linux-2.6-microblaze.git
8 years agoIB/qib: Use address handle in rdmavt and remove from qib
Dennis Dalessandro [Fri, 22 Jan 2016 20:46:07 +0000 (12:46 -0800)]
IB/qib: Use address handle in rdmavt and remove from qib

Original patch from Kamal Heib <kamalh@mellanox.com>, split
apart from original.

Remove AH from qib and use rdmavt version.

Reviewed-by: Ira Weiny <ira.weiny@intel.com>
Signed-off-by: Kamal Heib <kamalh@mellanox.com>
Signed-off-by: Dennis Dalessandro <dennis.dalessandro@intel.com>
Signed-off-by: Doug Ledford <dledford@redhat.com>
8 years agoIB/qib: Remove qp and mr functionality from qib
Dennis Dalessandro [Fri, 22 Jan 2016 20:45:59 +0000 (12:45 -0800)]
IB/qib: Remove qp and mr functionality from qib

Remove qp and mr support from qib and use rdmavt. These two changes
cannot be reasonably be split apart into separate patches because they
depend on each other in mulitple places. This paves the way to remove
even more functions in subsequent patches.

Reviewed-by: Ira Weiny <ira.weiny@intel.com>
Reviewed-by: Mike Marciniszyn <mike.marciniszyn@intel.com>
Signed-off-by: Dennis Dalessandro <dennis.dalessandro@intel.com>
Signed-off-by: Doug Ledford <dledford@redhat.com>
8 years agoIB/qib: Add device specific info prints
Dennis Dalessandro [Fri, 22 Jan 2016 20:45:20 +0000 (12:45 -0800)]
IB/qib: Add device specific info prints

Implement get_card_name and get_pci_dev helper functions for rdmavt
for qib.

Reviewed-by: Mike Marciniszyn <mike.marciniszyn@intel.com>
Reviewed-by: Ira Weiny <ira.weiny@intel.com>
Signed-off-by: Dennis Dalessandro <dennis.dalessandro@intel.com>
Signed-off-by: Doug Ledford <dledford@redhat.com>
8 years agoIB/qib: Remove driver specific members from qib qp type
Dennis Dalessandro [Fri, 22 Jan 2016 20:45:11 +0000 (12:45 -0800)]
IB/qib: Remove driver specific members from qib qp type

In preparation for moving the queue pair data structure to rdmavt the
members of the driver specific queue pairs which are not common need to be
pushed off to a private driver structure. This structure will be available
in the queue pair once moved to rdmavt as a void pointer. This patch while
not adding a lot of value in and of itself is a prerequisite to move the
queue pair out of the drivers and into rdmavt.

The driver specific, private queue pair data structure should condense as
more of the send side code moves to rdmavt.

Reviewed-by: Ira Weiny <ira.weiny@intel.com>
Reviewed-by: Mike Marciniszyn <mike.marciniszyn@intel.com>
Signed-off-by: Dennis Dalessandro <dennis.dalessandro@intel.com>
Signed-off-by: Doug Ledford <dledford@redhat.com>
8 years agoIB/qib: Use rdmavt lid defines in qib
Dennis Dalessandro [Fri, 22 Jan 2016 20:45:02 +0000 (12:45 -0800)]
IB/qib: Use rdmavt lid defines in qib

Original patch for AH changes from Kamal Heib <kamalh@mellanox.com>, split
apart from original. This patch also removes the qib specific multicast
lid base and permissive lid defines since they are no longer needed.

Use common LID defines in qib driver.

Reviewed-by: Ira Weiny <ira.weiny@intel.com>
Signed-off-by: Kamal Heib <kamalh@mellanox.com>
Signed-off-by: Dennis Dalessandro <dennis.dalessandro@intel.com>
Signed-off-by: Doug Ledford <dledford@redhat.com>
8 years agoIB/qib: Remove most uses of QIB_PERMISSIVE_LID and QIB_MULTICAST_LID_BASE
Dennis Dalessandro [Fri, 22 Jan 2016 20:44:53 +0000 (12:44 -0800)]
IB/qib: Remove most uses of QIB_PERMISSIVE_LID and QIB_MULTICAST_LID_BASE

This patch removes most of the uses of QIB_PERMISSIBVE_LID and
QIB_MULTICAST_LID_BASE in favor of the recently added IB_* versions.
There are still minor uses in AH functions as well as the QIB_* defines
but those will be removed in a follow on patch.

Reviewed-by: Ira Weiny <ira.weiny@intel.com>
Reviewed-by: Mike Marciniszyn <mike.marciniszyn@intel.com>
Signed-off-by: Dennis Dalessandro <dennis.dalessandro@intel.com>
Signed-off-by: Doug Ledford <dledford@redhat.com>
8 years agoIB/qib: Use rdmavt protection domain
Dennis Dalessandro [Fri, 22 Jan 2016 20:44:44 +0000 (12:44 -0800)]
IB/qib: Use rdmavt protection domain

Remove protection domain datastructure from qib and use rdmavts version.

Reviewed-by: Ira Weiny <ira.weiny@intel.com>
Reviewed-by: Mike Marciniszyn <mike.marciniszyn@intel.com>
Signed-off-by: Dennis Dalessandro <dennis.dalessandro@intel.com>
Signed-off-by: Doug Ledford <dledford@redhat.com>
8 years agoIB/qib: Remove dma.c and use rdmavt version of dma functions
Dennis Dalessandro [Fri, 22 Jan 2016 20:44:36 +0000 (12:44 -0800)]
IB/qib: Remove dma.c and use rdmavt version of dma functions

This patch removes the qib_dma.c file and uses the version which has been
added to rdmavt.

Reviewed-by: Ira Weiny <ira.weiny@intel.com>
Reviewed-by: Mike Marciniszyn <mike.marciniszyn@intel.com>
Signed-off-by: Dennis Dalessandro <dennis.dalessandro@intel.com>
Signed-off-by: Doug Ledford <dledford@redhat.com>
8 years agoIB/qib: Begin to use rdmavt for verbs
Dennis Dalessandro [Fri, 22 Jan 2016 20:44:29 +0000 (12:44 -0800)]
IB/qib: Begin to use rdmavt for verbs

This patch begins to make use of rdmavt by registering with it and
providing access to the header files. This is just the beginning of
rdmavt support in qib.

Most functionality is still being done in the driver, set flags so that
rdmavt will let qib continue to handle mr, qp, and cq init.

Reviewed-by: Ira Weiny <ira.weiny@intel.com>
Reviewed-by: Mike Marciniszyn <mike.marciniszyn@intel.com>
Signed-off-by: Dennis Dalessandro <dennis.dalessandro@intel.com>
Signed-off-by: Doug Ledford <dledford@redhat.com>
8 years agoIB/rdmavt: Add pkey support
Dennis Dalessandro [Wed, 6 Jan 2016 18:05:12 +0000 (10:05 -0800)]
IB/rdmavt: Add pkey support

Add pkey table in rdi per port data structure. Also bring in related pkey
functions. Drivers will still be responsible for allocating and
maintaining the pkey table. However they need to tell rdmavt where to find
the pkey table. We can not move the pkey table up into rdmavt because
drivers need to manipulate this long before registering with it.

Reviewed-by: Ira Weiny <ira.weiny@intel.com>
Reviewed-by: Mike Marciniszyn <mike.marciniszyn@intel.com>
Signed-off-by: Dennis Dalessandro <dennis.dalessandro@intel.com>
Signed-off-by: Doug Ledford <dledford@redhat.com>
8 years agoIB/rdmavt: Add mmap related functions
Dennis Dalessandro [Wed, 6 Jan 2016 18:04:57 +0000 (10:04 -0800)]
IB/rdmavt: Add mmap related functions

The mmap data structure was moved in a previous commit. This patch now
pulls in the related functions.

Reviewed-by: Ira Weiny <ira.weiny@intel.com>
Reviewed-by: Mike Marciniszyn <mike.marciniszyn@intel.com>
Signed-off-by: Dennis Dalessandro <dennis.dalessandro@intel.com>
Signed-off-by: Doug Ledford <dledford@redhat.com>
8 years agoIB/rdmavt: Initialize and teardown of qpn table
Dennis Dalessandro [Wed, 6 Jan 2016 18:04:46 +0000 (10:04 -0800)]
IB/rdmavt: Initialize and teardown of qpn table

Add table init as well as teardown for handling qpn maps. Drivers can still
provide this functionality by setting the QP_INIT_DRIVER bit.

Reviewed-by: Ira Weiny <ira.weiny@intel.com>
Reviewed-by: Mike Marciniszyn <mike.marciniszyn@intel.com>
Signed-off-by: Dennis Dalessandro <dennis.dalessandro@intel.com>
Signed-off-by: Doug Ledford <dledford@redhat.com>
8 years agoIB/rdmavt: Break rdma_vt main include header file up
Dennis Dalessandro [Wed, 6 Jan 2016 18:04:31 +0000 (10:04 -0800)]
IB/rdmavt: Break rdma_vt main include header file up

Until all functionality is moved over to rdmavt drivers still need to
access a number of fields in data structures that are predominantly
meant to be used by rdmavt. Once these rdmavt_<ibta_object>.h header
files are no longer being touched by drivers their content should be
moved to rdmavt/<ibta_object>.h. While here move a couple #defines
over to more general IB verbs header files because they fit better.

Reviewed-by: Ira Weiny <ira.weiny@intel.com>
Reviewed-by: Mike Marciniszyn <mike.marciniszyn@intel.com>
Signed-off-by: Dennis Dalessandro <dennis.dalessandro@intel.com>
Signed-off-by: Doug Ledford <dledford@redhat.com>
8 years agoIB/rdmavt: Add driver notification for new AH
Dennis Dalessandro [Wed, 6 Jan 2016 18:04:23 +0000 (10:04 -0800)]
IB/rdmavt: Add driver notification for new AH

Drivers may need to do some work once an address handle has been
created. Add a driver function for this purpose.

Reviewed-by: Ira Weiny <ira.weiny@intel.com>
Reviewed-by: Mike Marciniszyn <mike.marciniszyn@intel.com>
Reviewed-by: Harish Chegondi <harish.chegondi@intel.com>
Signed-off-by: Dennis Dalessandro <dennis.dalessandro@intel.com>
Signed-off-by: Doug Ledford <dledford@redhat.com>
8 years agoIB/rdmavt: Add an ibport data structure to rdmavt
Dennis Dalessandro [Wed, 6 Jan 2016 18:04:13 +0000 (10:04 -0800)]
IB/rdmavt: Add an ibport data structure to rdmavt

Converge the ibport data structures of qib and hfi1 into a common ib
port structure. Also provides a place to keep track of these ports
in case rdmavt needs it. Along with this goes an attach and detach
function for drivers to use to notify rdmavt of the ports.

