* [PATCH v5 1/8] accel/qaic: Add documentation for AIC100 accelerator driver
2023-03-27 15:54 [PATCH v5 0/8] QAIC accel driver Jeffrey Hugo
@ 2023-03-27 15:54 ` Jeffrey Hugo
2023-03-27 15:54 ` [PATCH v5 2/8] accel/qaic: Add uapi and core driver file Jeffrey Hugo
` (7 subsequent siblings)
8 siblings, 0 replies; 16+ messages in thread
From: Jeffrey Hugo @ 2023-03-27 15:54 UTC (permalink / raw)
To: ogabbay, airlied, daniel
Cc: jacek.lawrynowicz, stanislaw.gruszka, dri-devel, quic_pkanojiy,
quic_carlv, quic_ajitpals, linux-doc, linux-arm-msm, bagasdotme,
mani, Jeffrey Hugo
The Qualcomm Cloud AI 100 (AIC100) device is an Artificial Intelligence
accelerator PCIe card. It contains a number of components both in the
SoC and on the card which facilitate running workloads:
QSM: management processor
NSPs: workload compute units
DMA Bridge: dedicated data mover for the workloads
MHI: multiplexed communication channels
DDR: workload storage and memory
The Linux kernel driver for AIC100 is called "QAIC" and is located in the
accel subsystem.
Signed-off-by: Jeffrey Hugo <quic_jhugo@quicinc.com>
Reviewed-by: Carl Vanderlip <quic_carlv@quicinc.com>
Reviewed-by: Pranjal Ramajor Asha Kanojiya <quic_pkanojiy@quicinc.com>
Reviewed-by: Stanislaw Gruszka <stanislaw.gruszka@linux.intel.com>
Reviewed-by: Jacek Lawrynowicz <jacek.lawrynowicz@linux.intel.com>
---
Documentation/accel/index.rst | 1 +
Documentation/accel/qaic/aic100.rst | 510 ++++++++++++++++++++++++++++++++++++
Documentation/accel/qaic/index.rst | 13 +
Documentation/accel/qaic/qaic.rst | 170 ++++++++++++
4 files changed, 694 insertions(+)
create mode 100644 Documentation/accel/qaic/aic100.rst
create mode 100644 Documentation/accel/qaic/index.rst
create mode 100644 Documentation/accel/qaic/qaic.rst
diff --git a/Documentation/accel/index.rst b/Documentation/accel/index.rst
index 2b43c9a..e94a016 100644
--- a/Documentation/accel/index.rst
+++ b/Documentation/accel/index.rst
@@ -8,6 +8,7 @@ Compute Accelerators
:maxdepth: 1
introduction
+ qaic/index
.. only:: subproject and html
diff --git a/Documentation/accel/qaic/aic100.rst b/Documentation/accel/qaic/aic100.rst
new file mode 100644
index 0000000..c80d0f1
--- /dev/null
+++ b/Documentation/accel/qaic/aic100.rst
@@ -0,0 +1,510 @@
+.. SPDX-License-Identifier: GPL-2.0-only
+
+===============================
+ Qualcomm Cloud AI 100 (AIC100)
+===============================
+
+Overview
+========
+
+The Qualcomm Cloud AI 100/AIC100 family of products (including SA9000P - part of
+Snapdragon Ride) are PCIe adapter cards which contain a dedicated SoC ASIC for
+the purpose of efficiently running Artificial Intelligence (AI) Deep Learning
+inference workloads. They are AI accelerators.
+
+The PCIe interface of AIC100 is capable of PCIe Gen4 speeds over eight lanes
+(x8). An individual SoC on a card can have up to 16 NSPs for running workloads.
+Each SoC has an A53 management CPU. On card, there can be up to 32 GB of DDR.
+
+Multiple AIC100 cards can be hosted in a single system to scale overall
+performance. AIC100 cards are multi-user capable and able to execute workloads
+from multiple users in a concurrent manner.
+
+Hardware Description
+====================
+
+An AIC100 card consists of an AIC100 SoC, on-card DDR, and a set of misc
+peripherals (PMICs, etc).
+
+An AIC100 card can either be a PCIe HHHL form factor (a traditional PCIe card),
+or a Dual M.2 card. Both use PCIe to connect to the host system.
+
+As a PCIe endpoint/adapter, AIC100 uses the standard VendorID(VID)/
+DeviceID(DID) combination to uniquely identify itself to the host. AIC100
+uses the standard Qualcomm VID (0x17cb). All AIC100 SKUs use the same
+AIC100 DID (0xa100).
+
+AIC100 does not implement FLR (function level reset).
+
+AIC100 implements MSI but does not implement MSI-X. AIC100 requires 17 MSIs to
+operate (1 for MHI, 16 for the DMA Bridge).
+
+As a PCIe device, AIC100 utilizes BARs to provide host interfaces to the device
+hardware. AIC100 provides 3, 64-bit BARs.
+
+* The first BAR is 4K in size, and exposes the MHI interface to the host.
+
+* The second BAR is 2M in size, and exposes the DMA Bridge interface to the
+ host.
+
+* The third BAR is variable in size based on an individual AIC100's
+ configuration, but defaults to 64K. This BAR currently has no purpose.
+
+From the host perspective, AIC100 has several key hardware components -
+
+* MHI (Modem Host Interface)
+* QSM (QAIC Service Manager)
+* NSPs (Neural Signal Processor)
+* DMA Bridge
+* DDR
+
+MHI
+---
+
+AIC100 has one MHI interface over PCIe. MHI itself is documented at
+Documentation/mhi/index.rst MHI is the mechanism the host uses to communicate
+with the QSM. Except for workload data via the DMA Bridge, all interaction with
+the device occurs via MHI.
+
+QSM
+---
+
+QAIC Service Manager. This is an ARM A53 CPU that runs the primary
+firmware of the card and performs on-card management tasks. It also
+communicates with the host via MHI. Each AIC100 has one of
+these.
+
+NSP
+---
+
+Neural Signal Processor. Each AIC100 has up to 16 of these. These are
+the processors that run the workloads on AIC100. Each NSP is a Qualcomm Hexagon
+(Q6) DSP with HVX and HMX. Each NSP can only run one workload at a time, but
+multiple NSPs may be assigned to a single workload. Since each NSP can only run
+one workload, AIC100 is limited to 16 concurrent workloads. Workload
+"scheduling" is under the purview of the host. AIC100 does not automatically
+timeslice.
+
+DMA Bridge
+----------
+
+The DMA Bridge is custom DMA engine that manages the flow of data
+in and out of workloads. AIC100 has one of these. The DMA Bridge has 16
+channels, each consisting of a set of request/response FIFOs. Each active
+workload is assigned a single DMA Bridge channel. The DMA Bridge exposes
+hardware registers to manage the FIFOs (head/tail pointers), but requires host
+memory to store the FIFOs.
+
+DDR
+---
+
+AIC100 has on-card DDR. In total, an AIC100 can have up to 32 GB of DDR.
+This DDR is used to store workloads, data for the workloads, and is used by the
+QSM for managing the device. NSPs are granted access to sections of the DDR by
+the QSM. The host does not have direct access to the DDR, and must make
+requests to the QSM to transfer data to the DDR.
+
+High-level Use Flow
+===================
+
+AIC100 is a multi-user, programmable accelerator typically used for running
+neural networks in inferencing mode to efficiently perform AI operations.
+AIC100 is not intended for training neural networks. AIC100 can be utilized
+for generic compute workloads.
+
+Assuming a user wants to utilize AIC100, they would follow these steps:
+
+1. Compile the workload into an ELF targeting the NSP(s)
+2. Make requests to the QSM to load the workload and related artifacts into the
+ device DDR
+3. Make a request to the QSM to activate the workload onto a set of idle NSPs
+4. Make requests to the DMA Bridge to send input data to the workload to be
+ processed, and other requests to receive processed output data from the
+ workload.
+5. Once the workload is no longer required, make a request to the QSM to
+ deactivate the workload, thus putting the NSPs back into an idle state.
+6. Once the workload and related artifacts are no longer needed for future
+ sessions, make requests to the QSM to unload the data from DDR. This frees
+ the DDR to be used by other users.
+
+
+Boot Flow
+=========
+
+AIC100 uses a flashless boot flow, derived from Qualcomm MSMs.
+
+When AIC100 is first powered on, it begins executing PBL (Primary Bootloader)
+from ROM. PBL enumerates the PCIe link, and initializes the BHI (Boot Host
+Interface) component of MHI.
+
+Using BHI, the host points PBL to the location of the SBL (Secondary Bootloader)
+image. The PBL pulls the image from the host, validates it, and begins
+execution of SBL.
+
+SBL initializes MHI, and uses MHI to notify the host that the device has entered
+the SBL stage. SBL performs a number of operations:
+
+* SBL initializes the majority of hardware (anything PBL left uninitialized),
+ including DDR.
+* SBL offloads the bootlog to the host.
+* SBL synchronizes timestamps with the host for future logging.
+* SBL uses the Sahara protocol to obtain the runtime firmware images from the
+ host.
+
+Once SBL has obtained and validated the runtime firmware, it brings the NSPs out
+of reset, and jumps into the QSM.
+
+The QSM uses MHI to notify the host that the device has entered the QSM stage
+(AMSS in MHI terms). At this point, the AIC100 device is fully functional, and
+ready to process workloads.
+
+Userspace components
+====================
+
+Compiler
+--------
+
+An open compiler for AIC100 based on upstream LLVM can be found at:
+https://github.com/quic/software-kit-for-qualcomm-cloud-ai-100-cc
+
+Usermode Driver (UMD)
+---------------------
+
+An open UMD that interfaces with the qaic kernel driver can be found at:
+https://github.com/quic/software-kit-for-qualcomm-cloud-ai-100
+
+Sahara loader
+-------------
+
+An open implementation of the Sahara protocol called kickstart can be found at:
+https://github.com/andersson/qdl
+
+MHI Channels
+============
+
+AIC100 defines a number of MHI channels for different purposes. This is a list
+of the defined channels, and their uses.
+
++----------------+---------+----------+----------------------------------------+
+| Channel name | IDs | EEs | Purpose |
++================+=========+==========+========================================+
+| QAIC_LOOPBACK | 0 & 1 | AMSS | Any data sent to the device on this |
+| | | | channel is sent back to the host. |
++----------------+---------+----------+----------------------------------------+
+| QAIC_SAHARA | 2 & 3 | SBL | Used by SBL to obtain the runtime |
+| | | | firmware from the host. |
++----------------+---------+----------+----------------------------------------+
+| QAIC_DIAG | 4 & 5 | AMSS | Used to communicate with QSM via the |
+| | | | DIAG protocol. |
++----------------+---------+----------+----------------------------------------+
+| QAIC_SSR | 6 & 7 | AMSS | Used to notify the host of subsystem |
+| | | | restart events, and to offload SSR |
+| | | | crashdumps. |
++----------------+---------+----------+----------------------------------------+
+| QAIC_QDSS | 8 & 9 | AMSS | Used for the Qualcomm Debug Subsystem. |
++----------------+---------+----------+----------------------------------------+
+| QAIC_CONTROL | 10 & 11 | AMSS | Used for the Neural Network Control |
+| | | | (NNC) protocol. This is the primary |
+| | | | channel between host and QSM for |
+| | | | managing workloads. |
++----------------+---------+----------+----------------------------------------+
+| QAIC_LOGGING | 12 & 13 | SBL | Used by the SBL to send the bootlog to |
+| | | | the host. |
++----------------+---------+----------+----------------------------------------+
+| QAIC_STATUS | 14 & 15 | AMSS | Used to notify the host of Reliability,|
+| | | | Accessibility, Serviceability (RAS) |
+| | | | events. |
++----------------+---------+----------+----------------------------------------+
+| QAIC_TELEMETRY | 16 & 17 | AMSS | Used to get/set power/thermal/etc |
+| | | | attributes. |
++----------------+---------+----------+----------------------------------------+
+| QAIC_DEBUG | 18 & 19 | AMSS | Not used. |
++----------------+---------+----------+----------------------------------------+
+| QAIC_TIMESYNC | 20 & 21 | SBL/AMSS | Used to synchronize timestamps in the |
+| | | | device side logs with the host time |
+| | | | source. |
++----------------+---------+----------+----------------------------------------+
+
+DMA Bridge
+==========
+
+Overview
+--------
+
+The DMA Bridge is one of the main interfaces to the host from the device
+(the other being MHI). As part of activating a workload to run on NSPs, the QSM
+assigns that network a DMA Bridge channel. A workload's DMA Bridge channel
+(DBC for short) is solely for the use of that workload and is not shared with
+other workloads.
+
+Each DBC is a pair of FIFOs that manage data in and out of the workload. One
+FIFO is the request FIFO. The other FIFO is the response FIFO.
+
+Each DBC contains 4 registers in hardware:
+
+* Request FIFO head pointer (offset 0x0). Read only by the host. Indicates the
+ latest item in the FIFO the device has consumed.
+* Request FIFO tail pointer (offset 0x4). Read/write by the host. Host
+ increments this register to add new items to the FIFO.
+* Response FIFO head pointer (offset 0x8). Read/write by the host. Indicates
+ the latest item in the FIFO the host has consumed.
+* Response FIFO tail pointer (offset 0xc). Read only by the host. Device
+ increments this register to add new items to the FIFO.
+
+The values in each register are indexes in the FIFO. To get the location of the
+FIFO element pointed to by the register: FIFO base address + register * element
+size.
+
+DBC registers are exposed to the host via the second BAR. Each DBC consumes
+4KB of space in the BAR.
+
+The actual FIFOs are backed by host memory. When sending a request to the QSM
+to activate a network, the host must donate memory to be used for the FIFOs.
+Due to internal mapping limitations of the device, a single contiguous chunk of
+memory must be provided per DBC, which hosts both FIFOs. The request FIFO will
+consume the beginning of the memory chunk, and the response FIFO will consume
+the end of the memory chunk.
+
+Request FIFO
+------------
+
+A request FIFO element has the following structure:
+
+.. code-block:: c
+
+ struct request_elem {
+ u16 req_id;
+ u8 seq_id;
+ u8 pcie_dma_cmd;
+ u32 reserved;
+ u64 pcie_dma_source_addr;
+ u64 pcie_dma_dest_addr;
+ u32 pcie_dma_len;
+ u32 reserved;
+ u64 doorbell_addr;
+ u8 doorbell_attr;
+ u8 reserved;
+ u16 reserved;
+ u32 doorbell_data;
+ u32 sem_cmd0;
+ u32 sem_cmd1;
+ u32 sem_cmd2;
+ u32 sem_cmd3;
+ };
+
+Request field descriptions:
+
+req_id
+ request ID. A request FIFO element and a response FIFO element with
+ the same request ID refer to the same command.
+
+seq_id
+ sequence ID within a request. Ignored by the DMA Bridge.
+
+pcie_dma_cmd
+ describes the DMA element of this request.
+
+ * Bit(7) is the force msi flag, which overrides the DMA Bridge MSI logic
+ and generates a MSI when this request is complete, and QSM
+ configures the DMA Bridge to look at this bit.
+ * Bits(6:5) are reserved.
+ * Bit(4) is the completion code flag, and indicates that the DMA Bridge
+ shall generate a response FIFO element when this request is
+ complete.
+ * Bit(3) indicates if this request is a linked list transfer(0) or a bulk
+ transfer(1).
+ * Bit(2) is reserved.
+ * Bits(1:0) indicate the type of transfer. No transfer(0), to device(1),
+ from device(2). Value 3 is illegal.
+
+pcie_dma_source_addr
+ source address for a bulk transfer, or the address of the linked list.
+
+pcie_dma_dest_addr
+ destination address for a bulk transfer.
+
+pcie_dma_len
+ length of the bulk transfer. Note that the size of this field
+ limits transfers to 4G in size.
+
+doorbell_addr
+ address of the doorbell to ring when this request is complete.
+
+doorbell_attr
+ doorbell attributes.
+
+ * Bit(7) indicates if a write to a doorbell is to occur.
+ * Bits(6:2) are reserved.
+ * Bits(1:0) contain the encoding of the doorbell length. 0 is 32-bit,
+ 1 is 16-bit, 2 is 8-bit, 3 is reserved. The doorbell address
+ must be naturally aligned to the specified length.
+
+doorbell_data
+ data to write to the doorbell. Only the bits corresponding to
+ the doorbell length are valid.
+
+sem_cmdN
+ semaphore command.
+
+ * Bit(31) indicates this semaphore command is enabled.
+ * Bit(30) is the to-device DMA fence. Block this request until all
+ to-device DMA transfers are complete.
+ * Bit(29) is the from-device DMA fence. Block this request until all
+ from-device DMA transfers are complete.
+ * Bits(28:27) are reserved.
+ * Bits(26:24) are the semaphore command. 0 is NOP. 1 is init with the
+ specified value. 2 is increment. 3 is decrement. 4 is wait
+ until the semaphore is equal to the specified value. 5 is wait
+ until the semaphore is greater or equal to the specified value.
+ 6 is "P", wait until semaphore is greater than 0, then
+ decrement by 1. 7 is reserved.
+ * Bit(23) is reserved.
+ * Bit(22) is the semaphore sync. 0 is post sync, which means that the
+ semaphore operation is done after the DMA transfer. 1 is
+ presync, which gates the DMA transfer. Only one presync is
+ allowed per request.
+ * Bit(21) is reserved.
+ * Bits(20:16) is the index of the semaphore to operate on.
+ * Bits(15:12) are reserved.
+ * Bits(11:0) are the semaphore value to use in operations.
+
+Overall, a request is processed in 4 steps:
+
+1. If specified, the presync semaphore condition must be true
+2. If enabled, the DMA transfer occurs
+3. If specified, the postsync semaphore conditions must be true
+4. If enabled, the doorbell is written
+
+By using the semaphores in conjunction with the workload running on the NSPs,
+the data pipeline can be synchronized such that the host can queue multiple
+requests of data for the workload to process, but the DMA Bridge will only copy
+the data into the memory of the workload when the workload is ready to process
+the next input.
+
+Response FIFO
+-------------
+
+Once a request is fully processed, a response FIFO element is generated if
+specified in pcie_dma_cmd. The structure of a response FIFO element:
+
+.. code-block:: c
+
+ struct response_elem {
+ u16 req_id;
+ u16 completion_code;
+ };
+
+req_id
+ matches the req_id of the request that generated this element.
+
+completion_code
+ status of this request. 0 is success. Non-zero is an error.
+
+The DMA Bridge will generate a MSI to the host as a reaction to activity in the
+response FIFO of a DBC. The DMA Bridge hardware has an IRQ storm mitigation
+algorithm, where it will only generate a MSI when the response FIFO transitions
+from empty to non-empty (unless force MSI is enabled and triggered). In
+response to this MSI, the host is expected to drain the response FIFO, and must
+take care to handle any race conditions between draining the FIFO, and the
+device inserting elements into the FIFO.
+
+Neural Network Control (NNC) Protocol
+=====================================
+
+The NNC protocol is how the host makes requests to the QSM to manage workloads.
+It uses the QAIC_CONTROL MHI channel.
+
+Each NNC request is packaged into a message. Each message is a series of
+transactions. A passthrough type transaction can contain elements known as
+commands.
+
+QSM requires NNC messages be little endian encoded and the fields be naturally
+aligned. Since there are 64-bit elements in some NNC messages, 64-bit alignment
+must be maintained.
+
+A message contains a header and then a series of transactions. A message may be
+at most 4K in size from QSM to the host. From the host to the QSM, a message
+can be at most 64K (maximum size of a single MHI packet), but there is a
+continuation feature where message N+1 can be marked as a continuation of
+message N. This is used for exceedingly large DMA xfer transactions.
+
+Transaction descriptions
+------------------------
+
+passthrough
+ Allows userspace to send an opaque payload directly to the QSM.
+ This is used for NNC commands. Userspace is responsible for managing
+ the QSM message requirements in the payload.
+
+dma_xfer
+ DMA transfer. Describes an object that the QSM should DMA into the
+ device via address and size tuples.
+
+activate
+ Activate a workload onto NSPs. The host must provide memory to be
+ used by the DBC.
+
+deactivate
+ Deactivate an active workload and return the NSPs to idle.
+
+status
+ Query the QSM about it's NNC implementation. Returns the NNC version,
+ and if CRC is used.
+
+terminate
+ Release a user's resources.
+
+dma_xfer_cont
+ Continuation of a previous DMA transfer. If a DMA transfer
+ cannot be specified in a single message (highly fragmented), this
+ transaction can be used to specify more ranges.
+
+validate_partition
+ Query to QSM to determine if a partition identifier is valid.
+
+Each message is tagged with a user id, and a partition id. The user id allows
+QSM to track resources, and release them when the user goes away (eg the process
+crashes). A partition id identifies the resource partition that QSM manages,
+which this message applies to.
+
+Messages may have CRCs. Messages should have CRCs applied until the QSM
+reports via the status transaction that CRCs are not needed. The QSM on the
+SA9000P requires CRCs for black channel safing.
+
+Subsystem Restart (SSR)
+=======================
+
+SSR is the concept of limiting the impact of an error. An AIC100 device may
+have multiple users, each with their own workload running. If the workload of
+one user crashes, the fallout of that should be limited to that workload and not
+impact other workloads. SSR accomplishes this.
+
+If a particular workload crashes, QSM notifies the host via the QAIC_SSR MHI
+channel. This notification identifies the workload by it's assigned DBC. A
+multi-stage recovery process is then used to cleanup both sides, and get the
+DBC/NSPs into a working state.
+
+When SSR occurs, any state in the workload is lost. Any inputs that were in
+process, or queued by not yet serviced, are lost. The loaded artifacts will
+remain in on-card DDR, but the host will need to re-activate the workload if
+it desires to recover the workload.
+
+Reliability, Accessibility, Serviceability (RAS)
+================================================
+
+AIC100 is expected to be deployed in server systems where RAS ideology is
+applied. Simply put, RAS is the concept of detecting, classifying, and
+reporting errors. While PCIe has AER (Advanced Error Reporting) which factors
+into RAS, AER does not allow for a device to report details about internal
+errors. Therefore, AIC100 implements a custom RAS mechanism. When a RAS event
+occurs, QSM will report the event with appropriate details via the QAIC_STATUS
+MHI channel. A sysadmin may determine that a particular device needs
+additional service based on RAS reports.
+
+Telemetry
+=========
+
+QSM has the ability to report various physical attributes of the device, and in
+some cases, to allow the host to control them. Examples include thermal limits,
+thermal readings, and power readings. These items are communicated via the
+QAIC_TELEMETRY MHI channel.
diff --git a/Documentation/accel/qaic/index.rst b/Documentation/accel/qaic/index.rst
new file mode 100644
index 0000000..ad19b88
--- /dev/null
+++ b/Documentation/accel/qaic/index.rst
@@ -0,0 +1,13 @@
+.. SPDX-License-Identifier: GPL-2.0-only
+
+=====================================
+ accel/qaic Qualcomm Cloud AI driver
+=====================================
+
+The accel/qaic driver supports the Qualcomm Cloud AI machine learning
+accelerator cards.
+
+.. toctree::
+
+ qaic
+ aic100
diff --git a/Documentation/accel/qaic/qaic.rst b/Documentation/accel/qaic/qaic.rst
new file mode 100644
index 0000000..72a70ab
--- /dev/null
+++ b/Documentation/accel/qaic/qaic.rst
@@ -0,0 +1,170 @@
+.. SPDX-License-Identifier: GPL-2.0-only
+
+=============
+ QAIC driver
+=============
+
+The QAIC driver is the Kernel Mode Driver (KMD) for the AIC100 family of AI
+accelerator products.
+
+Interrupts
+==========
+
+While the AIC100 DMA Bridge hardware implements an IRQ storm mitigation
+mechanism, it is still possible for an IRQ storm to occur. A storm can happen
+if the workload is particularly quick, and the host is responsive. If the host
+can drain the response FIFO as quickly as the device can insert elements into
+it, then the device will frequently transition the response FIFO from empty to
+non-empty and generate MSIs at a rate equivalent to the speed of the
+workload's ability to process inputs. The lprnet (license plate reader network)
+workload is known to trigger this condition, and can generate in excess of 100k
+MSIs per second. It has been observed that most systems cannot tolerate this
+for long, and will crash due to some form of watchdog due to the overhead of
+the interrupt controller interrupting the host CPU.
+
+To mitigate this issue, the QAIC driver implements specific IRQ handling. When
+QAIC receives an IRQ, it disables that line. This prevents the interrupt
+controller from interrupting the CPU. Then AIC drains the FIFO. Once the FIFO
+is drained, QAIC implements a "last chance" polling algorithm where QAIC will
+sleep for a time to see if the workload will generate more activity. The IRQ
+line remains disabled during this time. If no activity is detected, QAIC exits
+polling mode and reenables the IRQ line.
+
+This mitigation in QAIC is very effective. The same lprnet usecase that
+generates 100k IRQs per second (per /proc/interrupts) is reduced to roughly 64
+IRQs over 5 minutes while keeping the host system stable, and having the same
+workload throughput performance (within run to run noise variation).
+
+
+Neural Network Control (NNC) Protocol
+=====================================
+
+The implementation of NNC is split between the KMD (QAIC) and UMD. In general
+QAIC understands how to encode/decode NNC wire protocol, and elements of the
+protocol which require kernel space knowledge to process (for example, mapping
+host memory to device IOVAs). QAIC understands the structure of a message, and
+all of the transactions. QAIC does not understand commands (the payload of a
+passthrough transaction).
+
+QAIC handles and enforces the required little endianness and 64-bit alignment,
+to the degree that it can. Since QAIC does not know the contents of a
+passthrough transaction, it relies on the UMD to satisfy the requirements.
+
+The terminate transaction is of particular use to QAIC. QAIC is not aware of
+the resources that are loaded onto a device since the majority of that activity
+occurs within NNC commands. As a result, QAIC does not have the means to
+roll back userspace activity. To ensure that a userspace client's resources
+are fully released in the case of a process crash, or a bug, QAIC uses the
+terminate command to let QSM know when a user has gone away, and the resources
+can be released.
+
+QSM can report a version number of the NNC protocol it supports. This is in the
+form of a Major number and a Minor number.
+
+Major number updates indicate changes to the NNC protocol which impact the
+message format, or transactions (impacts QAIC).
+
+Minor number updates indicate changes to the NNC protocol which impact the
+commands (does not impact QAIC).
+
+uAPI
+====
+
+QAIC defines a number of driver specific IOCTLs as part of the userspace API.
+This section describes those APIs.
+
+DRM_IOCTL_QAIC_MANAGE
+ This IOCTL allows userspace to send a NNC request to the QSM. The call will
+ block until a response is received, or the request has timed out.
+
+DRM_IOCTL_QAIC_CREATE_BO
+ This IOCTL allows userspace to allocate a buffer object (BO) which can send
+ or receive data from a workload. The call will return a GEM handle that
+ represents the allocated buffer. The BO is not usable until it has been
+ sliced (see DRM_IOCTL_QAIC_ATTACH_SLICE_BO).
+
+DRM_IOCTL_QAIC_MMAP_BO
+ This IOCTL allows userspace to prepare an allocated BO to be mmap'd into the
+ userspace process.
+
+DRM_IOCTL_QAIC_ATTACH_SLICE_BO
+ This IOCTL allows userspace to slice a BO in preparation for sending the BO
+ to the device. Slicing is the operation of describing what portions of a BO
+ get sent where to a workload. This requires a set of DMA transfers for the
+ DMA Bridge, and as such, locks the BO to a specific DBC.
+
+DRM_IOCTL_QAIC_EXECUTE_BO
+ This IOCTL allows userspace to submit a set of sliced BOs to the device. The
+ call is non-blocking. Success only indicates that the BOs have been queued
+ to the device, but does not guarantee they have been executed.
