability OSC_MSI_SUPPORT is set when the root bridge is added
with pci_acpi_osc_support(), so we no longer need to do it in the PCI
MSI driver. Also adds the function pci_msi_enabled, which returns true
if pci=nomsi is not on the kernel command-line.
Signed-off-by: Andrew Patterson <andrew.patterson@hp.com>
Signed-off-by: Jesse Barnes <jbarnes@virtuousgeek.org>
Impact: simplify code
Pass irq_desc and cfg around, instead of raw IRQ numbers - this way
we dont have to look it up again and again.
Signed-off-by: Yinghai Lu <yinghai@kernel.org>
Signed-off-by: Ingo Molnar <mingo@elte.hu>
Currently linux doesn't have any code to set the "MSI supported" bit in
Support Fireld of _OSC. This patch adds the code for that.
Signed-off-by: Kenji Kaneshige <kaneshige.kenji@jp.fujitsu.com>
Signed-off-by: Taku Izumi <izumi.taku@jp.fujitsu.com>
Signed-off-by: Jesse Barnes <jbarnes@virtuousgeek.org>
In msi_capability_init, we can make use of the calculated results
instead of calling is_mask_bit_support and is_64bit_address twice.
Signed-off-by: Jike Song <albcamus@gmail.com>
Signed-off-by: Jesse Barnes <jbarnes@virtuousgeek.org>
With the recent change to avoid masking MSIs using the MSI enable bit, devices
without an MSI mask bit will have their MSI capability always enabled when MSI
is in use, so we need to restore it regardless of the mask bit state.
Fixes kernel bz 11178.
Acked-by: Matthew Wilcox <willy@linux.intel.com>
Signed-off-by: Alan Jenkins <alan-jenkins@tuffmail.co.uk>
Signed-off-by: Jesse Barnes <jbarnes@virtuousgeek.org>
David Vrabel has a device which generates an interrupt storm on the INTx
pin if we disable MSI interrupts altogether. Masking interrupts is only
a performance optimisation, so we can ignore the request to mask the
interrupt.
Signed-off-by: Matthew Wilcox <willy@linux.intel.com>
Signed-off-by: Jesse Barnes <jbarnes@virtuousgeek.org>
Convert printks to use dev_printk().
I converted pr_debug() to dev_dbg(). Both use KERN_DEBUG and are enabled
only when DEBUG is defined.
I converted printk(KERN_DEBUG) to dev_printk(KERN_DEBUG), not to dev_dbg(),
because dev_dbg() is only enabled when DEBUG is defined.
I converted DBG(KERN_INFO) (only in setup-bus.c) to dev_info(). The DBG()
name makes it sound like debug, but it's been enabled forever, so dev_info()
preserves the previous behavior.
I tried to make the resource assignment formats more consistent, e.g.,
"BAR %d: got res [%#llx-%#llx] bus [%#llx-%#llx] flags %#lx\n"
instead of sometimes using "start-end" and sometimes using "size@start".
I'm not attached to one or the other; I'd just like them consistent.
Signed-off-by: Bjorn Helgaas <bjorn.helgaas@hp.com>
Signed-off-by: Jesse Barnes <jbarnes@virtuousgeek.org>
The position of MSI capability is already cached in the msi_desc when
we enter the msi_set_mask_bits(). Use it instead.
Signed-off-by: Hidetoshi Seto <seto.hidetoshi@jp.fujitsu.com>
Acked-by: Arnaldo Carvalho de Melo <acme@redhat.com>
Signed-off-by: Jesse Barnes <jbarnes@virtuousgeek.org>
[PATCH 2/2] pci/irq: let pci_device_shutdown to call pci_msi_shutdown v2
this change
| commit 23a274c8a5
| Author: Prakash, Sathya <sathya.prakash@lsi.com>
| Date: Fri Mar 7 15:53:21 2008 +0530
|
| [SCSI] mpt fusion: Enable MSI by default for SAS controllers
|
| This patch modifies the driver to enable MSI by default for all SAS chips.
|
| Signed-off-by: Sathya Prakash <sathya.prakash@lsi.com>
| Signed-off-by: James Bottomley <James.Bottomley@HansenPartnership.com>
|
Causes the kexec of a RHEL 5.1 kernel to fail.