Reviewed-by: Ira Weiny <ira.weiny@intel.com>
Reviewed-by: Harish Chegondi <harish.chegondi@intel.com>
Signed-off-by: Dennis Dalessandro <dennis.dalessandro@intel.com>
Signed-off-by: Doug Ledford <dledford@redhat.com>
8 years agoIB/rdmavt: Move SRQ data structure into rdmavt
Dennis Dalessandro [Wed, 6 Jan 2016 18:04:06 +0000 (10:04 -0800)]
IB/rdmavt: Move SRQ data structure into rdmavt

Patch moves the srq data structure into rdmavt in preparation for
removal from qib and hfi1 which will follow in subsequent patches.

Reviewed-by: Ira Weiny <ira.weiny@intel.com>
Reviewed-by: Harish Chegondi <harish.chegondi@intel.com>
Signed-off-by: Dennis Dalessandro <dennis.dalessandro@intel.com>
Signed-off-by: Doug Ledford <dledford@redhat.com>
8 years agoIB/rdmavt: Add AH to rdmavt
Kamal Heib [Wed, 6 Jan 2016 18:03:59 +0000 (10:03 -0800)]
IB/rdmavt: Add AH to rdmavt

Original patch is from Kamal Heib <kamalh@mellanox.com>. It has
been split into three separate patches. This one for rdmavt,
a follow on for qib, and one for hfi1.

Create datastructure for address handle and implement the
create/destroy/modify/query of address handle for rdmavt.

Reviewed-by: Ira Weiny <ira.weiny@intel.com>
Signed-off-by: Kamal Heib <kamalh@mellanox.com>
Signed-off-by: Dennis Dalessandro <dennis.dalessandro@intel.com>
Signed-off-by: Doug Ledford <dledford@redhat.com>
8 years agoIB/rdmavt: Add common LID defines to rdmavt
Kamal Heib [Wed, 6 Jan 2016 18:03:47 +0000 (10:03 -0800)]
IB/rdmavt: Add common LID defines to rdmavt

Original patch is from Kamal Heib <kamalh@mellanox.com>. It has
been split into separate patches.

This patch adds RVT_PERMISSIVE_LID and RVT_MULTICAST_LID_BASE
to rdmavt.

Reviewed-by: Ira Weiny <ira.weiny@intel.com>
Signed-off-by: Kamal Heib <kamalh@mellanox.com>
Signed-off-by: Dennis Dalessandro <dennis.dalessandro@intel.com>
Signed-off-by: Doug Ledford <dledford@redhat.com>
8 years agoIB/rdmavt: Do not use rvt prints which rely on driver too early
Dennis Dalessandro [Wed, 6 Jan 2016 18:03:39 +0000 (10:03 -0800)]
IB/rdmavt: Do not use rvt prints which rely on driver too early

Trying to print debug and error messages with the rdmavt helpers will not
work out so well if the drivers have not provided the get_card and get
pci functions. Use the normal pr_error instead until we can check this.

Reviewed-by: Ira Weiny <ira.weiny@intel.com>
Reviewed-by: Mike Marciniszyn <mike.marciniszyn@intel.com>
Signed-off-by: Dennis Dalessandro <dennis.dalessandro@intel.com>
Signed-off-by: Doug Ledford <dledford@redhat.com>
8 years agoIB/rdmavt: Move memory registration into rdmavt
Dennis Dalessandro [Wed, 6 Jan 2016 18:03:31 +0000 (10:03 -0800)]
IB/rdmavt: Move memory registration into rdmavt

Use the memory registration routines in hfi1 and move them to rdmavt.
A follow on patch will address removing the duplicated code in the
hfi1 and qib drivers.

Reviewed-by: Ira Weiny <ira.weiny@intel.com>
Reviewed-by: Mike Marciniszyn <mike.marciniszyn@intel.com>
Signed-off-by: Dennis Dalessandro <dennis.dalessandro@intel.com>
Signed-off-by: Doug Ledford <dledford@redhat.com>
8 years agoIB/rdmavt: Add the start of capability flags
Dennis Dalessandro [Wed, 6 Jan 2016 18:03:07 +0000 (10:03 -0800)]
IB/rdmavt: Add the start of capability flags

Drivers will need a set of flags to dictate behavior to rdmavt. This patch
adds a placeholder and a spot for it to live, as well as a few flags
that will be used.

Reviewed-by: Ira Weiny <ira.weiny@intel.com>
Reviewed-by: Mike Marciniszyn <mike.marciniszyn@intel.com>
Signed-off-by: Dennis Dalessandro <dennis.dalessandro@intel.com>
Signed-off-by: Doug Ledford <dledford@redhat.com>
8 years agoIB/rdmavt: Add device specific info prints
Dennis Dalessandro [Wed, 6 Jan 2016 18:02:59 +0000 (10:02 -0800)]
IB/rdmavt: Add device specific info prints

Follow hfi1's example for printing information about the driver and
incorporate into rdmavt. This requires two new functions to be
provided by the driver, one to get_card_name and one to get_pci_dev.

Reviewed-by: Mike Marciniszyn <mike.marciniszyn@intel.com>
Reviewed-by: Ira Weiny <ira.weiny@intel.com>
Signed-off-by: Dennis Dalessandro <dennis.dalessandro@intel.com>
Signed-off-by: Doug Ledford <dledford@redhat.com>
8 years agoIB/rdmavt: Move driver helper functions to a common structure
Dennis Dalessandro [Wed, 6 Jan 2016 18:02:52 +0000 (10:02 -0800)]
IB/rdmavt: Move driver helper functions to a common structure

Drivers are going to need to provide multiple functions for rdmavt to
call in to. We already have one, so go ahead and push this into a
data structure designated for driver supplied functions.

Reviewed-by: Mike Marciniszyn <mike.marciniszyn@intel.com>
Reviewed-by: Ira Weiny <ira.weiny@intel.com>
Signed-off-by: Dennis Dalessandro <dennis.dalessandro@intel.com>
Signed-off-by: Doug Ledford <dledford@redhat.com>
8 years agoIB/rdmavt: Add queue pair data structure to rdmavt
Dennis Dalessandro [Wed, 6 Jan 2016 18:02:41 +0000 (10:02 -0800)]
IB/rdmavt: Add queue pair data structure to rdmavt

Add queue pair data structure as well as supporting structures to rdmavt.

Reviewed-by: Ira Weiny <ira.weiny@intel.com>
Reviewed-by: Mike Marciniszyn <mike.marciniszyn@intel.com>
Signed-off-by: Dennis Dalessandro <dennis.dalessandro@intel.com>
Signed-off-by: Doug Ledford <dledford@redhat.com>
8 years agoIB/rdmavt: Move MR datastructures into rvt
Dennis Dalessandro [Wed, 6 Jan 2016 18:01:42 +0000 (10:01 -0800)]
IB/rdmavt: Move MR datastructures into rvt

This patch adds the MR datastructures based on hfi1 into rvt. For now the
data structures are defined in include/rdma/rdma_vt.h but once all MR
functionality has been moved from the drivers into rvt these should move to
rdmavt/mr.h

Reviewed-by: Ira Weiny <ira.weiny@intel.com>
Reviewed-by: Dean Luick <dean.luick@intel.com>
Reviewed-by: Mike Marciniszyn <mike.marciniszyn@intel.com>
Signed-off-by: Dennis Dalessandro <dennis.dalessandro@intel.com>
Signed-off-by: Doug Ledford <dledford@redhat.com>
8 years agoIB/rdmavt: Add post send and recv stubs
Dennis Dalessandro [Wed, 6 Jan 2016 18:01:17 +0000 (10:01 -0800)]
IB/rdmavt: Add post send and recv stubs

This adds the post sned and recv function stubs.

Reviewed-by: Mike Marciniszyn <mike.marciniszyn@intel.com>
Reviewed-by: Ira Weiny <ira.weiny@intel.com>
Signed-off-by: Dennis Dalessandro <dennis.dalessandro@intel.com>
Signed-off-by: Doug Ledford <dledford@redhat.com>
8 years agoIB/rdmavt: Add completion queue function stubs
Dennis Dalessandro [Wed, 6 Jan 2016 18:00:42 +0000 (10:00 -0800)]
IB/rdmavt: Add completion queue function stubs

Create stubs for completion queue creation, polling,
resizing, calling for notification, and destroying.

Reviewed-by: Mike Marciniszyn <mike.marciniszyn@intel.com>
Reviewed-by: Ira Weiny <ira.weiny@intel.com>
Signed-off-by: Dennis Dalessandro <dennis.dalessandro@intel.com>
Signed-off-by: Doug Ledford <dledford@redhat.com>
8 years agoIB/rdmavt: Add get port immutable stub
Dennis Dalessandro [Wed, 6 Jan 2016 17:59:38 +0000 (09:59 -0800)]
IB/rdmavt: Add get port immutable stub

This adds the get port immutable verbs call.

Reviewed-by: Mike Marciniszyn <mike.marciniszyn@intel.com>
Reviewed-by: Ira Weiny <ira.weiny@intel.com>
Signed-off-by: Dennis Dalessandro <dennis.dalessandro@intel.com>
Signed-off-by: Doug Ledford <dledford@redhat.com>
8 years agoIB/rdmavt: Add mmap stub
Dennis Dalessandro [Wed, 6 Jan 2016 17:59:04 +0000 (09:59 -0800)]
IB/rdmavt: Add mmap stub

Adds the stub for the mmap verbs call.

Reviewed-by: Mike Marciniszyn <mike.marciniszyn@intel.com>
Reviewed-by: Ira Weiny <ira.weiny@intel.com>
Signed-off-by: Dennis Dalessandro <dennis.dalessandro@intel.com>
Signed-off-by: Doug Ledford <dledford@redhat.com>
8 years agoIB/rdmavt: Add process MAD stub
Dennis Dalessandro [Wed, 6 Jan 2016 17:58:37 +0000 (09:58 -0800)]
IB/rdmavt: Add process MAD stub

This adds the stub for process mad. More study is needed to determine the
final MAD interaction between the driver and rvt.

