The PMD uses a public interface to allow applications to
control the token pop mode. Supported token pop modes are
as follows, and they impact core scheduling affinity for
ldb ports.
AUTO_POP: Pop the CQ tokens immediately after dequeueing.
DELAYED_POP: Pop CQ tokens after (dequeue_depth - 1) events
are released. Supported on load-balanced ports
only.
DEFERRED_POP: Pop the CQ tokens during next dequeue operation.
DLB does not currently support interrupts, but instead uses
umonitor/umwait if supported by the processor. This allows
the software to monitor and wait on writes to a cache-line.
DLB supports normal and sparse cq mode. In normal mode the
hardware will pack 4 QEs into each cache line. In sparse cq
mode, the hardware will only populate one QE per cache line.
Software must be aware of the cq mode, and take the appropriate
actions, based on the mode.
Add support for the eventdev start entry point.
DLB delays setting up single link resources until
eventdev start, because it is only then that it can
ascertain which ports have just one linked queue.
event/dlb: add port unlink and unlinks in progress
Add supports for the port unlink(s) eventdev entry points.
The unlink operation is an asynchronous operation executed by
a control thread, and the unlinks-in-progress function reads
a counter shared with the control thread. Port QE and memzone
memory is freed here.
Add port link entry point. Directed queues are identified and created
at this stage. Their setup deferred until link-time, at which
point we know the directed port ID. Directed queue setup
will only fail if this queue is already setup or there are
no directed queues left to configure.
Load balanced (ldb) queues are setup here.
Directed queues are not set up until link time, at which
point we know the directed port ID. Directed queue setup
will only fail if this queue is already setup or there are
no directed queues left to configure.
Add support for configuring the DLB hardware.
In particular, this patch configures the DLB
hardware's scheduling domain, such that it is provisioned with
the requested number of ports and queues, provided sufficient
resources are available. Individual queues and ports are
configured later in port setup and eventdev start.
This commit introduces the flexible interface. This
interface allows the core code to operate in PF mode (direct
hardware access) or bifurcated mode (hardware configured via
kernel driver). This driver currently only supports PF modei,
but bifurcated mode will be added in a future patch-set.
Note that the flexible interface is not used for data path
operations, and thus there are no performance concerns
related to the use of function pointers.
Add miscellaneous inline functions that may be called
from multiple files. These functions include inline
assembly of new x86 instructions, such as movdir64b,
since they are not available as builtin functions in
the minimum supported GCC version.
event/dlb: add definitions shared with LKM or shared code
Add headers containing structs and constants shared between
the PMD and the shared code. The term shared code refers to
the code that implements the hardware interface. The shared code
is introduced in the probe patch, and then is extended as
additional eventdev PMD entry points are added to the patchset.
In the case of the bifurcated PMD (to be introduced in the
future), the shared code is contained in the Linux kernel
module itself.
event/dlb: add private data structures and constants
Add headers used internally by the PMD. They include constants,
macros for device resources, structure definitions for hardware interfaces
and software state, and various forward-declarations.
DLB does not support reconfiguring individual queues
or ports on the fly. The entire device must be reconfigured.
Previously allocated port QE and memzone memory
is freed in this patch.
The PMD uses a public interface to allow applications to
control the token pop mode. Supported token pop modes are
as follows, and they impact core scheduling affinity for
ldb ports.
AUTO_POP: Pop the CQ tokens immediately after dequeueing.
DELAYED_POP: Pop CQ tokens after (dequeue_depth - 1) events
are released. Supported on load-balanced ports
only.
DEFERRED_POP: Pop the CQ tokens during next dequeue operation.
DLB2 does not currently support interrupts, but instead use
umonitor/umwait if supported by the processor. This allows
the software to monitor and wait on writes to a cache-line.
DLB2 supports normal and sparse cq mode. In normal mode the
hardware will pack 4 QEs into each cache line. In sparse cq
mode, the hardware will only populate one QE per cache line.
Software must be aware of the cq mode, and take the appropriate
actions, based on the mode.
Add support for the eventdev start entry point.
We delay initializing some resources until
eventdev start, since the number of linked queues can be
used to determine if we are dealing with a ldb or dir resource.
If this is a device restart, then the previous configuration
will be reapplied.
event/dlb2: add port unlink and unlinks in progress
Add supports for the port unlink(s) eventdev entry points.