+
+DRM_IOCTL_QAIC_PARTIAL_EXECUTE_BO
+ This IOCTL operates like DRM_IOCTL_QAIC_EXECUTE_BO, but it allows userspace
+ to shrink the BOs sent to the device for this specific call. If a BO
+ typically has N inputs, but only a subset of those is available, this IOCTL
+ allows userspace to indicate that only the first M bytes of the BO should be
+ sent to the device to minimize data transfer overhead. This IOCTL dynamically
+ recomputes the slicing, and therefore has some processing overhead before the
+ BOs can be queued to the device.
+
+DRM_IOCTL_QAIC_WAIT_BO
+ This IOCTL allows userspace to determine when a particular BO has been
+ processed by the device. The call will block until either the BO has been
+ processed and can be re-queued to the device, or a timeout occurs.
+
+DRM_IOCTL_QAIC_PERF_STATS_BO
+ This IOCTL allows userspace to collect performance statistics on the most
+ recent execution of a BO. This allows userspace to construct an end to end
+ timeline of the BO processing for a performance analysis.
+
+DRM_IOCTL_QAIC_PART_DEV
+ This IOCTL allows userspace to request a duplicate "shadow device". This extra
+ accelN device is associated with a specific partition of resources on the
+ AIC100 device and can be used for limiting a process to some subset of
+ resources.
+
+Userspace Client Isolation
+==========================
+
+AIC100 supports multiple clients. Multiple DBCs can be consumed by a single
+client, and multiple clients can each consume one or more DBCs. Workloads
+may contain sensitive information therefore only the client that owns the
+workload should be allowed to interface with the DBC.
+
+Clients are identified by the instance associated with their open(). A client
+may only use memory they allocate, and DBCs that are assigned to their
+workloads. Attempts to access resources assigned to other clients will be
+rejected.
+
+Module parameters
+=================
+
+QAIC supports the following module parameters:
+
+**datapath_polling (bool)**
+
+Configures QAIC to use a polling thread for datapath events instead of relying
+on the device interrupts. Useful for platforms with broken multiMSI. Must be
+set at QAIC driver initialization. Default is 0 (off).
+
+**mhi_timeout_ms (unsigned int)**
+
+Sets the timeout value for MHI operations in milliseconds (ms). Must be set
+at the time the driver detects a device. Default is 2000 (2 seconds).
+
+**control_resp_timeout_s (unsigned int)**
+
+Sets the timeout value for QSM responses to NNC messages in seconds (s). Must
+be set at the time the driver is sending a request to QSM. Default is 60 (one
+minute).
+
+**wait_exec_default_timeout_ms (unsigned int)**
+
+Sets the default timeout for the wait_exec ioctl in milliseconds (ms). Must be
+set prior to the waic_exec ioctl call. A value specified in the ioctl call
+overrides this for that call. Default is 5000 (5 seconds).
+
+**datapath_poll_interval_us (unsigned int)**
+
+Sets the polling interval in microseconds (us) when datapath polling is active.
+Takes effect at the next polling interval. Default is 100 (100 us).
--
2.7.4
^ permalink raw reply related [flat|nested] 16+ messages in thread* [PATCH v5 2/8] accel/qaic: Add uapi and core driver file
2023-03-27 15:54 [PATCH v5 0/8] QAIC accel driver Jeffrey Hugo
2023-03-27 15:54 ` [PATCH v5 1/8] accel/qaic: Add documentation for AIC100 accelerator driver Jeffrey Hugo
@ 2023-03-27 15:54 ` Jeffrey Hugo
2023-03-27 15:54 ` [PATCH v5 3/8] accel/qaic: Add MHI controller Jeffrey Hugo
` (6 subsequent siblings)
8 siblings, 0 replies; 16+ messages in thread
From: Jeffrey Hugo @ 2023-03-27 15:54 UTC (permalink / raw)
To: ogabbay, airlied, daniel
Cc: jacek.lawrynowicz, stanislaw.gruszka, dri-devel, quic_pkanojiy,
quic_carlv, quic_ajitpals, linux-doc, linux-arm-msm, bagasdotme,
mani, Jeffrey Hugo
Add the QAIC driver uapi file and core driver file that binds to the PCIe
device. The core driver file also creates the accel device and manages
all the interconnections between the different parts of the driver.
The driver can be built as a module. If so, it will be called "qaic.ko".
Signed-off-by: Jeffrey Hugo <quic_jhugo@quicinc.com>
Reviewed-by: Carl Vanderlip <quic_carlv@quicinc.com>
Reviewed-by: Pranjal Ramajor Asha Kanojiya <quic_pkanojiy@quicinc.com>
Reviewed-by: Stanislaw Gruszka <stanislaw.gruszka@linux.intel.com>
Reviewed-by: Jacek Lawrynowicz <jacek.lawrynowicz@linux.intel.com>
---
drivers/accel/qaic/qaic.h | 282 ++++++++++++++++++
drivers/accel/qaic/qaic_drv.c | 647 ++++++++++++++++++++++++++++++++++++++++++
include/uapi/drm/qaic_accel.h | 397 ++++++++++++++++++++++++++
3 files changed, 1326 insertions(+)
create mode 100644 drivers/accel/qaic/qaic.h
create mode 100644 drivers/accel/qaic/qaic_drv.c
create mode 100644 include/uapi/drm/qaic_accel.h
diff --git a/drivers/accel/qaic/qaic.h b/drivers/accel/qaic/qaic.h
new file mode 100644
index 0000000..f2bd637
--- /dev/null
+++ b/drivers/accel/qaic/qaic.h
@@ -0,0 +1,282 @@
+/* SPDX-License-Identifier: GPL-2.0-only
+ *
+ * Copyright (c) 2019-2021, The Linux Foundation. All rights reserved.
+ * Copyright (c) 2021-2023 Qualcomm Innovation Center, Inc. All rights reserved.
+ */
+
+#ifndef _QAIC_H_
+#define _QAIC_H_
+
+#include <linux/interrupt.h>
+#include <linux/kref.h>
+#include <linux/mhi.h>
+#include <linux/mutex.h>
+#include <linux/pci.h>
+#include <linux/spinlock.h>
+#include <linux/srcu.h>
+#include <linux/wait.h>
+#include <linux/workqueue.h>
+#include <drm/drm_device.h>
+#include <drm/drm_gem.h>
+
+#define QAIC_DBC_BASE SZ_128K
+#define QAIC_DBC_SIZE SZ_4K
+
+#define QAIC_NO_PARTITION -1
+
+#define QAIC_DBC_OFF(i) ((i) * QAIC_DBC_SIZE + QAIC_DBC_BASE)
+
+#define to_qaic_bo(obj) container_of(obj, struct qaic_bo, base)
+
+extern bool datapath_polling;
+
+struct qaic_user {
+ /* Uniquely identifies this user for the device */
+ int handle;
+ struct kref ref_count;
+ /* Char device opened by this user */
+ struct qaic_drm_device *qddev;
+ /* Node in list of users that opened this drm device */
+ struct list_head node;
+ /* SRCU used to synchronize this user during cleanup */
+ struct srcu_struct qddev_lock;
+ atomic_t chunk_id;
+};
+
+struct dma_bridge_chan {
+ /* Pointer to device strcut maintained by driver */
+ struct qaic_device *qdev;
+ /* ID of this DMA bridge channel(DBC) */
+ unsigned int id;
+ /* Synchronizes access to xfer_list */
+ spinlock_t xfer_lock;
+ /* Base address of request queue */
+ void *req_q_base;
+ /* Base address of response queue */
+ void *rsp_q_base;
+ /*
+ * Base bus address of request queue. Response queue bus address can be
+ * calculated by adding request queue size to this variable
+ */
+ dma_addr_t dma_addr;
+ /* Total size of request and response queue in byte */
+ u32 total_size;
+ /* Capacity of request/response queue */
+ u32 nelem;
+ /* The user that opened this DBC */
+ struct qaic_user *usr;
+ /*
+ * Request ID of next memory handle that goes in request queue. One
+ * memory handle can enqueue more than one request elements, all
+ * this requests that belong to same memory handle have same request ID
+ */
+ u16 next_req_id;
+ /* true: DBC is in use; false: DBC not in use */
+ bool in_use;
+ /*
+ * Base address of device registers. Used to read/write request and
+ * response queue's head and tail pointer of this DBC.
+ */
+ void __iomem *dbc_base;
+ /* Head of list where each node is a memory handle queued in request queue */
+ struct list_head xfer_list;
+ /* Synchronizes DBC readers during cleanup */
+ struct srcu_struct ch_lock;
+ /*
+ * When this DBC is released, any thread waiting on this wait queue is
+ * woken up
+ */
+ wait_queue_head_t dbc_release;
+ /* Head of list where each node is a bo associated with this DBC */
+ struct list_head bo_lists;
+ /* The irq line for this DBC. Used for polling */
+ unsigned int irq;
+ /* Polling work item to simulate interrupts */
+ struct work_struct poll_work;
+};
+
+struct qaic_device {
+ /* Pointer to base PCI device struct of our physical device */
+ struct pci_dev *pdev;
+ /* Req. ID of request that will be queued next in MHI control device */
+ u32 next_seq_num;
+ /* Base address of bar 0 */
+ void __iomem *bar_0;
+ /* Base address of bar 2 */
+ void __iomem *bar_2;
+ /* Controller structure for MHI devices */
+ struct mhi_controller *mhi_cntrl;
+ /* MHI control channel device */
+ struct mhi_device *cntl_ch;
+ /* List of requests queued in MHI control device */
+ struct list_head cntl_xfer_list;
+ /* Synchronizes MHI control device transactions and its xfer list */
+ struct mutex cntl_mutex;
+ /* Array of DBC struct of this device */
+ struct dma_bridge_chan *dbc;
+ /* Work queue for tasks related to MHI control device */
+ struct workqueue_struct *cntl_wq;
+ /* Synchronizes all the users of device during cleanup */
+ struct srcu_struct dev_lock;
+ /* true: Device under reset; false: Device not under reset */
+ bool in_reset;
+ /*
+ * true: A tx MHI transaction has failed and a rx buffer is still queued
+ * in control device. Such a buffer is considered lost rx buffer
+ * false: No rx buffer is lost in control device
+ */
+ bool cntl_lost_buf;
+ /* Maximum number of DBC supported by this device */
+ u32 num_dbc;
+ /* Reference to the drm_device for this device when it is created */
+ struct qaic_drm_device *qddev;
+ /* Generate the CRC of a control message */
+ u32 (*gen_crc)(void *msg);
+ /* Validate the CRC of a control message */
+ bool (*valid_crc)(void *msg);
+};
+
+struct qaic_drm_device {
+ /* Pointer to the root device struct driven by this driver */
+ struct qaic_device *qdev;
+ /*
+ * The physical device can be partition in number of logical devices.
+ * And each logical device is given a partition id. This member stores
+ * that id. QAIC_NO_PARTITION is a sentinel used to mark that this drm
+ * device is the actual physical device
+ */
+ s32 partition_id;
+ /* Pointer to the drm device struct of this drm device */
+ struct drm_device *ddev;
+ /* Head in list of users who have opened this drm device */
+ struct list_head users;
+ /* Synchronizes access to users list */
+ struct mutex users_mutex;
+};
+
+struct qaic_bo {
+ struct drm_gem_object base;
+ /* Scatter/gather table for allocate/imported BO */
+ struct sg_table *sgt;
+ /* BO size requested by user. GEM object might be bigger in size. */
+ u64 size;
+ /* Head in list of slices of this BO */
+ struct list_head slices;
+ /* Total nents, for all slices of this BO */
+ int total_slice_nents;
+ /*
+ * Direction of transfer. It can assume only two value DMA_TO_DEVICE and
+ * DMA_FROM_DEVICE.
+ */
+ int dir;
+ /* The pointer of the DBC which operates on this BO */
+ struct dma_bridge_chan *dbc;
+ /* Number of slice that belongs to this buffer */
+ u32 nr_slice;
+ /* Number of slice that have been transferred by DMA engine */
+ u32 nr_slice_xfer_done;
+ /* true = BO is queued for execution, true = BO is not queued */
+ bool queued;
+ /*
+ * If true then user has attached slicing information to this BO by
+ * calling DRM_IOCTL_QAIC_ATTACH_SLICE_BO ioctl.
+ */
+ bool sliced;
+ /* Request ID of this BO if it is queued for execution */
+ u16 req_id;
+ /* Handle assigned to this BO */
+ u32 handle;
+ /* Wait on this for completion of DMA transfer of this BO */
+ struct completion xfer_done;
+ /*
+ * Node in linked list where head is dbc->xfer_list.
+ * This link list contain BO's that are queued for DMA transfer.
+ */
+ struct list_head xfer_list;
+ /*
+ * Node in linked list where head is dbc->bo_lists.
+ * This link list contain BO's that are associated with the DBC it is
+ * linked to.
+ */
+ struct list_head bo_list;
+ struct {
+ /*
+ * Latest timestamp(ns) at which kernel received a request to
+ * execute this BO
+ */
+ u64 req_received_ts;
+ /*
+ * Latest timestamp(ns) at which kernel enqueued requests of
+ * this BO for execution in DMA queue
+ */
+ u64 req_submit_ts;
+ /*
+ * Latest timestamp(ns) at which kernel received a completion
+ * interrupt for requests of this BO
+ */
+ u64 req_processed_ts;
+ /*
+ * Number of elements already enqueued in DMA queue before
+ * enqueuing requests of this BO
+ */
+ u32 queue_level_before;
+ } perf_stats;
+
+};
+
+struct bo_slice {
+ /* Mapped pages */
+ struct sg_table *sgt;
+ /* Number of requests required to queue in DMA queue */
+ int nents;
+ /* See enum dma_data_direction */
+ int dir;
+ /* Actual requests that will be copied in DMA queue */
+ struct dbc_req *reqs;
+ struct kref ref_count;
+ /* true: No DMA transfer required */
+ bool no_xfer;
+ /* Pointer to the parent BO handle */
+ struct qaic_bo *bo;
+ /* Node in list of slices maintained by parent BO */
+ struct list_head slice;
+ /* Size of this slice in bytes */
+ u64 size;
+ /* Offset of this slice in buffer */
+ u64 offset;
+};
+
+int get_dbc_req_elem_size(void);
+int get_dbc_rsp_elem_size(void);
+int get_cntl_version(struct qaic_device *qdev, struct qaic_user *usr, u16 *major, u16 *minor);
+int qaic_manage_ioctl(struct drm_device *dev, void *data, struct drm_file *file_priv);
+void qaic_mhi_ul_xfer_cb(struct mhi_device *mhi_dev, struct mhi_result *mhi_result);
+
+void qaic_mhi_dl_xfer_cb(struct mhi_device *mhi_dev, struct mhi_result *mhi_result);
+
+int qaic_control_open(struct qaic_device *qdev);
+void qaic_control_close(struct qaic_device *qdev);
+void qaic_release_usr(struct qaic_device *qdev, struct qaic_user *usr);
+
+irqreturn_t dbc_irq_threaded_fn(int irq, void *data);
+irqreturn_t dbc_irq_handler(int irq, void *data);
+int disable_dbc(struct qaic_device *qdev, u32 dbc_id, struct qaic_user *usr);
+void enable_dbc(struct qaic_device *qdev, u32 dbc_id, struct qaic_user *usr);
+void wakeup_dbc(struct qaic_device *qdev, u32 dbc_id);
+void release_dbc(struct qaic_device *qdev, u32 dbc_id);
+
+void wake_all_cntl(struct qaic_device *qdev);
+void qaic_dev_reset_clean_local_state(struct qaic_device *qdev, bool exit_reset);
+
+struct drm_gem_object *qaic_gem_prime_import(struct drm_device *dev, struct dma_buf *dma_buf);
+
+int qaic_create_bo_ioctl(struct drm_device *dev, void *data, struct drm_file *file_priv);
+int qaic_mmap_bo_ioctl(struct drm_device *dev, void *data, struct drm_file *file_priv);
+int qaic_attach_slice_bo_ioctl(struct drm_device *dev, void *data, struct drm_file *file_priv);
+int qaic_execute_bo_ioctl(struct drm_device *dev, void *data, struct drm_file *file_priv);
+int qaic_partial_execute_bo_ioctl(struct drm_device *dev, void *data, struct drm_file *file_priv);
+int qaic_wait_bo_ioctl(struct drm_device *dev, void *data, struct drm_file *file_priv);
+int qaic_perf_stats_bo_ioctl(struct drm_device *dev, void *data, struct drm_file *file_priv);
+void irq_polling_work(struct work_struct *work);
+
+#endif /* _QAIC_H_ */
diff --git a/drivers/accel/qaic/qaic_drv.c b/drivers/accel/qaic/qaic_drv.c
new file mode 100644
index 0000000..1106ad8
--- /dev/null
+++ b/drivers/accel/qaic/qaic_drv.c
@@ -0,0 +1,647 @@
+// SPDX-License-Identifier: GPL-2.0-only
+
+/* Copyright (c) 2019-2021, The Linux Foundation. All rights reserved. */
+/* Copyright (c) 2021-2023 Qualcomm Innovation Center, Inc. All rights reserved. */
+
+#include <linux/delay.h>
+#include <linux/dma-mapping.h>
+#include <linux/idr.h>
+#include <linux/interrupt.h>
+#include <linux/list.h>
+#include <linux/kref.h>
+#include <linux/mhi.h>
+#include <linux/module.h>
+#include <linux/msi.h>
+#include <linux/mutex.h>
+#include <linux/pci.h>
+#include <linux/spinlock.h>
+#include <linux/workqueue.h>
+#include <linux/wait.h>
+#include <drm/drm_accel.h>
+#include <drm/drm_drv.h>
+#include <drm/drm_file.h>
+#include <drm/drm_gem.h>
+#include <drm/drm_ioctl.h>
+#include <uapi/drm/qaic_accel.h>
+
+#include "mhi_controller.h"
+#include "mhi_qaic_ctrl.h"
+#include "qaic.h"
+
+MODULE_IMPORT_NS(DMA_BUF);
+
+#define PCI_DEV_AIC100 0xa100
+#define QAIC_NAME "qaic"
+#define QAIC_DESC "Qualcomm Cloud AI Accelerators"
+#define CNTL_MAJOR 5
+#define CNTL_MINOR 0
+
+bool datapath_polling;
+module_param(datapath_polling, bool, 0400);
+MODULE_PARM_DESC(datapath_polling, "Operate the datapath in polling mode");
+static bool link_up;
+static DEFINE_IDA(qaic_usrs);
+
+static int qaic_create_drm_device(struct qaic_device *qdev, s32 partition_id);
+static void qaic_destroy_drm_device(struct qaic_device *qdev, s32 partition_id);
+
+static void free_usr(struct kref *kref)
+{
+ struct qaic_user *usr = container_of(kref, struct qaic_user, ref_count);
+
+ cleanup_srcu_struct(&usr->qddev_lock);
+ ida_free(&qaic_usrs, usr->handle);
+ kfree(usr);
+}
+
+static int qaic_open(struct drm_device *dev, struct drm_file *file)
+{
+ struct qaic_drm_device *qddev = dev->dev_private;
+ struct qaic_device *qdev = qddev->qdev;
+ struct qaic_user *usr;
+ int rcu_id;
+ int ret;
+
+ rcu_id = srcu_read_lock(&qdev->dev_lock);
+ if (qdev->in_reset) {
+ ret = -ENODEV;
+ goto dev_unlock;
+ }
+
+ usr = kmalloc(sizeof(*usr), GFP_KERNEL);
+ if (!usr) {
+ ret = -ENOMEM;
+ goto dev_unlock;
+ }
+
+ usr->handle = ida_alloc(&qaic_usrs, GFP_KERNEL);
+ if (usr->handle < 0) {
+ ret = usr->handle;
+ goto free_usr;
+ }
+ usr->qddev = qddev;
+ atomic_set(&usr->chunk_id, 0);
+ init_srcu_struct(&usr->qddev_lock);
+ kref_init(&usr->ref_count);
+
+ ret = mutex_lock_interruptible(&qddev->users_mutex);
+ if (ret)
+ goto cleanup_usr;
+
+ list_add(&usr->node, &qddev->users);
+ mutex_unlock(&qddev->users_mutex);
+
+ file->driver_priv = usr;
+
+ srcu_read_unlock(&qdev->dev_lock, rcu_id);
+ return 0;
+
+cleanup_usr:
+ cleanup_srcu_struct(&usr->qddev_lock);
+free_usr:
+ kfree(usr);
+dev_unlock:
+ srcu_read_unlock(&qdev->dev_lock, rcu_id);
+ return ret;
+}
+
+static void qaic_postclose(struct drm_device *dev, struct drm_file *file)
+{
+ struct qaic_user *usr = file->driver_priv;
+ struct qaic_drm_device *qddev;
+ struct qaic_device *qdev;
+ int qdev_rcu_id;
+ int usr_rcu_id;
+ int i;
+
+ qddev = usr->qddev;
+ usr_rcu_id = srcu_read_lock(&usr->qddev_lock);
+ if (qddev) {
+ qdev = qddev->qdev;
+ qdev_rcu_id = srcu_read_lock(&qdev->dev_lock);
+ if (!qdev->in_reset) {
+ qaic_release_usr(qdev, usr);
+ for (i = 0; i < qdev->num_dbc; ++i)
+ if (qdev->dbc[i].usr && qdev->dbc[i].usr->handle == usr->handle)
+ release_dbc(qdev, i);
+ }
+ srcu_read_unlock(&qdev->dev_lock, qdev_rcu_id);
+
+ mutex_lock(&qddev->users_mutex);
+ if (!list_empty(&usr->node))
+ list_del_init(&usr->node);
+ mutex_unlock(&qddev->users_mutex);
+ }
+
+ srcu_read_unlock(&usr->qddev_lock, usr_rcu_id);
+ kref_put(&usr->ref_count, free_usr);
+
+ file->driver_priv = NULL;
+}
+
+DEFINE_DRM_ACCEL_FOPS(qaic_accel_fops);
+
+static const struct drm_ioctl_desc qaic_drm_ioctls[] = {
+ DRM_IOCTL_DEF_DRV(QAIC_MANAGE, qaic_manage_ioctl, 0),
+ DRM_IOCTL_DEF_DRV(QAIC_CREATE_BO, qaic_create_bo_ioctl, 0),
+ DRM_IOCTL_DEF_DRV(QAIC_MMAP_BO, qaic_mmap_bo_ioctl, 0),
+ DRM_IOCTL_DEF_DRV(QAIC_ATTACH_SLICE_BO, qaic_attach_slice_bo_ioctl, 0),
+ DRM_IOCTL_DEF_DRV(QAIC_EXECUTE_BO, qaic_execute_bo_ioctl, 0),
+ DRM_IOCTL_DEF_DRV(QAIC_PARTIAL_EXECUTE_BO, qaic_partial_execute_bo_ioctl, 0),
+ DRM_IOCTL_DEF_DRV(QAIC_WAIT_BO, qaic_wait_bo_ioctl, 0),
+ DRM_IOCTL_DEF_DRV(QAIC_PERF_STATS_BO, qaic_perf_stats_bo_ioctl, 0),
+};
+
+static const struct drm_driver qaic_accel_driver = {
+ .driver_features = DRIVER_GEM | DRIVER_COMPUTE_ACCEL,
+
+ .name = QAIC_NAME,
+ .desc = QAIC_DESC,
+ .date = "20190618",
+
+ .fops = &qaic_accel_fops,
+ .open = qaic_open,
+ .postclose = qaic_postclose,
+
+ .ioctls = qaic_drm_ioctls,
+ .num_ioctls = ARRAY_SIZE(qaic_drm_ioctls),
+ .prime_fd_to_handle = drm_gem_prime_fd_to_handle,
+ .gem_prime_import = qaic_gem_prime_import,
+};
+
+static int qaic_create_drm_device(struct qaic_device *qdev, s32 partition_id)
+{
+ struct qaic_drm_device *qddev;
+ struct drm_device *ddev;
+ struct device *pdev;
+ int ret;
+
+ /* Hold off implementing partitions until the uapi is determined */
+ if (partition_id != QAIC_NO_PARTITION)
+ return -EINVAL;
+
+ pdev = &qdev->pdev->dev;
+
+ qddev = kzalloc(sizeof(*qddev), GFP_KERNEL);
+ if (!qddev)
+ return -ENOMEM;
+
+ ddev = drm_dev_alloc(&qaic_accel_driver, pdev);
+ if (IS_ERR(ddev)) {
+ ret = PTR_ERR(ddev);
+ goto ddev_fail;
+ }
+
+ ddev->dev_private = qddev;
+ qddev->ddev = ddev;
+
+ qddev->qdev = qdev;
+ qddev->partition_id = partition_id;
+ INIT_LIST_HEAD(&qddev->users);
+ mutex_init(&qddev->users_mutex);
+
+ qdev->qddev = qddev;
+
+ ret = drm_dev_register(ddev, 0);
+ if (ret) {
+ pci_dbg(qdev->pdev, "%s: drm_dev_register failed %d\n", __func__, ret);
+ goto drm_reg_fail;
+ }
+
+ return 0;
+
+drm_reg_fail:
+ mutex_destroy(&qddev->users_mutex);
+ qdev->qddev = NULL;
+ drm_dev_put(ddev);
+ddev_fail:
+ kfree(qddev);
+ return ret;
+}
+
+static void qaic_destroy_drm_device(struct qaic_device *qdev, s32 partition_id)
+{
+ struct qaic_drm_device *qddev;
+ struct qaic_user *usr;
+
+ qddev = qdev->qddev;
+
+ /*
+ * Existing users get unresolvable errors till they close FDs.
+ * Need to sync carefully with users calling close(). The
+ * list of users can be modified elsewhere when the lock isn't
+ * held here, but the sync'ing the srcu with the mutex held
+ * could deadlock. Grab the mutex so that the list will be
+ * unmodified. The user we get will exist as long as the
+ * lock is held. Signal that the qcdev is going away, and
+ * grab a reference to the user so they don't go away for
+ * synchronize_srcu(). Then release the mutex to avoid
+ * deadlock and make sure the user has observed the signal.
+ * With the lock released, we cannot maintain any state of the
+ * user list.
+ */
+ mutex_lock(&qddev->users_mutex);
+ while (!list_empty(&qddev->users)) {
+ usr = list_first_entry(&qddev->users, struct qaic_user, node);
+ list_del_init(&usr->node);
+ kref_get(&usr->ref_count);
+ usr->qddev = NULL;
+ mutex_unlock(&qddev->users_mutex);
+ synchronize_srcu(&usr->qddev_lock);
+ kref_put(&usr->ref_count, free_usr);
+ mutex_lock(&qddev->users_mutex);
+ }
+ mutex_unlock(&qddev->users_mutex);
+
+ if (qddev->ddev) {
+ drm_dev_unregister(qddev->ddev);
+ drm_dev_put(qddev->ddev);
+ }
+
+ kfree(qddev);
+}
+
+static int qaic_mhi_probe(struct mhi_device *mhi_dev, const struct mhi_device_id *id)
+{
+ struct qaic_device *qdev;
+ u16 major, minor;
+ int ret;
+
+ /*
+ * Invoking this function indicates that the control channel to the
+ * device is available. We use that as a signal to indicate that
+ * the device side firmware has booted. The device side firmware
+ * manages the device resources, so we need to communicate with it
+ * via the control channel in order to utilize the device. Therefore
+ * we wait until this signal to create the drm dev that userspace will
+ * use to control the device, because without the device side firmware,
+ * userspace can't do anything useful.