Reviewed-by: Mike Marciniszyn <mike.marciniszyn@intel.com>
Reviewed-by: Ira Weiny <ira.weiny@intel.com>
Signed-off-by: Dennis Dalessandro <dennis.dalessandro@intel.com>
Signed-off-by: Doug Ledford <dledford@redhat.com>
8 years agoIB/rdmavt: Add multicast stubs
Dennis Dalessandro [Wed, 6 Jan 2016 17:58:23 +0000 (09:58 -0800)]
IB/rdmavt: Add multicast stubs

Adds the function stubs for attach and detach multicast.

Reviewed-by: Mike Marciniszyn <mike.marciniszyn@intel.com>
Reviewed-by: Ira Weiny <ira.weiny@intel.com>
Signed-off-by: Dennis Dalessandro <dennis.dalessandro@intel.com>
Signed-off-by: Doug Ledford <dledford@redhat.com>
8 years agoIB/rdmavt: Add SRQ stubs
Dennis Dalessandro [Wed, 6 Jan 2016 17:57:58 +0000 (09:57 -0800)]
IB/rdmavt: Add SRQ stubs

Adds the stubs for create, modify, query, and destory for shared
request queues.

Reviewed-by: Mike Marciniszyn <mike.marciniszyn@intel.com>
Reviewed-by: Ira Weiny <ira.weiny@intel.com>
Signed-off-by: Dennis Dalessandro <dennis.dalessandro@intel.com>
Signed-off-by: Doug Ledford <dledford@redhat.com>
8 years agoIB/rdmavt: Add memory region stubs
Dennis Dalessandro [Wed, 6 Jan 2016 17:57:21 +0000 (09:57 -0800)]
IB/rdmavt: Add memory region stubs

Adds the function stubs for allocating, and registering memory regions, as
well as deregistering them.

Reviewed-by: Mike Marciniszyn <mike.marciniszyn@intel.com>
Reviewed-by: Ira Weiny <ira.weiny@intel.com>
Signed-off-by: Dennis Dalessandro <dennis.dalessandro@intel.com>
Signed-off-by: Doug Ledford <dledford@redhat.com>
8 years agoIB/rdmavt: Add address handle stubs
Dennis Dalessandro [Wed, 6 Jan 2016 17:56:41 +0000 (09:56 -0800)]
IB/rdmavt: Add address handle stubs

Adds the stubs for create, destroy, modify, and query of the
address handle.

Reviewed-by: Mike Marciniszyn <mike.marciniszyn@intel.com>
Reviewed-by: Ira Weiny <ira.weiny@intel.com>
Signed-off-by: Dennis Dalessandro <dennis.dalessandro@intel.com>
Signed-off-by: Doug Ledford <dledford@redhat.com>
8 years agoIB/rdmavt: Add queue pair function stubs
Dennis Dalessandro [Wed, 6 Jan 2016 17:56:15 +0000 (09:56 -0800)]
IB/rdmavt: Add queue pair function stubs

Adds the stubs for create, modify, destroy and query functions for queue
pairs.

Reviewed-by: Mike Marciniszyn <mike.marciniszyn@intel.com>
Reviewed-by: Ira Weiny <ira.weiny@intel.com>
Signed-off-by: Dennis Dalessandro <dennis.dalessandro@intel.com>
Signed-off-by: Doug Ledford <dledford@redhat.com>
8 years agoIB/rdmavt: Alloc and dealloc ucontexts
Dennis Dalessandro [Wed, 6 Jan 2016 17:55:39 +0000 (09:55 -0800)]
IB/rdmavt: Alloc and dealloc ucontexts

Add the stubs to allocate and deallocate user contexts. This will be
handled completely by rvt.

Reviewed-by: Mike Marciniszyn <mike.marciniszyn@intel.com>
Reviewed-by: Ira Weiny <ira.weiny@intel.com>
Signed-off-by: Dennis Dalessandro <dennis.dalessandro@intel.com>
Signed-off-by: Doug Ledford <dledford@redhat.com>
8 years agoIB/rdmavt: Add query gid stub
Dennis Dalessandro [Wed, 6 Jan 2016 17:54:50 +0000 (09:54 -0800)]
IB/rdmavt: Add query gid stub

The handler for query gid operates along the same lines as the query pkey
handler. The driver will take care to keep the guid table updated.

Reviewed-by: Mike Marciniszyn <mike.marciniszyn@intel.com>
Reviewed-by: Ira Weiny <ira.weiny@intel.com>
Signed-off-by: Dennis Dalessandro <dennis.dalessandro@intel.com>
Signed-off-by: Doug Ledford <dledford@redhat.com>
8 years agoIB/rdmavt: Add pkey query stub
Dennis Dalessandro [Wed, 6 Jan 2016 17:54:16 +0000 (09:54 -0800)]
IB/rdmavt: Add pkey query stub

The pkey table will reside in the rvt structure but it will be modified
only when the driver requests then rvt will simply read the value to return
in the query.

Reviewed-by: Mike Marciniszyn <mike.marciniszyn@intel.com>
Reviewed-by: Ira Weiny <ira.weiny@intel.com>
Signed-off-by: Dennis Dalessandro <dennis.dalessandro@intel.com>
Signed-off-by: Doug Ledford <dledford@redhat.com>
8 years agoIB/rdmavt: Add query and modify port stubs
Dennis Dalessandro [Wed, 6 Jan 2016 17:54:07 +0000 (09:54 -0800)]
IB/rdmavt: Add query and modify port stubs

This patch adds the query and modify port stubs. The query will mostly
entail the driver returning everything in the ib_port_attr which will get
handed back to the verbs layer. The modify will need some API helpers in
the driver. The send_trap and post_mad_send are still issues to address.

Reviewed-by: Mike Marciniszyn <mike.marciniszyn@intel.com>
Reviewed-by: Ira Weiny <ira.weiny@intel.com>
Signed-off-by: Dennis Dalessandro <dennis.dalessandro@intel.com>
Signed-off-by: Doug Ledford <dledford@redhat.com>
8 years agoIB/rdmavt: Add query and modify device stubs
Dennis Dalessandro [Wed, 6 Jan 2016 17:53:05 +0000 (09:53 -0800)]
IB/rdmavt: Add query and modify device stubs

Adds the stubs which will handle the query and modify device functions. At
this time the only intention is to support changing the node desc and the
guid via these calls.

Reviewed-by: Mike Marciniszyn <mike.marciniszyn@intel.com>
Reviewed-by: Ira Weiny <ira.weiny@intel.com>
Signed-off-by: Dennis Dalessandro <dennis.dalessandro@intel.com>
Signed-off-by: Doug Ledford <dledford@redhat.com>
8 years agoIB/rdmavt: Macroize override checks during driver registration
Dennis Dalessandro [Wed, 6 Jan 2016 17:52:40 +0000 (09:52 -0800)]
IB/rdmavt: Macroize override checks during driver registration

Add a macro to cut down on writing the same lines over and over again for
what will be a large number of functions that will be supported.

Reviewed-by: Mike Marciniszyn <mike.marciniszyn@intel.com>
Reviewed-by: Ira Weiny <ira.weiny@intel.com>
Signed-off-by: Dennis Dalessandro <dennis.dalessandro@intel.com>
Signed-off-by: Doug Ledford <dledford@redhat.com>
8 years agoIB/rdmavt: Add ib core device attributes to rvt driver params list
Dennis Dalessandro [Wed, 6 Jan 2016 17:52:19 +0000 (09:52 -0800)]
IB/rdmavt: Add ib core device attributes to rvt driver params list

Instead of trying to handle each parameter separately, add ib_device_attr
to rvt_driver_params. This means drivers will fill this in and pass to the
rvt registration function.

Reviewed-by: Mike Marciniszyn <mike.marciniszyn@intel.com>
Reviewed-by: Ira Weiny <ira.weiny@intel.com>
Signed-off-by: Dennis Dalessandro <dennis.dalessandro@intel.com>
Signed-off-by: Doug Ledford <dledford@redhat.com>
8 years agoIB/rdmavt: Add protection domain to rdmavt.
Dennis Dalessandro [Wed, 6 Jan 2016 17:51:48 +0000 (09:51 -0800)]
IB/rdmavt: Add protection domain to rdmavt.

Add datastructure for and allocation/deallocation of protection domains for
RDMAVT.

Reviewed-by: Ira Weiny <ira.weiny@intel.com>
Reviewed-by: Mike Marciniszyn <mike.marciniszyn@intel.com>
Signed-off-by: Dennis Dalessandro <dennis.dalessandro@intel.com>
Signed-off-by: Doug Ledford <dledford@redhat.com>
8 years agoIB/rdmavt: Consolidate dma ops in rdmavt.
Dennis Dalessandro [Wed, 6 Jan 2016 17:51:18 +0000 (09:51 -0800)]
IB/rdmavt: Consolidate dma ops in rdmavt.

This patch adds dma functions to rdmavt. The source is hfi1's version of
dma.c which will be removed by a subsequent hfi1 patch.

Reviewed-by: Ira Weiny <ira.weiny@intel.com>
Reviewed-by: Mike Marciniszyn <mike.marciniszyn@intel.com>
Signed-off-by: Dennis Dalessandro <dennis.dalessandro@intel.com>
Signed-off-by: Doug Ledford <dledford@redhat.com>
8 years agoIB/rdmavt: Create module framework and handle driver registration
Dennis Dalessandro [Wed, 6 Jan 2016 17:50:24 +0000 (09:50 -0800)]
IB/rdmavt: Create module framework and handle driver registration

This patch introduces the basics for a new module called rdma_vt. This new
driver is a software implementation of the InfiniBand verbs and aims to
replace the multiple implementations that exist and duplicate each others'
code.

While the call to actually register the device with the IB core happens in
rdma_vt, most of the work is still done in the drivers themselves. This
will be changing in a follow on patch this is just laying the groundwork
for this infrastructure.

Reviewed-by: Ira Weiny <ira.weiny@intel.com>
Reviewed-by: Mike Marciniszyn <mike.marciniszyn@intel.com>
Signed-off-by: Dennis Dalessandro <dennis.dalessandro@intel.com>
Signed-off-by: Doug Ledford <dledford@redhat.com>
8 years agostaging/hfi1: check for ARMED->ACTIVE change in recv int
Jim Snow [Mon, 11 Jan 2016 23:32:21 +0000 (18:32 -0500)]
staging/hfi1: check for ARMED->ACTIVE change in recv int

The link state will transition from ARMED to ACTIVE when a non-SC15
packet arrives, but the driver might not notice the change.  With this
fix, if the slowpath receive interrupt handler sees a non-SC15 packet
while in the ARMED state, we queue work to call linkstate_active_work
from process context to promote it to ACTIVE.

