Currently there is a regression and the ipc sysctls don't show up in the
binary sysctl namespace.
This patch adds sysctl_ipc_data to read data/write from the appropriate
namespace and deliver it in the expected manner.
[[email protected]: warning fix]
Signed-off-by: Eric W. Biederman <[email protected]>
Signed-off-by: Andrew Morton <[email protected]>
Signed-off-by: Linus Torvalds <[email protected]>
Refactor the ipc sysctl support so that it is simpler, more readable, and
prepares for fixing the bug with the wrong values being returned in the
sys_sysctl interface.
The function proc_do_ipc_string() was misnamed as it never handled strings.
It's magic of when to work with strings and when to work with longs belonged
in the sysctl table. I couldn't tell if the code would work if you disabled
the ipc namespace but it certainly looked like it would have problems.
Signed-off-by: Eric W. Biederman <[email protected]>
Signed-off-by: Andrew Morton <[email protected]>
Signed-off-by: Linus Torvalds <[email protected]>
The problem: When using sys_sysctl we don't read the proper values for the
variables exported from the uts namespace, nor do we do the proper locking.
This patch introduces sysctl_uts_string which properly fetches the values and
does the proper locking.
Signed-off-by: Eric W. Biederman <[email protected]>
Signed-off-by: Andrew Morton <[email protected]>
Signed-off-by: Linus Torvalds <[email protected]>
The binary interface to the namespace sysctls was never implemented resulting
in some really weird things if you attempted to use sys_sysctl to read your
hostname for example.
This patch series simples the code a little and implements the binary sysctl
interface.
In testing this patch series I discovered that our 32bit compatibility for the
binary sysctl interface is imperfect. In particular KERN_SHMMAX and
KERN_SMMALL are size_t sized quantities and are returned as 8 bytes on to
32bit binaries using a x86_64 kernel. However this has existing for a long
time so it is not a new regression with the namespace work.
Gads the whole sysctl thing needs work before it stops being easy to shoot
yourself in the foot.
Looking forward a little bit we need a better way to handle sysctls and
namespaces as our current technique will not work for the network namespace.
I think something based on the current overlapping sysctl trees will work but
the proc side needs to be redone before we can use it.
This patch:
Introduce get_uts() and put_uts() (used later) and remove most of the special
cases for when UTS namespace is compiled in.
Signed-off-by: Eric W. Biederman <[email protected]>
Signed-off-by: Andrew Morton <[email protected]>
Signed-off-by: Linus Torvalds <[email protected]>
This patch modifies the i386 kernel so that if CONFIG_RELOCATABLE is
selected it will be able to be loaded at any 4K aligned address below
1G. The technique used is to compile the decompressor with -fPIC and
modify it so the decompressor is fully relocatable. For the main
kernel relocations are generated. Resulting in a kernel that is relocatable
with no runtime overhead and no need to modify the source code.
A reserved 32bit word in the parameters has been assigned
to serve as a stack so we figure out where are running.
Signed-off-by: Eric W. Biederman <[email protected]>
Signed-off-by: Vivek Goyal <[email protected]>
Signed-off-by: Andi Kleen <[email protected]>
Print the addresses of non-absolute symbols relative to _text
so that ld will generate relocations. Allowing a relocatable
kernel to relocate them. We can't actually use the symbol names
because kallsyms includes static symbols that are not exported
from their object files.
Add the _text symbol definitions to the architectures which don't
define it otherwise linker will fail.
Signed-off-by: Eric W. Biederman <[email protected]>
Signed-off-by: Vivek Goyal <[email protected]>
Signed-off-by: Andi Kleen <[email protected]>
Defining __PHYSICAL_START and __KERNEL_START in asm-i386/page.h works but
it triggers a full kernel rebuild for the silliest of reasons. This
modifies the users to directly use CONFIG_PHYSICAL_START and linux/config.h
which prevents the full rebuild problem, which makes the code much
more maintainer and hopefully user friendly.
Signed-off-by: Eric W. Biederman <[email protected]>
Signed-off-by: Vivek Goyal <[email protected]>
Signed-off-by: Andi Kleen <[email protected]>
Currently when we are reserving the memory the kernel text
resides in we start at __PHYSICAL_START which happens to be
correct but not very obvious. In addition when we start relocating
the kernel __PHYSICAL_START is the wrong value, as it is an
absolute symbol that does not get relocated.
By starting the reservation at __pa_symbol(_text)
the code is clearer and will be correct when relocated.
Signed-off-by: Eric W. Biederman <[email protected]>
Signed-off-by: Vivek Goyal <[email protected]>
Signed-off-by: Andi Kleen <[email protected]>
On x86_64 we have to be careful with calculating the physical
address of kernel symbols. Both because of compiler odditities
and because the symbols live in a different range of the virtual
address space.
Having a defintition of __pa_symbol that works on both x86_64 and
i386 simplifies writing code that works for both x86_64 and
i386 that has these kinds of dependencies.
So this patch adds the trivial i386 __pa_symbol definition.
Added assembly magic similar to RELOC_HIDE as suggested by Andi Kleen.
Just picked it up from x86_64.
Signed-off-by: Eric W. Biederman <[email protected]>
Signed-off-by: Vivek Goyal <[email protected]>
Signed-off-by: Andi Kleen <[email protected]>
Komuro reports that ISA interrupts do not work after a disable_irq(),
causing some PCMCIA drivers to not work, with messages like
eth0: Asix AX88190: io 0x300, irq 3, hw_addr xx:xx:xx:xx:xx:xx
eth0: found link beat
eth0: autonegotiation complete: 100baseT-FD selected
eth0: interrupt(s) dropped!
eth0: interrupt(s) dropped!
eth0: interrupt(s) dropped!
...
Linus Torvalds <[email protected]> said:
"Now, edge-triggered interrupts are a _lot_ harder to mask, because the
Intel APIC is an unbelievable piece of sh*t, and has the edge-detect logic
_before_ the mask logic, so if a edge happens _while_ the device is
masked, you'll never ever see the edge ever again (unmasking will not
cause a new edge, so you simply lost the interrupt).
So when you "mask" an edge-triggered IRQ, you can't really mask it at all,
because if you did that, you'd lose it forever if the IRQ comes in while
you masked it. Instead, we're supposed to leave it active, and set a flag,
and IF the IRQ comes in, we just remember it, and mask it at that point
instead, and then on unmasking, we have to replay it by sending a
self-IPI."
This trivial patch solves the problem.
Signed-off-by: Eric W. Biederman <[email protected]>
Cc: Ingo Molnar <[email protected]>
Acked-by: Komuro <[email protected]>
Signed-off-by: Linus Torvalds <[email protected]>
This patch adds a variant of ht_create_irq __ht_create_irq that takes an
aditional parameter update that is a function that is called whenever we want
to write to a drivers htirq configuration registers.
This is needed to support the ipath_iba6110 because it's registers in the
proper location are not actually conected to the hardware that controlls
interrupt delivery.
[[email protected]: fixes]
Signed-off-by: Eric W. Biederman <[email protected]>
Cc: Andi Kleen <[email protected]>
Cc: <[email protected]>
Cc: Roland Dreier <[email protected]>
Signed-off-by: Bryan O'Sullivan <[email protected]>
Signed-off-by: Andrew Morton <[email protected]>
Signed-off-by: Linus Torvalds <[email protected]>
This refactoring actually optimizes the code a little by caching the value
that we think the device is programmed with instead of reading it back from
the hardware. Which simplifies the code a little and should speed things up a
bit.
This patch introduces the concept of a ht_irq_msg and modifies the
architecture read/write routines to update this code.
There is a minor consistency fix here as well as x86_64 forgot to initialize
the htirq as masked.