The unlink operation is an asynchronous operation executed by
a control thread, and the unlinks-in-progress function reads
a counter shared with the control thread.
Add port link entry point. Directed queues are identified and created
at this stage. Their setup deferred until link-time, at which
point we know the directed port ID. Directed queue setup
will only fail if this queue is already setup or there are
no directed queues left to configure.
Load balanced (ldb) queues are setup here.
Directed queues are not set up until link time, at which
point we know the directed port ID. Directed queue setup
will only fail if this queue is already setup or there are
no directed queues left to configure.
Add support for configuring the DLB2 hardware.
In particular, this patch configures the DLB2
hardware's scheduling domain, such that it is provisioned with
the requested number of ports and queues, provided sufficient
resources are available. Individual queues and ports are
configured later in port setup and eventdev start.
This commit introduces the flexible interface. This
interface allows the core code to operate in PF mode (direct
hardware access) or bifurcated mode (hardware configured via
kernel driver). This driver currently only supports PF mode
but bifurcated mode will be added in a future DPDK patch-set.
Note that the flexible interface is not used for data path
operations, and thus there are no performance concerns
related to the use of function pointers.
Add miscellaneous inline functions that may be called
from multiple files. These functions include inline
assembly of new x86 instructions, such as movdir64b,
since they are not available as builtin functions in
the minimum supported GCC version.
event/dlb2: add definitions shared with LKM or shared code
Add headers containing structs and constants shared between
the PMD and the shared code. The term shared code refers to
the code that implements the hardware interface. The shared code
is introduced in the probe patch, and then is extended as
additional eventdev PMD entry points are added to the patchset.
In the case of the bifurcated PMD (to be introduced in the
future), the shared code is contained in the Linux kernel
module itself.
event/dlb2: add private data structures and constants
The header file dlb2_priv.h is used internally by the PMD.
It include constants, macros for device resources,
structure definitions for hardware interfaces and
software state, and various forward-declarations.
The header file rte_pmd_dlb2.h will be exported in a
subsequent patch, but is included here due to a data
structure dependency.
event/dlb2: add documentation and build infrastructure
Adds the meson build infrastructure, which includes
compile-time constants in rte_config.h. DLB2 is
only supported on Linux 64 bit X86 platforms at this time.
Adds announcement of availability for the new driver
for Intel Dynamic Load Balancer 2.0 hardware.
Timothy McDaniel [Mon, 26 Oct 2020 16:01:08 +0000 (11:01 -0500)]
eventdev: increase max queues configuration
DLB supports a total of 256 queues, 128 load balanced queues
and 128 directed queues. Increase RTE_EVENT_MAX_QUEUES_PER_DEV
to max possible uint_8_t max value.
Anatoly Burakov [Fri, 2 Oct 2020 12:07:09 +0000 (13:07 +0100)]
examples/l3fwd-power: fix race on interrupt wakeup log
Currently, the interrupt status notification prevents log spam by
remembering whether previous interrupt wakeup was due to traffic or due
to timeout expiring. However, it is a single variable that can
potentially be accessed from multiple threads, so it is not thread-safe.
Signed-off-by: David Marchand <david.marchand@redhat.com> Reviewed-by: Andrew Rybchenko <andrew.rybchenko@oktetlabs.ru> Acked-by: Thomas Monjalon <thomas@monjalon.net>
David Marchand [Wed, 28 Oct 2020 12:20:11 +0000 (13:20 +0100)]
eventdev: switch sequence number to dynamic mbuf field
The eventdev drivers have been hacking the deprecated field seqn for
internal test usage.
It is moved to a dynamic mbuf field in order to allow removal of seqn.
Signed-off-by: David Marchand <david.marchand@redhat.com>
David Marchand [Wed, 28 Oct 2020 12:20:10 +0000 (13:20 +0100)]
bus/fslmc: switch sequence number to dynamic mbuf field
The dpaa2 drivers have been hacking the deprecated field seqn for
internal features.
It is moved to a dynamic mbuf field in order to allow removal of seqn.
Signed-off-by: David Marchand <david.marchand@redhat.com>
David Marchand [Wed, 28 Oct 2020 12:20:09 +0000 (13:20 +0100)]
bus/dpaa: switch sequence number to dynamic mbuf field
The dpaa drivers have been hacking the deprecated field seqn for
internal features.
It is moved to a dynamic mbuf field in order to allow removal of seqn.