Reviewed-by: Dean Luick <dean.luick@intel.com>
Reviewed-by: Ira Weiny <ira.weiny@intel.com>
Reviewed-by: Mike Marciniszyn <mike.marciniszyn@intel.com>
Signed-off-by: Jim Snow <jim.m.snow@intel.com>
Signed-off-by: Brendan Cunningham <brendan.cunningham@intel.com>
Signed-off-by: Jubin John <jubin.john@intel.com>
Signed-off-by: Doug Ledford <dledford@redhat.com>
8 years agostaging/hfi1: Fix Xmit Wait calculation
Ira Weiny [Mon, 11 Jan 2016 23:31:44 +0000 (18:31 -0500)]
staging/hfi1: Fix Xmit Wait calculation

Total XMIT wait needs to sum the xmit wait values of all the VLs not just
those requested in the query.  Also, make the algorithm used for both
PortStatus and PortDataCounters the same.

Reviewed-by: Arthur Kepner <arthur.kepner@intel.com>
Reviewed-by: Breyer, Scott J <scott.j.breyer@intel.com>
Signed-off-by: Ira Weiny <iweiny@gmail.com>
Signed-off-by: Jubin John <jubin.john@intel.com>
Signed-off-by: Doug Ledford <dledford@redhat.com>
8 years agostaging/hfi1: Clean up comments
Edward Mascarenhas [Mon, 11 Jan 2016 23:31:43 +0000 (18:31 -0500)]
staging/hfi1: Clean up comments

Clean up comments by deleting numbering and terms internal to Intel.

The information on the actual bugs is not deleted.

Reviewed-by: Mike Marciniszyn <mike.marciniszyn@intel.com>
Signed-off-by: Edward Mascarenhas <edward.mascarenhas@intel.com>
Signed-off-by: Jubin John <jubin.john@intel.com>
Signed-off-by: Doug Ledford <dledford@redhat.com>
8 years agostaging/hfi1: Remove unneeded variable index
Dean Luick [Mon, 11 Jan 2016 23:30:57 +0000 (18:30 -0500)]
staging/hfi1: Remove unneeded variable index

The variable "index" increments the same as dd->ndevcntrs.
Just use the later.  Remove uneeded usage of "index" in the
fill loop - it is not used there or later in the function.

Reviewed-by: Dennis Dalessandro <dennis.dalessandro@intel.com>
Signed-off-by: Dean Luick <dean.luick@intel.com>
Signed-off-by: Jubin John <jubin.john@intel.com>
Signed-off-by: Doug Ledford <dledford@redhat.com>
8 years agostaging/hfi1: add per SDMA engine stats to hfistats
Vennila Megavannan [Mon, 11 Jan 2016 23:30:56 +0000 (18:30 -0500)]
staging/hfi1: add per SDMA engine stats to hfistats

Added the following per sdma engine stats:
  - SendDmaDescFetchedCnt
  - software maintained count of SDMA interrupts
 (SDmaInt, SDmaIdleInt, SDmaProgressInt)
  - software maintained counts of SDMA error cases

Reviewed-by: Dennis Dalessandro <dennis.dalessandro@intel.com>
Signed-off-by: Mike Marciniszyn <mike.marciniszyn@intel.com>
Signed-off-by: Vennila Megavannan <vennila.megavannan@intel.com>
Signed-off-by: Jubin John <jubin.john@intel.com>
Signed-off-by: Doug Ledford <dledford@redhat.com>
8 years agostaging/hfi1: Change default krcvqs
jubin.john@intel.com [Mon, 11 Jan 2016 23:30:55 +0000 (18:30 -0500)]
staging/hfi1: Change default krcvqs

Change the default number of krcvqs to number of numa nodes + 1
based on the performance data collected.

Reviewed-by: Mike Marciniszyn <mike.marciniszyn@intel.com>
Signed-off-by: Jubin John <jubin.john@intel.com>
Signed-off-by: Doug Ledford <dledford@redhat.com>
8 years agostaging/hfi1: change krcvqs mod param from byte to uint
Mark F. Brown [Mon, 11 Jan 2016 23:30:54 +0000 (18:30 -0500)]
staging/hfi1: change krcvqs mod param from byte to uint

The krcvqs parameter is displayed incorrectly in sysfs.
The workaround is to set the param type as uint.

Reviewed-by: Mike Marciniszyn <mike.marciniszyn@intel.com>
Reviewed-by: Mitko Haralanov <mitko.haralanov@intel.com>
Signed-off-by: Mark F. Brown <mark.f.brown@intel.com>
Signed-off-by: Jubin John <jubin.john@intel.com>
Signed-off-by: Doug Ledford <dledford@redhat.com>
8 years agostaging/hfi1: Move s_sde to read mostly section of hfi1_qp
Harish Chegondi [Mon, 11 Jan 2016 23:30:53 +0000 (18:30 -0500)]
staging/hfi1: Move s_sde to read mostly section of hfi1_qp

This would reduce L2 cache misses on s_sde in the _hfi1_schedule_send
function when invoked from post_send thereby improving performance of
post_send.

Reviewed-by: Mike Marciniszyn <mike.marciniszyn@intel.com>
Signed-off-by: Harish Chegondi <harish.chegondi@intel.com>
Signed-off-by: Jubin John <jubin.john@intel.com>
Signed-off-by: Doug Ledford <dledford@redhat.com>
8 years agostaging/hfi1: Use BIT macro
jubin.john@intel.com [Mon, 11 Jan 2016 23:30:52 +0000 (18:30 -0500)]
staging/hfi1: Use BIT macro

This patch fixes the checkpatch issue:
CHECK: Prefer using the BIT macro

Use of BIT macro for HDRQ_INCREMENT in chip.h causes a change in
format specifier for error message in init.c in order to avoid a
build warning.

Reviewed-by: Dean Luick <dean.luick@intel.com>
Reviewed-by: Ira Weiny <ira.weiny@intel.com>
Reviewed-by: Mike Marciniszyn <mike.marciniszyn@intel.com>
Signed-off-by: Jubin John <jubin.john@intel.com>
Signed-off-by: Doug Ledford <dledford@redhat.com>
8 years agostaging/hfi1: set Gen3 half-swing for integrated devices
Dean Luick [Mon, 11 Jan 2016 18:04:33 +0000 (13:04 -0500)]
staging/hfi1: set Gen3 half-swing for integrated devices

Correctly set half-swing for integrated devices.  A0 needs all fields set for
CcePcieCtrl.  B0 and later only need a few fields set.

Reviewed-by: Stuart Summers <john.s.summers@intel.com>
Signed-off-by: Dean Luick <dean.luick@intel.com>
Signed-off-by: Ira Weiny <ira.weiny@intel.com>
Signed-off-by: Doug Ledford <dledford@redhat.com>
8 years agostaging/hfi1: add dd_dev_dbg
Ira Weiny [Mon, 11 Jan 2016 18:04:32 +0000 (13:04 -0500)]
staging/hfi1: add dd_dev_dbg

To be used in future patches add dd_dev_dbg.  dd_* functions properly decode
the hfi1_devdata structure used throughout the driver

Signed-off-by: Ira Weiny <ira.weiny@intel.com>
Signed-off-by: Doug Ledford <dledford@redhat.com>
8 years agostaging/hfi1: Enable TID caching feature
Mitko Haralanov [Fri, 5 Feb 2016 16:57:58 +0000 (11:57 -0500)]
staging/hfi1: Enable TID caching feature

This commit "flips the switch" on the TID caching feature
implemented in this patch series.

As well as enabling the new feature by tying the new function
with the PSM API, it also cleans up the old unneeded code,
data structure members, and variables.

Due to difference in operation and information, the tracing
functions related to expected receives had to be changed. This
patch include these changes.

The tracing function changes could not be split into a separate
commit without including both tracing variants at the same time.
This would have caused other complications and ugliness.

Signed-off-by: Mitko Haralanov <mitko.haralanov@intel.com>
Reviewed-by: Ira Weiny <ira.weiny@intel.com>
Signed-off-by: Doug Ledford <dledford@redhat.com>
8 years agostaging/hfi1: Add TID entry program function body
Mitko Haralanov [Fri, 5 Feb 2016 16:57:57 +0000 (11:57 -0500)]
staging/hfi1: Add TID entry program function body

The previous patch in the series added the free/invalidate
function bodies. Now, it's time for the programming side.

This large function takes the user's buffer, breaks it up
into manageable chunks, allocates enough RcvArray groups
and programs the chunks into the RcvArray entries in the
hardware.

With this function, the TID caching functionality is implemented.
However, it is still unused. The switch will come in a later
patch in the series, which will remove the old functionality and
switch the driver over to TID caching.

Signed-off-by: Mitko Haralanov <mitko.haralanov@intel.com>
Reviewed-by: Ira Weiny <ira.weiny@intel.com>
Signed-off-by: Doug Ledford <dledford@redhat.com>
8 years agostaging/hfi1: Add TID free/clear function bodies
Mitko Haralanov [Fri, 5 Feb 2016 16:57:56 +0000 (11:57 -0500)]
staging/hfi1: Add TID free/clear function bodies

Up to now, the functions which cleared the programmed
TID entries and gave PSM the list of invalidated TID entries
were just stubs. With this commit, the bodies of these
functions are added.

This commit is a bit asymmetric as it only contains the
free code path. This is done on purpose to help with patch
reviews as the programming code path is much longer.

Signed-off-by: Mitko Haralanov <mitko.haralanov@intel.com>
Reviewed-by: Ira Weiny <ira.weiny@intel.com>
Signed-off-by: Doug Ledford <dledford@redhat.com>
8 years agostaging/hfi1: Add MMU notifier callback function
Mitko Haralanov [Fri, 5 Feb 2016 16:57:55 +0000 (11:57 -0500)]
staging/hfi1: Add MMU notifier callback function

TID caching will rely on the MMU notifier to be told
when memory is being invalidated. When the callback
is called, the driver will find all RcvArray entries
that span the invalidated buffer and "schedule" them
to be freed by the PSM library.

This function is currently unused and is being added
in preparation for the TID caching feature.