root casue: the rhel 5.1 kernel still uses INTx emulation. and
mptscsih_shutdown doesn't call pci_disable_msi to reenable INTx on kexec path
So call pci_msi_shutdown in the shutdown path to do the same thing to msix
Signed-off-by: Yinghai Lu <yhlu.kernel@gmail.com>
Signed-off-by: Jesse Barnes <jbarnes@hobbes.lan>
[PATCH 1/2] pci/irq: restore mask_bits in msi shutdown -v3
Yinghai found that kexec'ing a RHEL 5.1 kernel with 2.6.25-rc3+ kernels
prevents his NIC from working. He bisected to
| commit 89d694b9db
| Author: Thomas Gleixner <tglx@linutronix.de>
| Date: Mon Feb 18 18:25:17 2008 +0100
|
| genirq: do not leave interupts enabled on free_irq
|
| The default_disable() function was changed in commit:
|
| 76d2160147
| genirq: do not mask interrupts by default
|
For MSI, default_shutdown will call mask_bit for msi device. All mask bits
will left disabled after free_irq. Then in the kexec case, the next kernel
can only use msi_enable bit, so all device's MSI can not be used.
So lets to restore the mask bit to its pci reset defined value (enabled) when
we disable the kernels use of msi to be a little friendlier to kexec'd kernels.
Extend msi_set_mask_bit to msi_set_mask_bits to take mask, so we can fully
restore that to 0x00 instead of 0xfe.
Signed-off-by: Yinghai Lu <yhlu.kernel@gmail.com>
Signed-off-by: Jesse Barnes <jbarnes@hobbes.lan>
This patch fixes the following problem present with older gcc versions:
<-- snip -->
...
CC drivers/pci/msi.o
/home/bunk/linux/kernel-2.6/git/linux-2.6/drivers/pci/msi.c:692: warning: weak declaration of `arch_msi_check_device' after first use results in unspecified behavior
/home/bunk/linux/kernel-2.6/git/linux-2.6/drivers/pci/msi.c:704: warning: weak declaration of `arch_setup_msi_irqs' after first use results in unspecified behavior
/home/bunk/linux/kernel-2.6/git/linux-2.6/drivers/pci/msi.c:724: warning: weak declaration of `arch_teardown_msi_irqs' after first use results in unspecified behavior
...
<-- snip -->
Signed-off-by: Adrian Bunk <bunk@kernel.org>
Signed-off-by: Greg Kroah-Hartman <gregkh@suse.de>
PCI error recovery usually involves the PCI adapter being reset.
If the device is using MSI, the reset will cause the MSI state
to be lost; the device driver needs to restore the MSI state.
The pci_restore_msi_state() routine is currently protected
by CONFIG_PM; remove this, and also export the symbol, so
that it can be used in a modle.
Signed-off-by: Linas Vepstas <linas@austin.ibm.com>
Signed-off-by: Greg Kroah-Hartman <gregkh@suse.de>
A reasonably common problem with some devices is that they will
disable MSI generation when the INTX_DISABLE bit is set in the
PCI_COMMAND register.
Quirk this explicitly, guarding the pci_intx() calls in msi.c with
this quirk indication.
The first entries for this quirk are for 5714 and 5780 Tigon3 chips,
and thus we can remove the workaround code from the tg3.c driver.
Signed-off-by: David S. Miller <davem@davemloft.net>
Acked-by: Michael Chan <mchan@broadcom.com>
Acked-by: Jeff Garzik <jgarzik@redhat.com>
Signed-off-by: Greg Kroah-Hartman <gregkh@suse.de>
While reading the MSI code trying to find a reason why MSI wouldn't
work for devices that have a 32-bit MSI address capability, I noticed
that read_msi_msg() seems to read the message data from the wrong
offset in this case.
Signed-off-by: Roland Dreier <roland@digitalvampire.org>
Acked-by: Eric W. Biederman <ebiederm@xmission.com>
Cc: stable <stable@kernel.org>
Signed-off-by: Greg Kroah-Hartman <gregkh@suse.de>
"Mike Miller (OS Dev)" <mikem@beardog.cca.cpqcorp.net> writes:
Found what seems the problem with our vectors being listed backward. In
drivers/pci/msi.c we should be using list_add_tail rather than list_add to
preserve the ordering across various kernels. Please consider this for
inclusion.
Signed-off-by: "Eric W. Biederman" <ebiederm@xmission.com>
Screwed-up-by: Michael Ellerman <michael@ellerman.id.au>
Cc: "Mike Miller (OS Dev)" <mikem@beardog.cca.cpqcorp.net>
Cc: Andi Kleen <ak@suse.de>
Cc: Greg KH <greg@kroah.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
In file included from drivers/pci/msi.c:22:
include/asm/smp.h:17:26: asm/arch/smp.h: No such file or directory
include/asm/smp.h:20:3: #error "<asm-arm/smp.h> included in non-SMP build"
include/asm/smp.h:23:1: warning: "raw_smp_processor_id" redefined
In file included from include/linux/sched.h:65,
from include/linux/mm.h:4,
from drivers/pci/msi.c:10:
include/linux/smp.h:85:1: warning: this is the location of the previous
definition
Tested on powerpc, i386, and x86_64.