Signed-off-by: Eric W. Biederman <[email protected]>
Cc: Andi Kleen <[email protected]>
Acked-by: Bryan O'Sullivan <[email protected]>
Cc: <[email protected]>
Cc: Roland Dreier <[email protected]>
Signed-off-by: Andrew Morton <[email protected]>
Signed-off-by: Linus Torvalds <[email protected]>
The basic issue is that despite have been deprecated and warned about as a
very bad thing in the man pages since its inception there are a few real
users of sys_sysctl. It was my assumption that because sysctl had been
deprecated for all of 2.6 there would be no user space users by this point,
so I initially gave sys_sysctl a very short deprecation period.
Now that I know there are a few real users the only sane way to proceed
with deprecation is to push the time limit out to a year or two work and
work with distributions that have big testing pools like fedora core to
find these last remaining users.
Which means that the sys_sysctl interface needs to be maintained in the
meantime.
Since I have provided a technical measure that allows us to add new sysctl
entries without reserving more binary numbers I believe that is enough to
fix the sys_sysctl binary interface maintenance problems, because there is
no longer a need to change the binary interface at all.
Since the sys_sysctl implementation needs to stay around for a while and
the worst of the maintenance issues that caused us to occasionally break
the ABI have been addressed I don't see any advantage in continuing with
the removal of sys_sysctl.
So instead of merely increasing the deprecation period this patch removes
the deprecation of sys_sysctl and modifies the kernel to compile the code
in by default.
With committing to maintain sys_sysctl we get all of the advantages of a
fast interface for anything that needs it. Currently sys_sysctl is about
5x faster than /proc/sys, for the same string data.
Signed-off-by: Eric W. Biederman <[email protected]>
Acked-by: Alan Cox <[email protected]>
Signed-off-by: Andrew Morton <[email protected]>
Signed-off-by: Linus Torvalds <[email protected]>
This patch takes the CTL_UNNUMBERD concept from NFS and makes it available to
all new sysctl users.
At the same time the sysctl binary interface maintenance documentation is
updated to mention and to describe what is needed to successfully maintain the
sysctl binary interface.
Signed-off-by: Eric W. Biederman <[email protected]>
Acked-by: Alan Cox <[email protected]>
Signed-off-by: Andrew Morton <[email protected]>
Signed-off-by: Linus Torvalds <[email protected]>
Since it is becoming clear that there are just enough users of the binary
sysctl interface that completely removing the binary interface from the kernel
will not be an option for foreseeable future, we need to find a way to address
the sysctl maintenance issues.
The basic problem is that sysctl requires one central authority to allocate
sysctl numbers, or else conflicts and ABI breakage occur. The proc interface
to sysctl does not have that problem, as names are not densely allocated.
By not terminating a sysctl table until I have neither a ctl_name nor a
procname, it becomes simple to add sysctl entries that don't show up in the
binary sysctl interface. Which allows people to avoid allocating a binary
sysctl value when not needed.
I have audited the kernel code and in my reading I have not found a single
sysctl table that wasn't terminated by a completely zero filled entry. So
this change in behavior should not affect anything.
I think this mechanism eases the pain enough that combined with a little
disciple we can solve the reoccurring sysctl ABI breakage.
Signed-off-by: Eric W. Biederman <[email protected]>
Acked-by: Alan Cox <[email protected]>
Signed-off-by: Andrew Morton <[email protected]>
Signed-off-by: Linus Torvalds <[email protected]>
Don't warn about libpthread's access to kernel.version. When it receives
-ENOSYS it will read /proc/sys/kernel/version.
If anything else shows up print the sysctl number string.
Signed-off-by: Eric W. Biederman <[email protected]>
Cc: Cal Peake <[email protected]>
Cc: Alan Cox <[email protected]>
Signed-off-by: Andrew Morton <[email protected]>
Signed-off-by: Linus Torvalds <[email protected]>
When I generalized __assign_irq_vector I failed to pay attention
to what happens when you access a per cpu data structure for
a cpu that is not online. It is an undefined case making any
code that does it have undefined behavior as well.
The code still needs to be able to allocate a vector across cpus
that are not online to properly handle combinations like lowest
priority interrupt delivery and cpu_hotplug. Not that we can do
that today but the infrastructure shouldn't prevent it.
So this patch updates the places where we touch per cpu data
to only touch online cpus, it makes cpu vector allocation
an atomic operation with respect to cpu hotplug, and it updates
the cpu start code to properly initialize vector_irq so we
don't have inconsistencies.
Signed-off-by: Eric W. Biederman <[email protected]>
Signed-off-by: Andi Kleen <[email protected]>
There is no reason to remember a per cpu position of which vector
to try. Keeping a global position is simpler and more likely to
result in a global vector allocation even if I don't need or require
it. For level triggered interrupts this means we are less likely to
acknowledge another cpus irq, and cause the level triggered irq to
harmlessly refire.
This simplification makes it easier to only access data structures
of online cpus, by having fewer special cases to deal with.
Signed-off-by: Eric W. Biederman <[email protected]>
Signed-off-by: Andi Kleen <[email protected]>
TARGET_CPUS is the default irq routing poicy. It specifies which cpus the
kernel should aim an irq at. In physflat delivery mode we can route an irq to
a single cpu. But that doesn't mean our default policy should only be a
single cpu is allowed.
By allowing the irq routing code to select from multiple cpus this enables
systems with more irqs then we can service on a single processor to actually
work.
I just audited and tested the code and irqbalance doesn't care, and the
io_apic.c doesn't care if we have extra cpus in the mask. Everything will use
or assume we are using the lowest numbered cpu in the mask if we can't use
them all.
So this should result in no behavior changes except on systems that need it.
Thanks for YH Lu for spotting this problem in his testing.
Cc: Yinghai Lu <[email protected]>
Signed-off-by: Eric W. Biederman <[email protected]>
Signed-off-by: Andrew Morton <[email protected]>
Signed-off-by: Andi Kleen <[email protected]>
Thanks to YH Lu for spotting this. It appears I missed this function when I
refactored allocate_irq_vector and introduced irq_domain, with the result that
all retriggered irqs would go to cpu 0 even if we were not prepared to receive
them there.
While reviewing YH's patch I also noticed that this function was missing
locking, and since I am now reading two values from two diffrent arrays that
looks like a race we might be able to hit in the real world.
Cc: Yinghai Lu <[email protected]>
Signed-off-by: Eric W. Biederman <[email protected]>
Signed-off-by: Andrew Morton <[email protected]>
Signed-off-by: Andi Kleen <[email protected]>
This patch fixes my one line thinko where I was clearing
the vector_irq entries on the wrong cpus.
Signed-off-by: Eric W. Biederman <[email protected]>
Signed-off-by: Linus Torvalds <[email protected]>
Due to code bugs or misbehaving hardware it is possible that we can
receive an interrupt that we have not mapped into a linux irq. Calling
BUG when that happens is very rude, and if the problem is mild enough
prevents anything else from getting done.
So instead of calling BUG just scream loudly about the problem and
continue running. We don't have enough knowledge to know which
interrupt triggered this behavior so we don't acknowledge it. This will
likely prevent a recurrence of the problem by jamming up the works with
an unacknowledged interrupt.
If the interrupt was something important it is quite possible that
nothing productive will happen past this point. But it is now at least
possible to keep working if the kernel can survive without the interrupt
we dropped on the floor.
Solutions like irqpoll should generally make dropped irqs non-fatal.
Signed-off-by: Eric W. Biederman <[email protected]>
Signed-off-by: Linus Torvalds <[email protected]>
The problem we can't take advantage of lowest priority delivery mode if
the vectors are allocated for only one cpu at a time. Nor can we work
around hardware that assumes lowest priority delivery mode is always
used with several cpus.
So this patch introduces the concept of a vector_allocation_domain. A
set of cpus that will receive an irq on the same vector. Currently the
code for implementing this is placed in the genapic structure so we can
vary this depending on how we are using the io_apics.
This allows us to restore the previous behaviour of genapic_flat without
removing the benefits of having separate vector allocation for large
machines.