Signed-off-by: Mitko Haralanov <mitko.haralanov@intel.com>
Reviewed-by: Ira Weiny <ira.weiny@intel.com>
Signed-off-by: Doug Ledford <dledford@redhat.com>
8 years agostaging/hfi1: Add TID cache receive init and free funcs
Mitko Haralanov [Fri, 5 Feb 2016 16:57:54 +0000 (11:57 -0500)]
staging/hfi1: Add TID cache receive init and free funcs

The upcoming TID caching feature requires different data
structures and, by extension, different initialization for each
of the MPI processes.

The two new functions (currently unused) perform the required
initialization and freeing of required resources and structures.

Signed-off-by: Mitko Haralanov <mitko.haralanov@intel.com>
Reviewed-by: Ira Weiny <ira.weiny@intel.com>
Signed-off-by: Doug Ledford <dledford@redhat.com>
8 years agostaging/hfi1: Convert lock to mutex
Mitko Haralanov [Fri, 5 Feb 2016 16:57:53 +0000 (11:57 -0500)]
staging/hfi1: Convert lock to mutex

The exp_lock lock does not need to be a spinlock as
all its uses are in process context and allowing the
process to sleep when the mutex is contended might
be beneficial.

Signed-off-by: Mitko Haralanov <mitko.haralanov@intel.com>
Reviewed-by: Ira Weiny <ira.weiny@intel.com>
Signed-off-by: Doug Ledford <dledford@redhat.com>
8 years agostaging/hfi1: Add building blocks for TID caching
Mitko Haralanov [Fri, 5 Feb 2016 16:57:52 +0000 (11:57 -0500)]
staging/hfi1: Add building blocks for TID caching

Functions added by this patch are building blocks for the upcoming
TID caching functionality. The functions added are currently unsed
(and marked as such.)

The functions' purposes are to find physically contigous pages in
the user's virtual buffer, program the RcvArray group entries with
these physical chunks, and unprogram the RcvArray groups.

Reviewed-by: Ira Weiny <ira.weiny@intel.com>
Signed-off-by: Mitko Haralanov <mitko.haralanov@intel.com>
Signed-off-by: Doug Ledford <dledford@redhat.com>
8 years agostaging/hfi1: TID group definitions and support funcs
Mitko Haralanov [Fri, 5 Feb 2016 16:57:51 +0000 (11:57 -0500)]
staging/hfi1: TID group definitions and support funcs

Definitions and functions use to manage sets of TID/RcvArray groups.
These will be used by the TID cacheline functionality coming with
later patches.

TID groups (or RcvArray groups) are groups of TID/RcvArray entries
organized in sets of 8 and aligned on cacheline boundaries. The
TID/RcvArray entries are managed in this way to make taking
advantage of write-combining easier - each group is a entire
cacheline.

rcv_array_wc_fill() is provided to allow of generating writes to
TIDs which are not currently being used in order to cause the
flush of the write-combining buffer.

Signed-off-by: Mitko Haralanov <mitko.haralanov@intel.com>
Reviewed-by: Ira Weiny <ira.weiny@intel.com>
Signed-off-by: Doug Ledford <dledford@redhat.com>
8 years agostaging/hfi1: Remove un-needed variable
Mitko Haralanov [Fri, 5 Feb 2016 16:57:50 +0000 (11:57 -0500)]
staging/hfi1: Remove un-needed variable

There is no need to use a separate variable for a
return value and a label when returning right away
would do just as well.

Signed-off-by: Mitko Haralanov <mitko.haralanov@intel.com>
Reviewed-by: Ira Weiny <ira.weiny@intel.com>
Signed-off-by: Doug Ledford <dledford@redhat.com>
8 years agostaging/hfi1: Add definitions needed for TID cache
Mitko Haralanov [Fri, 5 Feb 2016 16:57:49 +0000 (11:57 -0500)]
staging/hfi1: Add definitions needed for TID cache

In preparation for adding the TID caching support, there is a set
of headers, structures, and variables which will be needed. This
commit adds them to the hfi.h header file.

Signed-off-by: Mitko Haralanov <mitko.haralanov@intel.com>
Reviewed-by: Ira Weiny <ira.weiny@intel.com>
Signed-off-by: Doug Ledford <dledford@redhat.com>
8 years agouapi/hfi1_user: Add command and event for TID caching
Mitko Haralanov [Fri, 5 Feb 2016 16:57:48 +0000 (11:57 -0500)]
uapi/hfi1_user: Add command and event for TID caching

TID caching will use a new event to signal userland that cache
invalidation has occurred and needs a matching command code that
will be used to read the invalidated TIDs.

Add the event bit and the new command to the exported header file.

The command is also added to the switch() statement in file_ops.c
for completeness and in preparation for its usage later.

Signed-off-by: Mitko Haralanov <mitko.haralanov@intel.com>
Reviewed-by: Ira Weiny <ira.weiny@intel.com>
Signed-off-by: Doug Ledford <dledford@redhat.com>
8 years agouapi/hfi1_user: Correct comment for capability bit
Mitko Haralanov [Fri, 5 Feb 2016 16:57:47 +0000 (11:57 -0500)]
uapi/hfi1_user: Correct comment for capability bit

The HFI1_CAP_TID_UNMAP comment was incorrectly implying the
opposite of what capability actually did. Correct this error.

Reviewed-by: Ira Weiny <ira.weiny@intel.com>
Signed-off-by: Mitko Haralanov <mitko.haralanov@intel.com>
Signed-off-by: Doug Ledford <dledford@redhat.com>
8 years agostaging/hfi1: Add function stubs for TID caching
Mitko Haralanov [Fri, 5 Feb 2016 16:57:46 +0000 (11:57 -0500)]
staging/hfi1: Add function stubs for TID caching

Add mmu notify helper functions and TID caching function
stubs in preparation for the TID caching implementation.

TID caching makes use of the MMU notifier to allow the driver
to respond to the user freeing memory which is allocated to
the HFI.

This patch implements the basic MMU notifier functions to insert,
find and remove buffer pages from memory based on the mmu_notifier
being invoked.

In addition it places stubs in place for the main entry points by
follow on code.

Follow up patches will complete the implementation of the interaction
with user space and makes use of these functions.

Signed-off-by: Mitko Haralanov <mitko.haralanov@intel.com>
Signed-off-by: Ira Weiny <ira.weiny@intel.com>
Signed-off-by: Doug Ledford <dledford@redhat.com>
8 years agoLinux 4.5-rc6
Linus Torvalds [Sun, 28 Feb 2016 16:41:20 +0000 (08:41 -0800)]
Linux 4.5-rc6

8 years agoMerge branch 'perf-urgent-for-linus' of git://git.kernel.org/pub/scm/linux/kernel...
Linus Torvalds [Sun, 28 Feb 2016 15:52:00 +0000 (07:52 -0800)]
Merge branch 'perf-urgent-for-linus' of git://git./linux/kernel/git/tip/tip

Pull perf fixes from Thomas Gleixner:
 "A rather largish series of 12 patches addressing a maze of race
  conditions in the perf core code from Peter Zijlstra"

* 'perf-urgent-for-linus' of git://git.kernel.org/pub/scm/linux/kernel/git/tip/tip:
  perf: Robustify task_function_call()
  perf: Fix scaling vs. perf_install_in_context()
  perf: Fix scaling vs. perf_event_enable()
  perf: Fix scaling vs. perf_event_enable_on_exec()
  perf: Fix ctx time tracking by introducing EVENT_TIME
  perf: Cure event->pending_disable race
  perf: Fix race between event install and jump_labels
  perf: Fix cloning
  perf: Only update context time when active
  perf: Allow perf_release() with !event->ctx
  perf: Do not double free
  perf: Close install vs. exit race

8 years agoMerge branch 'x86-urgent-for-linus' of git://git.kernel.org/pub/scm/linux/kernel...
Linus Torvalds [Sun, 28 Feb 2016 15:49:23 +0000 (07:49 -0800)]
Merge branch 'x86-urgent-for-linus' of git://git./linux/kernel/git/tip/tip

Pull x86 fixes from Thomas Gleixner:
 "This update contains:

   - Hopefully the last ASM CLAC fixups

   - A fix for the Quark family related to the IMR lock which makes
     kexec work again

   - A off-by-one fix in the MPX code.  Ironic, isn't it?

   - A fix for X86_PAE which addresses once more an unsigned long vs
     phys_addr_t hickup"

* 'x86-urgent-for-linus' of git://git.kernel.org/pub/scm/linux/kernel/git/tip/tip:
  x86/mpx: Fix off-by-one comparison with nr_registers
  x86/mm: Fix slow_virt_to_phys() for X86_PAE again
  x86/entry/compat: Add missing CLAC to entry_INT80_32
  x86/entry/32: Add an ASM_CLAC to entry_SYSENTER_32
  x86/platform/intel/quark: Change the kernel's IMR lock bit to false

8 years agoMerge branch 'sched-urgent-for-linus' of git://git.kernel.org/pub/scm/linux/kernel...
Linus Torvalds [Sun, 28 Feb 2016 15:48:01 +0000 (07:48 -0800)]
Merge branch 'sched-urgent-for-linus' of git://git./linux/kernel/git/tip/tip

Pull scheduler fixlet from Thomas Gleixner:
 "A trivial printk typo fix"

* 'sched-urgent-for-linus' of git://git.kernel.org/pub/scm/linux/kernel/git/tip/tip:
  sched/deadline: Fix trivial typo in printk() message

8 years agoMerge branch 'irq-urgent-for-linus' of git://git.kernel.org/pub/scm/linux/kernel...
Linus Torvalds [Sun, 28 Feb 2016 15:45:58 +0000 (07:45 -0800)]
Merge branch 'irq-urgent-for-linus' of git://git./linux/kernel/git/tip/tip

Pull irq fixes from Thomas Gleixner:
 "Four small fixes for irqchip drivers:

   - Add missing low level irq handler initialization on mxs, so
     interrupts can acutally be delivered

   - Add a missing barrier to the GIC driver

   - Two fixes for the GIC-V3-ITS driver, addressing a double EOI write
     and a cache flush beyond the actual region"

* 'irq-urgent-for-linus' of git://git.kernel.org/pub/scm/linux/kernel/git/tip/tip:
  irqchip/gic-v3: Add missing barrier to 32bit version of gic_read_iar()
  irqchip/mxs: Add missing set_handle_irq()
  irqchip/gicv3-its: Avoid cache flush beyond ITS_BASERn memory size
  irqchip/gic-v3-its: Fix double ICC_EOIR write for LPI in EOImode==1

8 years agoMerge tag 'staging-4.5-rc6' of git://git.kernel.org/pub/scm/linux/kernel/git/gregkh...
Linus Torvalds [Sun, 28 Feb 2016 15:39:15 +0000 (07:39 -0800)]
Merge tag 'staging-4.5-rc6' of git://git./linux/kernel/git/gregkh/staging

Pull staging/android fix from Greg KH:
 "Here is one patch, for the android binder driver, to resolve a
  reported problem.  Turns out it has been around for a while (since
  3.15), so it is good to finally get it resolved.