Signed-off-by: Dan Williams <dan.j.williams@intel.com>
Acked-by: Eric W. Biederman <ebiederm@xmission.com>
Signed-off-by: Greg Kroah-Hartman <gregkh@suse.de>
Today's find is a triggered assertion in msi_free_irqs() when the system
doesn't support MSI, in which case arch_setup_msi_irqs() always returns
an error.
The problem is that when this happens we branch into msi_free_irqs(), to
which you added the following assertion loop:
list_for_each_entry(entry, &dev->msi_list, list)
BUG_ON(irq_has_action(entry->irq));
Well, if arch_setup_msi_irqs() fails, entry->irq will be zero and
although that's never assigned to any normal devices we use that IRQ
number for the timer interrupt on sparc64 so this assertion triggers.
Better to test for zero before doing the irq_has_action() assertion
thing.
Signed-off-by: David S. Miller <davem@davemloft.net>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
Remove includes of <linux/smp_lock.h> where it is not used/needed.
Suggested by Al Viro.
Builds cleanly on x86_64, i386, alpha, ia64, powerpc, sparc,
sparc64, and arm (all 59 defconfigs).
Signed-off-by: Randy Dunlap <randy.dunlap@oracle.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
This patch introduces an optional function, arch_teardown_msi_irqs(),
which gives an arch the opportunity to do per-device teardown for
MSI/X. If that's not required, the default version simply calls
arch_teardown_msi_irq() for each msi irq required.
arch_teardown_msi_irqs() is simply passed a pdev, attached to the pdev
is a list of msi_descs, it is up to the arch to free the irq associated
with each of these as appropriate.
For archs that _don't_ implement arch_teardown_msi_irqs(), all msi_descs
with irq == 0 are considered unallocated, and the arch teardown routine
is not called on them.
Signed-off-by: Michael Ellerman <michael@ellerman.id.au>
Signed-off-by: Greg Kroah-Hartman <gregkh@suse.de>
This patch introduces an optional function, arch_setup_msi_irqs(),
(note the plural) which gives an arch the opportunity to do per-device
setup for MSI/X and then allocate all the requested MSI/Xs at once.
If that's not required by the arch, the default version simply calls
arch_setup_msi_irq() for each MSI irq required.
arch_setup_msi_irqs() is passed a pdev, attached to the pdev is a list
of msi_descs with irq == 0, it is up to the arch to connect these up to
an irq (via set_irq_msi()) or return an error. For convenience the number
of vectors and the type are passed also.
All msi_descs with irq != 0 are considered allocated, and the arch
teardown routine will be called on them when necessary.
The existing semantics of pci_enable_msix() are that if the requested
number of irqs can not be allocated, the maximum number that _could_ be
allocated is returned. To support that, we define that in case of an
error from arch_setup_msi_irqs(), the number of msi_descs with irq != 0
are considered allocated, and are counted toward the "max that could be
allocated".
Signed-off-by: Michael Ellerman <michael@ellerman.id.au>
Signed-off-by: Greg Kroah-Hartman <gregkh@suse.de>
set_irq_msi() currently connects an irq_desc to an msi_desc. The archs call
it at some point in their setup routine, and then the generic code sets up the
reverse mapping from the msi_desc back to the irq.
set_irq_msi() should do both connections, making it the one and only call
required to connect an irq with it's MSI desc and vice versa.
The arch code MUST call set_irq_msi(), and it must do so only once it's sure
it's not going to fail the irq allocation.
Given that there's no need for the arch to return the irq anymore, the return
value from the arch setup routine just becomes 0 for success and anything else
for failure.
Signed-off-by: Michael Ellerman <michael@ellerman.id.au>
Signed-off-by: Greg Kroah-Hartman <gregkh@suse.de>
Now that we keep a list of msi descriptors, we don't need first_msi_irq
in the pci dev.
If we somehow have zero MSIs configured list_entry() will give us weird
oopes or nice memory corruption bugs. So be paranoid. Add BUG_ONs and also
a check in pci_msi_check_device() to make sure nvec > 0.
Signed-off-by: Michael Ellerman <michael@ellerman.id.au>
Signed-off-by: Greg Kroah-Hartman <gregkh@suse.de>
The msi descriptors are linked together with what looks a lot like
a linked list, but isn't a struct list_head list. Make it one.