This should also fix the problem report where a hyperthreaded cpu was
receving the irq on the wrong hyperthread when in logical delivery mode
because the previous behaviour is restored.
This patch properly records our allocation of the first 16 irqs to the
first 16 available vectors on all cpus. This should be fine but it may
run into problems with multiple interrupts at the same interrupt level.
Except for some badly maintained comments in the code and the behaviour
of the interrupt allocator I have no real understanding of that problem.
Signed-off-by: Eric W. Biederman <[email protected]>
Acked-by: Muli Ben-Yehuda <[email protected]>
Signed-off-by: Linus Torvalds <[email protected]>
Which vector an irq is assigned to now varies dynamically and is
not needed outside of io_apic.c. So remove the possibility
of accessing the information outside of io_apic.c and remove
the silly macro that makes looking for users of irq_vector
difficult.
The fact this compiles ensures there aren't any more pieces
of the old CONFIG_PCI_MSI weirdness that I failed to remove.
Signed-off-by: Eric W. Biederman <[email protected]>
Signed-off-by: Linus Torvalds <[email protected]>
In commit ace80ab796 I removed the weird
logic that used the vector number as the irq number when MSI was
defined. However pci_enable_irq was using a different test in the
io_apic_assign_irqs path and I missed it :(
This patch removes the wrong code so no one hits this problem.
This code is only active when a specific set of boot command line
parameters is specified which likely explains why no one has notices
this earlier.
Signed-off-by: Eric W. Biederman <[email protected]>
Signed-off-by: Linus Torvalds <[email protected]>
This moves the declarations for the architecture helpers into
include/linux/htirq.h from the generic include/linux/pci.h. Hopefully this
will make this distinction clearer.
htirq.h is included where it is needed.
The dependency on the msi code is fixed and removed.
The Makefile is tidied up.
Signed-off-by: Eric W. Biederman <[email protected]>
Cc: Ingo Molnar <[email protected]>
Cc: Tony Luck <[email protected]>
Cc: Andi Kleen <[email protected]>
Cc: Thomas Gleixner <[email protected]>
Cc: Greg KH <[email protected]>
Cc: Benjamin Herrenschmidt <[email protected]>
Signed-off-by: Andrew Morton <[email protected]>
Signed-off-by: Linus Torvalds <[email protected]>
It turns out msi_ops was simply not enough to abstract the architecture
specific details of msi. So I have moved the resposibility of constructing
the struct irq_chip to the architectures, and have two architecture specific
functions arch_setup_msi_irq, and arch_teardown_msi_irq.
For simple architectures those functions can do all of the work. For
architectures with platform dependencies they can call into the appropriate
platform code.
With this msi.c is finally free of assuming you have an apic, and this
actually takes less code.
The helpers for the architecture specific code are declared in the linux/msi.h
to keep them separate from the msi functions used by drivers in linux/pci.h
Signed-off-by: Eric W. Biederman <[email protected]>
Cc: Ingo Molnar <[email protected]>
Cc: Tony Luck <[email protected]>
Cc: Andi Kleen <[email protected]>
Cc: Thomas Gleixner <[email protected]>
Cc: Greg KH <[email protected]>
Cc: Benjamin Herrenschmidt <[email protected]>
Signed-off-by: Andrew Morton <[email protected]>
Signed-off-by: Linus Torvalds <[email protected]>
The logic works like this.
Since we no longer track the state logic by hand in msi.c startup and shutdown
are no longer needed.
By updating msi_set_mask_bit to work on msi devices that do not implement a
mask bit we can always call the mask/unmask functions.
What we really have are mask and unmask so we use them to implement the .mask
and .unmask functions instead of .enable and .disable.
By switching to the handle_edge_irq handler we only need an ack function that
moves the irq if necessary. Which removes the old end and ack functions and
their peculiar logic of sometimes disabling an irq.
This removes the reliance on pre genirq irq handling methods.
Signed-off-by: Eric W. Biederman <[email protected]>
Cc: Ingo Molnar <[email protected]>
Cc: Tony Luck <[email protected]>
Cc: Andi Kleen <[email protected]>
Cc: Thomas Gleixner <[email protected]>
Cc: Greg KH <[email protected]>
Cc: Benjamin Herrenschmidt <[email protected]>
Signed-off-by: Andrew Morton <[email protected]>
Signed-off-by: Linus Torvalds <[email protected]>
Currently msi.c is doing sanity checks that make certain before an irq is
destroyed it has no more users.
By adding irq_has_action I can perform the test is a generic way, instead of
relying on a msi specific data structure.
By performing the core check in dynamic_irq_cleanup I ensure every user of
dynamic irqs has a test present and we don't free resources that are in use.
In msi.c this allows me to kill the attrib.state member of msi_desc and all of
the assciated code to maintain it.
To keep from freeing data structures when irq cleanup code is called to soon
changing dyanamic_irq_cleanup is insufficient because there are msi specific
data structures that are also not safe to free.
Signed-off-by: Eric W. Biederman <[email protected]>
Cc: Ingo Molnar <[email protected]>
Cc: Tony Luck <[email protected]>
Cc: Andi Kleen <[email protected]>
Cc: Thomas Gleixner <[email protected]>
Cc: Greg KH <[email protected]>
Cc: Benjamin Herrenschmidt <[email protected]>
Signed-off-by: Andrew Morton <[email protected]>
Signed-off-by: Linus Torvalds <[email protected]>
This patch implements two functions ht_create_irq and ht_destroy_irq for
use by drivers. Several other functions are implemented as helpers for
arch specific irq_chip handlers.
The driver for the card I tested this on isn't yet ready to be merged.
However this code is and hypertransport irqs are in use in a few other
places in the kernel. Not that any of this will get merged before 2.6.19
Because the ipath-ht400 is slightly out of spec this code will need to be
generalized to work there.
I think all of the powerpc uses are for a plain interrupt controller in a
chipset so support for native hypertransport devices is a little less
interesting.
However I think this is a half way decent model on how to separate arch
specific and generic helper code, and I think this is a functional model of
how to get the architecture dependencies out of the msi code.
[[email protected]: Kconfig fix]
Signed-off-by: Eric W. Biederman <[email protected]>
Cc: Greg KH <[email protected]>
Cc: Andi Kleen <[email protected]>
Cc: Benjamin Herrenschmidt <[email protected]>
Signed-off-by: Andrew Morton <[email protected]>
Signed-off-by: Linus Torvalds <[email protected]>
This patch adds support for systems that cannot receive every interrupt on a
single cpu simultaneously, in the check to see if we have enough HARDIRQ_BITS.
MAX_HARDIRQS_PER_CPU becomes the count of the maximum number of hardare
generated interrupts per cpu.
On architectures that support per cpu interrupt delivery this can be a
significant space savings and scalability bonus.
This patch adds support for systems that cannot receive every interrupt on
Signed-off-by: Eric W. Biederman <[email protected]>
Cc: Ingo Molnar <[email protected]>
Cc: Thomas Gleixner <[email protected]>
Cc: Benjamin Herrenschmidt <[email protected]>
Cc: Rajesh Shah <[email protected]>
Cc: Andi Kleen <[email protected]>
Cc: "Protasevich, Natalie" <[email protected]>
Cc: "Luck, Tony" <[email protected]>
Signed-off-by: Andrew Morton <[email protected]>
Signed-off-by: Linus Torvalds <[email protected]>
Because of the nasty way that CONFIG_PCI_MSI was implemented we wound up with
set_irq_info and set_native_irq_info, with move_irq and move_native_irq. Both
functions did the same thing but they were built and called under different
circumstances. Now that the msi hacks are gone we can kill move_irq and
set_irq_info.
