The host side of ring needs this logic too.
Signed-off-by: NRusty Russell <rusty@rustcorp.com.au>

Now noone relies on this behavior, we simplify virtqueue_add_buf() so it
return 0 or -errno.
Signed-off-by: NRusty Russell <rusty@rustcorp.com.au>
Acked-by: NMichael S. Tsirkin <mst@redhat.com>

They're generic concepts, so hoist them.  This also avoids accessor
functions (though kept around for merge with DaveM's net tree).

This goes even further than Jason Wang's 17bb6d40 patch
("virtio-ring: move queue_index to vring_virtqueue") which moved the
queue_index from the specific transport.
Acked-by: NMichael S. Tsirkin <mst@redhat.com>
Signed-off-by: NRusty Russell <rusty@rustcorp.com.au>

Virtio devices may attempt to add descriptors to a virtqueue from atomic
context using GFP_ATOMIC allocation. This is problematic because such
allocations can fall outside of the lowmem mapping, causing virt_to_phys
to report bogus physical addresses which are subsequently passed to
userspace via the buffers for the virtual device.

This patch masks out __GFP_HIGH and __GFP_HIGHMEM from the requested
flags when allocating descriptors for a virtqueue. If an atomic
allocation is requested and later fails, we will return -ENOSPC which
will be handled by the driver.

Cc: stable@kernel.org
Cc: Sasha Levin <levinsasha928@gmail.com>
Signed-off-by: NWill Deacon <will.deacon@arm.com>
Signed-off-by: NRusty Russell <rusty@rustcorp.com.au>

Instead of storing the queue index in transport-specific virtio structs,
this patch moves them to vring_virtqueue and introduces an helper to get
the value.  This lets drivers simplify their management and tracing of
virtqueues.
Signed-off-by: NJason Wang <jasowang@redhat.com>
Signed-off-by: NPaolo Bonzini <pbonzini@redhat.com>
Signed-off-by: NRusty Russell <rusty@rustcorp.com.au>

Use virtio_mb() to make sure the available index to be exposed before
checking the the avail event. Otherwise we may get stale value of
avail event in guest and never kick the host after.

Note: this fixes a bug introduced by ee7cd898.
Signed-off-by: NJason Wang <jasowang@redhat.com>
Acked-by: NMichael S. Tsirkin <mst@redhat.com>
Signed-off-by: NRusty Russell <rusty@rustcorp.com.au>
Cc: stable@kernel.org

Note: this fixes a bug introduced recently in
7b21e34f.
Signed-off-by: NJason Wang <jasowang@redhat.com>
Acked-by: NMichael S. Tsirkin <mst@redhat.com>
Signed-off-by: NRusty Russell <rusty@rustcorp.com.au>

Under the existing #ifdef DEBUG, check that they don't have more than
1/10 of a second between an add_buf() and a
virtqueue_notify()/virtqueue_kick_prepare() call.

We could get false positives on a really busy system, but good for
development.
Signed-off-by: NRusty Russell <rusty@rustcorp.com.au>

A virtio driver does virtqueue_add_buf() multiple times before finally
calling virtqueue_kick(); previously we only exposed the added buffers
in the virtqueue_kick() call.  This means we don't need a memory
barrier in virtqueue_add_buf(), but it reduces concurrency as the
device (ie. host) can't see the buffers until the kick.

In the unusual (but now possible) case where a driver does add_buf()
and get_buf() without doing a kick, we do need to insert one before
our counter wraps.  Otherwise we could wrap num_added, and later on
not realize that we have passed the marker where we should have
kicked.
Signed-off-by: NRusty Russell <rusty@rustcorp.com.au>

Since we know vq->vring.num is a power of 2, modulus is lazy (it's asserted
in vring_new_virtqueue()).
Signed-off-by: NRusty Russell <rusty@rustcorp.com.au>

Based on patch by Christoph for virtio_blk speedup:

	Split virtqueue_kick to be able to do the actual notification
	outside the lock protecting the virtqueue.  This patch was
	originally done by Stefan Hajnoczi, but I can't find the
	original one anymore and had to recreated it from memory.
	Pointers to the original or corrections for the commit message
	are welcome.