The only complication is that previously we walked a list of irqs, and
got the descriptor for each with get_irq_msi(). Now we have a list of
descriptors and need to get the irq out of it, so it needs to be in the
actual struct msi_desc. We use 0 to indicate no irq is setup.
Signed-off-by: Michael Ellerman <michael@ellerman.id.au>
Signed-off-by: Greg Kroah-Hartman <gregkh@suse.de>
Add an arch_check_device(), which gives archs a chance to check the input
to pci_enable_msi/x. The arch might be interested in the value of nvec so
pass it in. Propagate the error value returned from the arch routine out
to the caller.
Signed-off-by: Michael Ellerman <michael@ellerman.id.au>
Signed-off-by: Greg Kroah-Hartman <gregkh@suse.de>
As pointed out by Eric, the name pci_msi_supported() suggests it should
return a boolean value, however it doesn't. So update the name to be
a bit less confusing and update the doco too.
Signed-off-by: Michael Ellerman <michael@ellerman.id.au>
Signed-off-by: Greg Kroah-Hartman <gregkh@suse.de>
Consolidate precondition checks into a single if statement.
Signed-off-by: Michael Ellerman <michael@ellerman.id.au>
Acked-by: Eric W. Biederman <ebiederm@xmission.com>
Signed-off-by: Greg Kroah-Hartman <gregkh@suse.de>
pci_enable_msi() and pci_enable_msix() both search for the MSI/MSI-X
capability, we can fold this into pci_msi_supported() by passing the
type in.
Update the code to match the comment for pci_msi_supported(). That is
it returns 0 on success, and anything else indicates an error.
Signed-off-by: Michael Ellerman <michael@ellerman.id.au>
Signed-off-by: Greg Kroah-Hartman <gregkh@suse.de>
We don't need a special cache just for msi descriptors. They're not
particularly large, under 100 bytes for sure, and don't seem to require any
special alignment etc. On most systems there will be relatively few MSIs,
and hence we waste most of a page on the cache. Better to just kzalloc the
space for the few we do need.
Signed-off-by: Michael Ellerman <michael@ellerman.id.au>
Acked-by: Eric W. Biederman <ebiederm@xmission.com>
Signed-off-by: Greg Kroah-Hartman <gregkh@suse.de>
Move EXPORT_SYMBOL()s near their definition.
Signed-off-by: Michael Ellerman <michael@ellerman.id.au>
Acked-by: Eric W. Biederman <ebiederm@xmission.com>
Signed-off-by: Greg Kroah-Hartman <gregkh@suse.de>
When freeing MSIs and MSI-Xs, we BUG_ON() if the irq has not been
freed, ie. if it still has an action. We can consolidate all of these
BUG_ON()s into msi_free_irqs() as all the code paths lead there almost
immediately anyway.
Signed-off-by: Michael Ellerman <michael@ellerman.id.au>
Acked-by: Eric W. Biederman <ebiederm@xmission.com>
Signed-off-by: Greg Kroah-Hartman <gregkh@suse.de>
For the MSI-X case we do exactly the same logic in pci_disable_msix() and
msi_remove_pci_irq_vectors(), so consolidate them.
msi_remove_pci_irq_vectors() wasn't setting dev->first_msi_irq to 0, but
I think it should have been, so the consolidated version does.
Signed-off-by: Michael Ellerman <michael@ellerman.id.au>
Acked-by: Eric W. Biederman <ebiederm@xmission.com>
Signed-off-by: Greg Kroah-Hartman <gregkh@suse.de>
Although it might be nice to do a printk before BUG'ing, it's really not
necessary, and it complicates the code.
The behaviour has changed slightly, in that before we set a flag if the irq
had an action, and continued freeing the other irqs. But as I see it that's
all irrelevant because we end up BUG'ing anyway.
Signed-off-by: Michael Ellerman <michael@ellerman.id.au>
Acked-by: Eric W. Biederman <ebiederm@xmission.com>
Signed-off-by: Greg Kroah-Hartman <gregkh@suse.de>
Although it might be nice to do a printk before BUG'ing, it's really not
necessary, and it complicates the code.
Signed-off-by: Michael Ellerman <michael@ellerman.id.au>
Acked-by: Eric W. Biederman <ebiederm@xmission.com>
Signed-off-by: Greg Kroah-Hartman <gregkh@suse.de>
Although it might be nice to do a printk before BUG'ing, it's really not
necessary, and it complicates the code.
The behaviour has changed slightly, in that before we set a flag if the irq
had an action, and continued freeing the other irqs. But as I see it that's
all irrelevant because we end up BUG'ing anyway.