Signed-off-by: Eric W. Biederman <[email protected]>
Cc: Ingo Molnar <[email protected]>
Cc: Thomas Gleixner <[email protected]>
Cc: Benjamin Herrenschmidt <[email protected]>
Cc: Rajesh Shah <[email protected]>
Cc: Andi Kleen <[email protected]>
Cc: "Protasevich, Natalie" <[email protected]>
Cc: "Luck, Tony" <[email protected]>
Signed-off-by: Andrew Morton <[email protected]>
Signed-off-by: Linus Torvalds <[email protected]>
This patch removes the change in behavior of the irq allocation code when
CONFIG_PCI_MSI is defined. Removing all instances of the assumption that irq
== vector.
create_irq is rewritten to first allocate a free irq and then to assign that
irq a vector.
assign_irq_vector is made static and the AUTO_ASSIGN case which allocates an
vector not bound to an irq is removed.
The ioapic vector methods are removed, and everything now works with irqs.
The definition of NR_IRQS no longer depends on CONFIG_PCI_MSI
[[email protected]: cleanup]
Signed-off-by: Eric W. Biederman <[email protected]>
Cc: Ingo Molnar <[email protected]>
Cc: Thomas Gleixner <[email protected]>
Cc: Benjamin Herrenschmidt <[email protected]>
Cc: Rajesh Shah <[email protected]>
Cc: Andi Kleen <[email protected]>
Cc: "Protasevich, Natalie" <[email protected]>
Cc: "Luck, Tony" <[email protected]>
Signed-off-by: Andrew Morton <[email protected]>
Signed-off-by: Linus Torvalds <[email protected]>
This patch removes the change in behavior of the irq allocation code when
CONFIG_PCI_MSI is defined. Removing all instances of the assumption that irq
== vector.
create_irq is rewritten to first allocate a free irq and then to assign that
irq a vector.
assign_irq_vector is made static and the AUTO_ASSIGN case which allocates an
vector not bound to an irq is removed.
The ioapic vector methods are removed, and everything now works with irqs.
The definition of NR_IRQS no longer depends on CONFIG_PCI_MSI
Signed-off-by: Eric W. Biederman <[email protected]>
Cc: Ingo Molnar <[email protected]>
Cc: Thomas Gleixner <[email protected]>
Cc: Benjamin Herrenschmidt <[email protected]>
Cc: Rajesh Shah <[email protected]>
Cc: Andi Kleen <[email protected]>
Cc: "Protasevich, Natalie" <[email protected]>
Cc: "Luck, Tony" <[email protected]>
Signed-off-by: Andrew Morton <[email protected]>
Signed-off-by: Linus Torvalds <[email protected]>
This removes the hardcoded assumption that irq == vector in the msi
composition code, and it allows the msi message composition to setup logical
mode, or lowest priorirty delivery mode as we do for other apic interrupts,
and with the same selection criteria.
Basically this moves the problem of what is in the msi message into the
architecture irq management code where it belongs. Not in a generic layer
that doesn't have enough information to compose msi messages properly.
Signed-off-by: Eric W. Biederman <[email protected]>
Cc: Ingo Molnar <[email protected]>
Cc: Thomas Gleixner <[email protected]>
Cc: Benjamin Herrenschmidt <[email protected]>
Cc: Rajesh Shah <[email protected]>
Cc: Andi Kleen <[email protected]>
Cc: "Protasevich, Natalie" <[email protected]>
Cc: "Luck, Tony" <[email protected]>
Signed-off-by: Andrew Morton <[email protected]>
Signed-off-by: Linus Torvalds <[email protected]>
This removes the hardcoded assumption that irq == vector in the msi
composition code, and it allows the msi message composition to setup logical
mode, or lowest priorirty delivery mode as we do for other apic interrupts,
and with the same selection criteria.
Basically this moves the problem of what is in the msi message into the
architecture irq management code where it belongs. Not in a generic layer
that doesn't have enough information to compose msi messages properly.
Signed-off-by: Eric W. Biederman <[email protected]>
Cc: Ingo Molnar <[email protected]>
Cc: Thomas Gleixner <[email protected]>
Cc: Benjamin Herrenschmidt <[email protected]>
Cc: Rajesh Shah <[email protected]>
Cc: Andi Kleen <[email protected]>
Cc: "Protasevich, Natalie" <[email protected]>
Cc: "Luck, Tony" <[email protected]>
Signed-off-by: Andrew Morton <[email protected]>
Signed-off-by: Linus Torvalds <[email protected]>
The msi currently allocates irqs backwards. First it allocates a platform
dependent routing value for an interrupt the ``vector'' and then it figures
out from the vector which irq you are on.
For ia64 this is fine. For x86 and x86_64 this is complete nonsense and makes
an enourmous mess of the irq handling code and prevents some pretty
significant cleanups in the code for handling large numbers of irqs.
This patch refactors msi.c to work in terms of irqs and create_irq/destroy_irq
for dynamically managing irqs.
Hopefully this is finally a version of msi.c that is useful on more than just
x86 derivatives.
Signed-off-by: Eric W. Biederman <[email protected]>
Cc: Ingo Molnar <[email protected]>
Cc: Thomas Gleixner <[email protected]>
Cc: Benjamin Herrenschmidt <[email protected]>
Cc: Rajesh Shah <[email protected]>
Cc: Andi Kleen <[email protected]>
Cc: "Protasevich, Natalie" <[email protected]>
Cc: "Luck, Tony" <[email protected]>
Signed-off-by: Andrew Morton <[email protected]>
Signed-off-by: Linus Torvalds <[email protected]>
The current implementation of create_irq() is a hack but it is the current
hack that msi.c uses, and unfortunately the ``generic'' apic msi ops depend on
this hack. Thus we are this hack of assuming irq == vector until the
depencencies in the generic irq code are removed.
Signed-off-by: Eric W. Biederman <[email protected]>
Cc: Ingo Molnar <[email protected]>
Cc: Thomas Gleixner <[email protected]>
Cc: Benjamin Herrenschmidt <[email protected]>
Cc: Rajesh Shah <[email protected]>
Cc: Andi Kleen <[email protected]>
Cc: "Protasevich, Natalie" <[email protected]>
Cc: "Luck, Tony" <[email protected]>
Signed-off-by: Andrew Morton <[email protected]>
Signed-off-by: Linus Torvalds <[email protected]>
The current implementation of create_irq() is a hack but it is the current
hack that msi.c uses, and unfortunately the ``generic'' apic msi ops depend on
this hack. Thus we are stuck this hack of assuming irq == vector until the
depencencies in the generic msi code are removed.
Signed-off-by: Eric W. Biederman <[email protected]>
Cc: Ingo Molnar <[email protected]>
Cc: Thomas Gleixner <[email protected]>
Cc: Benjamin Herrenschmidt <[email protected]>
Cc: Rajesh Shah <[email protected]>
Cc: Andi Kleen <[email protected]>
Cc: "Protasevich, Natalie" <[email protected]>
Cc: "Luck, Tony" <[email protected]>
Signed-off-by: Andrew Morton <[email protected]>
Signed-off-by: Linus Torvalds <[email protected]>
With the msi support comes a new concept in irq handling, irqs that are
created dynamically at run time.
Currently the msi code allocates irqs backwards. First it allocates a
platform dependent routing value for an interrupt the ``vector'' and then it
figures out from the vector which irq you are on.
This msi backwards allocator suffers from two basic problems. The allocator
suffers because it is trying to do something that is architecture specific in
a generic way making it brittle, inflexible, and tied to tightly to the
architecture implementation. The alloctor also suffers from it's very
backwards nature as it has tied things together that should have no
dependencies.
To solve the basic dynamic irq allocation problem two new architecture
specific functions are added: create_irq and destroy_irq.
create_irq takes no input and returns an unused irq number, that won't be
reused until it is returned to the free poll with destroy_irq. The irq then
can be used for any purpose although the only initial consumer is the msi
code.
destroy_irq takes an irq number allocated with create_irq and returns it to
the free pool.
Making this functionality per architecture increases the simplicity of the irq
allocation code and increases it's flexibility.
dynamic_irq_init() and dynamic_irq_cleanup() are added to automate the
irq_desc initializtion that should happen for dynamic irqs.