Signed-off-by: David Marchand <david.marchand@redhat.com>
David Marchand [Wed, 28 Oct 2020 12:20:07 +0000 (13:20 +0100)]
net/ark: remove use of seqn for debug
The seqn mbuf field is deprecated.
It is currently hacked for debug purpose, it could be changed to a
dynamic field but I see little value in the debug info it offers.
Signed-off-by: David Marchand <david.marchand@redhat.com> Acked-by: Ed Czeck <ed.czeck@atomicrules.com>
Ciara Power [Fri, 30 Oct 2020 15:37:00 +0000 (15:37 +0000)]
examples/fips_validation: fix crash on allocation failure
The return value was not being checked when calling the
get_writeback_data function in the AES test case. On failure, this led
to a NULL dereference when using memcpy later. The return value is now
checked to avoid this NULL dereference.
Coverity issue: 363463 Fixes: 952e10cdad5e ("examples/fips_validation: support scatter gather list") Signed-off-by: Ciara Power <ciara.power@intel.com> Reviewed-by: David Marchand <david.marchand@redhat.com>
Thomas Monjalon [Mon, 26 Oct 2020 01:04:06 +0000 (02:04 +0100)]
net/vmxnet3: switch MSS hint to dynamic mbuf field
The segment count, used for MSS guess,
was stored in the deprecated mbuf field udata64.
It is moved to a dynamic field in order to allow removal of udata64.
Signed-off-by: Thomas Monjalon <thomas@monjalon.net>
Ed Czeck [Tue, 27 Oct 2020 21:01:07 +0000 (22:01 +0100)]
net/ark: switch user data to dynamic mbuf fields
The second field of metadata is reserved for user data
which was using a deprecated mbuf field.
It is moved to dynamic fields in order to allow removal of udata64.
The use of meta data must be enabled with a compile-time flag
RTE_PMD_ARK_{TX,RX}_USERDATA_ENABLE.
User data on Tx and Rx paths can be defined and used separately.
Signed-off-by: Thomas Monjalon <thomas@monjalon.net> Signed-off-by: Ed Czeck <ed.czeck@atomicrules.com>
Thomas Monjalon [Mon, 26 Oct 2020 01:27:44 +0000 (02:27 +0100)]
event/sw: switch test counter to dynamic mbuf field
The test worker_loopback used the deprecated mbuf field udata64.
It is moved to a dynamic field in order to allow removal of udata64.
Signed-off-by: Thomas Monjalon <thomas@monjalon.net> Reviewed-by: Andrew Rybchenko <andrew.rybchenko@oktetlabs.ru> Acked-by: Harry van Haaren <harry.van.haaren@intel.com>
Thomas Monjalon [Mon, 26 Oct 2020 00:12:19 +0000 (01:12 +0100)]
security: switch metadata to dynamic mbuf field
The device-specific metadata was stored in the deprecated field udata64.
It is moved to a dynamic mbuf field in order to allow removal of udata64.
The name rte_security_dynfield is not very descriptive
but it should be replaced later by separate fields for each type of data
that drivers pass to the upper layer.
Signed-off-by: Thomas Monjalon <thomas@monjalon.net> Acked-by: Haiyue Wang <haiyue.wang@intel.com>
Thomas Monjalon [Sun, 25 Oct 2020 20:39:26 +0000 (21:39 +0100)]
mbuf: fix typo in dynamic field convention note
Replace "in a in PMD" with "in a PMD".
Fixes: 4958ca3a443a ("mbuf: support dynamic fields and flags") Cc: stable@dpdk.org Signed-off-by: Thomas Monjalon <thomas@monjalon.net> Acked-by: Andrew Rybchenko <andrew.rybchenko@oktetlabs.ru>
Thomas Monjalon [Sun, 25 Oct 2020 18:48:20 +0000 (19:48 +0100)]
kni: move header file from EAL
Since the kernel module is not part of EAL anymore,
there is no need to have the common KNI header file in EAL.
The file rte_kni_common.h is moved to librte_kni.
Signed-off-by: Thomas Monjalon <thomas@monjalon.net> Acked-by: Andrew Rybchenko <andrew.rybchenko@oktetlabs.ru> Acked-by: Bruce Richardson <bruce.richardson@intel.com>
Thomas Monjalon [Mon, 26 Oct 2020 04:18:57 +0000 (05:18 +0100)]
examples: enclose DPDK includes with angle brackets
In examples, DPDK header files are external,
so they must be enclosed with angle brackets, not quotes.