Stefan's patch was here:

	https://github.com/stefanha/linux/commit/a6d06644e3a58e57a774e77d7dc34c4a5a2e7496
	http://www.spinics.net/lists/linux-virtualization/msg14616.html

Third time's the charm!
Signed-off-by: NRusty Russell <rusty@rustcorp.com.au>

Remove wrapper functions. This makes the allocation type explicit in
all callers; I used GPF_KERNEL where it seemed obvious, left it at
GFP_ATOMIC otherwise.
Signed-off-by: NRusty Russell <rusty@rustcorp.com.au>
Reviewed-by: NChristoph Hellwig <hch@lst.de>

The old documentation is left over from when we used a structure with
strategy pointers.

And move the documentation to the C file as per kernel practice.
Though I disagree...
Signed-off-by: NRusty Russell <rusty@rustcorp.com.au>
Reviewed-by: NChristoph Hellwig <hch@lst.de>

We were cheating with our barriers; using the smp ones rather than the
real device ones.  That was fine, until rpmsg came along, which is
used to talk to a real device (a non-SMP CPU).

Unfortunately, just putting back the real barriers (reverting
d57ed95d) causes a performance regression on virtio-pci.  In
particular, Amos reports netbench's TCP_RR over virtio_net CPU
utilization increased up to 35% while throughput went down by up to
14%.

By comparison, this branch is in the noise.

Reference: https://lkml.org/lkml/2011/12/11/22Signed-off-by: NRusty Russell <rusty@rustcorp.com.au>

Up to now, the module.h header was as hard to keep out as
sunlight.  But we are cleaning that up.  Fix the virtio users
who simply expect module.h to be there in every C file.
Signed-off-by: NPaul Gortmaker <paul.gortmaker@windriver.com>

Add support for reporting ring sizes via ethtool -g to the virtio_net
driver.
Signed-off-by: NRick Jones <rick.jones2@hp.com>
Acked-by: NRusty Russell <rusty@rustcorp.com.au>
Acked-by: NMichael S. Tsirkin <mst@redhat.com>
Signed-off-by: NDavid S. Miller <davem@davemloft.net>

Add an API that tells the other side that callbacks
should be delayed until a lot of work has been done.
Implement using the new event_idx feature.

Note: it might seem advantageous to let the drivers
ask for a callback after a specific capacity has
been reached. However, as a single head can
free many entries in the descriptor table,
we don't really have a clue about capacity
until get_buf is called. The API is the simplest
to implement at the moment, we'll see what kind of
hints drivers can pass when there's more than one
user of the feature.
Signed-off-by: NMichael S. Tsirkin <mst@redhat.com>
Signed-off-by: NRusty Russell <rusty@rustcorp.com.au>

Support for the new event idx feature:
1. When enabling interrupts, publish the current avail index
   value to the host to get interrupts on the next update.
2. Use the new avail_event feature to reduce the number
   of exits from the guest.

Simple test with the simulator:

[virtio]# time ./virtio_test
spurious wakeus: 0x7

real    0m0.169s
user    0m0.140s
sys     0m0.019s
[virtio]# time ./virtio_test --no-event-idx
spurious wakeus: 0x11

real    0m0.649s
user    0m0.295s
sys     0m0.335s
Signed-off-by: NMichael S. Tsirkin <mst@redhat.com>
Signed-off-by: NRusty Russell <rusty@rustcorp.com.au>

When detaching a buffer from a vq, the avail.idx value should be
decremented as well.

This was noticed by hot-unplugging a virtio console port and then
plugging in a new one on the same number (re-using the vqs which were
just 'disowned').  qemu reported

   'Guest moved used index from 0 to 256'

when any IO was attempted on the new port.

CC: stable@kernel.org
Reported-by: Njuzhang <juzhang@redhat.com>
Signed-off-by: NAmit Shah <amit.shah@redhat.com>
Signed-off-by: NRusty Russell <rusty@rustcorp.com.au>

We can't rely on indirect buffers for capacity
calculations because they need a memory allocation
which might fail.  In particular, virtio_net can get
into this situation under stress, and it drops packets
and performs badly.