Signed-off-by: Eric W. Biederman <[email protected]>
Cc: Ingo Molnar <[email protected]>
Cc: Thomas Gleixner <[email protected]>
Cc: Benjamin Herrenschmidt <[email protected]>
Cc: Rajesh Shah <[email protected]>
Cc: Andi Kleen <[email protected]>
Cc: "Protasevich, Natalie" <[email protected]>
Cc: "Luck, Tony" <[email protected]>
Signed-off-by: Andrew Morton <[email protected]>
Signed-off-by: Linus Torvalds <[email protected]>
Currently we attempt to predict how many irqs we will be able to allocate with
msi using pci_vector_resources and some complicated accounting, and then we
only allow each device as many irqs as we think are available on average.
Only the s2io driver even takes advantage of this feature all other drivers
have a fixed number of irqs they need and bail if they can't get them.
pci_vector_resources is inaccurate if anyone ever frees an irq. The whole
implmentation is racy. The current irq limit policy does not appear to make
sense with current drivers. So I have simplified things. We can revisit this
we we need a more sophisticated policy.
Signed-off-by: Eric W. Biederman <[email protected]>
Cc: Ingo Molnar <[email protected]>
Cc: Thomas Gleixner <[email protected]>
Cc: Benjamin Herrenschmidt <[email protected]>
Cc: Rajesh Shah <[email protected]>
Cc: Andi Kleen <[email protected]>
Cc: "Protasevich, Natalie" <[email protected]>
Cc: "Luck, Tony" <[email protected]>
Signed-off-by: Andrew Morton <[email protected]>
Signed-off-by: Linus Torvalds <[email protected]>
The current msi_ops are short sighted in a number of ways, this patch attempts
to fix the glaring deficiences.
- Report in msi_ops if a 64bit address is needed in the msi message, so we
can fail 32bit only msi structures.
- Send and receive a full struct msi_msg in both setup and target. This is
a little cleaner and allows for architectures that need to modify the data
to retarget the msi interrupt to a different cpu.
- In target pass in the full cpu mask instead of just the first cpu in case
we can make use of the full cpu mask.
- Operate in terms of irqs and not vectors, currently there is still a 1-1
relationship but on architectures other than ia64 I expect this will change.
Signed-off-by: Eric W. Biederman <[email protected]>
Cc: Ingo Molnar <[email protected]>
Cc: Thomas Gleixner <[email protected]>
Cc: Benjamin Herrenschmidt <[email protected]>
Cc: Rajesh Shah <[email protected]>
Cc: Andi Kleen <[email protected]>
Cc: "Protasevich, Natalie" <[email protected]>
Cc: "Luck, Tony" <[email protected]>
Signed-off-by: Andrew Morton <[email protected]>
Signed-off-by: Linus Torvalds <[email protected]>
In support of this I also add a struct msi_msg that captures the the two
address and one data field ina typical msi message, and I remember the pos and
if the address is 64bit in struct msi_desc.
This makes the code a little more readable and easier to maintain, and paves
the way to further simplfications.
Signed-off-by: Eric W. Biederman <[email protected]>
Cc: Ingo Molnar <[email protected]>
Cc: Thomas Gleixner <[email protected]>
Cc: Benjamin Herrenschmidt <[email protected]>
Cc: Rajesh Shah <[email protected]>
Cc: Andi Kleen <[email protected]>
Cc: "Protasevich, Natalie" <[email protected]>
Cc: "Luck, Tony" <[email protected]>
Signed-off-by: Andrew Morton <[email protected]>
Signed-off-by: Linus Torvalds <[email protected]>
The problem. Because the disable routines leave the msi interrupts in all
sorts of half enabled states the enable routines become impossible to
implement correctly, and almost impossible to understand.
Simplifing this allows me to simply kill the buggy reroute_msix_table, and
generally makes the code more maintainable.
Signed-off-by: Eric W. Biederman <[email protected]>
Cc: Ingo Molnar <[email protected]>
Cc: Thomas Gleixner <[email protected]>
Cc: Benjamin Herrenschmidt <[email protected]>
Cc: Rajesh Shah <[email protected]>
Cc: Andi Kleen <[email protected]>
Cc: "Protasevich, Natalie" <[email protected]>
Cc: "Luck, Tony" <[email protected]>
Cc: Rajesh Shah <[email protected]>
Signed-off-by: Andrew Morton <[email protected]>
Signed-off-by: Linus Torvalds <[email protected]>
Currently move_native_irq disables and renables the irq we are migrating to
ensure we don't take that irq when we are actually doing the migration
operation. Disabling the irq needs to happen but sometimes doing the work is
move_native_irq is too late.
On x86 with ioapics the irq move sequences needs to be:
edge_triggered:
mask irq.
move irq.
unmask irq.
ack irq.
level_triggered:
mask irq.
ack irq.
move irq.
unmask irq.
We can easily perform the edge triggered sequence, with the current defintion
of move_native_irq. However the level triggered case does not map well. For
that I have added move_masked_irq, to allow me to disable the irqs around both
the ack and the move.
Q: Why have we not seen this problem earlier?
A: The only symptom I have been able to reproduce is that if we change
the vector before acknowleding an irq the wrong irq is acknowledged.
Since we currently are not reprogramming the irq vector during
migration no problems show up.
We have to mask the irq before we acknowledge the irq or else we could
hit a window where an irq is asserted just before we acknowledge it.
Edge triggered irqs do not have this problem because acknowledgements
do not propogate in the same way.
Signed-off-by: Eric W. Biederman <[email protected]>
Cc: Ingo Molnar <[email protected]>
Cc: Thomas Gleixner <[email protected]>
Cc: Benjamin Herrenschmidt <[email protected]>
Cc: Rajesh Shah <[email protected]>
Cc: Andi Kleen <[email protected]>
Cc: "Protasevich, Natalie" <[email protected]>
Cc: "Luck, Tony" <[email protected]>
Signed-off-by: Andrew Morton <[email protected]>
Signed-off-by: Linus Torvalds <[email protected]>
The primary aim of this patchset is to remove maintenances problems caused by
the irq infrastructure. The two big issues I address are an artificially
small cap on the number of irqs, and that MSI assumes vector == irq. My
primary focus is on x86_64 but I have touched other architectures where
necessary to keep them from breaking.
- To increase the number of irqs I modify the code to look at the (cpu,
vector) pair instead of just looking at the vector.
With a large number of irqs available systems with a large irq count no
longer need to compress their irq numbers to fit. Removing a lot of brittle
special cases.
For acpi guys the result is that irq == gsi.
- Addressing the fact that MSI assumes irq == vector takes a few more
patches. But suffice it to say when I am done none of the generic irq code
even knows what a vector is.
In quick testing on a large Unisys x86_64 machine we stumbled over at least
one driver that assumed that NR_IRQS could always fit into an 8 bit number.
This driver is clearly buggy today. But this has become a class of bugs that
it is now much easier to hit.
This patch:
This is a minor space optimization. In practice I don't think this has any
affect because of our alignment constraints and the other fields but there is
not point in chewing up an uncessary word and since we already read the flag
field this should improve the cache hit ratio of the irq handler.
Signed-off-by: Eric W. Biederman <[email protected]>
Cc: Ingo Molnar <[email protected]>
Cc: Thomas Gleixner <[email protected]>
Cc: Benjamin Herrenschmidt <[email protected]>
Cc: Rajesh Shah <[email protected]>
Cc: Andi Kleen <[email protected]>
Cc: "Protasevich, Natalie" <[email protected]>
Cc: "Luck, Tony" <[email protected]>
Signed-off-by: Andrew Morton <[email protected]>
Signed-off-by: Linus Torvalds <[email protected]>
It was pointed out that since I am taking ARRAY_SIZE anyway the trailing empty
entry is silly and just wastes space.