  It has been in linux-next for a while with no reported issues"

* tag 'staging-4.5-rc6' of git://git.kernel.org/pub/scm/linux/kernel/git/gregkh/staging:
  drivers: android: correct the size of struct binder_uintptr_t for BC_DEAD_BINDER_DONE

8 years agoMerge tag 'usb-4.5-rc6' of git://git.kernel.org/pub/scm/linux/kernel/git/gregkh/usb
Linus Torvalds [Sun, 28 Feb 2016 15:37:30 +0000 (07:37 -0800)]
Merge tag 'usb-4.5-rc6' of git://git./linux/kernel/git/gregkh/usb

Pull USB fixes from Greg KH:
 "Here are a few USB fixes for 4.5-rc6

  They fix a reported bug for some USB 3 devices by reverting the recent
  patch, a MAINTAINERS change for some drivers, some new device ids, and
  of course, the usual bunch of USB gadget driver fixes.

  All have been in linux-next for a while with no reported issues"

* tag 'usb-4.5-rc6' of git://git.kernel.org/pub/scm/linux/kernel/git/gregkh/usb:
  MAINTAINERS: drop OMAP USB and MUSB maintainership
  usb: musb: fix DMA for host mode
  usb: phy: msm: Trigger USB state detection work in DRD mode
  usb: gadget: net2280: fix endpoint max packet for super speed connections
  usb: gadget: gadgetfs: unregister gadget only if it got successfully registered
  usb: gadget: remove driver from pending list on probe error
  Revert "usb: hub: do not clear BOS field during reset device"
  usb: chipidea: fix return value check in ci_hdrc_pci_probe()
  usb: chipidea: error on overflow for port_test_write
  USB: option: add "4G LTE usb-modem U901"
  USB: cp210x: add IDs for GE B650V3 and B850V3 boards
  USB: option: add support for SIM7100E
  usb: musb: Fix DMA desired mode for Mentor DMA engine
  usb: gadget: fsl_qe_udc: fix IS_ERR_VALUE usage
  usb: dwc2: USB_DWC2 should depend on HAS_DMA
  usb: dwc2: host: fix the data toggle error in full speed descriptor dma
  usb: dwc2: host: fix logical omissions in dwc2_process_non_isoc_desc
  usb: dwc3: Fix assignment of EP transfer resources
  usb: dwc2: Add extra delay when forcing dr_mode

8 years agoMerge branch 'for-linus' of git://git.kernel.org/pub/scm/linux/kernel/git/viro/vfs
Linus Torvalds [Sun, 28 Feb 2016 01:10:32 +0000 (17:10 -0800)]
Merge branch 'for-linus' of git://git./linux/kernel/git/viro/vfs

Pull vfs fixes from Al Viro.

* 'for-linus' of git://git.kernel.org/pub/scm/linux/kernel/git/viro/vfs:
  do_last(): ELOOP failure exit should be done after leaving RCU mode
  should_follow_link(): validate ->d_seq after having decided to follow
  namei: ->d_inode of a pinned dentry is stable only for positives
  do_last(): don't let a bogus return value from ->open() et.al. to confuse us
  fs: return -EOPNOTSUPP if clone is not supported
  hpfs: don't truncate the file when delete fails

8 years agoMerge tag 'armsoc-fixes' of git://git.kernel.org/pub/scm/linux/kernel/git/arm/arm-soc
Linus Torvalds [Sun, 28 Feb 2016 00:58:32 +0000 (16:58 -0800)]
Merge tag 'armsoc-fixes' of git://git./linux/kernel/git/arm/arm-soc

Pull ARM SoC fixes from Olof Johansson:
 "We didn't have a batch last week, so this one is slightly larger.

  None of them are scary though, a handful of fixes for small DT pieces,
  replacing properties with newer conventions.

  Highlights:
   - N900 fix for setting system revision
   - onenand init fix to avoid filesystem corruption
   - Clock fix for audio on Beaglebone-x15
   - Fixes on shmobile to deal with CONFIG_DEBUG_RODATA (default y in 4.6)

  + misc smaller stuff"

* tag 'armsoc-fixes' of git://git.kernel.org/pub/scm/linux/kernel/git/arm/arm-soc:
  MAINTAINERS: Extend info, add wiki and ml for meson arch
  MAINTAINERS: alpine: add a new maintainer and update the entry
  ARM: at91/dt: fix typo in sama5d2 pinmux descriptions
  ARM: OMAP2+: Fix onenand initialization to avoid filesystem corruption
  Revert "regulator: tps65217: remove tps65217.dtsi file"
  ARM: shmobile: Remove shmobile_boot_arg
  ARM: shmobile: Move shmobile_smp_{mpidr, fn, arg}[] from .text to .bss
  ARM: shmobile: r8a7779: Remove remainings of removed SCU boot setup code
  ARM: shmobile: Move shmobile_scu_base from .text to .bss
  ARM: OMAP2+: Fix omap_device for module reload on PM runtime forbid
  ARM: OMAP2+: Improve omap_device error for driver writers
  ARM: DTS: am57xx-beagle-x15: Select SYS_CLK2 for audio clocks
  ARM: dts: am335x/am57xx: replace gpio-key,wakeup with wakeup-source property
  ARM: OMAP2+: Set system_rev from ATAGS for n900
  ARM: dts: orion5x: fix the missing mtd flash on linkstation lswtgl
  ARM: dts: kirkwood: use unique machine name for ds112
  ARM: dts: imx6: remove bogus interrupt-parent from CAAM node

8 years agodo_last(): ELOOP failure exit should be done after leaving RCU mode
Al Viro [Sun, 28 Feb 2016 00:37:37 +0000 (19:37 -0500)]
do_last(): ELOOP failure exit should be done after leaving RCU mode

... or we risk seeing a bogus value of d_is_symlink() there.

Cc: stable@vger.kernel.org # v4.2+
Signed-off-by: Al Viro <viro@zeniv.linux.org.uk>
8 years agoshould_follow_link(): validate ->d_seq after having decided to follow
Al Viro [Sun, 28 Feb 2016 00:31:01 +0000 (19:31 -0500)]
should_follow_link(): validate ->d_seq after having decided to follow

... otherwise d_is_symlink() above might have nothing to do with
the inode value we've got.

Cc: stable@vger.kernel.org # v4.2+
Signed-off-by: Al Viro <viro@zeniv.linux.org.uk>
8 years agonamei: ->d_inode of a pinned dentry is stable only for positives
Al Viro [Sun, 28 Feb 2016 00:23:16 +0000 (19:23 -0500)]
namei: ->d_inode of a pinned dentry is stable only for positives

both do_last() and walk_component() risk picking a NULL inode out
of dentry about to become positive, *then* checking its flags and
seeing that it's not negative anymore and using (already stale by
then) value they'd fetched earlier.  Usually ends up oopsing soon
after that...

Cc: stable@vger.kernel.org # v3.13+
Signed-off-by: Al Viro <viro@zeniv.linux.org.uk>
8 years agodo_last(): don't let a bogus return value from ->open() et.al. to confuse us
Al Viro [Sun, 28 Feb 2016 00:17:33 +0000 (19:17 -0500)]
do_last(): don't let a bogus return value from ->open() et.al. to confuse us

... into returning a positive to path_openat(), which would interpret that
as "symlink had been encountered" and proceed to corrupt memory, etc.
It can only happen due to a bug in some ->open() instance or in some LSM
hook, etc., so we report any such event *and* make sure it doesn't trick
us into further unpleasantness.

Cc: stable@vger.kernel.org # v3.6+, at least
Signed-off-by: Al Viro <viro@zeniv.linux.org.uk>
8 years agofs: return -EOPNOTSUPP if clone is not supported
Christoph Hellwig [Fri, 26 Feb 2016 17:53:12 +0000 (18:53 +0100)]
fs: return -EOPNOTSUPP if clone is not supported

-EBADF is a rather confusing error if an operations is not supported,
and nfsd gets rather upset about it.

Signed-off-by: Christoph Hellwig <hch@lst.de>
Signed-off-by: Al Viro <viro@zeniv.linux.org.uk>
8 years agohpfs: don't truncate the file when delete fails
Mikulas Patocka [Thu, 25 Feb 2016 17:17:38 +0000 (18:17 +0100)]
hpfs: don't truncate the file when delete fails

The delete opration can allocate additional space on the HPFS filesystem
due to btree split. The HPFS driver checks in advance if there is
available space, so that it won't corrupt the btree if we run out of space
during splitting.

If there is not enough available space, the HPFS driver attempted to
truncate the file, but this results in a deadlock since the commit
7dd29d8d865efdb00c0542a5d2c87af8c52ea6c7 ("HPFS: Introduce a global mutex
and lock it on every callback from VFS").

This patch removes the code that tries to truncate the file and -ENOSPC is
returned instead. If the user hits -ENOSPC on delete, he should try to
delete other files (that are stored in a leaf btree node), so that the
delete operation will make some space for deleting the file stored in
non-leaf btree node.