So return the number of buffers we can guarantee users.
Signed-off-by: NMichael S. Tsirkin <mst@redhat.com>
Signed-off-by: NRusty Russell <rusty@rustcorp.com.au>
Reported-By: NKrishna Kumar2 <krkumar2@in.ibm.com>

virtio ring was changed to return an error code on OOM,
but one caller was missed and still checks for vq->vring.num.
The fix is just to check for <0 error code.

Long term it might make sense to change goto add_head to
just return an error on oom instead, but let's apply
a minimal fix for 2.6.35.
Reported-by: NChris Mason <chris.mason@oracle.com>
Signed-off-by: NMichael S. Tsirkin <mst@redhat.com>
Signed-off-by: NRusty Russell <rusty@rustcorp.com.au>
Tested-by: NChris Mason <chris.mason@oracle.com>
Cc: stable@kernel.org # .34.x
Signed-off-by: NLinus Torvalds <torvalds@linux-foundation.org>

add_buf returns ring size on out of memory,
this is not what devices expect.
Signed-off-by: NMichael S. Tsirkin <mst@redhat.com>
Acked-by: NAmit Shah <amit.shah@redhat.com>
Signed-off-by: NRusty Russell <rusty@rustcorp.com.au>
Cc: stable@kernel.org # .34.x

Add an add_buf variant that gets gfp parameter. Use that
to allocate indirect buffers.
Signed-off-by: NMichael S. Tsirkin <mst@redhat.com>
Signed-off-by: NRusty Russell <rusty@rustcorp.com.au>

We have a single virtqueue_ops implementation,
and it seems unlikely we'll get another one
at this point. So let's remove an unnecessary
level of indirection: it would be very easy to
re-add it if another implementation surfaces.
Signed-off-by: NMichael S. Tsirkin <mst@redhat.com>
Signed-off-by: NRusty Russell <rusty@rustcorp.com.au>

include cleanup: Update gfp.h and slab.h includes to prepare for breaking implicit slab.h inclusion from percpu.h

percpu.h is included by sched.h and module.h and thus ends up being
included when building most .c files.  percpu.h includes slab.h which
in turn includes gfp.h making everything defined by the two files
universally available and complicating inclusion dependencies.

percpu.h -> slab.h dependency is about to be removed.  Prepare for
this change by updating users of gfp and slab facilities include those
headers directly instead of assuming availability.  As this conversion
needs to touch large number of source files, the following script is
used as the basis of conversion.

  http://userweb.kernel.org/~tj/misc/slabh-sweep.py

The script does the followings.

* Scan files for gfp and slab usages and update includes such that
  only the necessary includes are there.  ie. if only gfp is used,
  gfp.h, if slab is used, slab.h.

* When the script inserts a new include, it looks at the include
  blocks and try to put the new include such that its order conforms
  to its surrounding.  It's put in the include block which contains
  core kernel includes, in the same order that the rest are ordered -
  alphabetical, Christmas tree, rev-Xmas-tree or at the end if there
  doesn't seem to be any matching order.

* If the script can't find a place to put a new include (mostly
  because the file doesn't have fitting include block), it prints out
  an error message indicating which .h file needs to be added to the
  file.

The conversion was done in the following steps.

1. The initial automatic conversion of all .c files updated slightly
   over 4000 files, deleting around 700 includes and adding ~480 gfp.h
   and ~3000 slab.h inclusions.  The script emitted errors for ~400
   files.

2. Each error was manually checked.  Some didn't need the inclusion,
   some needed manual addition while adding it to implementation .h or
   embedding .c file was more appropriate for others.  This step added
   inclusions to around 150 files.

3. The script was run again and the output was compared to the edits
   from #2 to make sure no file was left behind.

4. Several build tests were done and a couple of problems were fixed.
   e.g. lib/decompress_*.c used malloc/free() wrappers around slab
   APIs requiring slab.h to be added manually.