Signed-off-by: Eric W. Biederman <[email protected]>
Cc: Oleg Nesterov <[email protected]>
Signed-off-by: Andrew Morton <[email protected]>
Signed-off-by: Linus Torvalds <[email protected]>
The value doesn't change but this ensures I will have the proper value when
other files are added to proc_base_stuff.
Signed-off-by: Eric W. Biederman <[email protected]>
Cc: Oleg Nesterov <[email protected]>
Signed-off-by: Andrew Morton <[email protected]>
Signed-off-by: Linus Torvalds <[email protected]>
The implementation is exactly the same and there is currently nothing to
distinguish proc_tid_attr, and proc_tgid_attr. So it is pointless to have two
separate implementations.
Signed-off-by: Eric W. Biederman <[email protected]>
Signed-off-by: Andrew Morton <[email protected]>
Signed-off-by: Linus Torvalds <[email protected]>
The hard coded inode numbers in proc currently limit its maintainability,
its flexibility, and what can be done with the rest of system. /proc limits
pid-max to 32768 on 32 bit systems it limits fd-max to 32768 on all systems,
and placing the pid in the inode number really gets in the way of implementing
subdirectories of per process information.
Ever since people started adding to the middle of the file type enumeration we
haven't been maintaing the historical inode numbers, all we have really
succeeded in doing is keeping the pid in the proc inode number. The pid is
already available in the directory name so no information is lost removing it
from the inode number.
So if something in user space cares if we remove the inode number from the
/proc inode it is almost certainly broken.
Signed-off-by: Eric W. Biederman <[email protected]>
Signed-off-by: Andrew Morton <[email protected]>
Signed-off-by: Linus Torvalds <[email protected]>
To remove the hard coded proc inode numbers it is necessary to be able to
create the proc inodes during readdir. The instantiate methods are the subset
of lookup that is needed to accomplish that.
This first step just splits the lookup methods into 2 functions.
Signed-off-by: Eric W. Biederman <[email protected]>
Signed-off-by: Andrew Morton <[email protected]>
Signed-off-by: Linus Torvalds <[email protected]>
This patch generalizes the concept of files in /proc that are related to
processes but live in the root directory of /proc
Ideally this would reuse infrastructure from the rest of the process specific
parts of proc but unfortunately security_task_to_inode must not be called on
files that are not strictly per process. security_task_to_inode really needs
to be reexamined as the security label can change in important places that we
are not currently catching, but I'm not certain that simplifies this problem.
By at least matching the structure of the rest of proc we get more idiom reuse
and it becomes easier to spot problems in the way things are put together.
Later things like /proc/mounts are likely to be moved into proc_base as well.
If union mounts are ever supported we may be able to make /proc a union mount,
and properly split it into 2 filesystems.
Signed-off-by: Eric W. Biederman <[email protected]>
Signed-off-by: Andrew Morton <[email protected]>
Signed-off-by: Linus Torvalds <[email protected]>
The problem with remembering a user space process by its pid is that it is
possible that the process will exit, pid wrap around will occur.
Converting to a struct pid avoid that problem, and paves the way for
implementing a pid namespace.
Also since usb is the only user of kill_proc_info_as_uid rename
kill_proc_info_as_uid to kill_pid_info_as_uid and have the new version take
a struct pid.
Signed-off-by: Eric W. Biederman <[email protected]>
Acked-by: Greg Kroah-Hartman <[email protected]>
Signed-off-by: Andrew Morton <[email protected]>
Signed-off-by: Linus Torvalds <[email protected]>
This has been needed for a long time, but now with the advent of a
reference counted struct pid there are real consequences for getting this
wrong.
Someone I think it was Oleg Nesterov pointed out that this construct was
missing locking, when I introduced struct pid. After taking time to review
the locking construct already present I figured out which lock needs to be
taken. The other paths that access f_owner.pid take either the f_owner
read or the write lock.
Signed-off-by: Eric W. Biederman <[email protected]>
Cc: Oleg Nesterov <[email protected]>
Signed-off-by: Andrew Morton <[email protected]>
Signed-off-by: Linus Torvalds <[email protected]>
This updates my proc: readdir race fix (take 3) patch
to account for the changes made by: Sukadev Bhattiprolu <[email protected]>
to introduce struct pspace.
Signed-off-by: Eric W. Biederman <[email protected]>
Cc: KAMEZAWA Hiroyuki <[email protected]>
Signed-off-by: Andrew Morton <[email protected]>
Signed-off-by: Linus Torvalds <[email protected]>
Use struct pidmap instead of pidmap_t.
This updates my proc: readdir race fix (take 3) patch
to account for the changes made by: Sukadev Bhattiprolu <[email protected]>
to kill pidmap_t.
Signed-off-by: Eric W. Biederman <[email protected]>
Cc: KAMEZAWA Hiroyuki <[email protected]>
Signed-off-by: Andrew Morton <[email protected]>
Signed-off-by: Linus Torvalds <[email protected]>
File handles can be requested to send sigio and sigurg to processes. By
tracking the destination processes using struct pid instead of pid_t we make
the interface safe from all potential pid wrap around problems.
Signed-off-by: Eric W. Biederman <[email protected]>
Signed-off-by: Andrew Morton <[email protected]>
Signed-off-by: Linus Torvalds <[email protected]>
I took a good hard look at the locking and it appears the locking on vt_pid
is the console semaphore. Every modified path is called under the console
semaphore except reset_vc when it is called from fn_SAK or do_SAK both of
which appear to be in interrupt context. In addition I need to be careful
because in the presence of an oops the console_sem may be arbitrarily
dropped.
Which leads me to conclude the current locking is inadequate for my needs.
Given the weird cases we could hit because of oops printing instead of
introducing an extra spin lock to protect the data and keep the pid to
signal and the signal to send in sync, I have opted to use xchg on just the
struct pid * pointer instead.
Due to console_sem we will stay in sync between vt_pid and vt_mode except
for a small window during a SAK, or oops handling. SAK handling should
kill any user space process that care, and oops handling we are broken
anyway. Besides the worst that can happen is that I try to send the wrong
signal.
Signed-off-by: Eric W. Biederman <[email protected]>
Cc: Oleg Nesterov <[email protected]>
Signed-off-by: Andrew Morton <[email protected]>
Signed-off-by: Linus Torvalds <[email protected]>
This is such a rare path it took me a while to figure out how to test
this after soring out the locking.
This patch does several things.
- The variables used are moved into a structure and declared in vt_kern.h
- A spinlock is added so we don't have SMP races updating the values.
- Instead of raw pid_t value a struct_pid is used to guard against
pid wrap around issues, if the daemon to spawn a new console dies.
Signed-off-by: Eric W. Biederman <[email protected]>
Cc: Oleg Nesterov <[email protected]>
Signed-off-by: Andrew Morton <[email protected]>
Signed-off-by: Linus Torvalds <[email protected]>
As we stop storing pid_t's and move to storing struct pid *. We need a way to
get the pid_t from the struct pid to report to user space what we have stored.
Having a clean well defined way to do this is especially important as we move
to multiple pid spaces as may need to report a different value to the caller
depending on which pid space the caller is in.
Signed-off-by: Eric W. Biederman <[email protected]>
Signed-off-by: Andrew Morton <[email protected]>
Signed-off-by: Linus Torvalds <[email protected]>
pids aren't something that drivers should care about. However there are a lot
of helper layers in the kernel that do care, and are built as modules. Before
I can convert them to using struct pid instead of pid_t I need to export the
appropriate symbols so they can continue to be built.
Signed-off-by: Eric W. Biederman <[email protected]>
Signed-off-by: Andrew Morton <[email protected]>
Signed-off-by: Linus Torvalds <[email protected]>
Currently the signal functions all either take a task or a pid_t argument.
This patch implements variants that take a struct pid *. After all of the
users have been update it is my intention to remove the variants that take a
pid_t as using pid_t can be more work (an extra hash table lookup) and
difficult to get right in the presence of multiple pid namespaces.