Signed-off-by: Thomas Monjalon <thomas@monjalon.net> Acked-by: Andrew Rybchenko <andrew.rybchenko@oktetlabs.ru> Acked-by: Bruce Richardson <bruce.richardson@intel.com>
Thomas Monjalon [Sat, 31 Oct 2020 14:27:03 +0000 (15:27 +0100)]
test/mbuf: skip field registration at busy offset
There is a test for dynamic field registration at a specific offset.
Depending on which driver is probed, some dynamic fields may be
already registered at this offset.
This failure is skipped with a warning.
Fixes: 4958ca3a443a ("mbuf: support dynamic fields and flags") Cc: stable@dpdk.org Reported-by: David Marchand <david.marchand@redhat.com> Signed-off-by: Thomas Monjalon <thomas@monjalon.net> Reviewed-by: David Marchand <david.marchand@redhat.com>
Bruce Richardson [Wed, 28 Oct 2020 12:45:34 +0000 (12:45 +0000)]
raw/ioat: fix work-queue config size
According to latest DSA spec[1], the work-queue config register size
should be based off a value read from the WQ capabilities register.
Update driver to read this value and base the start of each WQ config
off that value.
Bruce Richardson [Wed, 28 Oct 2020 16:27:01 +0000 (16:27 +0000)]
examples/multi_process: fix build on Ubuntu 20.04
Two warnings are reported by gcc 9.3.0 on Ubuntu 20.04.
When producing a printable mac address the buffer was appropriately sized
for holding the mac address exactly, but the actual snprintf included a
'\n' character at the end, which means that the snprintf technically is
getting truncated i.e. the \n would not be added due to lack of space.
This gets flagged as a problem by modern versions of gcc, e.g. on Ubuntu
20.04.
main.c:77:37: warning: ‘__builtin___snprintf_chk’ output truncated
before the last format character [-Wformat-truncation=]
77 | "%02x:%02x:%02x:%02x:%02x:%02x\n",
| ^
Since the \n is getting stripped anyway, we can fix the issue by just
removing it. In the process we can switch to using the standard ethernet
address formatting function from rte_ether.h.
The other warning is about possible string truncation when getting the
RX queue name:
In file included from init.c:36:
init.c: In function ‘init’:
../shared/common.h:38:28: warning: ‘%u’ directive output may be truncated
writing between 1 and 10 bytes into a region of size 8
[-Wformat-truncation=]
38 | #define MP_CLIENT_RXQ_NAME "MProc_Client_%u_RX"
| ^~~~~~~~~~~~~~~~~~~~
../shared/common.h:52:35: note: in expansion of macro ‘MP_CLIENT_RXQ_NAME’
52 | snprintf(buffer, sizeof(buffer), MP_CLIENT_RXQ_NAME, id);
| ^~~~~~~~~~~~~~~~~~
This is a false positive, as the value of the "id" is limited to 255,
being stored in the app as a uint8_t value, removing the possibility of
the %u being replaced by anything other then 3 characters max (rather than
up to 10 as thought by the compiler). Therefore, the warning can be easily
removed by changing the type of the "id" parameter to the local function
from "unsigned" to "uint8_t" also, ensuring the compiler is aware of the
range limit.
Fixes: af75078fece3 ("first public release") Cc: stable@dpdk.org Signed-off-by: Bruce Richardson <bruce.richardson@intel.com> Acked-by: Radu Nicolau <radu.nicolau@intel.com>
Bruce Richardson [Wed, 28 Oct 2020 16:27:00 +0000 (16:27 +0000)]
examples/vm_power: fix build on Ubuntu 20.04
When compiling on Ubuntu 20.04, a warning was issued about possible
truncation of the path string for the power management socket.
channel_manager.c: In function ‘add_all_channels’:
channel_manager.c:470:41: warning: ‘%s’ directive output may be
truncated writing up to 255 bytes into a region of size 90
[-Wformat-truncation=]
470 | sizeof(chan_info->channel_path), "%s%s",
| ^~
This can be fixed by adding in an explicit truncation check to the code
and handling it appropriately.