5. The script was run on all .h files but without automatically
   editing them as sprinkling gfp.h and slab.h inclusions around .h
   files could easily lead to inclusion dependency hell.  Most gfp.h
   inclusion directives were ignored as stuff from gfp.h was usually
   wildly available and often used in preprocessor macros.  Each
   slab.h inclusion directive was examined and added manually as
   necessary.

6. percpu.h was updated not to include slab.h.

7. Build test were done on the following configurations and failures
   were fixed.  CONFIG_GCOV_KERNEL was turned off for all tests (as my
   distributed build env didn't work with gcov compiles) and a few
   more options had to be turned off depending on archs to make things
   build (like ipr on powerpc/64 which failed due to missing writeq).

   * x86 and x86_64 UP and SMP allmodconfig and a custom test config.
   * powerpc and powerpc64 SMP allmodconfig
   * sparc and sparc64 SMP allmodconfig
   * ia64 SMP allmodconfig
   * s390 SMP allmodconfig
   * alpha SMP allmodconfig
   * um on x86_64 SMP allmodconfig

8. percpu.h modifications were reverted so that it could be applied as
   a separate patch and serve as bisection point.

Given the fact that I had only a couple of failures from tests on step
6, I'm fairly confident about the coverage of this conversion patch.
If there is a breakage, it's likely to be something in one of the arch
headers which should be easily discoverable easily on most builds of
the specific arch.
Signed-off-by: NTejun Heo <tj@kernel.org>
Guess-its-ok-by: NChristoph Lameter <cl@linux-foundation.org>
Cc: Ingo Molnar <mingo@redhat.com>
Cc: Lee Schermerhorn <Lee.Schermerhorn@hp.com>

vq operations depend on vq->data[i] being NULL to figure out if the vq
entry is in use (since the previous patch).

We have to initialize them to NULL to ensure we don't work with junk
data and trigger false BUG_ONs.
Signed-off-by: NAmit Shah <amit.shah@redhat.com>
Signed-off-by: NRusty Russell <rusty@rustcorp.com.au>
Cc: Shirley Ma <xma@us.ibm.com>

There's currently no way for a virtio driver to ask for unused
buffers, so it has to keep a list itself to reclaim them at shutdown.
This is redundant, since virtio_ring stores that information.  So
add a new hook to do this.
Signed-off-by: NShirley Ma <xma@us.ibm.com>
Signed-off-by: NAmit Shah <amit.shah@redhat.com>
Signed-off-by: NRusty Russell <rusty@rustcorp.com.au>

virtio is communicating with a virtual "device" that actually runs on
another host processor. Thus SMP barriers can be used to control
memory access ordering.

Where possible, we should use SMP barriers which are more lightweight than
mandatory barriers, because mandatory barriers also control MMIO effects on
accesses through relaxed memory I/O windows (which virtio does not use)
(compare specifically smp_rmb and rmb on x86_64).

We can't just use smp_mb and friends though, because
we must force memory ordering even if guest is UP since host could be
running on another CPU, but SMP barriers are defined to barrier() in
that configuration. So, for UP fall back to mandatory barriers instead.
Signed-off-by: NMichael S. Tsirkin <mst@redhat.com>
Signed-off-by: NRusty Russell <rusty@rustcorp.com.au>

With DEBUG defined, we add an ->in_use flag to detect if the caller
invokes two virtio methods in parallel.  The barriers attempt to ensure
timely update of the ->in_use flag.

But they're voodoo: if we need these barriers it implies that the
calling code doesn't have sufficient synchronization to ensure the
code paths aren't invoked at the same time anyway, and we want to
detect it.

Also, adding barriers changes timing, so turning on debug has more
chance of hiding real problems.

Thanks to MST for drawing my attention to this code...