There are two kinds of functions introduced in this patch. The are the
general use functions kill_pgrp and kill_pid which take a priv argument that
is ultimately used to create the appropriate siginfo information, Then there
are _kill_pgrp_info, kill_pgrp_info, kill_pid_info the internal implementation
helpers that take an explicit siginfo.
The distinction is made because filling out an explcit siginfo is tricky, and
will be even more tricky when pid namespaces are introduced.
Signed-off-by: Eric W. Biederman <[email protected]>
Signed-off-by: Andrew Morton <[email protected]>
Signed-off-by: Linus Torvalds <[email protected]>
To avoid pid rollover confusion the kernel needs to work with struct pid *
instead of pid_t. Currently there is not an iterator that walks through all
of the tasks of a given pid type starting with a struct pid. This prevents us
replacing some pid_t instances with struct pid. So this patch adds
do_each_pid_task which walks through the set of task for a given pid type
starting with a struct pid.
Signed-off-by: Eric W. Biederman <[email protected]>
Signed-off-by: Andrew Morton <[email protected]>
Signed-off-by: Linus Torvalds <[email protected]>
In the last round of cleaning up the pid hash table a more general struct pid
was introduced, that can be referenced counted.
With the more general struct pid most if not all places where we store a pid_t
we can now store a struct pid * and remove the need for a hash table lookup,
and avoid any possible problems with pid roll over.
Looking forward to the pid namespaces struct pid * gives us an absolute form a
pid so we can compare and use them without caring which pid namespace we are
in.
This patchset introduces the infrastructure needed to use struct pid instead
of pid_t, and then it goes on to convert two different kernel users that
currently store a pid_t value.
There are a lot more places to go but this is enough to get the basic idea.
Before we can merge a pid namespace patch all of the kernel pid_t users need
to be examined. Those that deal with user space processes need to be
converted to using a struct pid *. Those that deal with kernel processes need
to converted to using the kthread api. A rare few that only use their current
processes pid values get to be left alone.
This patch:
task_session returns the struct pid of a tasks session.
task_pgrp returns the struct pid of a tasks process group.
task_tgid returns the struct pid of a tasks thread group.
task_pid returns the struct pid of a tasks process id.
These can be used to avoid unnecessary hash table lookups, and to implement
safe pid comparisions in the face of a pid namespace.
Signed-off-by: Eric W. Biederman <[email protected]>
Signed-off-by: Andrew Morton <[email protected]>
Signed-off-by: Linus Torvalds <[email protected]>
Helper functions in base.c like proc_pident_readdir and proc_pident_lookup
assume the directories have an associated task, and cannot currently be used
on the /proc root directory because it does not have such a task.
This small changes allows for base.c to be simplified and later when multiple
pid spaces are introduced it makes getting the needed context information
trivial.
Signed-off-by: Eric W. Biederman <[email protected]>
Signed-off-by: Andrew Morton <[email protected]>
Signed-off-by: Linus Torvalds <[email protected]>
Currently proc_pident_lookup gets the names and types from a table and then
has a huge switch statement to get the inode and file operations it needs.
That is silly and is becoming increasingly hard to maintain so I just put all
of the information in the table.
Signed-off-by: Eric W. Biederman <[email protected]>
Signed-off-by: Andrew Morton <[email protected]>
Signed-off-by: Linus Torvalds <[email protected]>
The problem: An opendir, readdir, closedir sequence can fail to report
process ids that are continually in use throughout the sequence of system
calls. For this race to trigger the process that proc_pid_readdir stops at
must exit before readdir is called again.
This can cause ps to fail to report processes, and it is in violation of
posix guarantees and normal application expectations with respect to
readdir.
Currently there is no way to work around this problem in user space short
of providing a gargantuan buffer to user space so the directory read all
happens in on system call.
This patch implements the normal directory semantics for proc, that
guarantee that a directory entry that is neither created nor destroyed
while reading the directory entry will be returned. For directory that are
either created or destroyed during the readdir you may or may not see them.
Furthermore you may seek to a directory offset you have previously seen.
These are the guarantee that ext[23] provides and that posix requires, and
more importantly that user space expects. Plus it is a simple semantic to
implement reliable service. It is just a matter of calling readdir a
second time if you are wondering if something new has show up.
These better semantics are implemented by scanning through the pids in
numerical order and by making the file offset a pid plus a fixed offset.
The pid scan happens on the pid bitmap, which when you look at it is
remarkably efficient for a brute force algorithm. Given that a typical
cache line is 64 bytes and thus covers space for 64*8 == 200 pids. There
are only 40 cache lines for the entire 32K pid space. A typical system
will have 100 pids or more so this is actually fewer cache lines we have to
look at to scan a linked list, and the worst case of having to scan the
entire pid bitmap is pretty reasonable.
If we need something more efficient we can go to a more efficient data
structure for indexing the pids, but for now what we have should be
sufficient.
In addition this takes no additional locks and is actually less code than
what we are doing now.
Also another very subtle bug in this area has been fixed. It is possible
to catch a task in the middle of de_thread where a thread is assuming the
thread of it's thread group leader. This patch carefully handles that case
so if we hit it we don't fail to return the pid, that is undergoing the
de_thread dance.
Thanks to KAMEZAWA Hiroyuki <[email protected]> for
providing the first fix, pointing this out and working on it.
[[email protected]: fix it]
Signed-off-by: Eric W. Biederman <[email protected]>
Acked-by: KAMEZAWA Hiroyuki <[email protected]>
Signed-off-by: Oleg Nesterov <[email protected]>
Cc: Jean Delvare <[email protected]>
Signed-off-by: Andrew Morton <[email protected]>
Signed-off-by: Linus Torvalds <[email protected]>
This appears to be the only usage of is_init in the kernel besides the
usage in sched.h. On ia64 the same function is called in_init. So to
remove the conflict and make the kernel more consistent rename is_init
is_core is_local and is_local_section to in_init in_core in_local and
in_local_section respectively.
Thanks to Adrian Bunk who spotted this, and to Matthew Wilcox
who suggested this fix.
Signed-off-by: Eric Biederman <[email protected]>
Cc: Kyle McMartin <[email protected]>
Cc: Matthew Wilcox <[email protected]>
Cc: Adrian Bunk <[email protected]>
Signed-off-by: Andrew Morton <[email protected]>
Signed-off-by: Linus Torvalds <[email protected]>
Ingo Oeser pointed out that because current expands to an inline function
it is more space efficient and somewhat faster to simply keep a cached copy
of current in another variable. This patch implements that for the
de_thread function.
(akpm: saves nearly 100 bytes of text on x86)
Signed-off-by: Eric W. Biederman <[email protected]>
Signed-off-by: Andrew Morton <[email protected]>
Signed-off-by: Linus Torvalds <[email protected]>
With the patches flying between Oleg and myself somehow this temporary
debug code got left in pid.c. It was never intended to make it to the
stable kernel.
Signed-off-by: Eric W. Biederman <[email protected]>
Cc: Oleg Nesterov <[email protected]>
Signed-off-by: Andrew Morton <[email protected]>
Signed-off-by: Linus Torvalds <[email protected]>
In de_thread we move pids from one process to another, a rather ugly case.
The function transfer_pid makes it clear what we are doing, and makes the
action atomic. This is useful we ever want to atomically traverse the
process group and session lists, in a rcu safe manner.
Even if the atomic properties this change should be a win as transfer_pid
should be less code to execute than executing both attach_pid and
detach_pid, and this should make de_thread slightly smaller as only a
single function call needs to be emitted. The only downside is that the
code might be slower to execute as the odds are against transfer_pid being
in cache.
Signed-off-by: Eric W. Biederman <[email protected]>
Cc: Oleg Nesterov <[email protected]>
Signed-off-by: Andrew Morton <[email protected]>
Signed-off-by: Linus Torvalds <[email protected]>
Since sys_sysctl is deprecated start allow it to be compiled out. This
should catch any remaining user space code that cares, and paves the way
for further sysctl cleanups.