+ */
+
+ qdev = pci_get_drvdata(to_pci_dev(mhi_dev->mhi_cntrl->cntrl_dev));
+
+ qdev->in_reset = false;
+
+ dev_set_drvdata(&mhi_dev->dev, qdev);
+ qdev->cntl_ch = mhi_dev;
+
+ ret = qaic_control_open(qdev);
+ if (ret) {
+ pci_dbg(qdev->pdev, "%s: control_open failed %d\n", __func__, ret);
+ return ret;
+ }
+
+ ret = get_cntl_version(qdev, NULL, &major, &minor);
+ if (ret || major != CNTL_MAJOR || minor > CNTL_MINOR) {
+ pci_err(qdev->pdev, "%s: Control protocol version (%d.%d) not supported. Supported version is (%d.%d). Ret: %d\n",
+ __func__, major, minor, CNTL_MAJOR, CNTL_MINOR, ret);
+ ret = -EINVAL;
+ goto close_control;
+ }
+
+ ret = qaic_create_drm_device(qdev, QAIC_NO_PARTITION);
+
+ return ret;
+
+close_control:
+ qaic_control_close(qdev);
+ return ret;
+}
+
+static void qaic_mhi_remove(struct mhi_device *mhi_dev)
+{
+/* This is redundant since we have already observed the device crash */
+}
+
+static void qaic_notify_reset(struct qaic_device *qdev)
+{
+ int i;
+
+ qdev->in_reset = true;
+ /* wake up any waiters to avoid waiting for timeouts at sync */
+ wake_all_cntl(qdev);
+ for (i = 0; i < qdev->num_dbc; ++i)
+ wakeup_dbc(qdev, i);
+ synchronize_srcu(&qdev->dev_lock);
+}
+
+void qaic_dev_reset_clean_local_state(struct qaic_device *qdev, bool exit_reset)
+{
+ int i;
+
+ qaic_notify_reset(qdev);
+
+ /* remove drmdevs to prevent new users from coming in */
+ qaic_destroy_drm_device(qdev, QAIC_NO_PARTITION);
+
+ /* start tearing things down */
+ for (i = 0; i < qdev->num_dbc; ++i)
+ release_dbc(qdev, i);
+
+ if (exit_reset)
+ qdev->in_reset = false;
+}
+
+static struct qaic_device *create_qdev(struct pci_dev *pdev, const struct pci_device_id *id)
+{
+ struct qaic_device *qdev;
+ int i;
+
+ qdev = devm_kzalloc(&pdev->dev, sizeof(*qdev), GFP_KERNEL);
+ if (!qdev)
+ return NULL;
+
+ if (id->device == PCI_DEV_AIC100) {
+ qdev->num_dbc = 16;
+ qdev->dbc = devm_kcalloc(&pdev->dev, qdev->num_dbc, sizeof(*qdev->dbc), GFP_KERNEL);
+ if (!qdev->dbc)
+ return NULL;
+ }
+
+ qdev->cntl_wq = alloc_workqueue("qaic_cntl", WQ_UNBOUND, 0);
+ if (!qdev->cntl_wq)
+ return NULL;
+
+ pci_set_drvdata(pdev, qdev);
+ qdev->pdev = pdev;
+
+ mutex_init(&qdev->cntl_mutex);
+ INIT_LIST_HEAD(&qdev->cntl_xfer_list);
+ init_srcu_struct(&qdev->dev_lock);
+
+ for (i = 0; i < qdev->num_dbc; ++i) {
+ spin_lock_init(&qdev->dbc[i].xfer_lock);
+ qdev->dbc[i].qdev = qdev;
+ qdev->dbc[i].id = i;
+ INIT_LIST_HEAD(&qdev->dbc[i].xfer_list);
+ init_srcu_struct(&qdev->dbc[i].ch_lock);
+ init_waitqueue_head(&qdev->dbc[i].dbc_release);
+ INIT_LIST_HEAD(&qdev->dbc[i].bo_lists);
+ }
+
+ return qdev;
+}
+
+static void cleanup_qdev(struct qaic_device *qdev)
+{
+ int i;
+
+ for (i = 0; i < qdev->num_dbc; ++i)
+ cleanup_srcu_struct(&qdev->dbc[i].ch_lock);
+ cleanup_srcu_struct(&qdev->dev_lock);
+ pci_set_drvdata(qdev->pdev, NULL);
+ destroy_workqueue(qdev->cntl_wq);
+}
+
+static int init_pci(struct qaic_device *qdev, struct pci_dev *pdev)
+{
+ int bars;
+ int ret;
+
+ bars = pci_select_bars(pdev, IORESOURCE_MEM);
+
+ /* make sure the device has the expected BARs */
+ if (bars != (BIT(0) | BIT(2) | BIT(4))) {
+ pci_dbg(pdev, "%s: expected BARs 0, 2, and 4 not found in device. Found 0x%x\n",
+ __func__, bars);
+ return -EINVAL;
+ }
+
+ ret = pcim_enable_device(pdev);
+ if (ret)
+ return ret;
+
+ ret = dma_set_mask_and_coherent(&pdev->dev, DMA_BIT_MASK(64));
+ if (ret)
+ return ret;
+ ret = dma_set_max_seg_size(&pdev->dev, UINT_MAX);
+ if (ret)
+ return ret;
+
+ qdev->bar_0 = devm_ioremap_resource(&pdev->dev, &pdev->resource[0]);
+ if (IS_ERR(qdev->bar_0))
+ return PTR_ERR(qdev->bar_0);
+
+ qdev->bar_2 = devm_ioremap_resource(&pdev->dev, &pdev->resource[2]);
+ if (IS_ERR(qdev->bar_2))
+ return PTR_ERR(qdev->bar_2);
+
+ /* Managed release since we use pcim_enable_device above */
+ pci_set_master(pdev);
+
+ return 0;
+}
+
+static int init_msi(struct qaic_device *qdev, struct pci_dev *pdev)
+{
+ int mhi_irq;
+ int ret;
+ int i;
+
+ /* Managed release since we use pcim_enable_device */
+ ret = pci_alloc_irq_vectors(pdev, 1, 32, PCI_IRQ_MSI);
+ if (ret < 0)
+ return ret;
+
+ if (ret < 32) {
+ pci_err(pdev, "%s: Requested 32 MSIs. Obtained %d MSIs which is less than the 32 required.\n",
+ __func__, ret);
+ return -ENODEV;
+ }
+
+ mhi_irq = pci_irq_vector(pdev, 0);
+ if (mhi_irq < 0)
+ return mhi_irq;
+
+ for (i = 0; i < qdev->num_dbc; ++i) {
+ ret = devm_request_threaded_irq(&pdev->dev, pci_irq_vector(pdev, i + 1),
+ dbc_irq_handler, dbc_irq_threaded_fn, IRQF_SHARED,
+ "qaic_dbc", &qdev->dbc[i]);
+ if (ret)
+ return ret;
+
+ if (datapath_polling) {
+ qdev->dbc[i].irq = pci_irq_vector(pdev, i + 1);
+ disable_irq_nosync(qdev->dbc[i].irq);
+ INIT_WORK(&qdev->dbc[i].poll_work, irq_polling_work);
+ }
+ }
+
+ return mhi_irq;
+}
+
+static int qaic_pci_probe(struct pci_dev *pdev, const struct pci_device_id *id)
+{
+ struct qaic_device *qdev;
+ int mhi_irq;
+ int ret;
+ int i;
+
+ qdev = create_qdev(pdev, id);
+ if (!qdev)
+ return -ENOMEM;
+
+ ret = init_pci(qdev, pdev);
+ if (ret)
+ goto cleanup_qdev;
+
+ for (i = 0; i < qdev->num_dbc; ++i)
+ qdev->dbc[i].dbc_base = qdev->bar_2 + QAIC_DBC_OFF(i);
+
+ mhi_irq = init_msi(qdev, pdev);
+ if (mhi_irq < 0) {
+ ret = mhi_irq;
+ goto cleanup_qdev;
+ }
+
+ qdev->mhi_cntrl = qaic_mhi_register_controller(pdev, qdev->bar_0, mhi_irq);
+ if (IS_ERR(qdev->mhi_cntrl)) {
+ ret = PTR_ERR(qdev->mhi_cntrl);
+ goto cleanup_qdev;
+ }
+
+ return 0;
+
+cleanup_qdev:
+ cleanup_qdev(qdev);
+ return ret;
+}
+
+static void qaic_pci_remove(struct pci_dev *pdev)
+{
+ struct qaic_device *qdev = pci_get_drvdata(pdev);
+
+ if (!qdev)
+ return;
+
+ qaic_dev_reset_clean_local_state(qdev, false);
+ qaic_mhi_free_controller(qdev->mhi_cntrl, link_up);
+ cleanup_qdev(qdev);
+}
+
+static void qaic_pci_shutdown(struct pci_dev *pdev)
+{
+ /* see qaic_exit for what link_up is doing */
+ link_up = true;
+ qaic_pci_remove(pdev);
+}
+
+static pci_ers_result_t qaic_pci_error_detected(struct pci_dev *pdev, pci_channel_state_t error)
+{
+ return PCI_ERS_RESULT_NEED_RESET;
+}
+
+static void qaic_pci_reset_prepare(struct pci_dev *pdev)
+{
+ struct qaic_device *qdev = pci_get_drvdata(pdev);
+
+ qaic_notify_reset(qdev);
+ qaic_mhi_start_reset(qdev->mhi_cntrl);
+ qaic_dev_reset_clean_local_state(qdev, false);
+}
+
+static void qaic_pci_reset_done(struct pci_dev *pdev)
+{
+ struct qaic_device *qdev = pci_get_drvdata(pdev);
+
+ qdev->in_reset = false;
+ qaic_mhi_reset_done(qdev->mhi_cntrl);
+}
+
+static const struct mhi_device_id qaic_mhi_match_table[] = {
+ { .chan = "QAIC_CONTROL", },
+ {},
+};
+
+static struct mhi_driver qaic_mhi_driver = {
+ .id_table = qaic_mhi_match_table,
+ .remove = qaic_mhi_remove,
+ .probe = qaic_mhi_probe,
+ .ul_xfer_cb = qaic_mhi_ul_xfer_cb,
+ .dl_xfer_cb = qaic_mhi_dl_xfer_cb,
+ .driver = {
+ .name = "qaic_mhi",
+ },
+};
+
+static const struct pci_device_id qaic_ids[] = {
+ { PCI_DEVICE(PCI_VENDOR_ID_QCOM, PCI_DEV_AIC100), },
+ { }
+};
+MODULE_DEVICE_TABLE(pci, qaic_ids);
+
+static const struct pci_error_handlers qaic_pci_err_handler = {
+ .error_detected = qaic_pci_error_detected,
+ .reset_prepare = qaic_pci_reset_prepare,
+ .reset_done = qaic_pci_reset_done,
+};
+
+static struct pci_driver qaic_pci_driver = {
+ .name = QAIC_NAME,
+ .id_table = qaic_ids,
+ .probe = qaic_pci_probe,
+ .remove = qaic_pci_remove,
+ .shutdown = qaic_pci_shutdown,
+ .err_handler = &qaic_pci_err_handler,
+};
+
+static int __init qaic_init(void)
+{
+ int ret;
+
+ ret = mhi_driver_register(&qaic_mhi_driver);
+ if (ret) {
+ pr_debug("qaic: mhi_driver_register failed %d\n", ret);
+ return ret;
+ }
+
+ ret = pci_register_driver(&qaic_pci_driver);
+ if (ret) {
+ pr_debug("qaic: pci_register_driver failed %d\n", ret);
+ goto free_mhi;
+ }
+
+ ret = mhi_qaic_ctrl_init();
+ if (ret) {
+ pr_debug("qaic: mhi_qaic_ctrl_init failed %d\n", ret);
+ goto free_pci;
+ }
+
+ return 0;
+
+free_pci:
+ pci_unregister_driver(&qaic_pci_driver);
+free_mhi:
+ mhi_driver_unregister(&qaic_mhi_driver);
+ return ret;
+}
+
+static void __exit qaic_exit(void)
+{
+ /*
+ * We assume that qaic_pci_remove() is called due to a hotplug event
+ * which would mean that the link is down, and thus
+ * qaic_mhi_free_controller() should not try to access the device during
+ * cleanup.
+ * We call pci_unregister_driver() below, which also triggers
+ * qaic_pci_remove(), but since this is module exit, we expect the link
+ * to the device to be up, in which case qaic_mhi_free_controller()
+ * should try to access the device during cleanup to put the device in
+ * a sane state.
+ * For that reason, we set link_up here to let qaic_mhi_free_controller
+ * know the expected link state. Since the module is going to be
+ * removed at the end of this, we don't need to worry about
+ * reinitializing the link_up state after the cleanup is done.
+ */
+ link_up = true;
+ mhi_qaic_ctrl_deinit();
+ pci_unregister_driver(&qaic_pci_driver);
+ mhi_driver_unregister(&qaic_mhi_driver);
+}
+
+module_init(qaic_init);
+module_exit(qaic_exit);
+
+MODULE_AUTHOR(QAIC_DESC " Kernel Driver Team");
+MODULE_DESCRIPTION(QAIC_DESC " Accel Driver");
+MODULE_LICENSE("GPL");
diff --git a/include/uapi/drm/qaic_accel.h b/include/uapi/drm/qaic_accel.h
new file mode 100644
index 0000000..2d34874
--- /dev/null
+++ b/include/uapi/drm/qaic_accel.h
@@ -0,0 +1,397 @@
+/* SPDX-License-Identifier: GPL-2.0 WITH Linux-syscall-note
+ *
+ * Copyright (c) 2019-2020, The Linux Foundation. All rights reserved.
+ * Copyright (c) 2021-2023 Qualcomm Innovation Center, Inc. All rights reserved.
+ */
+
+#ifndef QAIC_ACCEL_H_
+#define QAIC_ACCEL_H_
+
+#include "drm.h"
+
+#if defined(__cplusplus)
+extern "C" {
+#endif
+
+/* The length(4K) includes len and count fields of qaic_manage_msg */
+#define QAIC_MANAGE_MAX_MSG_LENGTH SZ_4K
+
+/* semaphore flags */
+#define QAIC_SEM_INSYNCFENCE 2
+#define QAIC_SEM_OUTSYNCFENCE 1
+
+/* Semaphore commands */
+#define QAIC_SEM_NOP 0
+#define QAIC_SEM_INIT 1
+#define QAIC_SEM_INC 2
+#define QAIC_SEM_DEC 3
+#define QAIC_SEM_WAIT_EQUAL 4
+#define QAIC_SEM_WAIT_GT_EQ 5 /* Greater than or equal */
+#define QAIC_SEM_WAIT_GT_0 6 /* Greater than 0 */
+
+#define QAIC_TRANS_UNDEFINED 0
+#define QAIC_TRANS_PASSTHROUGH_FROM_USR 1
+#define QAIC_TRANS_PASSTHROUGH_TO_USR 2
+#define QAIC_TRANS_PASSTHROUGH_FROM_DEV 3
+#define QAIC_TRANS_PASSTHROUGH_TO_DEV 4
+#define QAIC_TRANS_DMA_XFER_FROM_USR 5
+#define QAIC_TRANS_DMA_XFER_TO_DEV 6
+#define QAIC_TRANS_ACTIVATE_FROM_USR 7
+#define QAIC_TRANS_ACTIVATE_FROM_DEV 8
+#define QAIC_TRANS_ACTIVATE_TO_DEV 9
+#define QAIC_TRANS_DEACTIVATE_FROM_USR 10
+#define QAIC_TRANS_DEACTIVATE_FROM_DEV 11
+#define QAIC_TRANS_STATUS_FROM_USR 12
+#define QAIC_TRANS_STATUS_TO_USR 13
+#define QAIC_TRANS_STATUS_FROM_DEV 14
+#define QAIC_TRANS_STATUS_TO_DEV 15
+#define QAIC_TRANS_TERMINATE_FROM_DEV 16
+#define QAIC_TRANS_TERMINATE_TO_DEV 17
+#define QAIC_TRANS_DMA_XFER_CONT 18
+#define QAIC_TRANS_VALIDATE_PARTITION_FROM_DEV 19
+#define QAIC_TRANS_VALIDATE_PARTITION_TO_DEV 20
+
+/**
+ * struct qaic_manage_trans_hdr - Header for a transaction in a manage message.
+ * @type: In. Identifies this transaction. See QAIC_TRANS_* defines.
+ * @len: In. Length of this transaction, including this header.
+ */
+struct qaic_manage_trans_hdr {
+ __u32 type;
+ __u32 len;
+};
+
+/**
+ * struct qaic_manage_trans_passthrough - Defines a passthrough transaction.
+ * @hdr: In. Header to identify this transaction.
+ * @data: In. Payload of this ransaction. Opaque to the driver. Userspace must
+ * encode in little endian and align/pad to 64-bit.
+ */
+struct qaic_manage_trans_passthrough {
+ struct qaic_manage_trans_hdr hdr;
+ __u8 data[];
+};
+
+/**
+ * struct qaic_manage_trans_dma_xfer - Defines a DMA transfer transaction.
+ * @hdr: In. Header to identify this transaction.
+ * @tag: In. Identified this transfer in other transactions. Opaque to the
+ * driver.
+ * @pad: Structure padding.
+ * @addr: In. Address of the data to DMA to the device.
+ * @size: In. Length of the data to DMA to the device.
+ */
+struct qaic_manage_trans_dma_xfer {
+ struct qaic_manage_trans_hdr hdr;
+ __u32 tag;
+ __u32 pad;
+ __u64 addr;
+ __u64 size;
+};
+
+/**
+ * struct qaic_manage_trans_activate_to_dev - Defines an activate request.
+ * @hdr: In. Header to identify this transaction.
+ * @queue_size: In. Number of elements for DBC request and response queues.
+ * @eventfd: Unused.
+ * @options: In. Device specific options for this activate.
+ * @pad: Structure padding. Must be 0.
+ */
+struct qaic_manage_trans_activate_to_dev {
+ struct qaic_manage_trans_hdr hdr;
+ __u32 queue_size;
+ __u32 eventfd;
+ __u32 options;
+ __u32 pad;
+};
+
+/**
+ * struct qaic_manage_trans_activate_from_dev - Defines an activate response.
+ * @hdr: Out. Header to identify this transaction.
+ * @status: Out. Return code of the request from the device.
+ * @dbc_id: Out. Id of the assigned DBC for successful request.
+ * @options: Out. Device specific options for this activate.
+ */
+struct qaic_manage_trans_activate_from_dev {
+ struct qaic_manage_trans_hdr hdr;
+ __u32 status;
+ __u32 dbc_id;
+ __u64 options;
+};
+
+/**
+ * struct qaic_manage_trans_deactivate - Defines a deactivate request.
+ * @hdr: In. Header to identify this transaction.
+ * @dbc_id: In. Id of assigned DBC.
+ * @pad: Structure padding. Must be 0.
+ */
+struct qaic_manage_trans_deactivate {
+ struct qaic_manage_trans_hdr hdr;
+ __u32 dbc_id;
+ __u32 pad;
+};
+
+/**
+ * struct qaic_manage_trans_status_to_dev - Defines a status request.
+ * @hdr: In. Header to identify this transaction.
+ */
+struct qaic_manage_trans_status_to_dev {
+ struct qaic_manage_trans_hdr hdr;
+};
+
+/**
+ * struct qaic_manage_trans_status_from_dev - Defines a status response.
+ * @hdr: Out. Header to identify this transaction.
+ * @major: Out. NNC protocol version major number.
+ * @minor: Out. NNC protocol version minor number.
+ * @status: Out. Return code from device.
+ * @status_flags: Out. Flags from device. Bit 0 indicates if CRCs are required.
+ */
+struct qaic_manage_trans_status_from_dev {
+ struct qaic_manage_trans_hdr hdr;
+ __u16 major;
+ __u16 minor;
+ __u32 status;
+ __u64 status_flags;
+};
+
+/**
+ * struct qaic_manage_msg - Defines a message to the device.
+ * @len: In. Length of all the transactions contained within this message.
+ * @count: In. Number of transactions in this message.
+ * @data: In. Address to an array where the transactions can be found.
+ */
+struct qaic_manage_msg {
+ __u32 len;
+ __u32 count;
+ __u64 data;
+};
+
+/**
+ * struct qaic_create_bo - Defines a request to create a buffer object.
+ * @size: In. Size of the buffer in bytes.
+ * @handle: Out. GEM handle for the BO.
+ * @pad: Structure padding. Must be 0.
+ */
+struct qaic_create_bo {
+ __u64 size;
+ __u32 handle;
+ __u32 pad;
+};
+
+/**
+ * struct qaic_mmap_bo - Defines a request to prepare a BO for mmap().
+ * @handle: In. Handle of the GEM BO to prepare for mmap().
+ * @pad: Structure padding. Must be 0.
+ * @offset: Out. Offset value to provide to mmap().
+ */
+struct qaic_mmap_bo {
+ __u32 handle;
+ __u32 pad;
+ __u64 offset;
+};
+
+/**
+ * struct qaic_sem - Defines a semaphore command for a BO slice.
+ * @val: In. Only lower 12 bits are valid.
+ * @index: In. Only lower 5 bits are valid.
+ * @presync: In. 1 if presync operation, 0 if postsync.
+ * @cmd: In. One of QAIC_SEM_*.
+ * @flags: In. Bitfield. See QAIC_SEM_INSYNCFENCE and QAIC_SEM_OUTSYNCFENCE
+ * @pad: Structure padding. Must be 0.
+ */
+struct qaic_sem {
+ __u16 val;
+ __u8 index;
+ __u8 presync;
+ __u8 cmd;
+ __u8 flags;
+ __u16 pad;
+};
+
+/**
+ * struct qaic_attach_slice_entry - Defines a single BO slice.
+ * @size: In. Size of this slice in bytes.
+ * @sem0: In. Semaphore command 0. Must be 0 is not valid.
+ * @sem1: In. Semaphore command 1. Must be 0 is not valid.
+ * @sem2: In. Semaphore command 2. Must be 0 is not valid.
+ * @sem3: In. Semaphore command 3. Must be 0 is not valid.
+ * @dev_addr: In. Device address this slice pushes to or pulls from.
+ * @db_addr: In. Address of the doorbell to ring.
+ * @db_data: In. Data to write to the doorbell.
+ * @db_len: In. Size of the doorbell data in bits - 32, 16, or 8. 0 is for
+ * inactive doorbells.
+ * @offset: In. Start of this slice as an offset from the start of the BO.
+ */
+struct qaic_attach_slice_entry {
+ __u64 size;
+ struct qaic_sem sem0;
+ struct qaic_sem sem1;
+ struct qaic_sem sem2;
+ struct qaic_sem sem3;
+ __u64 dev_addr;
+ __u64 db_addr;
+ __u32 db_data;
+ __u32 db_len;
+ __u64 offset;
+};
+
+/**
+ * struct qaic_attach_slice_hdr - Defines metadata for a set of BO slices.
+ * @count: In. Number of slices for this BO.
+ * @dbc_id: In. Associate the sliced BO with this DBC.
+ * @handle: In. GEM handle of the BO to slice.
+ * @dir: In. Direction of data flow. 1 = DMA_TO_DEVICE, 2 = DMA_FROM_DEVICE
+ * @size: In. Total length of the BO.
+ * If BO is imported (DMABUF/PRIME) then this size
+ * should not exceed the size of DMABUF provided.
+ * If BO is allocated using DRM_IOCTL_QAIC_CREATE_BO
+ * then this size should be exactly same as the size
+ * provided during DRM_IOCTL_QAIC_CREATE_BO.
+ * @dev_addr: In. Device address this slice pushes to or pulls from.
+ * @db_addr: In. Address of the doorbell to ring.
+ * @db_data: In. Data to write to the doorbell.
+ * @db_len: In. Size of the doorbell data in bits - 32, 16, or 8. 0 is for
+ * inactive doorbells.
+ * @offset: In. Start of this slice as an offset from the start of the BO.
+ */
+struct qaic_attach_slice_hdr {
+ __u32 count;
+ __u32 dbc_id;
+ __u32 handle;
+ __u32 dir;
+ __u64 size;
+};
+
+/**
+ * struct qaic_attach_slice - Defines a set of BO slices.
+ * @hdr: In. Metadata of the set of slices.
+ * @data: In. Pointer to an array containing the slice definitions.
+ */
+struct qaic_attach_slice {
+ struct qaic_attach_slice_hdr hdr;
+ __u64 data;
+};
+
+/**
+ * struct qaic_execute_entry - Defines a BO to submit to the device.
+ * @handle: In. GEM handle of the BO to commit to the device.
+ * @dir: In. Direction of data. 1 = to device, 2 = from device.
+ */
+struct qaic_execute_entry {
+ __u32 handle;
+ __u32 dir;
+};
+
+/**
+ * struct qaic_partial_execute_entry - Defines a BO to resize and submit.
+ * @handle: In. GEM handle of the BO to commit to the device.
+ * @dir: In. Direction of data. 1 = to device, 2 = from device.
+ * @resize: In. New size of the BO. Must be <= the original BO size. 0 is
+ * short for no resize.
+ */
+struct qaic_partial_execute_entry {
+ __u32 handle;
+ __u32 dir;
+ __u64 resize;
+};
+
+/**
+ * struct qaic_execute_hdr - Defines metadata for BO submission.
+ * @count: In. Number of BOs to submit.
+ * @dbc_id: In. DBC to submit the BOs on.
+ */
+struct qaic_execute_hdr {
+ __u32 count;
+ __u32 dbc_id;
+};
+
+/**
+ * struct qaic_execute - Defines a list of BOs to submit to the device.
+ * @hdr: In. BO list metadata.
+ * @data: In. Pointer to an array of BOs to submit.
+ */
+struct qaic_execute {
+ struct qaic_execute_hdr hdr;
+ __u64 data;
+};
+
+/**
+ * struct qaic_wait - Defines a blocking wait for BO execution.
+ * @handle: In. GEM handle of the BO to wait on.
+ * @timeout: In. Maximum time in ms to wait for the BO.
+ * @dbc_id: In. DBC the BO is submitted to.
+ * @pad: Structure padding. Must be 0.
+ */
+struct qaic_wait {
+ __u32 handle;
+ __u32 timeout;
+ __u32 dbc_id;
+ __u32 pad;
+};
+
+/**
+ * struct qaic_perf_stats_hdr - Defines metadata for getting BO perf info.
+ * @count: In. Number of BOs requested.
+ * @pad: Structure padding. Must be 0.
+ * @dbc_id: In. DBC the BO are associated with.
+ */
+struct qaic_perf_stats_hdr {
+ __u16 count;
+ __u16 pad;
+ __u32 dbc_id;
+};
+
+/**
+ * struct qaic_perf_stats - Defines a request for getting BO perf info.
+ * @hdr: In. Request metadata
+ * @data: In. Pointer to array of stats structures that will receive the data.
+ */
+struct qaic_perf_stats {
+ struct qaic_perf_stats_hdr hdr;
+ __u64 data;
+};
+
+/**
+ * struct qaic_perf_stats_entry - Defines a BO perf info.
+ * @handle: In. GEM handle of the BO to get perf stats for.
+ * @queue_level_before: Out. Number of elements in the queue before this BO
+ * was submitted.
+ * @num_queue_element: Out. Number of elements added to the queue to submit
+ * this BO.
+ * @submit_latency_us: Out. Time taken by the driver to submit this BO.
+ * @device_latency_us: Out. Time taken by the device to execute this BO.
+ * @pad: Structure padding. Must be 0.