CC: Michael S. Tsirkin <mst@redhat.com>
Signed-off-by: NRusty Russell <rusty@rustcorp.com.au>

There's currently no way for a virtio driver to ask for unused
buffers, so it has to keep a list itself to reclaim them at shutdown.
This is redundant, since virtio_ring stores that information.  So
add a new hook to do this.
Signed-off-by: NShirley Ma <xma@us.ibm.com>
Signed-off-by: NAmit Shah <amit.shah@redhat.com>
Signed-off-by: NRusty Russell <rusty@rustcorp.com.au>
Signed-off-by: NDavid S. Miller <davem@davemloft.net>

On SMP guests, reads from the ring might bypass used index reads. This
causes guest crashes because host writes to used index to signal ring
data readiness.  Fix this by inserting rmb before used ring reads.
Signed-off-by: NMichael S. Tsirkin <mst@redhat.com>
Signed-off-by: NRusty Russell <rusty@rustcorp.com.au>
Cc: stable@kernel.org

This API change means that virtio_net can tell how much capacity
remains for buffers.  It's necessarily fuzzy, since
VIRTIO_RING_F_INDIRECT_DESC means we can fit any number of descriptors
in one, *if* we can kmalloc.
Signed-off-by: NRusty Russell <rusty@rustcorp.com.au>
Cc: Dinesh Subhraveti <dineshs@us.ibm.com>

Add a new feature flag for indirect ring entries. These are ring
entries which point to a table of buffer descriptors.

The idea here is to increase the ring capacity by allowing a larger
effective ring size whereby the ring size dictates the number of
requests that may be outstanding, rather than the size of those
requests.

This should be most effective in the case of block I/O where we can
potentially benefit by concurrently dispatching a large number of
large requests. Even in the simple case of single segment block
requests, this results in a threefold increase in ring capacity.
Signed-off-by: NMark McLoughlin <markmc@redhat.com>
Signed-off-by: NRusty Russell <rusty@rustcorp.com.au>

Add a linked list of all virtqueues for a virtio device: this helps for
debugging and is also needed for upcoming interface change.

Also, add a "name" field for clearer debug messages.
Signed-off-by: NRusty Russell <rusty@rustcorp.com.au>

Impact: cleanup

Roel Kluin drew attention to these macros with his patch: here I
neaten them a little further:
1) Add a comment on what START_USE and END_USE are checking,
2) Brackets around _vq in BAD_RING,
3) Neaten formatting for START_USE so it's less than 80 cols.
Signed-off-by: NRusty Russell <rusty@rustcorp.com.au>

Impact: cleanup

fix BAD_RING, START_US and END_USE macros

When these macros aren't called with a variable named vq as first
argument, this would result in a build failure.
Signed-off-by: NRoel Kluin <roel.kluin@gmail.com>
Signed-off-by: NRusty Russell <rusty@rustcorp.com.au>

This allows each virtio user to hand in the alignment appropriate to
their virtio_ring structures.
Signed-off-by: NRusty Russell <rusty@rustcorp.com.au>
Acked-by: NChristian Borntraeger <borntraeger@de.ibm.com>

To prepare for virtio_ring transport feature bits, hook in a call in
all the users to manipulate them.  This currently just clears all the
bits, since it doesn't understand any features.
Signed-off-by: NRusty Russell <rusty@rustcorp.com.au>

We force notification when the ring is full, even if the host has
indicated it doesn't want to know.  This seemed like a good idea at
the time: if we fill the transmit ring, we should tell the host
immediately.

Unfortunately this logic also applies to the receiving ring, which is
refilled constantly.  We should introduce real notification thesholds
to replace this logic.  Meanwhile, removing the logic altogether breaks
the heuristics which KVM uses, so we use a hack: only notify if there are
outgoing parts of the new buffer.

Here are the number of exits with lguest's crappy network implementation:
Before:
	network xmit 7859051 recv 236420
After:
	network xmit 7858610 recv 118136
Signed-off-by: NRusty Russell <rusty@rustcorp.com.au>

virtio allows drivers to suppress callbacks (ie. interrupts) for
efficiency (no locking, it's just an optimization).

There's a similar mechanism for the host to suppress notifications
coming from the guest: in that case, we ignore the suppression if the
ring is completely full.

It turns out that life is simpler if the host similarly ignores
callback suppression when the ring is completely empty: the network
driver wants to free up old packets in a timely manner, and otherwise
has to use a timer to poll.

We have to remove the code which ignores interrupts when the driver
has disabled them (again, it had no locking and hence was unreliable
anyway).
Signed-off-by: NRusty Russell <rusty@rustcorp.com.au>