[[email protected]: If sys_sysctl() is not compiled-in, emit a warning]
Signed-off-by: Eric W. Biederman <[email protected]>
Signed-off-by: Andrew Morton <[email protected]>
Signed-off-by: Linus Torvalds <[email protected]>
When testing on a Unisys machine it was discovered that the megaraid driver
would not initialize as it was requesting irq 162 instead of irq 1442 it
was assigned. The problem was the irq number had been truncated by being
stored in an unsigned char.
This patches fixes that problem and the driver now appears to work.
The ioctl interface appears fundamentally broken as it exports the irq
number to user space in an unsigned char.
Signed-off-by: Eric W. Biederman <[email protected]>
Signed-off-by: Andrew Morton <[email protected]>
Signed-off-by: James Bottomley <[email protected]>
kexec has been marked experimental for a year now and all
of the serious problems have been worked through. So it
is time (if not past time) to remove the experimental mark.
Signed-off-by: Eric W. Biederman <[email protected]>
Signed-off-by: Andi Kleen <[email protected]>
kexec has been marked experimental for a year now and all
of the serious problems have been worked through. So it
is time (if not past time) to remove the experimental mark.
Signed-off-by: Eric W. Biederman <[email protected]>
Signed-off-by: Andi Kleen <[email protected]>
In long mode the %cs is largely a relic. However there are a few cases
like iret where it matters that we have a valid value. Without this
patch it is possible to enter the kernel in startup_64 without setting
%cs to a valid value. With this patch we don't care what %cs value
we enter the kernel with, so long as the cs shadow register indicates
it is a privileged code segment.
Thanks to Magnus Damm for finding this problem and posting the
first workable patch. I have moved the jump to set %cs down a
few instructions so we don't need to take an extra jump. Which
keeps the code simpler.
Signed-of-by: Eric W. Biederman <[email protected]>
Signed-off-by: Andi Kleen <[email protected]>
Now for a completely different but trivial approach.
I just boot tested it with 255 CPUS and everything worked.
Currently everything (except module data) we place in
the per cpu area we know about at compile time. So
instead of allocating a fixed size for the per_cpu area
allocate the number of bytes we need plus a fixed constant
for to be used for modules.
It isn't perfect but it is much less of a pain to
work with than what we are doing now.
AK: fixed warning
Signed-off-by: Eric W. Biederman <[email protected]>
Signed-off-by: Andi Kleen <[email protected]>
Allows easier extension of the GDT by using the proper C symbol
for the size in the descriptor.
Signed-off-by: Eric W. Biederman <[email protected]>
Signed-off-by: Andi Kleen <[email protected]>
This patch removes the foolish assumption that SMP implied local apics.
That assumption is not-true on the Voyager subarch. This makes that
dependency explicit, and allows the code to build.
What gets disabled is just an optimization to get better crash dumps so the
support should work if there is a kernel that will initialization on the
voyager subarch under those harsh conditions.
Hopefully we can figure out how to initialize apics in init_IRQ and remove
the need to disable io_apics and this dependency.
Signed-off-by: Eric W. Biederman <[email protected]>
Signed-off-by: Andrew Morton <[email protected]>
Signed-off-by: Linus Torvalds <[email protected]>
It looks like someone confused kmem_cache_create with a different allocator
and was attempting to give it knowledge of how many cache entries there
were.
With the unfortunate result that each slab entry was big enough to hold
every irq.
Signed-off-by: Eric W. Biederman <[email protected]>
Cc: Greg KH <[email protected]>
Signed-off-by: Andrew Morton <[email protected]>
Signed-off-by: Linus Torvalds <[email protected]>
This is just like my previous removal of tasklist_lock from first_tgid, and
next_tgid. It simply had to wait until it was rcu safe to walk the thread
list.
This should be the last instance of the tasklist_lock in proc. So user
processes should not be able to influence the tasklist lock hold times.
Signed-off-by: Eric W. Biederman <[email protected]>
Signed-off-by: Andrew Morton <[email protected]>
Signed-off-by: Linus Torvalds <[email protected]>
In process of getting proc_fd_access_allowed to work it has developed a few
warts. In particular the special case that always allows introspection and
the special case to allow inspection of kernel threads.
The special case for introspection is needed for /proc/self/mem.
The special case for kernel threads really should be overridable
by security modules.
So consolidate these checks into ptrace.c:may_attach().
The check to always allow introspection is trivial.
The check to allow access to kernel threads, and zombies is a little
trickier. mem_read and mem_write already verify an mm exists so it isn't
needed twice. proc_fd_access_allowed only doesn't want a check to verify
task->mm exits, s it prevents all access to kernel threads. So just move
the task->mm check into ptrace_attach where it is needed for practical
reasons.
I did a quick audit and none of the security modules in the kernel seem to
care if they are passed a task without an mm into security_ptrace. So the
above move should be safe and it allows security modules to come up with
more restrictive policy.
Signed-off-by: Eric W. Biederman <[email protected]>
Cc: Stephen Smalley <[email protected]>
Cc: Chris Wright <[email protected]>
Cc: James Morris <[email protected]>
Signed-off-by: Andrew Morton <[email protected]>
Signed-off-by: Linus Torvalds <[email protected]>
Since 2.2 we have been doing a chroot check to see if it is appropriate to
return a read or follow one of these magic symlinks. The chroot check was
asking a question about the visibility of files to the calling process and
it was actually checking the destination process, and not the files
themselves. That test was clearly bogus.
In my first pass through I simply fixed the test to check the visibility of
the files themselves. That naive approach to fixing the permissions was
too strict and resulted in cases where a task could not even see all of
it's file descriptors.
What has disturbed me about relaxing this check is that file descriptors
are per-process private things, and they are occasionaly used a user space
capability tokens. Looking a little farther into the symlink path on /proc
I did find userid checks and a check for capability (CAP_DAC_OVERRIDE) so
there were permissions checking this.
But I was still concerned about privacy. Besides /proc there is only one
other way to find out this kind of information, and that is ptrace. ptrace
has been around for a long time and it has a well established security
model.
So after thinking about it I finally realized that the permission checks
that make sense are the permission checks applied to ptrace_attach. The
checks are simple per process, and won't cause nasty surprises for people
coming from less capable unices.
Unfortunately there is one case that the current ptrace_attach test does
not cover: Zombies and kernel threads. Single stepping those kinds of
processes is impossible. Being able to see which file descriptors are open
on these tasks is important to lsof, fuser and friends. So for these
special processes I made the rule you can't find out unless you have
CAP_SYS_PTRACE.
These proc permission checks should now conform to the principle of least
surprise. As well as using much less code to implement :)
Signed-off-by: Eric W. Biederman <[email protected]>
Signed-off-by: Andrew Morton <[email protected]>
Signed-off-by: Linus Torvalds <[email protected]>
The code doesn't need to sleep to when making this check so I can just do the
comparison and not worry about the reference counts.
TODO: While looking at this I realized that my original cleanup did not push
the permission check far enough down into the stack. The call of
proc_check_dentry_visible needs to move out of the generic proc
readlink/follow link code and into the individual get_link instances.
Otherwise the shared resources checks are not quite correct (shared
files_struct does not require a shared fs_struct), and there are races with
unshare.
Signed-off-by: Eric W. Biederman <[email protected]>
Signed-off-by: Andrew Morton <[email protected]>
Signed-off-by: Linus Torvalds <[email protected]>
Incrementally update my proc-dont-lock-task_structs-indefinitely patches so
that they work with struct pid instead of struct task_ref.
Mostly this is a straight 1-1 substitution.
Signed-off-by: Eric W. Biederman <[email protected]>
Signed-off-by: Andrew Morton <[email protected]>
Signed-off-by: Linus Torvalds <[email protected]>
Eric W. Biederman