Reported-by: Al Viro <viro@ZenIV.linux.org.uk>
Signed-off-by: Mikulas Patocka <mikulas@artax.karlin.mff.cuni.cz>
Cc: stable@vger.kernel.org # 2.6.39+
Signed-off-by: Al Viro <viro@zeniv.linux.org.uk>
8 years agoMerge branch 'akpm' (patches from Andrew)
Linus Torvalds [Sat, 27 Feb 2016 20:46:16 +0000 (12:46 -0800)]
Merge branch 'akpm' (patches from Andrew)

Merge fixes from Andrew Morton:
 "10 fixes"

* emailed patches from Andrew Morton <akpm@linux-foundation.org>:
  dax: move writeback calls into the filesystems
  dax: give DAX clearing code correct bdev
  ext4: online defrag not supported with DAX
  ext2, ext4: only set S_DAX for regular inodes
  block: disable block device DAX by default
  ocfs2: unlock inode if deleting inode from orphan fails
  mm: ASLR: use get_random_long()
  drivers: char: random: add get_random_long()
  mm: numa: quickly fail allocations for NUMA balancing on full nodes
  mm: thp: fix SMP race condition between THP page fault and MADV_DONTNEED

8 years agoMerge tag 'tags/ext4_for_linus_stable' of git://git.kernel.org/pub/scm/linux/kernel...
Linus Torvalds [Sat, 27 Feb 2016 20:40:49 +0000 (12:40 -0800)]
Merge tag 'tags/ext4_for_linus_stable' of git://git./linux/kernel/git/tytso/ext4

Pull ext2/4 DAX fix from Ted Ts'o:
 "This fixes a file system corruption bug with DAX"

* tag 'tags/ext4_for_linus_stable' of git://git.kernel.org/pub/scm/linux/kernel/git/tytso/ext4:
  ext2, ext4: fix issue with missing journal entry in ext4_dax_mkwrite()

8 years agoMerge tag 'pci-v4.5-fixes-3' of git://git.kernel.org/pub/scm/linux/kernel/git/helgaas/pci
Linus Torvalds [Sat, 27 Feb 2016 20:33:42 +0000 (12:33 -0800)]
Merge tag 'pci-v4.5-fixes-3' of git://git./linux/kernel/git/helgaas/pci

Pull PCI fixes from Bjorn Helgaas:
 "Enumeration:
    Revert x86 pcibios_alloc_irq() to fix regression (Bjorn Helgaas)

  Marvell MVEBU host bridge driver:
    Restrict build to 32-bit ARM (Thierry Reding)"

* tag 'pci-v4.5-fixes-3' of git://git.kernel.org/pub/scm/linux/kernel/git/helgaas/pci:
  PCI: mvebu: Restrict build to 32-bit ARM
  Revert "PCI, x86: Implement pcibios_alloc_irq() and pcibios_free_irq()"
  Revert "PCI: Add helpers to manage pci_dev->irq and pci_dev->irq_managed"
  Revert "x86/PCI: Don't alloc pcibios-irq when MSI is enabled"

8 years agoext2, ext4: fix issue with missing journal entry in ext4_dax_mkwrite()
Ross Zwisler [Sat, 27 Feb 2016 19:01:13 +0000 (14:01 -0500)]
ext2, ext4: fix issue with missing journal entry in ext4_dax_mkwrite()

As it is currently written ext4_dax_mkwrite() assumes that the call into
__dax_mkwrite() will not have to do a block allocation so it doesn't create
a journal entry.  For a read that creates a zero page to cover a hole
followed by a write that actually allocates storage this is incorrect.  The
ext4_dax_mkwrite() -> __dax_mkwrite() -> __dax_fault() path calls
get_blocks() to allocate storage.

Fix this by having the ->page_mkwrite fault handler call ext4_dax_fault()
as this function already has all the logic needed to allocate a journal
entry and call __dax_fault().

Also update the ext2 fault handlers in this same way to remove duplicate
code and keep the logic between ext2 and ext4 the same.

Reviewed-by: Jan Kara <jack@suse.cz>
Signed-off-by: Ross Zwisler <ross.zwisler@linux.intel.com>
Signed-off-by: Theodore Ts'o <tytso@mit.edu>
8 years agoMerge tag 'clk-fixes-for-linus' of git://git.kernel.org/pub/scm/linux/kernel/git...
Linus Torvalds [Sat, 27 Feb 2016 18:30:14 +0000 (10:30 -0800)]
Merge tag 'clk-fixes-for-linus' of git://git./linux/kernel/git/clk/linux

Pull clk fix from Stephen Boyd:
 "One small fix to keep OMAP platforms working across a suspend/resume
  cycle"

* tag 'clk-fixes-for-linus' of git://git.kernel.org/pub/scm/linux/kernel/git/clk/linux:
  clk: ti: omap3+: dpll: use non-locking version of clk_get_rate

8 years agodax: move writeback calls into the filesystems
Ross Zwisler [Fri, 26 Feb 2016 23:19:55 +0000 (15:19 -0800)]
dax: move writeback calls into the filesystems

Previously calls to dax_writeback_mapping_range() for all DAX filesystems
(ext2, ext4 & xfs) were centralized in filemap_write_and_wait_range().

dax_writeback_mapping_range() needs a struct block_device, and it used
to get that from inode->i_sb->s_bdev.  This is correct for normal inodes
mounted on ext2, ext4 and XFS filesystems, but is incorrect for DAX raw
block devices and for XFS real-time files.

Instead, call dax_writeback_mapping_range() directly from the filesystem
->writepages function so that it can supply us with a valid block
device.  This also fixes DAX code to properly flush caches in response
to sync(2).

Signed-off-by: Ross Zwisler <ross.zwisler@linux.intel.com>
Signed-off-by: Jan Kara <jack@suse.cz>
Cc: Al Viro <viro@ftp.linux.org.uk>
Cc: Dan Williams <dan.j.williams@intel.com>
Cc: Dave Chinner <david@fromorbit.com>
Cc: Jens Axboe <axboe@fb.com>
Cc: Matthew Wilcox <matthew.r.wilcox@intel.com>
Cc: Theodore Ts'o <tytso@mit.edu>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
8 years agodax: give DAX clearing code correct bdev
Ross Zwisler [Fri, 26 Feb 2016 23:19:52 +0000 (15:19 -0800)]
dax: give DAX clearing code correct bdev

dax_clear_blocks() needs a valid struct block_device and previously it
was using inode->i_sb->s_bdev in all cases.  This is correct for normal
inodes on mounted ext2, ext4 and XFS filesystems, but is incorrect for
DAX raw block devices and for XFS real-time devices.

Instead, rename dax_clear_blocks() to dax_clear_sectors(), and change
its arguments to take a bdev and a sector instead of an inode and a
block.  This better reflects what the function does, and it allows the
filesystem and raw block device code to pass in an appropriate struct
block_device.

Signed-off-by: Ross Zwisler <ross.zwisler@linux.intel.com>
Suggested-by: Dan Williams <dan.j.williams@intel.com>
Reviewed-by: Jan Kara <jack@suse.cz>
Cc: Theodore Ts'o <tytso@mit.edu>
Cc: Al Viro <viro@ftp.linux.org.uk>
Cc: Dave Chinner <david@fromorbit.com>
Cc: Jens Axboe <axboe@fb.com>
Cc: Matthew Wilcox <matthew.r.wilcox@intel.com>
Cc: Ross Zwisler <ross.zwisler@linux.intel.com>
Cc: Theodore Ts'o <tytso@mit.edu>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
8 years agoext4: online defrag not supported with DAX
Ross Zwisler [Fri, 26 Feb 2016 23:19:49 +0000 (15:19 -0800)]
ext4: online defrag not supported with DAX

Online defrag operations for ext4 are hard coded to use the page cache.
See ext4_ioctl() -> ext4_move_extents() -> move_extent_per_page()

When combined with DAX I/O, which circumvents the page cache, this can
result in data corruption.  This was observed with xfstests ext4/307 and
ext4/308.

Fix this by only allowing online defrag for non-DAX files.

Signed-off-by: Ross Zwisler <ross.zwisler@linux.intel.com>
Reviewed-by: Jan Kara <jack@suse.cz>
Cc: Theodore Ts'o <tytso@mit.edu>
Cc: Al Viro <viro@ftp.linux.org.uk>
Cc: Dan Williams <dan.j.williams@intel.com>
Cc: Dave Chinner <david@fromorbit.com>
Cc: Jens Axboe <axboe@fb.com>
Cc: Matthew Wilcox <matthew.r.wilcox@intel.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
8 years agoext2, ext4: only set S_DAX for regular inodes
Ross Zwisler [Fri, 26 Feb 2016 23:19:46 +0000 (15:19 -0800)]
ext2, ext4: only set S_DAX for regular inodes

When S_DAX is set on an inode we assume that if there are pages attached
to the mapping (mapping->nrpages != 0), those pages are clean zero pages
that were used to service reads from holes.  Any dirty data associated
with the inode should be in the form of DAX exceptional entries
(mapping->nrexceptional) that is written back via
dax_writeback_mapping_range().

With the current code, though, this isn't always true.  For example,
ext2 and ext4 directory inodes can have S_DAX set, but have their dirty
data stored as dirty page cache entries.  For these types of inodes,
having S_DAX set doesn't really make sense since their I/O doesn't
actually happen through the DAX code path.

Instead, only allow S_DAX to be set for regular inodes for ext2 and
ext4.  This allows us to have strict DAX vs non-DAX paths in the
writeback code.

Signed-off-by: Ross Zwisler <ross.zwisler@linux.intel.com>
Reviewed-by: Jan Kara <jack@suse.cz>
Cc: Theodore Ts'o <tytso@mit.edu>
Cc: Al Viro <viro@ftp.linux.org.uk>
Cc: Dan Williams <dan.j.williams@intel.com>
Cc: Dave Chinner <david@fromorbit.com>
Cc: Jens Axboe <axboe@fb.com>
Cc: Matthew Wilcox <matthew.r.wilcox@intel.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
8 years agoblock: disable block device DAX by default
Dan Williams [Fri, 26 Feb 2016 23:19:43 +0000 (15:19 -0800)]
block: disable block device DAX by default

The recent *sync enabling discovered that we are inserting into the
block_device pagecache counter to the expectations of the dirty data
tracking for dax mappings.  This can lead to data corruption.

We want to support DAX for block devices eventually, but it requires
wider changes to properly manage the pagecache.

   dump_stack+0x85/0xc2
   dax_writeback_mapping_range+0x60/0xe0
   blkdev_writepages+0x3f/0x50
   do_writepages+0x21/0x30
   __filemap_fdatawrite_range+0xc6/0x100
   filemap_write_and_wait+0x4a/0xa0
   set_blocksize+0x70/0xd0
   sb_set_blocksize+0x1d/0x50
   ext4_fill_super+0x75b/0x3360
   mount_bdev+0x180/0x1b0
   ext4_mount+0x15/0x20
   mount_fs+0x38/0x170

Mark the support broken so its disabled by default, but otherwise still
available for testing.