+ */
+struct qaic_perf_stats_entry {
+ __u32 handle;
+ __u32 queue_level_before;
+ __u32 num_queue_element;
+ __u32 submit_latency_us;
+ __u32 device_latency_us;
+ __u32 pad;
+};
+
+#define DRM_QAIC_MANAGE 0x00
+#define DRM_QAIC_CREATE_BO 0x01
+#define DRM_QAIC_MMAP_BO 0x02
+#define DRM_QAIC_ATTACH_SLICE_BO 0x03
+#define DRM_QAIC_EXECUTE_BO 0x04
+#define DRM_QAIC_PARTIAL_EXECUTE_BO 0x05
+#define DRM_QAIC_WAIT_BO 0x06
+#define DRM_QAIC_PERF_STATS_BO 0x07
+
+#define DRM_IOCTL_QAIC_MANAGE DRM_IOWR(DRM_COMMAND_BASE + DRM_QAIC_MANAGE, struct qaic_manage_msg)
+#define DRM_IOCTL_QAIC_CREATE_BO DRM_IOWR(DRM_COMMAND_BASE + DRM_QAIC_CREATE_BO, struct qaic_create_bo)
+#define DRM_IOCTL_QAIC_MMAP_BO DRM_IOWR(DRM_COMMAND_BASE + DRM_QAIC_MMAP_BO, struct qaic_mmap_bo)
+#define DRM_IOCTL_QAIC_ATTACH_SLICE_BO DRM_IOW(DRM_COMMAND_BASE + DRM_QAIC_ATTACH_SLICE_BO, struct qaic_attach_slice)
+#define DRM_IOCTL_QAIC_EXECUTE_BO DRM_IOW(DRM_COMMAND_BASE + DRM_QAIC_EXECUTE_BO, struct qaic_execute)
+#define DRM_IOCTL_QAIC_PARTIAL_EXECUTE_BO DRM_IOW(DRM_COMMAND_BASE + DRM_QAIC_PARTIAL_EXECUTE_BO, struct qaic_execute)
+#define DRM_IOCTL_QAIC_WAIT_BO DRM_IOW(DRM_COMMAND_BASE + DRM_QAIC_WAIT_BO, struct qaic_wait)
+#define DRM_IOCTL_QAIC_PERF_STATS_BO DRM_IOWR(DRM_COMMAND_BASE + DRM_QAIC_PERF_STATS_BO, struct qaic_perf_stats)
+
+#if defined(__cplusplus)
+}
+#endif
+
+#endif /* QAIC_ACCEL_H_ */
--
2.7.4
^ permalink raw reply related [flat|nested] 16+ messages in thread* [PATCH v5 3/8] accel/qaic: Add MHI controller
2023-03-27 15:54 [PATCH v5 0/8] QAIC accel driver Jeffrey Hugo
2023-03-27 15:54 ` [PATCH v5 1/8] accel/qaic: Add documentation for AIC100 accelerator driver Jeffrey Hugo
2023-03-27 15:54 ` [PATCH v5 2/8] accel/qaic: Add uapi and core driver file Jeffrey Hugo
@ 2023-03-27 15:54 ` Jeffrey Hugo
2023-03-27 15:54 ` [PATCH v5 4/8] accel/qaic: Add control path Jeffrey Hugo
` (5 subsequent siblings)
8 siblings, 0 replies; 16+ messages in thread
From: Jeffrey Hugo @ 2023-03-27 15:54 UTC (permalink / raw)
To: ogabbay, airlied, daniel
Cc: jacek.lawrynowicz, stanislaw.gruszka, dri-devel, quic_pkanojiy,
quic_carlv, quic_ajitpals, linux-doc, linux-arm-msm, bagasdotme,
mani, Jeffrey Hugo
An AIC100 device contains a MHI interface with a number of different
channels for controlling different aspects of the device. The MHI
controller works with the MHI bus to enable and drive that interface.
AIC100 uses the BHI protocol in PBL to load SBL. The MHI controller
expects the SBL to be located at /lib/firmware/qcom/aic100/sbl.bin and
expects the MHI bus to manage the process of loading and sending SBL to
the device.
Signed-off-by: Jeffrey Hugo <quic_jhugo@quicinc.com>
Reviewed-by: Carl Vanderlip <quic_carlv@quicinc.com>
Reviewed-by: Pranjal Ramajor Asha Kanojiya <quic_pkanojiy@quicinc.com>
Reviewed-by: Stanislaw Gruszka <stanislaw.gruszka@linux.intel.com>
Reviewed-by: Jacek Lawrynowicz <jacek.lawrynowicz@linux.intel.com>
---
drivers/accel/qaic/mhi_controller.c | 563 ++++++++++++++++++++++++++++++++++++
drivers/accel/qaic/mhi_controller.h | 16 +
2 files changed, 579 insertions(+)
create mode 100644 drivers/accel/qaic/mhi_controller.c
create mode 100644 drivers/accel/qaic/mhi_controller.h
diff --git a/drivers/accel/qaic/mhi_controller.c b/drivers/accel/qaic/mhi_controller.c
new file mode 100644
index 0000000..5036e58
--- /dev/null
+++ b/drivers/accel/qaic/mhi_controller.c
@@ -0,0 +1,563 @@
+// SPDX-License-Identifier: GPL-2.0-only
+
+/* Copyright (c) 2019-2021, The Linux Foundation. All rights reserved. */
+/* Copyright (c) 2021-2023 Qualcomm Innovation Center, Inc. All rights reserved. */
+
+#include <linux/delay.h>
+#include <linux/err.h>
+#include <linux/memblock.h>
+#include <linux/mhi.h>
+#include <linux/moduleparam.h>
+#include <linux/pci.h>
+#include <linux/sizes.h>
+
+#include "mhi_controller.h"
+#include "qaic.h"
+
+#define MAX_RESET_TIME_SEC 25
+
+static unsigned int mhi_timeout_ms = 2000; /* 2 sec default */
+module_param(mhi_timeout_ms, uint, 0600);
+MODULE_PARM_DESC(mhi_timeout_ms, "MHI controller timeout value");
+
+static struct mhi_channel_config aic100_channels[] = {
+ {
+ .name = "QAIC_LOOPBACK",
+ .num = 0,
+ .num_elements = 32,
+ .local_elements = 0,
+ .event_ring = 0,
+ .dir = DMA_TO_DEVICE,
+ .ee_mask = MHI_CH_EE_AMSS,
+ .pollcfg = 0,
+ .doorbell = MHI_DB_BRST_DISABLE,
+ .lpm_notify = false,
+ .offload_channel = false,
+ .doorbell_mode_switch = false,
+ .auto_queue = false,
+ .wake_capable = false,
+ },
+ {
+ .name = "QAIC_LOOPBACK",
+ .num = 1,
+ .num_elements = 32,
+ .local_elements = 0,
+ .event_ring = 0,
+ .dir = DMA_FROM_DEVICE,
+ .ee_mask = MHI_CH_EE_AMSS,
+ .pollcfg = 0,
+ .doorbell = MHI_DB_BRST_DISABLE,
+ .lpm_notify = false,
+ .offload_channel = false,
+ .doorbell_mode_switch = false,
+ .auto_queue = false,
+ .wake_capable = false,
+ },
+ {
+ .name = "QAIC_SAHARA",
+ .num = 2,
+ .num_elements = 32,
+ .local_elements = 0,
+ .event_ring = 0,
+ .dir = DMA_TO_DEVICE,
+ .ee_mask = MHI_CH_EE_SBL,
+ .pollcfg = 0,
+ .doorbell = MHI_DB_BRST_DISABLE,
+ .lpm_notify = false,
+ .offload_channel = false,
+ .doorbell_mode_switch = false,
+ .auto_queue = false,
+ .wake_capable = false,
+ },
+ {
+ .name = "QAIC_SAHARA",
+ .num = 3,
+ .num_elements = 32,
+ .local_elements = 0,
+ .event_ring = 0,
+ .dir = DMA_FROM_DEVICE,
+ .ee_mask = MHI_CH_EE_SBL,
+ .pollcfg = 0,
+ .doorbell = MHI_DB_BRST_DISABLE,
+ .lpm_notify = false,
+ .offload_channel = false,
+ .doorbell_mode_switch = false,
+ .auto_queue = false,
+ .wake_capable = false,
+ },
+ {
+ .name = "QAIC_DIAG",
+ .num = 4,
+ .num_elements = 32,
+ .local_elements = 0,
+ .event_ring = 0,
+ .dir = DMA_TO_DEVICE,
+ .ee_mask = MHI_CH_EE_AMSS,
+ .pollcfg = 0,
+ .doorbell = MHI_DB_BRST_DISABLE,
+ .lpm_notify = false,
+ .offload_channel = false,
+ .doorbell_mode_switch = false,
+ .auto_queue = false,
+ .wake_capable = false,
+ },
+ {
+ .name = "QAIC_DIAG",
+ .num = 5,
+ .num_elements = 32,
+ .local_elements = 0,
+ .event_ring = 0,
+ .dir = DMA_FROM_DEVICE,
+ .ee_mask = MHI_CH_EE_AMSS,
+ .pollcfg = 0,
+ .doorbell = MHI_DB_BRST_DISABLE,
+ .lpm_notify = false,
+ .offload_channel = false,
+ .doorbell_mode_switch = false,
+ .auto_queue = false,
+ .wake_capable = false,
+ },
+ {
+ .name = "QAIC_SSR",
+ .num = 6,
+ .num_elements = 32,
+ .local_elements = 0,
+ .event_ring = 0,
+ .dir = DMA_TO_DEVICE,
+ .ee_mask = MHI_CH_EE_AMSS,
+ .pollcfg = 0,
+ .doorbell = MHI_DB_BRST_DISABLE,
+ .lpm_notify = false,
+ .offload_channel = false,
+ .doorbell_mode_switch = false,
+ .auto_queue = false,
+ .wake_capable = false,
+ },
+ {
+ .name = "QAIC_SSR",
+ .num = 7,
+ .num_elements = 32,
+ .local_elements = 0,
+ .event_ring = 0,
+ .dir = DMA_FROM_DEVICE,
+ .ee_mask = MHI_CH_EE_AMSS,
+ .pollcfg = 0,
+ .doorbell = MHI_DB_BRST_DISABLE,
+ .lpm_notify = false,
+ .offload_channel = false,
+ .doorbell_mode_switch = false,
+ .auto_queue = false,
+ .wake_capable = false,
+ },
+ {
+ .name = "QAIC_QDSS",
+ .num = 8,
+ .num_elements = 32,
+ .local_elements = 0,
+ .event_ring = 0,
+ .dir = DMA_TO_DEVICE,
+ .ee_mask = MHI_CH_EE_AMSS,
+ .pollcfg = 0,
+ .doorbell = MHI_DB_BRST_DISABLE,
+ .lpm_notify = false,
+ .offload_channel = false,
+ .doorbell_mode_switch = false,
+ .auto_queue = false,
+ .wake_capable = false,
+ },
+ {
+ .name = "QAIC_QDSS",
+ .num = 9,
+ .num_elements = 32,
+ .local_elements = 0,
+ .event_ring = 0,
+ .dir = DMA_FROM_DEVICE,
+ .ee_mask = MHI_CH_EE_AMSS,
+ .pollcfg = 0,
+ .doorbell = MHI_DB_BRST_DISABLE,
+ .lpm_notify = false,
+ .offload_channel = false,
+ .doorbell_mode_switch = false,
+ .auto_queue = false,
+ .wake_capable = false,
+ },
+ {
+ .name = "QAIC_CONTROL",
+ .num = 10,
+ .num_elements = 128,
+ .local_elements = 0,
+ .event_ring = 0,
+ .dir = DMA_TO_DEVICE,
+ .ee_mask = MHI_CH_EE_AMSS,
+ .pollcfg = 0,
+ .doorbell = MHI_DB_BRST_DISABLE,
+ .lpm_notify = false,
+ .offload_channel = false,
+ .doorbell_mode_switch = false,
+ .auto_queue = false,
+ .wake_capable = false,
+ },
+ {
+ .name = "QAIC_CONTROL",
+ .num = 11,
+ .num_elements = 128,
+ .local_elements = 0,
+ .event_ring = 0,
+ .dir = DMA_FROM_DEVICE,
+ .ee_mask = MHI_CH_EE_AMSS,
+ .pollcfg = 0,
+ .doorbell = MHI_DB_BRST_DISABLE,
+ .lpm_notify = false,
+ .offload_channel = false,
+ .doorbell_mode_switch = false,
+ .auto_queue = false,
+ .wake_capable = false,
+ },
+ {
+ .name = "QAIC_LOGGING",
+ .num = 12,
+ .num_elements = 32,
+ .local_elements = 0,
+ .event_ring = 0,
+ .dir = DMA_TO_DEVICE,
+ .ee_mask = MHI_CH_EE_SBL,
+ .pollcfg = 0,
+ .doorbell = MHI_DB_BRST_DISABLE,
+ .lpm_notify = false,
+ .offload_channel = false,
+ .doorbell_mode_switch = false,
+ .auto_queue = false,
+ .wake_capable = false,
+ },
+ {
+ .name = "QAIC_LOGGING",
+ .num = 13,
+ .num_elements = 32,
+ .local_elements = 0,
+ .event_ring = 0,
+ .dir = DMA_FROM_DEVICE,
+ .ee_mask = MHI_CH_EE_SBL,
+ .pollcfg = 0,
+ .doorbell = MHI_DB_BRST_DISABLE,
+ .lpm_notify = false,
+ .offload_channel = false,
+ .doorbell_mode_switch = false,
+ .auto_queue = false,
+ .wake_capable = false,
+ },
+ {
+ .name = "QAIC_STATUS",
+ .num = 14,
+ .num_elements = 32,
+ .local_elements = 0,
+ .event_ring = 0,
+ .dir = DMA_TO_DEVICE,
+ .ee_mask = MHI_CH_EE_AMSS,
+ .pollcfg = 0,
+ .doorbell = MHI_DB_BRST_DISABLE,
+ .lpm_notify = false,
+ .offload_channel = false,
+ .doorbell_mode_switch = false,
+ .auto_queue = false,
+ .wake_capable = false,
+ },
+ {
+ .name = "QAIC_STATUS",
+ .num = 15,
+ .num_elements = 32,
+ .local_elements = 0,
+ .event_ring = 0,
+ .dir = DMA_FROM_DEVICE,
+ .ee_mask = MHI_CH_EE_AMSS,
+ .pollcfg = 0,
+ .doorbell = MHI_DB_BRST_DISABLE,
+ .lpm_notify = false,
+ .offload_channel = false,
+ .doorbell_mode_switch = false,
+ .auto_queue = false,
+ .wake_capable = false,
+ },
+ {
+ .name = "QAIC_TELEMETRY",
+ .num = 16,
+ .num_elements = 32,
+ .local_elements = 0,
+ .event_ring = 0,
+ .dir = DMA_TO_DEVICE,
+ .ee_mask = MHI_CH_EE_AMSS,
+ .pollcfg = 0,
+ .doorbell = MHI_DB_BRST_DISABLE,
+ .lpm_notify = false,
+ .offload_channel = false,
+ .doorbell_mode_switch = false,
+ .auto_queue = false,
+ .wake_capable = false,
+ },
+ {
+ .name = "QAIC_TELEMETRY",
+ .num = 17,
+ .num_elements = 32,
+ .local_elements = 0,
+ .event_ring = 0,
+ .dir = DMA_FROM_DEVICE,
+ .ee_mask = MHI_CH_EE_AMSS,
+ .pollcfg = 0,
+ .doorbell = MHI_DB_BRST_DISABLE,
+ .lpm_notify = false,
+ .offload_channel = false,
+ .doorbell_mode_switch = false,
+ .auto_queue = false,
+ .wake_capable = false,
+ },
+ {
+ .name = "QAIC_DEBUG",
+ .num = 18,
+ .num_elements = 32,
+ .local_elements = 0,
+ .event_ring = 0,
+ .dir = DMA_TO_DEVICE,
+ .ee_mask = MHI_CH_EE_AMSS,
+ .pollcfg = 0,
+ .doorbell = MHI_DB_BRST_DISABLE,
+ .lpm_notify = false,
+ .offload_channel = false,
+ .doorbell_mode_switch = false,
+ .auto_queue = false,
+ .wake_capable = false,
+ },
+ {
+ .name = "QAIC_DEBUG",
+ .num = 19,
+ .num_elements = 32,
+ .local_elements = 0,
+ .event_ring = 0,
+ .dir = DMA_FROM_DEVICE,
+ .ee_mask = MHI_CH_EE_AMSS,
+ .pollcfg = 0,
+ .doorbell = MHI_DB_BRST_DISABLE,
+ .lpm_notify = false,
+ .offload_channel = false,
+ .doorbell_mode_switch = false,
+ .auto_queue = false,
+ .wake_capable = false,
+ },
+ {
+ .name = "QAIC_TIMESYNC",
+ .num = 20,
+ .num_elements = 32,
+ .local_elements = 0,
+ .event_ring = 0,
+ .dir = DMA_TO_DEVICE,
+ .ee_mask = MHI_CH_EE_SBL | MHI_CH_EE_AMSS,
+ .pollcfg = 0,
+ .doorbell = MHI_DB_BRST_DISABLE,
+ .lpm_notify = false,
+ .offload_channel = false,
+ .doorbell_mode_switch = false,
+ .auto_queue = false,
+ .wake_capable = false,
+ },
+ {
+ .num = 21,
+ .name = "QAIC_TIMESYNC",
+ .num_elements = 32,
+ .local_elements = 0,
+ .event_ring = 0,
+ .dir = DMA_FROM_DEVICE,
+ .ee_mask = MHI_CH_EE_SBL | MHI_CH_EE_AMSS,
+ .pollcfg = 0,
+ .doorbell = MHI_DB_BRST_DISABLE,
+ .lpm_notify = false,
+ .offload_channel = false,
+ .doorbell_mode_switch = false,
+ .auto_queue = false,
+ .wake_capable = false,
+ },
+};
+
+static struct mhi_event_config aic100_events[] = {
+ {
+ .num_elements = 32,
+ .irq_moderation_ms = 0,
+ .irq = 0,
+ .channel = U32_MAX,
+ .priority = 1,
+ .mode = MHI_DB_BRST_DISABLE,
+ .data_type = MHI_ER_CTRL,
+ .hardware_event = false,
+ .client_managed = false,
+ .offload_channel = false,
+ },
+};
+
+static struct mhi_controller_config aic100_config = {
+ .max_channels = 128,
+ .timeout_ms = 0, /* controlled by mhi_timeout */
+ .buf_len = 0,
+ .num_channels = ARRAY_SIZE(aic100_channels),
+ .ch_cfg = aic100_channels,
+ .num_events = ARRAY_SIZE(aic100_events),
+ .event_cfg = aic100_events,
+ .use_bounce_buf = false,
+ .m2_no_db = false,
+};
+
+static int mhi_read_reg(struct mhi_controller *mhi_cntrl, void __iomem *addr, u32 *out)
+{
+ u32 tmp = readl_relaxed(addr);
+
+ if (tmp == U32_MAX)
+ return -EIO;
+
+ *out = tmp;
+
+ return 0;
+}
+
+static void mhi_write_reg(struct mhi_controller *mhi_cntrl, void __iomem *addr, u32 val)
+{
+ writel_relaxed(val, addr);
+}
+
+static int mhi_runtime_get(struct mhi_controller *mhi_cntrl)
+{
+ return 0;
+}
+
+static void mhi_runtime_put(struct mhi_controller *mhi_cntrl)
+{
+}
+
+static void mhi_status_cb(struct mhi_controller *mhi_cntrl, enum mhi_callback reason)
+{
+ struct qaic_device *qdev = pci_get_drvdata(to_pci_dev(mhi_cntrl->cntrl_dev));
+
+ /* this event occurs in atomic context */
+ if (reason == MHI_CB_FATAL_ERROR)
+ pci_err(qdev->pdev, "Fatal error received from device. Attempting to recover\n");
+ /* this event occurs in non-atomic context */
+ if (reason == MHI_CB_SYS_ERROR)
+ qaic_dev_reset_clean_local_state(qdev, true);
+}
+
+static int mhi_reset_and_async_power_up(struct mhi_controller *mhi_cntrl)
+{
+ u8 time_sec = 1;
+ int current_ee;
+ int ret;
+
+ /* Reset the device to bring the device in PBL EE */
+ mhi_soc_reset(mhi_cntrl);
+
+ /*
+ * Keep checking the execution environment(EE) after every 1 second
+ * interval.
+ */
+ do {
+ msleep(1000);
+ current_ee = mhi_get_exec_env(mhi_cntrl);
+ } while (current_ee != MHI_EE_PBL && time_sec++ <= MAX_RESET_TIME_SEC);
+
+ /* If the device is in PBL EE retry power up */
+ if (current_ee == MHI_EE_PBL)
+ ret = mhi_async_power_up(mhi_cntrl);
+ else
+ ret = -EIO;
+
+ return ret;
+}
+
+struct mhi_controller *qaic_mhi_register_controller(struct pci_dev *pci_dev, void __iomem *mhi_bar,
+ int mhi_irq)
+{
+ struct mhi_controller *mhi_cntrl;
+ int ret;
+
+ mhi_cntrl = devm_kzalloc(&pci_dev->dev, sizeof(*mhi_cntrl), GFP_KERNEL);
+ if (!mhi_cntrl)
+ return ERR_PTR(-ENOMEM);
+
+ mhi_cntrl->cntrl_dev = &pci_dev->dev;
+
+ /*
+ * Covers the entire possible physical ram region. Remote side is
+ * going to calculate a size of this range, so subtract 1 to prevent
+ * rollover.
+ */
+ mhi_cntrl->iova_start = 0;
+ mhi_cntrl->iova_stop = PHYS_ADDR_MAX - 1;
+ mhi_cntrl->status_cb = mhi_status_cb;
+ mhi_cntrl->runtime_get = mhi_runtime_get;
+ mhi_cntrl->runtime_put = mhi_runtime_put;
+ mhi_cntrl->read_reg = mhi_read_reg;
+ mhi_cntrl->write_reg = mhi_write_reg;
+ mhi_cntrl->regs = mhi_bar;
+ mhi_cntrl->reg_len = SZ_4K;
+ mhi_cntrl->nr_irqs = 1;
+ mhi_cntrl->irq = devm_kmalloc(&pci_dev->dev, sizeof(*mhi_cntrl->irq), GFP_KERNEL);
+
+ if (!mhi_cntrl->irq)
+ return ERR_PTR(-ENOMEM);
+
+ mhi_cntrl->irq[0] = mhi_irq;
+ mhi_cntrl->fw_image = "qcom/aic100/sbl.bin";
+
+ /* use latest configured timeout */
+ aic100_config.timeout_ms = mhi_timeout_ms;
+ ret = mhi_register_controller(mhi_cntrl, &aic100_config);
+ if (ret) {
+ pci_err(pci_dev, "mhi_register_controller failed %d\n", ret);
+ return ERR_PTR(ret);
+ }
+
+ ret = mhi_prepare_for_power_up(mhi_cntrl);
+ if (ret) {
+ pci_err(pci_dev, "mhi_prepare_for_power_up failed %d\n", ret);
+ goto prepare_power_up_fail;
+ }
+
+ ret = mhi_async_power_up(mhi_cntrl);
+ /*
+ * If EIO is returned it is possible that device is in SBL EE, which is
+ * undesired. SOC reset the device and try to power up again.
+ */
+ if (ret == -EIO && MHI_EE_SBL == mhi_get_exec_env(mhi_cntrl)) {
+ pci_err(pci_dev, "Found device in SBL at MHI init. Attempting a reset.\n");
+ ret = mhi_reset_and_async_power_up(mhi_cntrl);
+ }
+
+ if (ret) {
+ pci_err(pci_dev, "mhi_async_power_up failed %d\n", ret);
+ goto power_up_fail;
+ }
+
+ return mhi_cntrl;
+
+power_up_fail:
+ mhi_unprepare_after_power_down(mhi_cntrl);
+prepare_power_up_fail:
+ mhi_unregister_controller(mhi_cntrl);
+ return ERR_PTR(ret);
+}
+
+void qaic_mhi_free_controller(struct mhi_controller *mhi_cntrl, bool link_up)
+{
+ mhi_power_down(mhi_cntrl, link_up);
+ mhi_unprepare_after_power_down(mhi_cntrl);
+ mhi_unregister_controller(mhi_cntrl);
+}
+
+void qaic_mhi_start_reset(struct mhi_controller *mhi_cntrl)
+{
+ mhi_power_down(mhi_cntrl, true);
+}
+
+void qaic_mhi_reset_done(struct mhi_controller *mhi_cntrl)
+{
+ struct pci_dev *pci_dev = container_of(mhi_cntrl->cntrl_dev, struct pci_dev, dev);
+ int ret;
+
+ ret = mhi_async_power_up(mhi_cntrl);
+ if (ret)
+ pci_err(pci_dev, "mhi_async_power_up failed after reset %d\n", ret);
+}
diff --git a/drivers/accel/qaic/mhi_controller.h b/drivers/accel/qaic/mhi_controller.h
new file mode 100644
index 0000000..2ae45d7
--- /dev/null
+++ b/drivers/accel/qaic/mhi_controller.h
@@ -0,0 +1,16 @@
+/* SPDX-License-Identifier: GPL-2.0-only
+ *
+ * Copyright (c) 2019-2020, The Linux Foundation. All rights reserved.
+ * Copyright (c) 2023 Qualcomm Innovation Center, Inc. All rights reserved.
+ */
+
+#ifndef MHICONTROLLERQAIC_H_
+#define MHICONTROLLERQAIC_H_
+
+struct mhi_controller *qaic_mhi_register_controller(struct pci_dev *pci_dev, void __iomem *mhi_bar,
+ int mhi_irq);
+void qaic_mhi_free_controller(struct mhi_controller *mhi_cntrl, bool link_up);
+void qaic_mhi_start_reset(struct mhi_controller *mhi_cntrl);
+void qaic_mhi_reset_done(struct mhi_controller *mhi_cntrl);
+
+#endif /* MHICONTROLLERQAIC_H_ */
--
2.7.4
^ permalink raw reply related [flat|nested] 16+ messages in thread* [PATCH v5 4/8] accel/qaic: Add control path
2023-03-27 15:54 [PATCH v5 0/8] QAIC accel driver Jeffrey Hugo
` (2 preceding siblings ...)
2023-03-27 15:54 ` [PATCH v5 3/8] accel/qaic: Add MHI controller Jeffrey Hugo
@ 2023-03-27 15:54 ` Jeffrey Hugo
2023-03-27 15:54 ` [PATCH v5 5/8] accel/qaic: Add datapath Jeffrey Hugo
` (4 subsequent siblings)
8 siblings, 0 replies; 16+ messages in thread
From: Jeffrey Hugo @ 2023-03-27 15:54 UTC (permalink / raw)
To: ogabbay, airlied, daniel
Cc: jacek.lawrynowicz, stanislaw.gruszka, dri-devel, quic_pkanojiy,
quic_carlv, quic_ajitpals, linux-doc, linux-arm-msm, bagasdotme,
mani, Jeffrey Hugo
Add the control path component that talks to the management processor (QSM)
to load workloads onto the AIC100 device. This implements the KMD portion
of the NNC protocol over the QAIC_CONTROL MHI channel and the
DRM_IOCTL_QAIC_MANAGE IOCTL to userspace. With this functionality, QAIC
clients are able to load, run, and cleanup their workloads on the device
but not interact with the workloads (run inferences).
Signed-off-by: Jeffrey Hugo <quic_jhugo@quicinc.com>
Reviewed-by: Carl Vanderlip <quic_carlv@quicinc.com>
Reviewed-by: Pranjal Ramajor Asha Kanojiya <quic_pkanojiy@quicinc.com>
Reviewed-by: Stanislaw Gruszka <stanislaw.gruszka@linux.intel.com>
Reviewed-by: Jacek Lawrynowicz <jacek.lawrynowicz@linux.intel.com>
---
drivers/accel/qaic/qaic_control.c | 1527 +++++++++++++++++++++++++++++++++++++
1 file changed, 1527 insertions(+)
create mode 100644 drivers/accel/qaic/qaic_control.c
diff --git a/drivers/accel/qaic/qaic_control.c b/drivers/accel/qaic/qaic_control.c
new file mode 100644
index 0000000..0f01b8c
--- /dev/null
+++ b/drivers/accel/qaic/qaic_control.c
@@ -0,0 +1,1527 @@
+// SPDX-License-Identifier: GPL-2.0-only
+
+/* Copyright (c) 2019-2021, The Linux Foundation. All rights reserved. */
+/* Copyright (c) 2021-2023 Qualcomm Innovation Center, Inc. All rights reserved. */
+
+#include <asm/byteorder.h>
+#include <linux/completion.h>
+#include <linux/crc32.h>
+#include <linux/delay.h>
+#include <linux/dma-mapping.h>
+#include <linux/kref.h>
+#include <linux/list.h>
+#include <linux/mhi.h>
+#include <linux/mm.h>
+#include <linux/moduleparam.h>
+#include <linux/mutex.h>
+#include <linux/pci.h>
+#include <linux/scatterlist.h>
+#include <linux/types.h>
+#include <linux/uaccess.h>
+#include <linux/workqueue.h>
+#include <linux/wait.h>
+#include <drm/drm_device.h>
+#include <drm/drm_file.h>
+#include <uapi/drm/qaic_accel.h>
+
+#include "qaic.h"
+
+#define MANAGE_MAGIC_NUMBER ((__force __le32)0x43494151) /* "QAIC" in little endian */
+#define QAIC_DBC_Q_GAP SZ_256
+#define QAIC_DBC_Q_BUF_ALIGN SZ_4K
+#define QAIC_MANAGE_EXT_MSG_LENGTH SZ_64K /* Max DMA message length */
+#define QAIC_WRAPPER_MAX_SIZE SZ_4K
+#define QAIC_MHI_RETRY_WAIT_MS 100
+#define QAIC_MHI_RETRY_MAX 20
+
+static unsigned int control_resp_timeout_s = 60; /* 60 sec default */
+module_param(control_resp_timeout_s, uint, 0600);
+MODULE_PARM_DESC(control_resp_timeout_s, "Timeout for NNC responses from QSM");
+
+struct manage_msg {
+ u32 len;
+ u32 count;
+ u8 data[];
+};
+
+/*
+ * wire encoding structures for the manage protocol.
