Add comment about minimum and maximum size of command buffer.
Add some text about the expected input of CS IOCTL.
Signed-off-by: Oded Gabbay <oded.gabbay@gmail.com>
Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
This patch increase the size field in the uapi structure of the Memory
IOCTL from 32-bit to 64-bit. This is to allow the user to allocate and/or
map memory in chunks that are larger then 4GB.
Goya's device memory (DRAM) can be up to 16GB, and for certain
topologies, the user may want an allocation that is larger than 4GB.
This change doesn't break current user-space because there was a "pad"
field in the uapi structure right after the size field. Changing the size
field to be 64-bit and removing the pad field maintains compatibility with
current user-space.
Signed-off-by: Oded Gabbay <oded.gabbay@gmail.com>
Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
This patch implements the INFO IOCTL. That IOCTL is used by the user to
query information that is relevant/needed by the user in order to submit
deep learning jobs to Goya.
The information is divided into several categories, such as H/W IP, Events
that happened, DDR usage and more.
Reviewed-by: Mike Rapoport <rppt@linux.ibm.com>
Signed-off-by: Oded Gabbay <oded.gabbay@gmail.com>
Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
This patch adds the Virtual Memory and MMU modules.
Goya has an internal MMU which provides process isolation on the internal
DDR. The internal MMU also performs translations for transactions that go
from Goya to the Host.
The driver is responsible for allocating and freeing memory on the DDR
upon user request. It also provides an interface to map and unmap DDR and
Host memory to the device address space.
The MMU in Goya supports 3-level and 4-level page tables. With 3-level, the
size of each page is 2MB, while with 4-level the size of each page is 4KB.
In the DDR, the physical pages are always 2MB.
Reviewed-by: Mike Rapoport <rppt@linux.ibm.com>
Signed-off-by: Omer Shpigelman <oshpigelman@habana.ai>
Signed-off-by: Oded Gabbay <oded.gabbay@gmail.com>
Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
This patch adds the main flow for the user to submit work to the device.
Each work is described by a command submission object (CS). The CS contains
3 arrays of command buffers: One for execution, and two for context-switch
(store and restore).
For each CB, the user specifies on which queue to put that CB. In case of
an internal queue, the entry doesn't contain a pointer to the CB but the
address in the on-chip memory that the CB resides at.
The driver parses some of the CBs to enforce security restrictions.
The user receives a sequence number that represents the CS object. The user
can then query the driver regarding the status of the CS, using that
sequence number.
In case the CS doesn't finish before the timeout expires, the driver will
perform a soft-reset of the device.
Reviewed-by: Mike Rapoport <rppt@linux.ibm.com>
Signed-off-by: Oded Gabbay <oded.gabbay@gmail.com>
Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
This patch adds the H/W queues module and the code to initialize Goya's
various compute and DMA engines and their queues.
Goya has 5 DMA channels, 8 TPC engines and a single MME engine. For each
channel/engine, there is a H/W queue logic which is used to pass commands
from the user to the H/W. That logic is called QMAN.
There are two types of QMANs: external and internal. The DMA QMANs are
considered external while the TPC and MME QMANs are considered internal.
For each external queue there is a completion queue, which is located on
the Host memory.
The differences between external and internal QMANs are:
1. The location of the queue's memory. External QMANs are located on the
Host memory while internal QMANs are located on the on-chip memory.
2. The external QMAN write an entry to a completion queue and sends an
MSI-X interrupt upon completion of a command buffer that was given to
it. The internal QMAN doesn't do that.
Reviewed-by: Mike Rapoport <rppt@linux.ibm.com>
Signed-off-by: Oded Gabbay <oded.gabbay@gmail.com>
Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
This patch adds the command buffer (CB) module, which allows the user to
create and destroy CBs and to map them to the user's process
address-space.
A command buffer is a memory blocks that reside in DMA-able address-space
and is physically contiguous so it can be accessed by the device without
MMU translation. The command buffer memory is allocated using the
coherent DMA API.
When creating a new CB, the IOCTL returns a handle of it, and the
user-space process needs to use that handle to mmap the buffer to get a VA
in the user's address-space.
Before destroying (freeing) a CB, the user must unmap the CB's VA using the
CB handle.
Each CB has a reference counter, which tracks its usage in command
submissions and also its mmaps (only a single mmap is allowed).
The driver maintains a pool of pre-allocated CBs in order to reduce
latency during command submissions. In case the pool is empty, the driver
will go to the slow-path of allocating a new CB, i.e. calling
dma_alloc_coherent.
Reviewed-by: Mike Rapoport <rppt@linux.ibm.com>
Signed-off-by: Oded Gabbay <oded.gabbay@gmail.com>
Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
This patch adds a basic support for the Goya device. The code initializes
the device's PCI controller and PCI bars. It also initializes various S/W
structures and adds some basic helper functions.
Reviewed-by: Mike Rapoport <rppt@linux.ibm.com>
Signed-off-by: Oded Gabbay <oded.gabbay@gmail.com>
Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