Fixes: e8ae9b662506 ("examples/vm_power: channel manager and monitor in host") Cc: stable@dpdk.org Signed-off-by: Bruce Richardson <bruce.richardson@intel.com> Acked-by: David Hunt <david.hunt@intel.com>
Bruce Richardson [Tue, 27 Oct 2020 17:19:53 +0000 (17:19 +0000)]
examples: fix linking against specific drivers
Some example apps rely on driver-specific functionality and link explicitly
against those drivers. These apps need their makefiles updated to take
account of the renaming of the driver libs.
Fixes: a20b2c01a7a1 ("build: standardize component names and defines") Signed-off-by: Bruce Richardson <bruce.richardson@intel.com> Reviewed-by: David Marchand <david.marchand@redhat.com>
Fix memory leak where variable oob_enable can go out of scope leaking
the storage it points to.
Coverity issue: 337674 Fixes: 95f648ff9ee ("examples/vm_power: make branch ratio threshold per core") Signed-off-by: Rory Sexton <rory.sexton@intel.com> Acked-by: David Hunt <david.hunt@intel.com>
David Marchand [Wed, 28 Oct 2020 21:02:46 +0000 (22:02 +0100)]
trace: fixup CTF event description at registration
CTF event description is currently built by appending all fields in a
single string at trace point registration.
When dumping the metadata, this string is split again and inspected to
fixup reserved keywords and special tokens like "." or "->".
Move this fixup per field at trace point registration time so that there
is no need for inspecting / string parsing when dumping metadata.
Use dynamic allocations to remove an artificial size limit on the CTF
event description manipulations.
Signed-off-by: David Marchand <david.marchand@redhat.com> Acked-by: Sunil Kumar Kori <skori@mavell.com>
Liang Ma [Tue, 27 Oct 2020 14:59:03 +0000 (14:59 +0000)]
eal: add intrinsics support check infrastructure
Currently, it is not possible to check support for intrinsics that
are platform-specific, cannot be abstracted in a generic way, or do not
have support on all architectures. The CPUID flags can be used to some
extent, but they are only defined for their platform, while intrinsics
will be available to all code as they are in generic headers.
This patch introduces infrastructure to check support for certain
platform-specific intrinsics, and adds support for checking support for
IA power management-related intrinsics for UMWAIT/UMONITOR and TPAUSE.
Signed-off-by: Anatoly Burakov <anatoly.burakov@intel.com> Signed-off-by: Liang Ma <liang.j.ma@intel.com> Acked-by: David Christensen <drc@linux.vnet.ibm.com> Acked-by: Jerin Jacob <jerinj@marvell.com> Acked-by: Ruifeng Wang <ruifeng.wang@arm.com> Acked-by: Ray Kinsella <mdr@ashroe.eu>
Liang Ma [Tue, 27 Oct 2020 14:59:02 +0000 (14:59 +0000)]
eal: add power management intrinsics
Add two new power management intrinsics, and provide an implementation
in eal/x86 based on UMONITOR/UMWAIT instructions. The instructions
are implemented as raw byte opcodes because there is not yet widespread
compiler support for these instructions.
The power management instructions provide an architecture-specific
function to either wait until a specified TSC timestamp is reached, or
optionally wait until either a TSC timestamp is reached or a memory
location is written to. The monitor function also provides an optional
comparison, to avoid sleeping when the expected write has already
happened, and no more writes are expected.
For more details, please refer to Intel(R) 64 and IA-32 Architectures
Software Developer's Manual, Volume 2.
Signed-off-by: Liang Ma <liang.j.ma@intel.com> Signed-off-by: Anatoly Burakov <anatoly.burakov@intel.com> Acked-by: David Christensen <drc@linux.vnet.ibm.com> Acked-by: Jerin Jacob <jerinj@marvell.com> Acked-by: Konstantin Ananyev <konstantin.ananyev@intel.com> Acked-by: Ruifeng Wang <ruifeng.wang@arm.com>
Yunjian Wang [Thu, 15 Oct 2020 08:42:30 +0000 (16:42 +0800)]
eal: fix interrupt trace point
This patch fixes (dereference after null check) coverity issue.
For this reason, we should add null check at the beginning of the
function and return error directly if the 'intr_handle' is null.
Add zero-copy APIs. These APIs provide the capability to
copy the data to/from the ring memory directly, without
having a temporary copy (for ex: an array of mbufs on
the stack). Use cases that involve copying large amount
of data to/from the ring can benefit from these APIs.
git.droids-corp.org / dpdk.git / log