+ * All fields are little endian on the wire
+ */
+struct wire_msg_hdr {
+ __le32 crc32; /* crc of everything following this field in the message */
+ __le32 magic_number;
+ __le32 sequence_number;
+ __le32 len; /* length of this message */
+ __le32 count; /* number of transactions in this message */
+ __le32 handle; /* unique id to track the resources consumed */
+ __le32 partition_id; /* partition id for the request (signed) */
+ __le32 padding; /* must be 0 */
+} __packed;
+
+struct wire_msg {
+ struct wire_msg_hdr hdr;
+ u8 data[];
+} __packed;
+
+struct wire_trans_hdr {
+ __le32 type;
+ __le32 len;
+} __packed;
+
+/* Each message sent from driver to device are organized in a list of wrapper_msg */
+struct wrapper_msg {
+ struct list_head list;
+ struct kref ref_count;
+ u32 len; /* length of data to transfer */
+ struct wrapper_list *head;
+ union {
+ struct wire_msg msg;
+ struct wire_trans_hdr trans;
+ };
+};
+
+struct wrapper_list {
+ struct list_head list;
+ spinlock_t lock;
+};
+
+struct wire_trans_passthrough {
+ struct wire_trans_hdr hdr;
+ u8 data[];
+} __packed;
+
+struct wire_addr_size_pair {
+ __le64 addr;
+ __le64 size;
+} __packed;
+
+struct wire_trans_dma_xfer {
+ struct wire_trans_hdr hdr;
+ __le32 tag;
+ __le32 count;
+ __le32 dma_chunk_id;
+ __le32 padding;
+ struct wire_addr_size_pair data[];
+} __packed;
+
+/* Initiated by device to continue the DMA xfer of a large piece of data */
+struct wire_trans_dma_xfer_cont {
+ struct wire_trans_hdr hdr;
+ __le32 dma_chunk_id;
+ __le32 padding;
+ __le64 xferred_size;
+} __packed;
+
+struct wire_trans_activate_to_dev {
+ struct wire_trans_hdr hdr;
+ __le64 req_q_addr;
+ __le64 rsp_q_addr;
+ __le32 req_q_size;
+ __le32 rsp_q_size;
+ __le32 buf_len;
+ __le32 options; /* unused, but BIT(16) has meaning to the device */
+} __packed;
+
+struct wire_trans_activate_from_dev {
+ struct wire_trans_hdr hdr;
+ __le32 status;
+ __le32 dbc_id;
+ __le64 options; /* unused */
+} __packed;
+
+struct wire_trans_deactivate_from_dev {
+ struct wire_trans_hdr hdr;
+ __le32 status;
+ __le32 dbc_id;
+} __packed;
+
+struct wire_trans_terminate_to_dev {
+ struct wire_trans_hdr hdr;
+ __le32 handle;
+ __le32 padding;
+} __packed;
+
+struct wire_trans_terminate_from_dev {
+ struct wire_trans_hdr hdr;
+ __le32 status;
+ __le32 padding;
+} __packed;
+
+struct wire_trans_status_to_dev {
+ struct wire_trans_hdr hdr;
+} __packed;
+
+struct wire_trans_status_from_dev {
+ struct wire_trans_hdr hdr;
+ __le16 major;
+ __le16 minor;
+ __le32 status;
+ __le64 status_flags;
+} __packed;
+
+struct wire_trans_validate_part_to_dev {
+ struct wire_trans_hdr hdr;
+ __le32 part_id;
+ __le32 padding;
+} __packed;
+
+struct wire_trans_validate_part_from_dev {
+ struct wire_trans_hdr hdr;
+ __le32 status;
+ __le32 padding;
+} __packed;
+
+struct xfer_queue_elem {
+ /*
+ * Node in list of ongoing transfer request on control channel.
+ * Maintained by root device struct.
+ */
+ struct list_head list;
+ /* Sequence number of this transfer request */
+ u32 seq_num;
+ /* This is used to wait on until completion of transfer request */
+ struct completion xfer_done;
+ /* Received data from device */
+ void *buf;
+};
+
+struct dma_xfer {
+ /* Node in list of DMA transfers which is used for cleanup */
+ struct list_head list;
+ /* SG table of memory used for DMA */
+ struct sg_table *sgt;
+ /* Array pages used for DMA */
+ struct page **page_list;
+ /* Number of pages used for DMA */
+ unsigned long nr_pages;
+};
+
+struct ioctl_resources {
+ /* List of all DMA transfers which is used later for cleanup */
+ struct list_head dma_xfers;
+ /* Base address of request queue which belongs to a DBC */
+ void *buf;
+ /*
+ * Base bus address of request queue which belongs to a DBC. Response
+ * queue base bus address can be calculated by adding size of request
+ * queue to base bus address of request queue.
+ */
+ dma_addr_t dma_addr;
+ /* Total size of request queue and response queue in byte */
+ u32 total_size;
+ /* Total number of elements that can be queued in each of request and response queue */
+ u32 nelem;
+ /* Base address of response queue which belongs to a DBC */
+ void *rsp_q_base;
+ /* Status of the NNC message received */
+ u32 status;
+ /* DBC id of the DBC received from device */
+ u32 dbc_id;
+ /*
+ * DMA transfer request messages can be big in size and it may not be
+ * possible to send them in one shot. In such cases the messages are
+ * broken into chunks, this field stores ID of such chunks.
+ */
+ u32 dma_chunk_id;
+ /* Total number of bytes transferred for a DMA xfer request */
+ u64 xferred_dma_size;
+ /* Header of transaction message received from user. Used during DMA xfer request. */
+ void *trans_hdr;
+};
+
+struct resp_work {
+ struct work_struct work;
+ struct qaic_device *qdev;
+ void *buf;
+};
+
+/*
+ * Since we're working with little endian messages, its useful to be able to
+ * increment without filling a whole line with conversions back and forth just
+ * to add one(1) to a message count.
+ */
+static __le32 incr_le32(__le32 val)
+{
+ return cpu_to_le32(le32_to_cpu(val) + 1);
+}
+
+static u32 gen_crc(void *msg)
+{
+ struct wrapper_list *wrappers = msg;
+ struct wrapper_msg *w;
+ u32 crc = ~0;
+
+ list_for_each_entry(w, &wrappers->list, list)
+ crc = crc32(crc, &w->msg, w->len);
+
+ return crc ^ ~0;
+}
+
+static u32 gen_crc_stub(void *msg)
+{
+ return 0;
+}
+
+static bool valid_crc(void *msg)
+{
+ struct wire_msg_hdr *hdr = msg;
+ bool ret;
+ u32 crc;
+
+ /*
+ * The output of this algorithm is always converted to the native
+ * endianness.
+ */
+ crc = le32_to_cpu(hdr->crc32);
+ hdr->crc32 = 0;
+ ret = (crc32(~0, msg, le32_to_cpu(hdr->len)) ^ ~0) == crc;
+ hdr->crc32 = cpu_to_le32(crc);
+ return ret;
+}
+
+static bool valid_crc_stub(void *msg)
+{
+ return true;
+}
+
+static void free_wrapper(struct kref *ref)
+{
+ struct wrapper_msg *wrapper = container_of(ref, struct wrapper_msg, ref_count);
+
+ list_del(&wrapper->list);
+ kfree(wrapper);
+}
+
+static void save_dbc_buf(struct qaic_device *qdev, struct ioctl_resources *resources,
+ struct qaic_user *usr)
+{
+ u32 dbc_id = resources->dbc_id;
+
+ if (resources->buf) {
+ wait_event_interruptible(qdev->dbc[dbc_id].dbc_release, !qdev->dbc[dbc_id].in_use);
+ qdev->dbc[dbc_id].req_q_base = resources->buf;
+ qdev->dbc[dbc_id].rsp_q_base = resources->rsp_q_base;
+ qdev->dbc[dbc_id].dma_addr = resources->dma_addr;
+ qdev->dbc[dbc_id].total_size = resources->total_size;
+ qdev->dbc[dbc_id].nelem = resources->nelem;
+ enable_dbc(qdev, dbc_id, usr);
+ qdev->dbc[dbc_id].in_use = true;
+ resources->buf = NULL;
+ }
+}
+
+static void free_dbc_buf(struct qaic_device *qdev, struct ioctl_resources *resources)
+{
+ if (resources->buf)
+ dma_free_coherent(&qdev->pdev->dev, resources->total_size, resources->buf,
+ resources->dma_addr);
+ resources->buf = NULL;
+}
+
+static void free_dma_xfers(struct qaic_device *qdev, struct ioctl_resources *resources)
+{
+ struct dma_xfer *xfer;
+ struct dma_xfer *x;
+ int i;
+
+ list_for_each_entry_safe(xfer, x, &resources->dma_xfers, list) {
+ dma_unmap_sgtable(&qdev->pdev->dev, xfer->sgt, DMA_TO_DEVICE, 0);
+ sg_free_table(xfer->sgt);
+ kfree(xfer->sgt);
+ for (i = 0; i < xfer->nr_pages; ++i)
+ put_page(xfer->page_list[i]);
+ kfree(xfer->page_list);
+ list_del(&xfer->list);
+ kfree(xfer);
+ }
+}
+
+static struct wrapper_msg *add_wrapper(struct wrapper_list *wrappers, u32 size)
+{
+ struct wrapper_msg *w = kzalloc(size, GFP_KERNEL);
+
+ if (!w)
+ return NULL;
+ list_add_tail(&w->list, &wrappers->list);
+ kref_init(&w->ref_count);
+ w->head = wrappers;
+ return w;
+}
+
+static int encode_passthrough(struct qaic_device *qdev, void *trans, struct wrapper_list *wrappers,
+ u32 *user_len)
+{
+ struct qaic_manage_trans_passthrough *in_trans = trans;
+ struct wire_trans_passthrough *out_trans;
+ struct wrapper_msg *trans_wrapper;
+ struct wrapper_msg *wrapper;
+ struct wire_msg *msg;
+ u32 msg_hdr_len;
+
+ wrapper = list_first_entry(&wrappers->list, struct wrapper_msg, list);
+ msg = &wrapper->msg;
+ msg_hdr_len = le32_to_cpu(msg->hdr.len);
+
+ if (in_trans->hdr.len % 8 != 0)
+ return -EINVAL;
+
+ if (msg_hdr_len + in_trans->hdr.len > QAIC_MANAGE_EXT_MSG_LENGTH)
+ return -ENOSPC;
+
+ trans_wrapper = add_wrapper(wrappers,
+ offsetof(struct wrapper_msg, trans) + in_trans->hdr.len);
+ if (!trans_wrapper)
+ return -ENOMEM;
+ trans_wrapper->len = in_trans->hdr.len;
+ out_trans = (struct wire_trans_passthrough *)&trans_wrapper->trans;
+
+ memcpy(out_trans->data, in_trans->data, in_trans->hdr.len - sizeof(in_trans->hdr));
+ msg->hdr.len = cpu_to_le32(msg_hdr_len + in_trans->hdr.len);
+ msg->hdr.count = incr_le32(msg->hdr.count);
+ *user_len += in_trans->hdr.len;
+ out_trans->hdr.type = cpu_to_le32(QAIC_TRANS_PASSTHROUGH_TO_DEV);
+ out_trans->hdr.len = cpu_to_le32(in_trans->hdr.len);
+
+ return 0;
+}
+
+/* returns error code for failure, 0 if enough pages alloc'd, 1 if dma_cont is needed */
+static int find_and_map_user_pages(struct qaic_device *qdev,
+ struct qaic_manage_trans_dma_xfer *in_trans,
+ struct ioctl_resources *resources, struct dma_xfer *xfer)
+{
+ unsigned long need_pages;
+ struct page **page_list;
+ unsigned long nr_pages;
+ struct sg_table *sgt;
+ u64 xfer_start_addr;
+ int ret;
+ int i;
+
+ xfer_start_addr = in_trans->addr + resources->xferred_dma_size;
+
+ need_pages = DIV_ROUND_UP(in_trans->size + offset_in_page(xfer_start_addr) -
+ resources->xferred_dma_size, PAGE_SIZE);
+
+ nr_pages = need_pages;
+
+ while (1) {
+ page_list = kmalloc_array(nr_pages, sizeof(*page_list), GFP_KERNEL | __GFP_NOWARN);
+ if (!page_list) {
+ nr_pages = nr_pages / 2;
+ if (!nr_pages)
+ return -ENOMEM;
+ } else {
+ break;
+ }
+ }
+
+ ret = get_user_pages_fast(xfer_start_addr, nr_pages, 0, page_list);
+ if (ret < 0 || ret != nr_pages) {
+ ret = -EFAULT;
+ goto free_page_list;
+ }
+
+ sgt = kmalloc(sizeof(*sgt), GFP_KERNEL);
+ if (!sgt) {
+ ret = -ENOMEM;
+ goto put_pages;
+ }
+
+ ret = sg_alloc_table_from_pages(sgt, page_list, nr_pages,
+ offset_in_page(xfer_start_addr),
+ in_trans->size - resources->xferred_dma_size, GFP_KERNEL);
+ if (ret) {
+ ret = -ENOMEM;
+ goto free_sgt;
+ }
+
+ ret = dma_map_sgtable(&qdev->pdev->dev, sgt, DMA_TO_DEVICE, 0);
+ if (ret)
+ goto free_table;
+
+ xfer->sgt = sgt;
+ xfer->page_list = page_list;
+ xfer->nr_pages = nr_pages;
+
+ return need_pages > nr_pages ? 1 : 0;
+
+free_table:
+ sg_free_table(sgt);
+free_sgt:
+ kfree(sgt);
+put_pages:
+ for (i = 0; i < nr_pages; ++i)
+ put_page(page_list[i]);
+free_page_list:
+ kfree(page_list);
+ return ret;
+}
+
+/* returns error code for failure, 0 if everything was encoded, 1 if dma_cont is needed */
+static int encode_addr_size_pairs(struct dma_xfer *xfer, struct wrapper_list *wrappers,
+ struct ioctl_resources *resources, u32 msg_hdr_len, u32 *size,
+ struct wire_trans_dma_xfer **out_trans)
+{
+ struct wrapper_msg *trans_wrapper;
+ struct sg_table *sgt = xfer->sgt;
+ struct wire_addr_size_pair *asp;
+ struct scatterlist *sg;
+ struct wrapper_msg *w;
+ unsigned int dma_len;
+ u64 dma_chunk_len;
+ void *boundary;
+ int nents_dma;
+ int nents;
+ int i;
+
+ nents = sgt->nents;
+ nents_dma = nents;
+ *size = QAIC_MANAGE_EXT_MSG_LENGTH - msg_hdr_len - sizeof(**out_trans);
+ for_each_sgtable_sg(sgt, sg, i) {
+ *size -= sizeof(*asp);
+ /* Save 1K for possible follow-up transactions. */
+ if (*size < SZ_1K) {
+ nents_dma = i;
+ break;
+ }
+ }
+
+ trans_wrapper = add_wrapper(wrappers, QAIC_WRAPPER_MAX_SIZE);
+ if (!trans_wrapper)
+ return -ENOMEM;
+ *out_trans = (struct wire_trans_dma_xfer *)&trans_wrapper->trans;
+
+ asp = (*out_trans)->data;
+ boundary = (void *)trans_wrapper + QAIC_WRAPPER_MAX_SIZE;
+ *size = 0;
+
+ dma_len = 0;
+ w = trans_wrapper;
+ dma_chunk_len = 0;
+ for_each_sg(sgt->sgl, sg, nents_dma, i) {
+ asp->size = cpu_to_le64(dma_len);
+ dma_chunk_len += dma_len;
+ if (dma_len) {
+ asp++;
+ if ((void *)asp + sizeof(*asp) > boundary) {
+ w->len = (void *)asp - (void *)&w->msg;
+ *size += w->len;
+ w = add_wrapper(wrappers, QAIC_WRAPPER_MAX_SIZE);
+ if (!w)
+ return -ENOMEM;
+ boundary = (void *)w + QAIC_WRAPPER_MAX_SIZE;
+ asp = (struct wire_addr_size_pair *)&w->msg;
+ }
+ }
+ asp->addr = cpu_to_le64(sg_dma_address(sg));
+ dma_len = sg_dma_len(sg);
+ }
+ /* finalize the last segment */
+ asp->size = cpu_to_le64(dma_len);
+ w->len = (void *)asp + sizeof(*asp) - (void *)&w->msg;
+ *size += w->len;
+ dma_chunk_len += dma_len;
+ resources->xferred_dma_size += dma_chunk_len;
+
+ return nents_dma < nents ? 1 : 0;
+}
+
+static void cleanup_xfer(struct qaic_device *qdev, struct dma_xfer *xfer)
+{
+ int i;
+
+ dma_unmap_sgtable(&qdev->pdev->dev, xfer->sgt, DMA_TO_DEVICE, 0);
+ sg_free_table(xfer->sgt);
+ kfree(xfer->sgt);
+ for (i = 0; i < xfer->nr_pages; ++i)
+ put_page(xfer->page_list[i]);
+ kfree(xfer->page_list);
+}
+
+static int encode_dma(struct qaic_device *qdev, void *trans, struct wrapper_list *wrappers,
+ u32 *user_len, struct ioctl_resources *resources, struct qaic_user *usr)
+{
+ struct qaic_manage_trans_dma_xfer *in_trans = trans;
+ struct wire_trans_dma_xfer *out_trans;
+ struct wrapper_msg *wrapper;
+ struct dma_xfer *xfer;
+ struct wire_msg *msg;
+ bool need_cont_dma;
+ u32 msg_hdr_len;
+ u32 size;
+ int ret;
+
+ wrapper = list_first_entry(&wrappers->list, struct wrapper_msg, list);
+ msg = &wrapper->msg;
+ msg_hdr_len = le32_to_cpu(msg->hdr.len);
+
+ if (msg_hdr_len > (UINT_MAX - QAIC_MANAGE_EXT_MSG_LENGTH))
+ return -EINVAL;
+
+ /* There should be enough space to hold at least one ASP entry. */
+ if (msg_hdr_len + sizeof(*out_trans) + sizeof(struct wire_addr_size_pair) >
+ QAIC_MANAGE_EXT_MSG_LENGTH)
+ return -ENOMEM;
+
+ if (in_trans->addr + in_trans->size < in_trans->addr || !in_trans->size)
+ return -EINVAL;
+
+ xfer = kmalloc(sizeof(*xfer), GFP_KERNEL);
+ if (!xfer)
+ return -ENOMEM;
+
+ ret = find_and_map_user_pages(qdev, in_trans, resources, xfer);
+ if (ret < 0)
+ goto free_xfer;
+
+ need_cont_dma = (bool)ret;
+
+ ret = encode_addr_size_pairs(xfer, wrappers, resources, msg_hdr_len, &size, &out_trans);
+ if (ret < 0)
+ goto cleanup_xfer;
+
+ need_cont_dma = need_cont_dma || (bool)ret;
+
+ msg->hdr.len = cpu_to_le32(msg_hdr_len + size);
+ msg->hdr.count = incr_le32(msg->hdr.count);
+
+ out_trans->hdr.type = cpu_to_le32(QAIC_TRANS_DMA_XFER_TO_DEV);
+ out_trans->hdr.len = cpu_to_le32(size);
+ out_trans->tag = cpu_to_le32(in_trans->tag);
+ out_trans->count = cpu_to_le32((size - sizeof(*out_trans)) /
+ sizeof(struct wire_addr_size_pair));
+
+ *user_len += in_trans->hdr.len;
+
+ if (resources->dma_chunk_id) {
+ out_trans->dma_chunk_id = cpu_to_le32(resources->dma_chunk_id);
+ } else if (need_cont_dma) {
+ while (resources->dma_chunk_id == 0)
+ resources->dma_chunk_id = atomic_inc_return(&usr->chunk_id);
+
+ out_trans->dma_chunk_id = cpu_to_le32(resources->dma_chunk_id);
+ }
+ resources->trans_hdr = trans;
+
+ list_add(&xfer->list, &resources->dma_xfers);
+ return 0;
+
+cleanup_xfer:
+ cleanup_xfer(qdev, xfer);
+free_xfer:
+ kfree(xfer);
+ return ret;
+}
+
+static int encode_activate(struct qaic_device *qdev, void *trans, struct wrapper_list *wrappers,
+ u32 *user_len, struct ioctl_resources *resources)
+{
+ struct qaic_manage_trans_activate_to_dev *in_trans = trans;
+ struct wire_trans_activate_to_dev *out_trans;
+ struct wrapper_msg *trans_wrapper;
+ struct wrapper_msg *wrapper;
+ struct wire_msg *msg;
+ dma_addr_t dma_addr;
+ u32 msg_hdr_len;
+ void *buf;
+ u32 nelem;
+ u32 size;
+ int ret;
+
+ wrapper = list_first_entry(&wrappers->list, struct wrapper_msg, list);
+ msg = &wrapper->msg;
+ msg_hdr_len = le32_to_cpu(msg->hdr.len);
+
+ if (msg_hdr_len + sizeof(*out_trans) > QAIC_MANAGE_MAX_MSG_LENGTH)
+ return -ENOSPC;
+
+ if (!in_trans->queue_size)
+ return -EINVAL;
+
+ if (in_trans->pad)
+ return -EINVAL;
+
+ nelem = in_trans->queue_size;
+ size = (get_dbc_req_elem_size() + get_dbc_rsp_elem_size()) * nelem;
+ if (size / nelem != get_dbc_req_elem_size() + get_dbc_rsp_elem_size())
+ return -EINVAL;
+
+ if (size + QAIC_DBC_Q_GAP + QAIC_DBC_Q_BUF_ALIGN < size)
+ return -EINVAL;
+
+ size = ALIGN((size + QAIC_DBC_Q_GAP), QAIC_DBC_Q_BUF_ALIGN);
+
+ buf = dma_alloc_coherent(&qdev->pdev->dev, size, &dma_addr, GFP_KERNEL);
+ if (!buf)
+ return -ENOMEM;
+
+ trans_wrapper = add_wrapper(wrappers,
+ offsetof(struct wrapper_msg, trans) + sizeof(*out_trans));
+ if (!trans_wrapper) {
+ ret = -ENOMEM;
+ goto free_dma;
+ }
+ trans_wrapper->len = sizeof(*out_trans);
+ out_trans = (struct wire_trans_activate_to_dev *)&trans_wrapper->trans;
+
+ out_trans->hdr.type = cpu_to_le32(QAIC_TRANS_ACTIVATE_TO_DEV);
+ out_trans->hdr.len = cpu_to_le32(sizeof(*out_trans));
+ out_trans->buf_len = cpu_to_le32(size);
+ out_trans->req_q_addr = cpu_to_le64(dma_addr);
+ out_trans->req_q_size = cpu_to_le32(nelem);
+ out_trans->rsp_q_addr = cpu_to_le64(dma_addr + size - nelem * get_dbc_rsp_elem_size());
+ out_trans->rsp_q_size = cpu_to_le32(nelem);
+ out_trans->options = cpu_to_le32(in_trans->options);
+
+ *user_len += in_trans->hdr.len;
+ msg->hdr.len = cpu_to_le32(msg_hdr_len + sizeof(*out_trans));
+ msg->hdr.count = incr_le32(msg->hdr.count);
+
+ resources->buf = buf;
+ resources->dma_addr = dma_addr;
+ resources->total_size = size;
+ resources->nelem = nelem;
+ resources->rsp_q_base = buf + size - nelem * get_dbc_rsp_elem_size();
+ return 0;
+
+free_dma:
+ dma_free_coherent(&qdev->pdev->dev, size, buf, dma_addr);
+ return ret;
+}
+
+static int encode_deactivate(struct qaic_device *qdev, void *trans,
+ u32 *user_len, struct qaic_user *usr)
+{
+ struct qaic_manage_trans_deactivate *in_trans = trans;
+
+ if (in_trans->dbc_id >= qdev->num_dbc || in_trans->pad)
+ return -EINVAL;
+
+ *user_len += in_trans->hdr.len;
+
+ return disable_dbc(qdev, in_trans->dbc_id, usr);
+}
+
+static int encode_status(struct qaic_device *qdev, void *trans, struct wrapper_list *wrappers,
+ u32 *user_len)
+{
+ struct qaic_manage_trans_status_to_dev *in_trans = trans;
+ struct wire_trans_status_to_dev *out_trans;
+ struct wrapper_msg *trans_wrapper;
+ struct wrapper_msg *wrapper;
+ struct wire_msg *msg;
+ u32 msg_hdr_len;
+
+ wrapper = list_first_entry(&wrappers->list, struct wrapper_msg, list);
+ msg = &wrapper->msg;
+ msg_hdr_len = le32_to_cpu(msg->hdr.len);
+
+ if (msg_hdr_len + in_trans->hdr.len > QAIC_MANAGE_MAX_MSG_LENGTH)
+ return -ENOSPC;
+
+ trans_wrapper = add_wrapper(wrappers, sizeof(*trans_wrapper));
+ if (!trans_wrapper)
+ return -ENOMEM;
+
+ trans_wrapper->len = sizeof(*out_trans);
+ out_trans = (struct wire_trans_status_to_dev *)&trans_wrapper->trans;
+
+ out_trans->hdr.type = cpu_to_le32(QAIC_TRANS_STATUS_TO_DEV);
+ out_trans->hdr.len = cpu_to_le32(in_trans->hdr.len);
+ msg->hdr.len = cpu_to_le32(msg_hdr_len + in_trans->hdr.len);
+ msg->hdr.count = incr_le32(msg->hdr.count);
+ *user_len += in_trans->hdr.len;
+
+ return 0;
+}
+
+static int encode_message(struct qaic_device *qdev, struct manage_msg *user_msg,
+ struct wrapper_list *wrappers, struct ioctl_resources *resources,
+ struct qaic_user *usr)
+{
+ struct qaic_manage_trans_hdr *trans_hdr;
+ struct wrapper_msg *wrapper;
+ struct wire_msg *msg;
+ u32 user_len = 0;
+ int ret;
+ int i;
+
+ if (!user_msg->count) {
+ ret = -EINVAL;
+ goto out;
+ }
+
+ wrapper = list_first_entry(&wrappers->list, struct wrapper_msg, list);
+ msg = &wrapper->msg;
+
+ msg->hdr.len = cpu_to_le32(sizeof(msg->hdr));
+
+ if (resources->dma_chunk_id) {
+ ret = encode_dma(qdev, resources->trans_hdr, wrappers, &user_len, resources, usr);
+ msg->hdr.count = cpu_to_le32(1);
+ goto out;
+ }
+
+ for (i = 0; i < user_msg->count; ++i) {
+ if (user_len >= user_msg->len) {
+ ret = -EINVAL;
+ break;
+ }
+ trans_hdr = (struct qaic_manage_trans_hdr *)(user_msg->data + user_len);
+ if (user_len + trans_hdr->len > user_msg->len) {
+ ret = -EINVAL;
+ break;
+ }
+
+ switch (trans_hdr->type) {
+ case QAIC_TRANS_PASSTHROUGH_FROM_USR:
+ ret = encode_passthrough(qdev, trans_hdr, wrappers, &user_len);
+ break;
+ case QAIC_TRANS_DMA_XFER_FROM_USR:
+ ret = encode_dma(qdev, trans_hdr, wrappers, &user_len, resources, usr);
+ break;
+ case QAIC_TRANS_ACTIVATE_FROM_USR:
+ ret = encode_activate(qdev, trans_hdr, wrappers, &user_len, resources);
+ break;
+ case QAIC_TRANS_DEACTIVATE_FROM_USR:
+ ret = encode_deactivate(qdev, trans_hdr, &user_len, usr);
+ break;
+ case QAIC_TRANS_STATUS_FROM_USR:
+ ret = encode_status(qdev, trans_hdr, wrappers, &user_len);
+ break;
+ default:
+ ret = -EINVAL;
+ break;
+ }
+
+ if (ret)
+ break;
+ }
+
+ if (user_len != user_msg->len)
+ ret = -EINVAL;
+out:
+ if (ret) {
+ free_dma_xfers(qdev, resources);
+ free_dbc_buf(qdev, resources);
+ return ret;
+ }
+
+ return 0;
+}
+
+static int decode_passthrough(struct qaic_device *qdev, void *trans, struct manage_msg *user_msg,
+ u32 *msg_len)
+{
+ struct qaic_manage_trans_passthrough *out_trans;
+ struct wire_trans_passthrough *in_trans = trans;
+ u32 len;
+
+ out_trans = (void *)user_msg->data + user_msg->len;
+
+ len = le32_to_cpu(in_trans->hdr.len);
+ if (len % 8 != 0)
+ return -EINVAL;
+
+ if (user_msg->len + len > QAIC_MANAGE_MAX_MSG_LENGTH)
+ return -ENOSPC;
+
+ memcpy(out_trans->data, in_trans->data, len - sizeof(in_trans->hdr));
+ user_msg->len += len;
+ *msg_len += len;
+ out_trans->hdr.type = le32_to_cpu(in_trans->hdr.type);
+ out_trans->hdr.len = len;
+
+ return 0;
+}
+
+static int decode_activate(struct qaic_device *qdev, void *trans, struct manage_msg *user_msg,
+ u32 *msg_len, struct ioctl_resources *resources, struct qaic_user *usr)
+{
+ struct qaic_manage_trans_activate_from_dev *out_trans;
+ struct wire_trans_activate_from_dev *in_trans = trans;
+ u32 len;
+
+ out_trans = (void *)user_msg->data + user_msg->len;
+
+ len = le32_to_cpu(in_trans->hdr.len);
+ if (user_msg->len + len > QAIC_MANAGE_MAX_MSG_LENGTH)
+ return -ENOSPC;
+
+ user_msg->len += len;
+ *msg_len += len;
+ out_trans->hdr.type = le32_to_cpu(in_trans->hdr.type);
+ out_trans->hdr.len = len;
+ out_trans->status = le32_to_cpu(in_trans->status);
+ out_trans->dbc_id = le32_to_cpu(in_trans->dbc_id);
+ out_trans->options = le64_to_cpu(in_trans->options);
+
+ if (!resources->buf)
+ /* how did we get an activate response without a request? */
+ return -EINVAL;
+
+ if (out_trans->dbc_id >= qdev->num_dbc)
+ /*
+ * The device assigned an invalid resource, which should never
+ * happen. Return an error so the user can try to recover.