Signed-off-by: Michael Ellerman <michael@ellerman.id.au>
Acked-by: Eric W. Biederman <ebiederm@xmission.com>
Signed-off-by: Greg Kroah-Hartman <gregkh@suse.de>
Although it might be nice to do a printk before BUG'ing, it's really not
necessary, and it complicates the code.
Signed-off-by: Michael Ellerman <michael@ellerman.id.au>
Acked-by: Eric W. Biederman <ebiederm@xmission.com>
Signed-off-by: Greg Kroah-Hartman <gregkh@suse.de>
This patch fixes a kernel bug which is triggered when using the
irqbalance daemon with MSI-X hardware.
Because both MSI-X interrupt messages and MSI-X table writes are posted,
it's possible for them to cross while in-flight. This results in
interrupts being received long after the kernel thinks they're disabled,
and in interrupts being sent to stale vectors after rebalancing.
This patch performs a read flush after writes to the MSI-X table for
mask and unmask operations. Since the SMP affinity is set while
the interrupt is masked, and since it's unmasked immediately after,
no additional flushes are required in the various affinity setting
routines.
This patch has been validated with (unreleased) network hardware which
uses MSI-X.
Revised with input from Eric Biederman.
Signed-off-by: Mitch Williams <mitch.a.williams@intel.com>
Acked-by: "Eric W. Biederman" <ebiederm@xmission.com>
Signed-off-by: Greg Kroah-Hartman <gregkh@suse.de>
This is a simplified and actually more comprehensive form of a bug
fix from Mitch Williams <mitch.a.williams@intel.com>.
When we mask or unmask a msi-x irqs the writes may be posted because
we are writing to memory mapped region. This means the mask and
unmask don't happen immediately but at some unspecified time in the
future. Which is out of sync with how the mask/unmask logic work
for ioapic irqs.
The practical result is that we get very subtle and hard to track down
irq migration bugs.
This patch performs a read flush after writes to the MSI-X table for mask
and unmask operations. Since the SMP affinity is set while the interrupt
is masked, and since it's unmasked immediately after, no additional flushes
are required in the various affinity setting routines.
The testing by Mitch Williams on his especially problematic system should
still be valid as I have only simplified the code, not changed the
functionality.
We currently have 7 drivers: cciss, mthca, cxgb3, forceth, s2io,
pcie/portdrv_core, and qla2xxx in 2.6.21 that are affected by this
problem when the hardware they driver is plugged into the right slot.
Given the difficulty of reproducing this bug and tracing it down to
anything that even remotely resembles a cause, even if people are
being affected we aren't likely to see many meaningful bug reports, and
the people who see this bug aren't likely to be able to reproduce this
bug in a timely fashion. So it is best to get this problem fixed
as soon as we can so people don't have problems.
Then if people do have a kernel message stating "No irq for vector" we
will know it is yet another novel cause that needs a complete new
investigation.
Cc: Greg KH <greg@kroah.com>
Cc: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: "Eric W. Biederman" <ebiederm@xmission.com>
Acked-by: Mitch Williams <mitch.a.williams@intel.com>
Acked-by: "Siddha, Suresh B" <suresh.b.siddha@intel.com>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
There are two ways pci_save_state and pci_restore_state are used. As
helper functions during suspend/resume, and as helper functions around
a hardware reset event. When used as helper functions around a hardware
reset event there is no reason to believe the calls will be paired, nor
is there a good reason to believe that if we restore the msi state from
before the reset that it will match the current msi state. Since arch
code may change the msi message without going through the driver, drivers
currently do not have enough information to even know when to call
pci_save_state to ensure they will have msi state in sync with the other
kernel irq reception data structures.
It turns out the solution is straight forward, cache the state in the
existing msi data structures (not the magic pci saved things) and
have the msi code update the cached state each time we write to the hardware.
This means we never need to read the hardware to figure out what the hardware
state should be.
By modifying the caching in this manner we get to remove our save_state
routines and only need to provide restore_state routines.
The only fields that were at all tricky to regenerate were the msi and msi-x
control registers and the way we regenerate them currently is a bit dependent
upon assumptions on how we use the allow msi registers to be configured and used
making the code a little bit brittle. If we ever change what cases we allow
or how we configure the msi bits we can address the fragility then.
Signed-off-by: Eric W. Biederman <ebiederm@xmission.com>
Signed-off-by: Greg Kroah-Hartman <gregkh@suse.de>
Acked-by: Auke Kok <auke-jan.h.kok@intel.com>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
Matthew Wilcox