Signed-off-by: Dan Williams <dan.j.williams@intel.com>
Signed-off-by: Ross Zwisler <ross.zwisler@linux.intel.com>
Reported-by: Ross Zwisler <ross.zwisler@linux.intel.com>
Suggested-by: Dave Chinner <david@fromorbit.com>
Reviewed-by: Jan Kara <jack@suse.cz>
Cc: Jens Axboe <axboe@fb.com>
Cc: Matthew Wilcox <matthew.r.wilcox@intel.com>
Cc: Al Viro <viro@ftp.linux.org.uk>
Cc: Theodore Ts'o <tytso@mit.edu>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
8 years agoocfs2: unlock inode if deleting inode from orphan fails
Guozhonghua [Fri, 26 Feb 2016 23:19:40 +0000 (15:19 -0800)]
ocfs2: unlock inode if deleting inode from orphan fails

When doing append direct io cleanup, if deleting inode fails, it goes
out without unlocking inode, which will cause the inode deadlock.

This issue was introduced by commit cf1776a9e834 ("ocfs2: fix a tiny
race when truncate dio orohaned entry").

Signed-off-by: Guozhonghua <guozhonghua@h3c.com>
Signed-off-by: Joseph Qi <joseph.qi@huawei.com>
Reviewed-by: Gang He <ghe@suse.com>
Cc: Mark Fasheh <mfasheh@suse.de>
Cc: Joel Becker <jlbec@evilplan.org>
Cc: Junxiao Bi <junxiao.bi@oracle.com>
Cc: <stable@vger.kernel.org> [4.2+]
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
8 years agomm: ASLR: use get_random_long()
Daniel Cashman [Fri, 26 Feb 2016 23:19:37 +0000 (15:19 -0800)]
mm: ASLR: use get_random_long()

Replace calls to get_random_int() followed by a cast to (unsigned long)
with calls to get_random_long().  Also address shifting bug which, in
case of x86 removed entropy mask for mmap_rnd_bits values > 31 bits.

Signed-off-by: Daniel Cashman <dcashman@android.com>
Acked-by: Kees Cook <keescook@chromium.org>
Cc: "Theodore Ts'o" <tytso@mit.edu>
Cc: Arnd Bergmann <arnd@arndb.de>
Cc: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
Cc: Catalin Marinas <catalin.marinas@arm.com>
Cc: Will Deacon <will.deacon@arm.com>
Cc: Ralf Baechle <ralf@linux-mips.org>
Cc: Benjamin Herrenschmidt <benh@kernel.crashing.org>
Cc: Paul Mackerras <paulus@samba.org>
Cc: Michael Ellerman <mpe@ellerman.id.au>
Cc: David S. Miller <davem@davemloft.net>
Cc: Thomas Gleixner <tglx@linutronix.de>
Cc: Ingo Molnar <mingo@redhat.com>
Cc: H. Peter Anvin <hpa@zytor.com>
Cc: Al Viro <viro@zeniv.linux.org.uk>
Cc: Nick Kralevich <nnk@google.com>
Cc: Jeff Vander Stoep <jeffv@google.com>
Cc: Mark Salyzyn <salyzyn@android.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
8 years agodrivers: char: random: add get_random_long()
Daniel Cashman [Fri, 26 Feb 2016 23:19:34 +0000 (15:19 -0800)]
drivers: char: random: add get_random_long()

Commit d07e22597d1d ("mm: mmap: add new /proc tunable for mmap_base
ASLR") added the ability to choose from a range of values to use for
entropy count in generating the random offset to the mmap_base address.

The maximum value on this range was set to 32 bits for 64-bit x86
systems, but this value could be increased further, requiring more than
the 32 bits of randomness provided by get_random_int(), as is already
possible for arm64.  Add a new function: get_random_long() which more
naturally fits with the mmap usage of get_random_int() but operates
exactly the same as get_random_int().

Also, fix the shifting constant in mmap_rnd() to be an unsigned long so
that values greater than 31 bits generate an appropriate mask without
overflow.  This is especially important on x86, as its shift instruction
uses a 5-bit mask for the shift operand, which meant that any value for
mmap_rnd_bits over 31 acts as a no-op and effectively disables mmap_base
randomization.

Finally, replace calls to get_random_int() with get_random_long() where
appropriate.

This patch (of 2):

Add get_random_long().

Signed-off-by: Daniel Cashman <dcashman@android.com>
Acked-by: Kees Cook <keescook@chromium.org>
Cc: "Theodore Ts'o" <tytso@mit.edu>
Cc: Arnd Bergmann <arnd@arndb.de>
Cc: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
Cc: Catalin Marinas <catalin.marinas@arm.com>
Cc: Will Deacon <will.deacon@arm.com>
Cc: Ralf Baechle <ralf@linux-mips.org>
Cc: Benjamin Herrenschmidt <benh@kernel.crashing.org>
Cc: Paul Mackerras <paulus@samba.org>
Cc: Michael Ellerman <mpe@ellerman.id.au>
Cc: David S. Miller <davem@davemloft.net>
Cc: Thomas Gleixner <tglx@linutronix.de>
Cc: Ingo Molnar <mingo@redhat.com>
Cc: H. Peter Anvin <hpa@zytor.com>
Cc: Al Viro <viro@zeniv.linux.org.uk>
Cc: Nick Kralevich <nnk@google.com>
Cc: Jeff Vander Stoep <jeffv@google.com>
Cc: Mark Salyzyn <salyzyn@android.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
8 years agomm: numa: quickly fail allocations for NUMA balancing on full nodes
Mel Gorman [Fri, 26 Feb 2016 23:19:31 +0000 (15:19 -0800)]
mm: numa: quickly fail allocations for NUMA balancing on full nodes

Commit 4167e9b2cf10 ("mm: remove GFP_THISNODE") removed the GFP_THISNODE
flag combination due to confusing semantics.  It noted that
alloc_misplaced_dst_page() was one such user after changes made by
commit e97ca8e5b864 ("mm: fix GFP_THISNODE callers and clarify").

Unfortunately when GFP_THISNODE was removed, users of
alloc_misplaced_dst_page() started waking kswapd and entering direct
reclaim because the wrong GFP flags are cleared.  The consequence is
that workloads that used to fit into memory now get reclaimed which is
addressed by this patch.

The problem can be demonstrated with "mutilate" that exercises memcached
which is software dedicated to memory object caching.  The configuration
uses 80% of memory and is run 3 times for varying numbers of clients.
The results on a 4-socket NUMA box are

mutilate
                            4.4.0                 4.4.0
                          vanilla           numaswap-v1
Hmean    1      8394.71 (  0.00%)     8395.32 (  0.01%)
Hmean    4     30024.62 (  0.00%)    34513.54 ( 14.95%)
Hmean    7     32821.08 (  0.00%)    70542.96 (114.93%)
Hmean    12    55229.67 (  0.00%)    93866.34 ( 69.96%)
Hmean    21    39438.96 (  0.00%)    85749.21 (117.42%)
Hmean    30    37796.10 (  0.00%)    50231.49 ( 32.90%)
Hmean    47    18070.91 (  0.00%)    38530.13 (113.22%)

The metric is queries/second with the more the better.  The results are
way outside of the noise and the reason for the improvement is obvious
from some of the vmstats

                                 4.4.0       4.4.0
                               vanillanumaswap-v1r1
Minor Faults                1929399272  2146148218
Major Faults                  19746529        3567
Swap Ins                      57307366        9913
Swap Outs                     50623229       17094
Allocation stalls                35909         443
DMA allocs                           0           0
DMA32 allocs                  72976349   170567396
Normal allocs               5306640898  5310651252
Movable allocs                       0           0
Direct pages scanned         404130893      799577
Kswapd pages scanned         160230174           0
Kswapd pages reclaimed        55928786           0
Direct pages reclaimed         1843936       41921
Page writes file                  2391           0
Page writes anon              50623229       17094

The vanilla kernel is swapping like crazy with large amounts of direct
reclaim and kswapd activity.  The figures are aggregate but it's known
that the bad activity is throughout the entire test.

Note that simple streaming anon/file memory consumers also see this
problem but it's not as obvious.  In those cases, kswapd is awake when
it should not be.

As there are at least two reclaim-related bugs out there, it's worth
spelling out the user-visible impact.  This patch only addresses bugs
related to excessive reclaim on NUMA hardware when the working set is
larger than a NUMA node.  There is a bug related to high kswapd CPU
usage but the reports are against laptops and other UMA hardware and is
not addressed by this patch.

Signed-off-by: Mel Gorman <mgorman@techsingularity.net>
Cc: Vlastimil Babka <vbabka@suse.cz>
Cc: Johannes Weiner <hannes@cmpxchg.org>
Cc: David Rientjes <rientjes@google.com>
Cc: <stable@vger.kernel.org> [4.1+]
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
8 years agomm: thp: fix SMP race condition between THP page fault and MADV_DONTNEED
Andrea Arcangeli [Fri, 26 Feb 2016 23:19:28 +0000 (15:19 -0800)]
mm: thp: fix SMP race condition between THP page fault and MADV_DONTNEED

pmd_trans_unstable()/pmd_none_or_trans_huge_or_clear_bad() were
introduced to locklessy (but atomically) detect when a pmd is a regular
(stable) pmd or when the pmd is unstable and can infinitely transition
from pmd_none() and pmd_trans_huge() from under us, while only holding
the mmap_sem for reading (for writing not).

While holding the mmap_sem only for reading, MADV_DONTNEED can run from
under us and so before we can assume the pmd to be a regular stable pmd
we need to compare it against pmd_none() and pmd_trans_huge() in an
atomic way, with pmd_trans_unstable().  The old pmd_trans_huge() left a
tiny window for a race.

Useful applications are unlikely to notice the difference as doing
MADV_DONTNEED concurrently with a page fault would lead to undefined
behavior.

[akpm@linux-foundation.org: tidy up comment grammar/layout]
Signed-off-by: Andrea Arcangeli <aarcange@redhat.com>
Reported-by: Kirill A. Shutemov <kirill.shutemov@linux.intel.com>
Cc: <stable@vger.kernel.org>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
8 years agoPCI: mvebu: Restrict build to 32-bit ARM
Thierry Reding [Thu, 18 Feb 2016 13:32:10 +0000 (14:32 +0100)]
PCI: mvebu: Restrict build to 32-bit ARM

This driver uses PCI glue that is only available on 32-bit ARM.  This used
to work fine as long as ARCH_MVEBU and ARCH_DOVE were exclusively 32-bit,
but there's a patch in the pipe to make ARCH_MVEBU also available on 64-bit
ARM.

[bhelgaas: changelog; patch is coming but not merged yet]
Signed-off-by: Thierry Reding <treding@nvidia.com>
Signed-off-by: Bjorn Helgaas <bhelgaas@google.com>
Acked-by: Thomas Petazzoni <thomas.petazzoni@free-electrons.com>