+ */
+ return -ENODEV;
+
+ if (out_trans->status)
+ /*
+ * Allocating resources failed on device side. This is not an
+ * expected behaviour, user is expected to handle this situation.
+ */
+ return -ECANCELED;
+
+ resources->status = out_trans->status;
+ resources->dbc_id = out_trans->dbc_id;
+ save_dbc_buf(qdev, resources, usr);
+
+ return 0;
+}
+
+static int decode_deactivate(struct qaic_device *qdev, void *trans, u32 *msg_len,
+ struct qaic_user *usr)
+{
+ struct wire_trans_deactivate_from_dev *in_trans = trans;
+ u32 dbc_id = le32_to_cpu(in_trans->dbc_id);
+ u32 status = le32_to_cpu(in_trans->status);
+
+ if (dbc_id >= qdev->num_dbc)
+ /*
+ * The device assigned an invalid resource, which should never
+ * happen. Inject an error so the user can try to recover.
+ */
+ return -ENODEV;
+
+ if (status) {
+ /*
+ * Releasing resources failed on the device side, which puts
+ * us in a bind since they may still be in use, so enable the
+ * dbc. User is expected to retry deactivation.
+ */
+ enable_dbc(qdev, dbc_id, usr);
+ return -ECANCELED;
+ }
+
+ release_dbc(qdev, dbc_id);
+ *msg_len += sizeof(*in_trans);
+
+ return 0;
+}
+
+static int decode_status(struct qaic_device *qdev, void *trans, struct manage_msg *user_msg,
+ u32 *user_len, struct wire_msg *msg)
+{
+ struct qaic_manage_trans_status_from_dev *out_trans;
+ struct wire_trans_status_from_dev *in_trans = trans;
+ u32 len;
+
+ out_trans = (void *)user_msg->data + user_msg->len;
+
+ len = le32_to_cpu(in_trans->hdr.len);
+ if (user_msg->len + len > QAIC_MANAGE_MAX_MSG_LENGTH)
+ return -ENOSPC;
+
+ out_trans->hdr.type = QAIC_TRANS_STATUS_FROM_DEV;
+ out_trans->hdr.len = len;
+ out_trans->major = le16_to_cpu(in_trans->major);
+ out_trans->minor = le16_to_cpu(in_trans->minor);
+ out_trans->status_flags = le64_to_cpu(in_trans->status_flags);
+ out_trans->status = le32_to_cpu(in_trans->status);
+ *user_len += le32_to_cpu(in_trans->hdr.len);
+ user_msg->len += len;
+
+ if (out_trans->status)
+ return -ECANCELED;
+ if (out_trans->status_flags & BIT(0) && !valid_crc(msg))
+ return -EPIPE;
+
+ return 0;
+}
+
+static int decode_message(struct qaic_device *qdev, struct manage_msg *user_msg,
+ struct wire_msg *msg, struct ioctl_resources *resources,
+ struct qaic_user *usr)
+{
+ u32 msg_hdr_len = le32_to_cpu(msg->hdr.len);
+ struct wire_trans_hdr *trans_hdr;
+ u32 msg_len = 0;
+ int ret;
+ int i;
+
+ if (msg_hdr_len > QAIC_MANAGE_MAX_MSG_LENGTH)
+ return -EINVAL;
+
+ user_msg->len = 0;
+ user_msg->count = le32_to_cpu(msg->hdr.count);
+
+ for (i = 0; i < user_msg->count; ++i) {
+ trans_hdr = (struct wire_trans_hdr *)(msg->data + msg_len);
+ if (msg_len + le32_to_cpu(trans_hdr->len) > msg_hdr_len)
+ return -EINVAL;
+
+ switch (le32_to_cpu(trans_hdr->type)) {
+ case QAIC_TRANS_PASSTHROUGH_FROM_DEV:
+ ret = decode_passthrough(qdev, trans_hdr, user_msg, &msg_len);
+ break;
+ case QAIC_TRANS_ACTIVATE_FROM_DEV:
+ ret = decode_activate(qdev, trans_hdr, user_msg, &msg_len, resources, usr);
+ break;
+ case QAIC_TRANS_DEACTIVATE_FROM_DEV:
+ ret = decode_deactivate(qdev, trans_hdr, &msg_len, usr);
+ break;
+ case QAIC_TRANS_STATUS_FROM_DEV:
+ ret = decode_status(qdev, trans_hdr, user_msg, &msg_len, msg);
+ break;
+ default:
+ return -EINVAL;
+ }
+
+ if (ret)
+ return ret;
+ }
+
+ if (msg_len != (msg_hdr_len - sizeof(msg->hdr)))
+ return -EINVAL;
+
+ return 0;
+}
+
+static void *msg_xfer(struct qaic_device *qdev, struct wrapper_list *wrappers, u32 seq_num,
+ bool ignore_signal)
+{
+ struct xfer_queue_elem elem;
+ struct wire_msg *out_buf;
+ struct wrapper_msg *w;
+ int retry_count;
+ long ret;
+
+ if (qdev->in_reset) {
+ mutex_unlock(&qdev->cntl_mutex);
+ return ERR_PTR(-ENODEV);
+ }
+
+ elem.seq_num = seq_num;
+ elem.buf = NULL;
+ init_completion(&elem.xfer_done);
+ if (likely(!qdev->cntl_lost_buf)) {
+ /*
+ * The max size of request to device is QAIC_MANAGE_EXT_MSG_LENGTH.
+ * The max size of response from device is QAIC_MANAGE_MAX_MSG_LENGTH.
+ */
+ out_buf = kmalloc(QAIC_MANAGE_MAX_MSG_LENGTH, GFP_KERNEL);
+ if (!out_buf) {
+ mutex_unlock(&qdev->cntl_mutex);
+ return ERR_PTR(-ENOMEM);
+ }
+
+ ret = mhi_queue_buf(qdev->cntl_ch, DMA_FROM_DEVICE, out_buf,
+ QAIC_MANAGE_MAX_MSG_LENGTH, MHI_EOT);
+ if (ret) {
+ mutex_unlock(&qdev->cntl_mutex);
+ return ERR_PTR(ret);
+ }
+ } else {
+ /*
+ * we lost a buffer because we queued a recv buf, but then
+ * queuing the corresponding tx buf failed. To try to avoid
+ * a memory leak, lets reclaim it and use it for this
+ * transaction.
+ */
+ qdev->cntl_lost_buf = false;
+ }
+
+ list_for_each_entry(w, &wrappers->list, list) {
+ kref_get(&w->ref_count);
+ retry_count = 0;
+retry:
+ ret = mhi_queue_buf(qdev->cntl_ch, DMA_TO_DEVICE, &w->msg, w->len,
+ list_is_last(&w->list, &wrappers->list) ? MHI_EOT : MHI_CHAIN);
+ if (ret) {
+ if (ret == -EAGAIN && retry_count++ < QAIC_MHI_RETRY_MAX) {
+ msleep_interruptible(QAIC_MHI_RETRY_WAIT_MS);
+ if (!signal_pending(current))
+ goto retry;
+ }
+
+ qdev->cntl_lost_buf = true;
+ kref_put(&w->ref_count, free_wrapper);
+ mutex_unlock(&qdev->cntl_mutex);
+ return ERR_PTR(ret);
+ }
+ }
+
+ list_add_tail(&elem.list, &qdev->cntl_xfer_list);
+ mutex_unlock(&qdev->cntl_mutex);
+
+ if (ignore_signal)
+ ret = wait_for_completion_timeout(&elem.xfer_done, control_resp_timeout_s * HZ);
+ else
+ ret = wait_for_completion_interruptible_timeout(&elem.xfer_done,
+ control_resp_timeout_s * HZ);
+ /*
+ * not using _interruptable because we have to cleanup or we'll
+ * likely cause memory corruption
+ */
+ mutex_lock(&qdev->cntl_mutex);
+ if (!list_empty(&elem.list))
+ list_del(&elem.list);
+ if (!ret && !elem.buf)
+ ret = -ETIMEDOUT;
+ else if (ret > 0 && !elem.buf)
+ ret = -EIO;
+ mutex_unlock(&qdev->cntl_mutex);
+
+ if (ret < 0) {
+ kfree(elem.buf);
+ return ERR_PTR(ret);
+ } else if (!qdev->valid_crc(elem.buf)) {
+ kfree(elem.buf);
+ return ERR_PTR(-EPIPE);
+ }
+
+ return elem.buf;
+}
+
+/* Add a transaction to abort the outstanding DMA continuation */
+static int abort_dma_cont(struct qaic_device *qdev, struct wrapper_list *wrappers, u32 dma_chunk_id)
+{
+ struct wire_trans_dma_xfer *out_trans;
+ u32 size = sizeof(*out_trans);
+ struct wrapper_msg *wrapper;
+ struct wrapper_msg *w;
+ struct wire_msg *msg;
+
+ wrapper = list_first_entry(&wrappers->list, struct wrapper_msg, list);
+ msg = &wrapper->msg;
+
+ /* Remove all but the first wrapper which has the msg header */
+ list_for_each_entry_safe(wrapper, w, &wrappers->list, list)
+ if (!list_is_first(&wrapper->list, &wrappers->list))
+ kref_put(&wrapper->ref_count, free_wrapper);
+
+ wrapper = add_wrapper(wrappers, offsetof(struct wrapper_msg, trans) + sizeof(*out_trans));
+
+ if (!wrapper)
+ return -ENOMEM;
+
+ out_trans = (struct wire_trans_dma_xfer *)&wrapper->trans;
+ out_trans->hdr.type = cpu_to_le32(QAIC_TRANS_DMA_XFER_TO_DEV);
+ out_trans->hdr.len = cpu_to_le32(size);
+ out_trans->tag = cpu_to_le32(0);
+ out_trans->count = cpu_to_le32(0);
+ out_trans->dma_chunk_id = cpu_to_le32(dma_chunk_id);
+
+ msg->hdr.len = cpu_to_le32(size + sizeof(*msg));
+ msg->hdr.count = cpu_to_le32(1);
+ wrapper->len = size;
+
+ return 0;
+}
+
+static struct wrapper_list *alloc_wrapper_list(void)
+{
+ struct wrapper_list *wrappers;
+
+ wrappers = kmalloc(sizeof(*wrappers), GFP_KERNEL);
+ if (!wrappers)
+ return NULL;
+ INIT_LIST_HEAD(&wrappers->list);
+ spin_lock_init(&wrappers->lock);
+
+ return wrappers;
+}
+
+static int qaic_manage_msg_xfer(struct qaic_device *qdev, struct qaic_user *usr,
+ struct manage_msg *user_msg, struct ioctl_resources *resources,
+ struct wire_msg **rsp)
+{
+ struct wrapper_list *wrappers;
+ struct wrapper_msg *wrapper;
+ struct wrapper_msg *w;
+ bool all_done = false;
+ struct wire_msg *msg;
+ int ret;
+
+ wrappers = alloc_wrapper_list();
+ if (!wrappers)
+ return -ENOMEM;
+
+ wrapper = add_wrapper(wrappers, sizeof(*wrapper));
+ if (!wrapper) {
+ kfree(wrappers);
+ return -ENOMEM;
+ }
+
+ msg = &wrapper->msg;
+ wrapper->len = sizeof(*msg);
+
+ ret = encode_message(qdev, user_msg, wrappers, resources, usr);
+ if (ret && resources->dma_chunk_id)
+ ret = abort_dma_cont(qdev, wrappers, resources->dma_chunk_id);
+ if (ret)
+ goto encode_failed;
+
+ ret = mutex_lock_interruptible(&qdev->cntl_mutex);
+ if (ret)
+ goto lock_failed;
+
+ msg->hdr.magic_number = MANAGE_MAGIC_NUMBER;
+ msg->hdr.sequence_number = cpu_to_le32(qdev->next_seq_num++);
+
+ if (usr) {
+ msg->hdr.handle = cpu_to_le32(usr->handle);
+ msg->hdr.partition_id = cpu_to_le32(usr->qddev->partition_id);
+ } else {
+ msg->hdr.handle = 0;
+ msg->hdr.partition_id = cpu_to_le32(QAIC_NO_PARTITION);
+ }
+
+ msg->hdr.padding = cpu_to_le32(0);
+ msg->hdr.crc32 = cpu_to_le32(qdev->gen_crc(wrappers));
+
+ /* msg_xfer releases the mutex */
+ *rsp = msg_xfer(qdev, wrappers, qdev->next_seq_num - 1, false);
+ if (IS_ERR(*rsp))
+ ret = PTR_ERR(*rsp);
+
+lock_failed:
+ free_dma_xfers(qdev, resources);
+encode_failed:
+ spin_lock(&wrappers->lock);
+ list_for_each_entry_safe(wrapper, w, &wrappers->list, list)
+ kref_put(&wrapper->ref_count, free_wrapper);
+ all_done = list_empty(&wrappers->list);
+ spin_unlock(&wrappers->lock);
+ if (all_done)
+ kfree(wrappers);
+
+ return ret;
+}
+
+static int qaic_manage(struct qaic_device *qdev, struct qaic_user *usr, struct manage_msg *user_msg)
+{
+ struct wire_trans_dma_xfer_cont *dma_cont = NULL;
+ struct ioctl_resources resources;
+ struct wire_msg *rsp = NULL;
+ int ret;
+
+ memset(&resources, 0, sizeof(struct ioctl_resources));
+
+ INIT_LIST_HEAD(&resources.dma_xfers);
+
+ if (user_msg->len > QAIC_MANAGE_MAX_MSG_LENGTH ||
+ user_msg->count > QAIC_MANAGE_MAX_MSG_LENGTH / sizeof(struct qaic_manage_trans_hdr))
+ return -EINVAL;
+
+dma_xfer_continue:
+ ret = qaic_manage_msg_xfer(qdev, usr, user_msg, &resources, &rsp);
+ if (ret)
+ return ret;
+ /* dma_cont should be the only transaction if present */
+ if (le32_to_cpu(rsp->hdr.count) == 1) {
+ dma_cont = (struct wire_trans_dma_xfer_cont *)rsp->data;
+ if (le32_to_cpu(dma_cont->hdr.type) != QAIC_TRANS_DMA_XFER_CONT)
+ dma_cont = NULL;
+ }
+ if (dma_cont) {
+ if (le32_to_cpu(dma_cont->dma_chunk_id) == resources.dma_chunk_id &&
+ le64_to_cpu(dma_cont->xferred_size) == resources.xferred_dma_size) {
+ kfree(rsp);
+ goto dma_xfer_continue;
+ }
+
+ ret = -EINVAL;
+ goto dma_cont_failed;
+ }
+
+ ret = decode_message(qdev, user_msg, rsp, &resources, usr);
+
+dma_cont_failed:
+ free_dbc_buf(qdev, &resources);
+ kfree(rsp);
+ return ret;
+}
+
+int qaic_manage_ioctl(struct drm_device *dev, void *data, struct drm_file *file_priv)
+{
+ struct qaic_manage_msg *user_msg;
+ struct qaic_device *qdev;
+ struct manage_msg *msg;
+ struct qaic_user *usr;
+ u8 __user *user_data;
+ int qdev_rcu_id;
+ int usr_rcu_id;
+ int ret;
+
+ usr = file_priv->driver_priv;
+
+ usr_rcu_id = srcu_read_lock(&usr->qddev_lock);
+ if (!usr->qddev) {
+ srcu_read_unlock(&usr->qddev_lock, usr_rcu_id);
+ return -ENODEV;
+ }
+
+ qdev = usr->qddev->qdev;
+
+ qdev_rcu_id = srcu_read_lock(&qdev->dev_lock);
+ if (qdev->in_reset) {
+ srcu_read_unlock(&qdev->dev_lock, qdev_rcu_id);
+ srcu_read_unlock(&usr->qddev_lock, usr_rcu_id);
+ return -ENODEV;
+ }
+
+ user_msg = data;
+
+ if (user_msg->len > QAIC_MANAGE_MAX_MSG_LENGTH) {
+ ret = -EINVAL;
+ goto out;
+ }
+
+ msg = kzalloc(QAIC_MANAGE_MAX_MSG_LENGTH + sizeof(*msg), GFP_KERNEL);
+ if (!msg) {
+ ret = -ENOMEM;
+ goto out;
+ }
+
+ msg->len = user_msg->len;
+ msg->count = user_msg->count;
+
+ user_data = u64_to_user_ptr(user_msg->data);
+
+ if (copy_from_user(msg->data, user_data, user_msg->len)) {
+ ret = -EFAULT;
+ goto free_msg;
+ }
+
+ ret = qaic_manage(qdev, usr, msg);
+
+ /*
+ * If the qaic_manage() is successful then we copy the message onto
+ * userspace memory but we have an exception for -ECANCELED.
+ * For -ECANCELED, it means that device has NACKed the message with a
+ * status error code which userspace would like to know.
+ */
+ if (ret == -ECANCELED || !ret) {
+ if (copy_to_user(user_data, msg->data, msg->len)) {
+ ret = -EFAULT;
+ } else {
+ user_msg->len = msg->len;
+ user_msg->count = msg->count;
+ }
+ }
+
+free_msg:
+ kfree(msg);
+out:
+ srcu_read_unlock(&qdev->dev_lock, qdev_rcu_id);
+ srcu_read_unlock(&usr->qddev_lock, usr_rcu_id);
+ return ret;
+}
+
+int get_cntl_version(struct qaic_device *qdev, struct qaic_user *usr, u16 *major, u16 *minor)
+{
+ struct qaic_manage_trans_status_from_dev *status_result;
+ struct qaic_manage_trans_status_to_dev *status_query;
+ struct manage_msg *user_msg;
+ int ret;
+
+ user_msg = kmalloc(sizeof(*user_msg) + sizeof(*status_result), GFP_KERNEL);
+ if (!user_msg) {
+ ret = -ENOMEM;
+ goto out;
+ }
+ user_msg->len = sizeof(*status_query);
+ user_msg->count = 1;
+
+ status_query = (struct qaic_manage_trans_status_to_dev *)user_msg->data;
+ status_query->hdr.type = QAIC_TRANS_STATUS_FROM_USR;
+ status_query->hdr.len = sizeof(status_query->hdr);
+
+ ret = qaic_manage(qdev, usr, user_msg);
+ if (ret)
+ goto kfree_user_msg;
+ status_result = (struct qaic_manage_trans_status_from_dev *)user_msg->data;
+ *major = status_result->major;
+ *minor = status_result->minor;
+
+ if (status_result->status_flags & BIT(0)) { /* device is using CRC */
+ /* By default qdev->gen_crc is programmed to generate CRC */
+ qdev->valid_crc = valid_crc;
+ } else {
+ /* By default qdev->valid_crc is programmed to bypass CRC */
+ qdev->gen_crc = gen_crc_stub;
+ }
+
+kfree_user_msg:
+ kfree(user_msg);
+out:
+ return ret;
+}
+
+static void resp_worker(struct work_struct *work)
+{
+ struct resp_work *resp = container_of(work, struct resp_work, work);
+ struct qaic_device *qdev = resp->qdev;
+ struct wire_msg *msg = resp->buf;
+ struct xfer_queue_elem *elem;
+ struct xfer_queue_elem *i;
+ bool found = false;
+
+ mutex_lock(&qdev->cntl_mutex);
+ list_for_each_entry_safe(elem, i, &qdev->cntl_xfer_list, list) {
+ if (elem->seq_num == le32_to_cpu(msg->hdr.sequence_number)) {
+ found = true;
+ list_del_init(&elem->list);
+ elem->buf = msg;
+ complete_all(&elem->xfer_done);
+ break;
+ }
+ }
+ mutex_unlock(&qdev->cntl_mutex);
+
+ if (!found)
+ /* request must have timed out, drop packet */
+ kfree(msg);
+
+ kfree(resp);
+}
+
+static void free_wrapper_from_list(struct wrapper_list *wrappers, struct wrapper_msg *wrapper)
+{
+ bool all_done = false;
+
+ spin_lock(&wrappers->lock);
+ kref_put(&wrapper->ref_count, free_wrapper);
+ all_done = list_empty(&wrappers->list);
+ spin_unlock(&wrappers->lock);
+
+ if (all_done)
+ kfree(wrappers);
+}
+
+void qaic_mhi_ul_xfer_cb(struct mhi_device *mhi_dev, struct mhi_result *mhi_result)
+{
+ struct wire_msg *msg = mhi_result->buf_addr;
+ struct wrapper_msg *wrapper = container_of(msg, struct wrapper_msg, msg);
+
+ free_wrapper_from_list(wrapper->head, wrapper);
+}
+
+void qaic_mhi_dl_xfer_cb(struct mhi_device *mhi_dev, struct mhi_result *mhi_result)
+{
+ struct qaic_device *qdev = dev_get_drvdata(&mhi_dev->dev);
+ struct wire_msg *msg = mhi_result->buf_addr;
+ struct resp_work *resp;
+
+ if (mhi_result->transaction_status || msg->hdr.magic_number != MANAGE_MAGIC_NUMBER) {
+ kfree(msg);
+ return;
+ }
+
+ resp = kmalloc(sizeof(*resp), GFP_ATOMIC);
+ if (!resp) {
+ pci_err(qdev->pdev, "dl_xfer_cb alloc fail, dropping message\n");
+ kfree(msg);
+ return;
+ }
+
+ INIT_WORK(&resp->work, resp_worker);
+ resp->qdev = qdev;
+ resp->buf = msg;
+ queue_work(qdev->cntl_wq, &resp->work);
+}
+
+int qaic_control_open(struct qaic_device *qdev)
+{
+ if (!qdev->cntl_ch)
+ return -ENODEV;
+
+ qdev->cntl_lost_buf = false;
+ /*
+ * By default qaic should assume that device has CRC enabled.
+ * Qaic comes to know if device has CRC enabled or disabled during the
+ * device status transaction, which is the first transaction performed
+ * on control channel.
+ *
+ * So CRC validation of first device status transaction response is
+ * ignored (by calling valid_crc_stub) and is done later during decoding
+ * if device has CRC enabled.
+ * Now that qaic knows whether device has CRC enabled or not it acts
+ * accordingly.
+ */
+ qdev->gen_crc = gen_crc;
+ qdev->valid_crc = valid_crc_stub;
+
+ return mhi_prepare_for_transfer(qdev->cntl_ch);
+}
+
+void qaic_control_close(struct qaic_device *qdev)
+{
+ mhi_unprepare_from_transfer(qdev->cntl_ch);
+}
+
+void qaic_release_usr(struct qaic_device *qdev, struct qaic_user *usr)
+{
+ struct wire_trans_terminate_to_dev *trans;
+ struct wrapper_list *wrappers;
+ struct wrapper_msg *wrapper;
+ struct wire_msg *msg;
+ struct wire_msg *rsp;
+
+ wrappers = alloc_wrapper_list();
+ if (!wrappers)
+ return;
+
+ wrapper = add_wrapper(wrappers, sizeof(*wrapper) + sizeof(*msg) + sizeof(*trans));
+ if (!wrapper)
+ return;
+
+ msg = &wrapper->msg;
+
+ trans = (struct wire_trans_terminate_to_dev *)msg->data;
+
+ trans->hdr.type = cpu_to_le32(QAIC_TRANS_TERMINATE_TO_DEV);
+ trans->hdr.len = cpu_to_le32(sizeof(*trans));
+ trans->handle = cpu_to_le32(usr->handle);
+
+ mutex_lock(&qdev->cntl_mutex);
+ wrapper->len = sizeof(msg->hdr) + sizeof(*trans);
+ msg->hdr.magic_number = MANAGE_MAGIC_NUMBER;
+ msg->hdr.sequence_number = cpu_to_le32(qdev->next_seq_num++);
+ msg->hdr.len = cpu_to_le32(wrapper->len);
+ msg->hdr.count = cpu_to_le32(1);
+ msg->hdr.handle = cpu_to_le32(usr->handle);
+ msg->hdr.padding = cpu_to_le32(0);
+ msg->hdr.crc32 = cpu_to_le32(qdev->gen_crc(wrappers));
+
+ /*
+ * msg_xfer releases the mutex
+ * We don't care about the return of msg_xfer since we will not do
+ * anything different based on what happens.
+ * We ignore pending signals since one will be set if the user is
+ * killed, and we need give the device a chance to cleanup, otherwise
+ * DMA may still be in progress when we return.
+ */
+ rsp = msg_xfer(qdev, wrappers, qdev->next_seq_num - 1, true);
+ if (!IS_ERR(rsp))
+ kfree(rsp);
+ free_wrapper_from_list(wrappers, wrapper);
+}
+
+void wake_all_cntl(struct qaic_device *qdev)
+{
+ struct xfer_queue_elem *elem;
+ struct xfer_queue_elem *i;
+
+ mutex_lock(&qdev->cntl_mutex);
+ list_for_each_entry_safe(elem, i, &qdev->cntl_xfer_list, list) {
+ list_del_init(&elem->list);
+ complete_all(&elem->xfer_done);
+ }
+ mutex_unlock(&qdev->cntl_mutex);
+}
--
2.7.4
^ permalink raw reply related [flat|nested] 16+ messages in thread* [PATCH v5 5/8] accel/qaic: Add datapath
2023-03-27 15:54 [PATCH v5 0/8] QAIC accel driver Jeffrey Hugo
` (3 preceding siblings ...)
2023-03-27 15:54 ` [PATCH v5 4/8] accel/qaic: Add control path Jeffrey Hugo
@ 2023-03-27 15:54 ` Jeffrey Hugo
2023-03-27 15:54 ` [PATCH v5 6/8] accel/qaic: Add mhi_qaic_cntl Jeffrey Hugo
` (3 subsequent siblings)
8 siblings, 0 replies; 16+ messages in thread
From: Jeffrey Hugo @ 2023-03-27 15:54 UTC (permalink / raw)
To: ogabbay, airlied, daniel
Cc: jacek.lawrynowicz, stanislaw.gruszka, dri-devel, quic_pkanojiy,
quic_carlv, quic_ajitpals, linux-doc, linux-arm-msm, bagasdotme,
mani, Jeffrey Hugo
Add the datapath component that manages BOs and submits them to running
workloads on the qaic device via the dma_bridge hardware. This allows
QAIC clients to interact with their workloads (run inferences) via the
following ioctls along with mmap():
DRM_IOCTL_QAIC_CREATE_BO
DRM_IOCTL_QAIC_MMAP_BO
DRM_IOCTL_QAIC_ATTACH_SLICE_BO
DRM_IOCTL_QAIC_EXECUTE_BO
DRM_IOCTL_QAIC_PARTIAL_EXECUTE_BO
DRM_IOCTL_QAIC_WAIT_BO
DRM_IOCTL_QAIC_PERF_STATS_BO
Signed-off-by: Jeffrey Hugo <quic_jhugo@quicinc.com>
Reviewed-by: Carl Vanderlip <quic_carlv@quicinc.com>
Reviewed-by: Pranjal Ramajor Asha Kanojiya <quic_pkanojiy@quicinc.com>
Reviewed-by: Stanislaw Gruszka <stanislaw.gruszka@linux.intel.com>
Reviewed-by: Jacek Lawrynowicz <jacek.lawrynowicz@linux.intel.com>
---
drivers/accel/qaic/qaic_data.c | 1903 ++++++++++++++++++++++++++++++++++++++++
1 file changed, 1903 insertions(+)
create mode 100644 drivers/accel/qaic/qaic_data.c
diff --git a/drivers/accel/qaic/qaic_data.c b/drivers/accel/qaic/qaic_data.c
new file mode 100644
index 0000000..c186e6c
--- /dev/null
+++ b/drivers/accel/qaic/qaic_data.c
@@ -0,0 +1,1903 @@
+// SPDX-License-Identifier: GPL-2.0-only
+
+/* Copyright (c) 2019-2021, The Linux Foundation. All rights reserved. */
+/* Copyright (c) 2021-2023 Qualcomm Innovation Center, Inc. All rights reserved. */
+
+#include <linux/bitfield.h>
+#include <linux/bits.h>
+#include <linux/completion.h>
+#include <linux/delay.h>
+#include <linux/dma-buf.h>
+#include <linux/dma-mapping.h>
+#include <linux/interrupt.h>
+#include <linux/kref.h>
+#include <linux/list.h>
+#include <linux/math64.h>
+#include <linux/mm.h>
+#include <linux/moduleparam.h>
+#include <linux/scatterlist.h>
+#include <linux/spinlock.h>
+#include <linux/srcu.h>
+#include <linux/types.h>
+#include <linux/uaccess.h>
+#include <linux/wait.h>
+#include <drm/drm_file.h>
+#include <drm/drm_gem.h>
+#include <drm/drm_print.h>
+#include <uapi/drm/qaic_accel.h>
+
+#include "qaic.h"
+
+#define SEM_VAL_MASK GENMASK_ULL(11, 0)
+#define SEM_INDEX_MASK GENMASK_ULL(4, 0)
+#define BULK_XFER BIT(3)
+#define GEN_COMPLETION BIT(4)
+#define INBOUND_XFER 1
+#define OUTBOUND_XFER 2
+#define REQHP_OFF 0x0 /* we read this */
+#define REQTP_OFF 0x4 /* we write this */
+#define RSPHP_OFF 0x8 /* we write this */
+#define RSPTP_OFF 0xc /* we read this */
+
+#define ENCODE_SEM(val, index, sync, cmd, flags) \
+ ({ \
+ FIELD_PREP(GENMASK(11, 0), (val)) | \
+ FIELD_PREP(GENMASK(20, 16), (index)) | \
+ FIELD_PREP(BIT(22), (sync)) | \
+ FIELD_PREP(GENMASK(26, 24), (cmd)) | \
+ FIELD_PREP(GENMASK(30, 29), (flags)) | \
+ FIELD_PREP(BIT(31), (cmd) ? 1 : 0); \
+ })
+#define NUM_EVENTS 128
+#define NUM_DELAYS 10
+
+static unsigned int wait_exec_default_timeout_ms = 5000; /* 5 sec default */
+module_param(wait_exec_default_timeout_ms, uint, 0600);
+MODULE_PARM_DESC(wait_exec_default_timeout_ms, "Default timeout for DRM_IOCTL_QAIC_WAIT_BO");
+
+static unsigned int datapath_poll_interval_us = 100; /* 100 usec default */
+module_param(datapath_poll_interval_us, uint, 0600);
+MODULE_PARM_DESC(datapath_poll_interval_us,
+ "Amount of time to sleep between activity when datapath polling is enabled");
+
+struct dbc_req {
+ /*
+ * A request ID is assigned to each memory handle going in DMA queue.
+ * As a single memory handle can enqueue multiple elements in DMA queue
+ * all of them will have the same request ID.
+ */
+ __le16 req_id;
+ /* Future use */
+ __u8 seq_id;
+ /*
+ * Special encoded variable
+ * 7 0 - Do not force to generate MSI after DMA is completed
+ * 1 - Force to generate MSI after DMA is completed
+ * 6:5 Reserved
+ * 4 1 - Generate completion element in the response queue
+ * 0 - No Completion Code
+ * 3 0 - DMA request is a Link list transfer
+ * 1 - DMA request is a Bulk transfer
+ * 2 Reserved
+ * 1:0 00 - No DMA transfer involved
+ * 01 - DMA transfer is part of inbound transfer
+ * 10 - DMA transfer has outbound transfer
+ * 11 - NA
+ */
+ __u8 cmd;
+ __le32 resv;
+ /* Source address for the transfer */
+ __le64 src_addr;
+ /* Destination address for the transfer */
+ __le64 dest_addr;
+ /* Length of transfer request */
+ __le32 len;
+ __le32 resv2;
+ /* Doorbell address */
+ __le64 db_addr;
+ /*
+ * Special encoded variable
+ * 7 1 - Doorbell(db) write
+ * 0 - No doorbell write
+ * 6:2 Reserved
+ * 1:0 00 - 32 bit access, db address must be aligned to 32bit-boundary
+ * 01 - 16 bit access, db address must be aligned to 16bit-boundary
+ * 10 - 8 bit access, db address must be aligned to 8bit-boundary
+ * 11 - Reserved
+ */
+ __u8 db_len;
+ __u8 resv3;
+ __le16 resv4;
+ /* 32 bit data written to doorbell address */
+ __le32 db_data;
+ /*
+ * Special encoded variable
+ * All the fields of sem_cmdX are passed from user and all are ORed
+ * together to form sem_cmd.
+ * 0:11 Semaphore value
+ * 15:12 Reserved
+ * 20:16 Semaphore index
+ * 21 Reserved
+ * 22 Semaphore Sync
+ * 23 Reserved
+ * 26:24 Semaphore command
+ * 28:27 Reserved
+ * 29 Semaphore DMA out bound sync fence
+ * 30 Semaphore DMA in bound sync fence
+ * 31 Enable semaphore command
+ */
+ __le32 sem_cmd0;
+ __le32 sem_cmd1;
+ __le32 sem_cmd2;
+ __le32 sem_cmd3;
+} __packed;
+
+struct dbc_rsp {
+ /* Request ID of the memory handle whose DMA transaction is completed */
+ __le16 req_id;
+ /* Status of the DMA transaction. 0 : Success otherwise failure */
+ __le16 status;
+} __packed;
+
+inline int get_dbc_req_elem_size(void)
+{
+ return sizeof(struct dbc_req);
+}
+
+inline int get_dbc_rsp_elem_size(void)
+{
+ return sizeof(struct dbc_rsp);
+}
+
+static void free_slice(struct kref *kref)
+{
+ struct bo_slice *slice = container_of(kref, struct bo_slice, ref_count);
+
+ list_del(&slice->slice);
+ drm_gem_object_put(&slice->bo->base);
+ sg_free_table(slice->sgt);
+ kfree(slice->sgt);
+ kfree(slice->reqs);
+ kfree(slice);
+}
+
+static int clone_range_of_sgt_for_slice(struct qaic_device *qdev, struct sg_table **sgt_out,
+ struct sg_table *sgt_in, u64 size, u64 offset)
+{
+ int total_len, len, nents, offf = 0, offl = 0;
+ struct scatterlist *sg, *sgn, *sgf, *sgl;
+ struct sg_table *sgt;
+ int ret, j;
+
+ /* find out number of relevant nents needed for this mem */
+ total_len = 0;
+ sgf = NULL;
+ sgl = NULL;
+ nents = 0;
+
+ size = size ? size : PAGE_SIZE;
+ for (sg = sgt_in->sgl; sg; sg = sg_next(sg)) {
+ len = sg_dma_len(sg);
+
+ if (!len)
+ continue;
+ if (offset >= total_len && offset < total_len + len) {
+ sgf = sg;
+ offf = offset - total_len;
+ }
+ if (sgf)
+ nents++;
+ if (offset + size >= total_len &&
+ offset + size <= total_len + len) {
+ sgl = sg;
+ offl = offset + size - total_len;
+ break;
+ }
+ total_len += len;
+ }
+
+ if (!sgf || !sgl) {
+ ret = -EINVAL;
+ goto out;
+ }
+
+ sgt = kzalloc(sizeof(*sgt), GFP_KERNEL);
+ if (!sgt) {
+ ret = -ENOMEM;
+ goto out;
+ }
+
+ ret = sg_alloc_table(sgt, nents, GFP_KERNEL);
+ if (ret)
+ goto free_sgt;
+
+ /* copy relevant sg node and fix page and length */
+ sgn = sgf;
+ for_each_sgtable_sg(sgt, sg, j) {
+ memcpy(sg, sgn, sizeof(*sg));
+ if (sgn == sgf) {
+ sg_dma_address(sg) += offf;
+ sg_dma_len(sg) -= offf;
+ sg_set_page(sg, sg_page(sgn), sg_dma_len(sg), offf);
+ } else {
+ offf = 0;
+ }
+ if (sgn == sgl) {
+ sg_dma_len(sg) = offl - offf;
+ sg_set_page(sg, sg_page(sgn), offl - offf, offf);
+ sg_mark_end(sg);
+ break;
+ }
+ sgn = sg_next(sgn);
+ }
+
+ *sgt_out = sgt;
+ return ret;
+
+free_sgt:
+ kfree(sgt);
+out:
+ *sgt_out = NULL;
+ return ret;
+}
+
+static int encode_reqs(struct qaic_device *qdev, struct bo_slice *slice,
+ struct qaic_attach_slice_entry *req)
+{
+ __le64 db_addr = cpu_to_le64(req->db_addr);
+ __le32 db_data = cpu_to_le32(req->db_data);
+ struct scatterlist *sg;
+ __u8 cmd = BULK_XFER;
+ int presync_sem;
+ u64 dev_addr;
+ __u8 db_len;
+ int i;
+
+ if (!slice->no_xfer)
+ cmd |= (slice->dir == DMA_TO_DEVICE ? INBOUND_XFER : OUTBOUND_XFER);
+
+ if (req->db_len && !IS_ALIGNED(req->db_addr, req->db_len / 8))
+ return -EINVAL;
+
+ presync_sem = req->sem0.presync + req->sem1.presync + req->sem2.presync + req->sem3.presync;
+ if (presync_sem > 1)
+ return -EINVAL;
+
+ presync_sem = req->sem0.presync << 0 | req->sem1.presync << 1 |
+ req->sem2.presync << 2 | req->sem3.presync << 3;
+
+ switch (req->db_len) {
+ case 32:
+ db_len = BIT(7);
+ break;
+ case 16:
+ db_len = BIT(7) | 1;
+ break;
+ case 8:
+ db_len = BIT(7) | 2;
+ break;
+ case 0:
+ db_len = 0; /* doorbell is not active for this command */
+ break;
+ default:
+ return -EINVAL; /* should never hit this */
+ }
+
+ /*
+ * When we end up splitting up a single request (ie a buf slice) into
+ * multiple DMA requests, we have to manage the sync data carefully.
+ * There can only be one presync sem. That needs to be on every xfer
+ * so that the DMA engine doesn't transfer data before the receiver is
+ * ready. We only do the doorbell and postsync sems after the xfer.
+ * To guarantee previous xfers for the request are complete, we use a
+ * fence.
+ */
+ dev_addr = req->dev_addr;
+ for_each_sgtable_sg(slice->sgt, sg, i) {
+ slice->reqs[i].cmd = cmd;
+ slice->reqs[i].src_addr = cpu_to_le64(slice->dir == DMA_TO_DEVICE ?
+ sg_dma_address(sg) : dev_addr);
+ slice->reqs[i].dest_addr = cpu_to_le64(slice->dir == DMA_TO_DEVICE ?
+ dev_addr : sg_dma_address(sg));
+ /*
+ * sg_dma_len(sg) returns size of a DMA segment, maximum DMA
+ * segment size is set to UINT_MAX by qaic and hence return
+ * values of sg_dma_len(sg) can never exceed u32 range. So,
+ * by down sizing we are not corrupting the value.
+ */
+ slice->reqs[i].len = cpu_to_le32((u32)sg_dma_len(sg));
+ switch (presync_sem) {
+ case BIT(0):
+ slice->reqs[i].sem_cmd0 = cpu_to_le32(ENCODE_SEM(req->sem0.val,
+ req->sem0.index,
+ req->sem0.presync,
+ req->sem0.cmd,
+ req->sem0.flags));
+ break;
+ case BIT(1):
+ slice->reqs[i].sem_cmd1 = cpu_to_le32(ENCODE_SEM(req->sem1.val,
+ req->sem1.index,
+ req->sem1.presync,
+ req->sem1.cmd,
+ req->sem1.flags));
+ break;
+ case BIT(2):
+ slice->reqs[i].sem_cmd2 = cpu_to_le32(ENCODE_SEM(req->sem2.val,
+ req->sem2.index,
+ req->sem2.presync,
+ req->sem2.cmd,
+ req->sem2.flags));
+ break;
+ case BIT(3):
+ slice->reqs[i].sem_cmd3 = cpu_to_le32(ENCODE_SEM(req->sem3.val,
+ req->sem3.index,
+ req->sem3.presync,
+ req->sem3.cmd,
+ req->sem3.flags));
+ break;
+ }
+ dev_addr += sg_dma_len(sg);
+ }
+ /* add post transfer stuff to last segment */
+ i--;
+ slice->reqs[i].cmd |= GEN_COMPLETION;
+ slice->reqs[i].db_addr = db_addr;
+ slice->reqs[i].db_len = db_len;
+ slice->reqs[i].db_data = db_data;
+ /*
+ * Add a fence if we have more than one request going to the hardware
+ * representing the entirety of the user request, and the user request
+ * has no presync condition.
+ * Fences are expensive, so we try to avoid them. We rely on the
+ * hardware behavior to avoid needing one when there is a presync
+ * condition. When a presync exists, all requests for that same
+ * presync will be queued into a fifo. Thus, since we queue the
+ * post xfer activity only on the last request we queue, the hardware
+ * will ensure that the last queued request is processed last, thus
+ * making sure the post xfer activity happens at the right time without
+ * a fence.
+ */
+ if (i && !presync_sem)
+ req->sem0.flags |= (slice->dir == DMA_TO_DEVICE ?
+ QAIC_SEM_INSYNCFENCE : QAIC_SEM_OUTSYNCFENCE);
+ slice->reqs[i].sem_cmd0 = cpu_to_le32(ENCODE_SEM(req->sem0.val, req->sem0.index,
+ req->sem0.presync, req->sem0.cmd,
+ req->sem0.flags));
+ slice->reqs[i].sem_cmd1 = cpu_to_le32(ENCODE_SEM(req->sem1.val, req->sem1.index,
+ req->sem1.presync, req->sem1.cmd,
+ req->sem1.flags));
+ slice->reqs[i].sem_cmd2 = cpu_to_le32(ENCODE_SEM(req->sem2.val, req->sem2.index,
+ req->sem2.presync, req->sem2.cmd,
+ req->sem2.flags));
+ slice->reqs[i].sem_cmd3 = cpu_to_le32(ENCODE_SEM(req->sem3.val, req->sem3.index,
+ req->sem3.presync, req->sem3.cmd,
+ req->sem3.flags));
+
+ return 0;
+}
+
+static int qaic_map_one_slice(struct qaic_device *qdev, struct qaic_bo *bo,
+ struct qaic_attach_slice_entry *slice_ent)
+{
+ struct sg_table *sgt = NULL;
+ struct bo_slice *slice;
+ int ret;
+
+ ret = clone_range_of_sgt_for_slice(qdev, &sgt, bo->sgt, slice_ent->size, slice_ent->offset);
+ if (ret)
+ goto out;
+
+ slice = kmalloc(sizeof(*slice), GFP_KERNEL);
+ if (!slice) {
+ ret = -ENOMEM;
+ goto free_sgt;
+ }
+
+ slice->reqs = kcalloc(sgt->nents, sizeof(*slice->reqs), GFP_KERNEL);
+ if (!slice->reqs) {
+ ret = -ENOMEM;
+ goto free_slice;
+ }
+
+ slice->no_xfer = !slice_ent->size;
+ slice->sgt = sgt;
+ slice->nents = sgt->nents;
+ slice->dir = bo->dir;
+ slice->bo = bo;
+ slice->size = slice_ent->size;
+ slice->offset = slice_ent->offset;
+
+ ret = encode_reqs(qdev, slice, slice_ent);
+ if (ret)
+ goto free_req;
+
+ bo->total_slice_nents += sgt->nents;
+ kref_init(&slice->ref_count);
+ drm_gem_object_get(&bo->base);
+ list_add_tail(&slice->slice, &bo->slices);
+
+ return 0;
+
+free_req:
+ kfree(slice->reqs);
+free_slice:
+ kfree(slice);
+free_sgt:
+ sg_free_table(sgt);
+ kfree(sgt);
+out:
+ return ret;
+}
+
+static int create_sgt(struct qaic_device *qdev, struct sg_table **sgt_out, u64 size)
+{
+ struct scatterlist *sg;
+ struct sg_table *sgt;
+ struct page **pages;
+ int *pages_order;
+ int buf_extra;
+ int max_order;
+ int nr_pages;
+ int ret = 0;
+ int i, j, k;
+ int order;
+
+ if (size) {
+ nr_pages = DIV_ROUND_UP(size, PAGE_SIZE);
+ /*
+ * calculate how much extra we are going to allocate, to remove
+ * later
+ */
+ buf_extra = (PAGE_SIZE - size % PAGE_SIZE) % PAGE_SIZE;
+ max_order = min(MAX_ORDER - 1, get_order(size));
+ } else {
+ /* allocate a single page for book keeping */
+ nr_pages = 1;
+ buf_extra = 0;
+ max_order = 0;
+ }
+
+ pages = kvmalloc_array(nr_pages, sizeof(*pages) + sizeof(*pages_order), GFP_KERNEL);
+ if (!pages) {
+ ret = -ENOMEM;
+ goto out;
+ }
+ pages_order = (void *)pages + sizeof(*pages) * nr_pages;
+
+ /*
+ * Allocate requested memory using alloc_pages. It is possible to allocate
+ * the requested memory in multiple chunks by calling alloc_pages
+ * multiple times. Use SG table to handle multiple allocated pages.
+ */
+ i = 0;
+ while (nr_pages > 0) {
+ order = min(get_order(nr_pages * PAGE_SIZE), max_order);
+ while (1) {
+ pages[i] = alloc_pages(GFP_KERNEL | GFP_HIGHUSER |
+ __GFP_NOWARN | __GFP_ZERO |
+ (order ? __GFP_NORETRY : __GFP_RETRY_MAYFAIL),
+ order);
+ if (pages[i])
+ break;
+ if (!order--) {
+ ret = -ENOMEM;
+ goto free_partial_alloc;
+ }
+ }
+
+ max_order = order;
+ pages_order[i] = order;
+
+ nr_pages -= 1 << order;
+ if (nr_pages <= 0)
+ /* account for over allocation */
+ buf_extra += abs(nr_pages) * PAGE_SIZE;
+ i++;
+ }
+
+ sgt = kmalloc(sizeof(*sgt), GFP_KERNEL);
+ if (!sgt) {
+ ret = -ENOMEM;
+ goto free_partial_alloc;
+ }
+
+ if (sg_alloc_table(sgt, i, GFP_KERNEL)) {
+ ret = -ENOMEM;
+ goto free_sgt;
+ }
+
+ /* Populate the SG table with the allocated memory pages */
+ sg = sgt->sgl;
+ for (k = 0; k < i; k++, sg = sg_next(sg)) {
+ /* Last entry requires special handling */
+ if (k < i - 1) {
+ sg_set_page(sg, pages[k], PAGE_SIZE << pages_order[k], 0);
+ } else {
+ sg_set_page(sg, pages[k], (PAGE_SIZE << pages_order[k]) - buf_extra, 0);
+ sg_mark_end(sg);
+ }
+
+ }
+
+ kvfree(pages);
+ *sgt_out = sgt;
+ return ret;
+
+free_sgt:
+ kfree(sgt);
+free_partial_alloc:
+ for (j = 0; j < i; j++)
+ __free_pages(pages[j], pages_order[j]);
+ kvfree(pages);
+out:
+ *sgt_out = NULL;
+ return ret;
+}
+
+static bool invalid_sem(struct qaic_sem *sem)
+{
+ if (sem->val & ~SEM_VAL_MASK || sem->index & ~SEM_INDEX_MASK ||
+ !(sem->presync == 0 || sem->presync == 1) || sem->pad ||
+ sem->flags & ~(QAIC_SEM_INSYNCFENCE | QAIC_SEM_OUTSYNCFENCE) ||
+ sem->cmd > QAIC_SEM_WAIT_GT_0)
+ return true;
+ return false;
+}
+
+static int qaic_validate_req(struct qaic_device *qdev, struct qaic_attach_slice_entry *slice_ent,
+ u32 count, u64 total_size)
+{
+ int i;
+
+ for (i = 0; i < count; i++) {
+ if (!(slice_ent[i].db_len == 32 || slice_ent[i].db_len == 16 ||
+ slice_ent[i].db_len == 8 || slice_ent[i].db_len == 0) ||
+ invalid_sem(&slice_ent[i].sem0) || invalid_sem(&slice_ent[i].sem1) ||
+ invalid_sem(&slice_ent[i].sem2) || invalid_sem(&slice_ent[i].sem3))
+ return -EINVAL;
+
+ if (slice_ent[i].offset + slice_ent[i].size > total_size)
+ return -EINVAL;
+ }
+
+ return 0;
+}
+
+static void qaic_free_sgt(struct sg_table *sgt)
+{
+ struct scatterlist *sg;
+
+ for (sg = sgt->sgl; sg; sg = sg_next(sg))
+ if (sg_page(sg))
+ __free_pages(sg_page(sg), get_order(sg->length));
+ sg_free_table(sgt);
+ kfree(sgt);
+}
+
+static void qaic_gem_print_info(struct drm_printer *p, unsigned int indent,
+ const struct drm_gem_object *obj)
+{
+ struct qaic_bo *bo = to_qaic_bo(obj);
+
+ drm_printf_indent(p, indent, "user requested size=%llu\n", bo->size);
+}
+
+static const struct vm_operations_struct drm_vm_ops = {
+ .open = drm_gem_vm_open,
+ .close = drm_gem_vm_close,
+};
+
+static int qaic_gem_object_mmap(struct drm_gem_object *obj, struct vm_area_struct *vma)
+{
+ struct qaic_bo *bo = to_qaic_bo(obj);
+ unsigned long offset = 0;
+ struct scatterlist *sg;
+ int ret;
+
+ if (obj->import_attach)
+ return -EINVAL;
+
+ for (sg = bo->sgt->sgl; sg; sg = sg_next(sg)) {
+ if (sg_page(sg)) {
+ ret = remap_pfn_range(vma, vma->vm_start + offset, page_to_pfn(sg_page(sg)),
+ sg->length, vma->vm_page_prot);
+ if (ret)
+ goto out;
+ offset += sg->length;
+ }
+ }
+
+out:
+ return ret;
+}
+
+static void qaic_free_object(struct drm_gem_object *obj)
+{
+ struct qaic_bo *bo = to_qaic_bo(obj);
+
+ if (obj->import_attach) {
+ /* DMABUF/PRIME Path */
+ dma_buf_detach(obj->import_attach->dmabuf, obj->import_attach);
+ dma_buf_put(obj->import_attach->dmabuf);
+ } else {
+ /* Private buffer allocation path */
+ qaic_free_sgt(bo->sgt);
+ }
+
+ drm_gem_object_release(obj);
+ kfree(bo);
+}
+
+static const struct drm_gem_object_funcs qaic_gem_funcs = {
+ .free = qaic_free_object,
+ .print_info = qaic_gem_print_info,
+ .mmap = qaic_gem_object_mmap,
+ .vm_ops = &drm_vm_ops,
+};
+
+static struct qaic_bo *qaic_alloc_init_bo(void)
+{
+ struct qaic_bo *bo;
+
+ bo = kzalloc(sizeof(*bo), GFP_KERNEL);
+ if (!bo)
+ return ERR_PTR(-ENOMEM);
+
+ INIT_LIST_HEAD(&bo->slices);
+ init_completion(&bo->xfer_done);
+ complete_all(&bo->xfer_done);
+
+ return bo;
+}
+
+int qaic_create_bo_ioctl(struct drm_device *dev, void *data, struct drm_file *file_priv)
+{
+ struct qaic_create_bo *args = data;
+ int usr_rcu_id, qdev_rcu_id;
+ struct drm_gem_object *obj;
+ struct qaic_device *qdev;
+ struct qaic_user *usr;
+ struct qaic_bo *bo;
+ size_t size;
+ int ret;
+
+ if (args->pad)
+ return -EINVAL;
+
+ usr = file_priv->driver_priv;
+ usr_rcu_id = srcu_read_lock(&usr->qddev_lock);
+ if (!usr->qddev) {
+ ret = -ENODEV;
+ goto unlock_usr_srcu;
+ }
+
+ qdev = usr->qddev->qdev;
+ qdev_rcu_id = srcu_read_lock(&qdev->dev_lock);
+ if (qdev->in_reset) {
+ ret = -ENODEV;
+ goto unlock_dev_srcu;
+ }
+
+ size = PAGE_ALIGN(args->size);
+ if (size == 0) {
+ ret = -EINVAL;
+ goto unlock_dev_srcu;
+ }
+
+ bo = qaic_alloc_init_bo();
+ if (IS_ERR(bo)) {
+ ret = PTR_ERR(bo);
+ goto unlock_dev_srcu;
+ }
+ obj = &bo->base;
+
+ drm_gem_private_object_init(dev, obj, size);
+
+ obj->funcs = &qaic_gem_funcs;
+ ret = create_sgt(qdev, &bo->sgt, size);
+ if (ret)
+ goto free_bo;
+
+ bo->size = args->size;
+
+ ret = drm_gem_handle_create(file_priv, obj, &args->handle);
+ if (ret)
+ goto free_sgt;
+
+ bo->handle = args->handle;
+ drm_gem_object_put(obj);
+ srcu_read_unlock(&qdev->dev_lock, qdev_rcu_id);
+ srcu_read_unlock(&usr->qddev_lock, usr_rcu_id);
+
+ return 0;
+
+free_sgt:
+ qaic_free_sgt(bo->sgt);
+free_bo:
+ kfree(bo);
+unlock_dev_srcu:
+ srcu_read_unlock(&qdev->dev_lock, qdev_rcu_id);
+unlock_usr_srcu:
+ srcu_read_unlock(&usr->qddev_lock, usr_rcu_id);
+ return ret;
+}
+
+int qaic_mmap_bo_ioctl(struct drm_device *dev, void *data, struct drm_file *file_priv)
+{
+ struct qaic_mmap_bo *args = data;
+ int usr_rcu_id, qdev_rcu_id;
+ struct drm_gem_object *obj;
+ struct qaic_device *qdev;
+ struct qaic_user *usr;
+ int ret;
+
+ usr = file_priv->driver_priv;
+ usr_rcu_id = srcu_read_lock(&usr->qddev_lock);
+ if (!usr->qddev) {
+ ret = -ENODEV;
+ goto unlock_usr_srcu;
+ }
+
+ qdev = usr->qddev->qdev;
+ qdev_rcu_id = srcu_read_lock(&qdev->dev_lock);
+ if (qdev->in_reset) {
+ ret = -ENODEV;
+ goto unlock_dev_srcu;
+ }
+
+ obj = drm_gem_object_lookup(file_priv, args->handle);
+ if (!obj) {
+ ret = -ENOENT;
+ goto unlock_dev_srcu;
+ }
+
+ ret = drm_gem_create_mmap_offset(obj);
+ if (ret == 0)
+ args->offset = drm_vma_node_offset_addr(&obj->vma_node);
+
+ drm_gem_object_put(obj);
+
+unlock_dev_srcu:
+ srcu_read_unlock(&qdev->dev_lock, qdev_rcu_id);
+unlock_usr_srcu:
+ srcu_read_unlock(&usr->qddev_lock, usr_rcu_id);
+ return ret;
+}
+
+struct drm_gem_object *qaic_gem_prime_import(struct drm_device *dev, struct dma_buf *dma_buf)
+{
+ struct dma_buf_attachment *attach;
+ struct drm_gem_object *obj;
+ struct qaic_bo *bo;
+ size_t size;
+ int ret;
+
+ bo = qaic_alloc_init_bo();
+ if (IS_ERR(bo)) {
+ ret = PTR_ERR(bo);
+ goto out;
+ }
+
+ obj = &bo->base;
+ get_dma_buf(dma_buf);
+
+ attach = dma_buf_attach(dma_buf, dev->dev);
+ if (IS_ERR(attach)) {
+ ret = PTR_ERR(attach);
+ goto attach_fail;
+ }
+
+ size = PAGE_ALIGN(attach->dmabuf->size);
+ if (size == 0) {
+ ret = -EINVAL;
+ goto size_align_fail;
+ }
+
+ drm_gem_private_object_init(dev, obj, size);
+ /*
+ * skipping dma_buf_map_attachment() as we do not know the direction
+ * just yet. Once the direction is known in the subsequent IOCTL to
+ * attach slicing, we can do it then.
+ */
+
+ obj->funcs = &qaic_gem_funcs;
+ obj->import_attach = attach;
+ obj->resv = dma_buf->resv;
+
+ return obj;
+
+size_align_fail:
+ dma_buf_detach(dma_buf, attach);
+attach_fail:
+ dma_buf_put(dma_buf);
+ kfree(bo);
+out:
+ return ERR_PTR(ret);
+}
+
+static int qaic_prepare_import_bo(struct qaic_bo *bo, struct qaic_attach_slice_hdr *hdr)
+{
+ struct drm_gem_object *obj = &bo->base;
+ struct sg_table *sgt;
+ int ret;
+
+ if (obj->import_attach->dmabuf->size < hdr->size)
+ return -EINVAL;
+
+ sgt = dma_buf_map_attachment(obj->import_attach, hdr->dir);
+ if (IS_ERR(sgt)) {
+ ret = PTR_ERR(sgt);
+ return ret;
+ }
+
+ bo->sgt = sgt;
+ bo->size = hdr->size;
+
+ return 0;
+}
+
+static int qaic_prepare_export_bo(struct qaic_device *qdev, struct qaic_bo *bo,
+ struct qaic_attach_slice_hdr *hdr)
+{
+ int ret;
+
+ if (bo->size != hdr->size)
+ return -EINVAL;
+
+ ret = dma_map_sgtable(&qdev->pdev->dev, bo->sgt, hdr->dir, 0);
+ if (ret)
+ return -EFAULT;
+
+ return 0;
+}
+
+static int qaic_prepare_bo(struct qaic_device *qdev, struct qaic_bo *bo,
+ struct qaic_attach_slice_hdr *hdr)
+{
+ int ret;
+
+ if (bo->base.import_attach)
+ ret = qaic_prepare_import_bo(bo, hdr);
+ else
+ ret = qaic_prepare_export_bo(qdev, bo, hdr);
+
+ if (ret == 0)
+ bo->dir = hdr->dir;
+
+ return ret;
+}
+
+static void qaic_unprepare_import_bo(struct qaic_bo *bo)
+{
+ dma_buf_unmap_attachment(bo->base.import_attach, bo->sgt, bo->dir);
+ bo->sgt = NULL;
+ bo->size = 0;
+}
+
+static void qaic_unprepare_export_bo(struct qaic_device *qdev, struct qaic_bo *bo)
+{
+ dma_unmap_sgtable(&qdev->pdev->dev, bo->sgt, bo->dir, 0);
+}
+
+static void qaic_unprepare_bo(struct qaic_device *qdev, struct qaic_bo *bo)
+{
+ if (bo->base.import_attach)
+ qaic_unprepare_import_bo(bo);
+ else
+ qaic_unprepare_export_bo(qdev, bo);
+
+ bo->dir = 0;
+}
+
+static void qaic_free_slices_bo(struct qaic_bo *bo)
+{
+ struct bo_slice *slice, *temp;
+
+ list_for_each_entry_safe(slice, temp, &bo->slices, slice)
+ kref_put(&slice->ref_count, free_slice);
+}
+
+static int qaic_attach_slicing_bo(struct qaic_device *qdev, struct qaic_bo *bo,
+ struct qaic_attach_slice_hdr *hdr,
+ struct qaic_attach_slice_entry *slice_ent)
+{
+ int ret, i;
+
+ for (i = 0; i < hdr->count; i++) {
+ ret = qaic_map_one_slice(qdev, bo, &slice_ent[i]);
+ if (ret) {
+ qaic_free_slices_bo(bo);
+ return ret;
+ }
+ }
+
+ if (bo->total_slice_nents > qdev->dbc[hdr->dbc_id].nelem) {
+ qaic_free_slices_bo(bo);
+ return -ENOSPC;
+ }
+
+ bo->sliced = true;
+ bo->nr_slice = hdr->count;
+ list_add_tail(&bo->bo_list, &qdev->dbc[hdr->dbc_id].bo_lists);
+
+ return 0;
+}
+
+int qaic_attach_slice_bo_ioctl(struct drm_device *dev, void *data, struct drm_file *file_priv)
+{
+ struct qaic_attach_slice_entry *slice_ent;
+ struct qaic_attach_slice *args = data;
+ struct dma_bridge_chan *dbc;
+ int usr_rcu_id, qdev_rcu_id;
+ struct drm_gem_object *obj;
+ struct qaic_device *qdev;
+ unsigned long arg_size;
+ struct qaic_user *usr;
+ u8 __user *user_data;
+ struct qaic_bo *bo;
+ int ret;
+
+ usr = file_priv->driver_priv;
+ usr_rcu_id = srcu_read_lock(&usr->qddev_lock);
+ if (!usr->qddev) {
+ ret = -ENODEV;
+ goto unlock_usr_srcu;
+ }
+
+ qdev = usr->qddev->qdev;
+ qdev_rcu_id = srcu_read_lock(&qdev->dev_lock);
+ if (qdev->in_reset) {
+ ret = -ENODEV;
+ goto unlock_dev_srcu;
+ }
+
+ if (args->hdr.count == 0) {
+ ret = -EINVAL;
+ goto unlock_dev_srcu;
+ }
+
+ arg_size = args->hdr.count * sizeof(*slice_ent);
+ if (arg_size / args->hdr.count != sizeof(*slice_ent)) {
+ ret = -EINVAL;
+ goto unlock_dev_srcu;
+ }
+
+ if (args->hdr.dbc_id >= qdev->num_dbc) {
+ ret = -EINVAL;
+ goto unlock_dev_srcu;
+ }
+
+ if (args->hdr.size == 0) {
+ ret = -EINVAL;
+ goto unlock_dev_srcu;
+ }
+
+ if (!(args->hdr.dir == DMA_TO_DEVICE || args->hdr.dir == DMA_FROM_DEVICE)) {
+ ret = -EINVAL;
+ goto unlock_dev_srcu;
+ }
+
+ dbc = &qdev->dbc[args->hdr.dbc_id];
+ if (dbc->usr != usr) {
+ ret = -EINVAL;
+ goto unlock_dev_srcu;
+ }
+
+ if (args->data == 0) {
+ ret = -EINVAL;
+ goto unlock_dev_srcu;
+ }
+
+ user_data = u64_to_user_ptr(args->data);
+
+ slice_ent = kzalloc(arg_size, GFP_KERNEL);
+ if (!slice_ent) {
+ ret = -EINVAL;
+ goto unlock_dev_srcu;
+ }
+
+ ret = copy_from_user(slice_ent, user_data, arg_size);
+ if (ret) {
+ ret = -EFAULT;
+ goto free_slice_ent;
+ }
+
+ ret = qaic_validate_req(qdev, slice_ent, args->hdr.count, args->hdr.size);
+ if (ret)
+ goto free_slice_ent;
+
+ obj = drm_gem_object_lookup(file_priv, args->hdr.handle);
+ if (!obj) {
+ ret = -ENOENT;
+ goto free_slice_ent;
+ }
+
+ bo = to_qaic_bo(obj);
+
+ ret = qaic_prepare_bo(qdev, bo, &args->hdr);
+ if (ret)
+ goto put_bo;
+
+ ret = qaic_attach_slicing_bo(qdev, bo, &args->hdr, slice_ent);
+ if (ret)
+ goto unprepare_bo;
+
+ if (args->hdr.dir == DMA_TO_DEVICE)
+ dma_sync_sgtable_for_cpu(&qdev->pdev->dev, bo->sgt, args->hdr.dir);
+
+ bo->dbc = dbc;
+ drm_gem_object_put(obj);
+ srcu_read_unlock(&qdev->dev_lock, qdev_rcu_id);
+ srcu_read_unlock(&usr->qddev_lock, usr_rcu_id);
+
+ return 0;
+
+unprepare_bo:
+ qaic_unprepare_bo(qdev, bo);
+put_bo:
+ drm_gem_object_put(obj);
+free_slice_ent:
+ kfree(slice_ent);
+unlock_dev_srcu:
+ srcu_read_unlock(&qdev->dev_lock, qdev_rcu_id);
+unlock_usr_srcu:
+ srcu_read_unlock(&usr->qddev_lock, usr_rcu_id);
+ return ret;
+}
+
+static inline int copy_exec_reqs(struct qaic_device *qdev, struct bo_slice *slice, u32 dbc_id,
+ u32 head, u32 *ptail)
+{
+ struct dma_bridge_chan *dbc = &qdev->dbc[dbc_id];
+ struct dbc_req *reqs = slice->reqs;
+ u32 tail = *ptail;
+ u32 avail;
+
+ avail = head - tail;
+ if (head <= tail)
+ avail += dbc->nelem;
+
+ --avail;
+
+ if (avail < slice->nents)
+ return -EAGAIN;
+
+ if (tail + slice->nents > dbc->nelem) {
+ avail = dbc->nelem - tail;
+ avail = min_t(u32, avail, slice->nents);
+ memcpy(dbc->req_q_base + tail * get_dbc_req_elem_size(), reqs,
+ sizeof(*reqs) * avail);
+ reqs += avail;
+ avail = slice->nents - avail;
+ if (avail)
+ memcpy(dbc->req_q_base, reqs, sizeof(*reqs) * avail);
+ } else {
+ memcpy(dbc->req_q_base + tail * get_dbc_req_elem_size(), reqs,
+ sizeof(*reqs) * slice->nents);
+ }
+
+ *ptail = (tail + slice->nents) % dbc->nelem;
+
+ return 0;
+}
+
+/*
+ * Based on the value of resize we may only need to transmit first_n
+ * entries and the last entry, with last_bytes to send from the last entry.
+ * Note that first_n could be 0.
+ */
+static inline int copy_partial_exec_reqs(struct qaic_device *qdev, struct bo_slice *slice,
+ u64 resize, u32 dbc_id, u32 head, u32 *ptail)
+{
+ struct dma_bridge_chan *dbc = &qdev->dbc[dbc_id];
+ struct dbc_req *reqs = slice->reqs;
+ struct dbc_req *last_req;
+ u32 tail = *ptail;
+ u64 total_bytes;
+ u64 last_bytes;
+ u32 first_n;
+ u32 avail;
+ int ret;
+ int i;
+
+ avail = head - tail;
+ if (head <= tail)
+ avail += dbc->nelem;
+
+ --avail;
+
+ total_bytes = 0;
+ for (i = 0; i < slice->nents; i++) {
+ total_bytes += le32_to_cpu(reqs[i].len);
+ if (total_bytes >= resize)
+ break;
+ }
+
+ if (total_bytes < resize) {
+ /* User space should have used the full buffer path. */
+ ret = -EINVAL;
+ return ret;
+ }
+
+ first_n = i;
+ last_bytes = i ? resize + le32_to_cpu(reqs[i].len) - total_bytes : resize;
+
+ if (avail < (first_n + 1))
+ return -EAGAIN;
+
+ if (first_n) {
+ if (tail + first_n > dbc->nelem) {
+ avail = dbc->nelem - tail;
+ avail = min_t(u32, avail, first_n);
+ memcpy(dbc->req_q_base + tail * get_dbc_req_elem_size(), reqs,
+ sizeof(*reqs) * avail);
+ last_req = reqs + avail;
+ avail = first_n - avail;
+ if (avail)
+ memcpy(dbc->req_q_base, last_req, sizeof(*reqs) * avail);
+ } else {
+ memcpy(dbc->req_q_base + tail * get_dbc_req_elem_size(), reqs,
+ sizeof(*reqs) * first_n);
+ }
+ }
+
+ /* Copy over the last entry. Here we need to adjust len to the left over
+ * size, and set src and dst to the entry it is copied to.
+ */
+ last_req = dbc->req_q_base + (tail + first_n) % dbc->nelem * get_dbc_req_elem_size();
+ memcpy(last_req, reqs + slice->nents - 1, sizeof(*reqs));
+
+ /*
+ * last_bytes holds size of a DMA segment, maximum DMA segment size is
+ * set to UINT_MAX by qaic and hence last_bytes can never exceed u32
+ * range. So, by down sizing we are not corrupting the value.
+ */
+ last_req->len = cpu_to_le32((u32)last_bytes);
+ last_req->src_addr = reqs[first_n].src_addr;
+ last_req->dest_addr = reqs[first_n].dest_addr;
+
+ *ptail = (tail + first_n + 1) % dbc->nelem;
+
+ return 0;
+}
+
+static int send_bo_list_to_device(struct qaic_device *qdev, struct drm_file *file_priv,
+ struct qaic_execute_entry *exec, unsigned int count,
+ bool is_partial, struct dma_bridge_chan *dbc, u32 head,
+ u32 *tail)
+{
+ struct qaic_partial_execute_entry *pexec = (struct qaic_partial_execute_entry *)exec;
+ struct drm_gem_object *obj;
+ struct bo_slice *slice;
+ unsigned long flags;
+ struct qaic_bo *bo;
+ bool queued;
+ int i, j;
+ int ret;
+
+ for (i = 0; i < count; i++) {
+ /*
+ * ref count will be decremented when the transfer of this
+ * buffer is complete. It is inside dbc_irq_threaded_fn().
+ */
+ obj = drm_gem_object_lookup(file_priv,
+ is_partial ? pexec[i].handle : exec[i].handle);
+ if (!obj) {
+ ret = -ENOENT;
+ goto failed_to_send_bo;
+ }
+
+ bo = to_qaic_bo(obj);
+
+ if (!bo->sliced) {
+ ret = -EINVAL;
+ goto failed_to_send_bo;
+ }
+
+ if (is_partial && pexec[i].resize > bo->size) {
+ ret = -EINVAL;
+ goto failed_to_send_bo;
+ }
+
+ spin_lock_irqsave(&dbc->xfer_lock, flags);
+ queued = bo->queued;
+ bo->queued = true;
+ if (queued) {
+ spin_unlock_irqrestore(&dbc->xfer_lock, flags);
+ ret = -EINVAL;
+ goto failed_to_send_bo;
+ }
+
+ bo->req_id = dbc->next_req_id++;
+
+ list_for_each_entry(slice, &bo->slices, slice) {
+ /*
+ * If this slice does not fall under the given
+ * resize then skip this slice and continue the loop
+ */
+ if (is_partial && pexec[i].resize && pexec[i].resize <= slice->offset)
+ continue;
+
+ for (j = 0; j < slice->nents; j++)
+ slice->reqs[j].req_id = cpu_to_le16(bo->req_id);
+
+ /*
+ * If it is a partial execute ioctl call then check if
+ * resize has cut this slice short then do a partial copy
+ * else do complete copy
+ */
+ if (is_partial && pexec[i].resize &&
+ pexec[i].resize < slice->offset + slice->size)
+ ret = copy_partial_exec_reqs(qdev, slice,
+ pexec[i].resize - slice->offset,
+ dbc->id, head, tail);
+ else
+ ret = copy_exec_reqs(qdev, slice, dbc->id, head, tail);
+ if (ret) {
+ bo->queued = false;
+ spin_unlock_irqrestore(&dbc->xfer_lock, flags);
+ goto failed_to_send_bo;
+ }
+ }
+ reinit_completion(&bo->xfer_done);
+ list_add_tail(&bo->xfer_list, &dbc->xfer_list);
+ spin_unlock_irqrestore(&dbc->xfer_lock, flags);
+ dma_sync_sgtable_for_device(&qdev->pdev->dev, bo->sgt, bo->dir);
+ }
+
+ return 0;
+
+failed_to_send_bo:
+ if (likely(obj))
+ drm_gem_object_put(obj);
+ for (j = 0; j < i; j++) {
+ spin_lock_irqsave(&dbc->xfer_lock, flags);
+ bo = list_last_entry(&dbc->xfer_list, struct qaic_bo, xfer_list);
+ obj = &bo->base;
+ bo->queued = false;
+ list_del(&bo->xfer_list);
+ spin_unlock_irqrestore(&dbc->xfer_lock, flags);
+ dma_sync_sgtable_for_cpu(&qdev->pdev->dev, bo->sgt, bo->dir);
+ drm_gem_object_put(obj);
+ }
+ return ret;
+}
+
+static void update_profiling_data(struct drm_file *file_priv,
+ struct qaic_execute_entry *exec, unsigned int count,
+ bool is_partial, u64 received_ts, u64 submit_ts, u32 queue_level)
+{
+ struct qaic_partial_execute_entry *pexec = (struct qaic_partial_execute_entry *)exec;
+ struct drm_gem_object *obj;
+ struct qaic_bo *bo;
+ int i;
+
+ for (i = 0; i < count; i++) {
+ /*
+ * Since we already committed the BO to hardware, the only way
+ * this should fail is a pending signal. We can't cancel the
+ * submit to hardware, so we have to just skip the profiling
+ * data. In case the signal is not fatal to the process, we
+ * return success so that the user doesn't try to resubmit.
+ */
+ obj = drm_gem_object_lookup(file_priv,
+ is_partial ? pexec[i].handle : exec[i].handle);
+ if (!obj)
+ break;
+ bo = to_qaic_bo(obj);
+ bo->perf_stats.req_received_ts = received_ts;
+ bo->perf_stats.req_submit_ts = submit_ts;
+ bo->perf_stats.queue_level_before = queue_level;
+ queue_level += bo->total_slice_nents;
+ drm_gem_object_put(obj);
+ }
+}
+
+static int __qaic_execute_bo_ioctl(struct drm_device *dev, void *data, struct drm_file *file_priv,
+ bool is_partial)
+{
+ struct qaic_partial_execute_entry *pexec;
+ struct qaic_execute *args = data;
+ struct qaic_execute_entry *exec;
+ struct dma_bridge_chan *dbc;
+ int usr_rcu_id, qdev_rcu_id;
+ struct qaic_device *qdev;
+ struct qaic_user *usr;
+ u8 __user *user_data;
+ unsigned long n;
+ u64 received_ts;
+ u32 queue_level;
+ u64 submit_ts;
+ int rcu_id;
+ u32 head;
+ u32 tail;
+ u64 size;
+ int ret;
+
+ received_ts = ktime_get_ns();
+
+ size = is_partial ? sizeof(*pexec) : sizeof(*exec);
+
+ n = (unsigned long)size * args->hdr.count;
+ if (args->hdr.count == 0 || n / args->hdr.count != size)
+ return -EINVAL;
+
+ user_data = u64_to_user_ptr(args->data);
+
+ exec = kcalloc(args->hdr.count, size, GFP_KERNEL);
+ pexec = (struct qaic_partial_execute_entry *)exec;
+ if (!exec)
+ return -ENOMEM;
+
+ if (copy_from_user(exec, user_data, n)) {
+ ret = -EFAULT;
+ goto free_exec;
+ }
+
+ usr = file_priv->driver_priv;
+ usr_rcu_id = srcu_read_lock(&usr->qddev_lock);
+ if (!usr->qddev) {
+ ret = -ENODEV;
+ goto unlock_usr_srcu;
+ }
+
+ qdev = usr->qddev->qdev;
+ qdev_rcu_id = srcu_read_lock(&qdev->dev_lock);
+ if (qdev->in_reset) {
+ ret = -ENODEV;
+ goto unlock_dev_srcu;
+ }
+
+ if (args->hdr.dbc_id >= qdev->num_dbc) {
+ ret = -EINVAL;
+ goto unlock_dev_srcu;
+ }
+
+ dbc = &qdev->dbc[args->hdr.dbc_id];
+
+ rcu_id = srcu_read_lock(&dbc->ch_lock);
+ if (!dbc->usr || dbc->usr->handle != usr->handle) {
+ ret = -EPERM;
+ goto release_ch_rcu;
+ }
+
+ head = readl(dbc->dbc_base + REQHP_OFF);
+ tail = readl(dbc->dbc_base + REQTP_OFF);
+
+ if (head == U32_MAX || tail == U32_MAX) {
+ /* PCI link error */
+ ret = -ENODEV;
+ goto release_ch_rcu;
+ }
+
+ queue_level = head <= tail ? tail - head : dbc->nelem - (head - tail);
+
+ ret = send_bo_list_to_device(qdev, file_priv, exec, args->hdr.count, is_partial, dbc,
+ head, &tail);
+ if (ret)
+ goto release_ch_rcu;
+
+ /* Finalize commit to hardware */
+ submit_ts = ktime_get_ns();
+ writel(tail, dbc->dbc_base + REQTP_OFF);
+
+ update_profiling_data(file_priv, exec, args->hdr.count, is_partial, received_ts,
+ submit_ts, queue_level);
+
+ if (datapath_polling)
+ schedule_work(&dbc->poll_work);
+
+release_ch_rcu:
+ srcu_read_unlock(&dbc->ch_lock, rcu_id);
+unlock_dev_srcu:
+ srcu_read_unlock(&qdev->dev_lock, qdev_rcu_id);
+unlock_usr_srcu:
+ srcu_read_unlock(&usr->qddev_lock, usr_rcu_id);
+free_exec:
+ kfree(exec);
+ return ret;
+}
+
+int qaic_execute_bo_ioctl(struct drm_device *dev, void *data, struct drm_file *file_priv)
+{
+ return __qaic_execute_bo_ioctl(dev, data, file_priv, false);
+}
+
+int qaic_partial_execute_bo_ioctl(struct drm_device *dev, void *data, struct drm_file *file_priv)
+{
+ return __qaic_execute_bo_ioctl(dev, data, file_priv, true);
+}
+
+/*
+ * Our interrupt handling is a bit more complicated than a simple ideal, but
+ * sadly necessary.
+ *
+ * Each dbc has a completion queue. Entries in the queue correspond to DMA
+ * requests which the device has processed. The hardware already has a built
+ * in irq mitigation. When the device puts an entry into the queue, it will
+ * only trigger an interrupt if the queue was empty. Therefore, when adding
+ * the Nth event to a non-empty queue, the hardware doesn't trigger an
+ * interrupt. This means the host doesn't get additional interrupts signaling
+ * the same thing - the queue has something to process.
+ * This behavior can be overridden in the DMA request.
+ * This means that when the host receives an interrupt, it is required to
+ * drain the queue.
+ *
+ * This behavior is what NAPI attempts to accomplish, although we can't use
+ * NAPI as we don't have a netdev. We use threaded irqs instead.
+ *
+ * However, there is a situation where the host drains the queue fast enough
+ * that every event causes an interrupt. Typically this is not a problem as
+ * the rate of events would be low. However, that is not the case with
+ * lprnet for example. On an Intel Xeon D-2191 where we run 8 instances of
+ * lprnet, the host receives roughly 80k interrupts per second from the device
+ * (per /proc/interrupts). While NAPI documentation indicates the host should
+ * just chug along, sadly that behavior causes instability in some hosts.
+ *
+ * Therefore, we implement an interrupt disable scheme similar to NAPI. The
+ * key difference is that we will delay after draining the queue for a small
+ * time to allow additional events to come in via polling. Using the above
+ * lprnet workload, this reduces the number of interrupts processed from
+ * ~80k/sec to about 64 in 5 minutes and appears to solve the system
+ * instability.
+ */
+irqreturn_t dbc_irq_handler(int irq, void *data)
+{
+ struct dma_bridge_chan *dbc = data;
+ int rcu_id;
+ u32 head;
+ u32 tail;
+
+ rcu_id = srcu_read_lock(&dbc->ch_lock);
+
+ if (!dbc->usr) {
+ srcu_read_unlock(&dbc->ch_lock, rcu_id);
+ return IRQ_HANDLED;
+ }
+
+ head = readl(dbc->dbc_base + RSPHP_OFF);
+ if (head == U32_MAX) { /* PCI link error */
+ srcu_read_unlock(&dbc->ch_lock, rcu_id);
+ return IRQ_NONE;
+ }
+
+ tail = readl(dbc->dbc_base + RSPTP_OFF);
+ if (tail == U32_MAX) { /* PCI link error */
+ srcu_read_unlock(&dbc->ch_lock, rcu_id);
+ return IRQ_NONE;
+ }
+
+ if (head == tail) { /* queue empty */
+ srcu_read_unlock(&dbc->ch_lock, rcu_id);
+ return IRQ_NONE;
+ }
+
+ disable_irq_nosync(irq);
+ srcu_read_unlock(&dbc->ch_lock, rcu_id);
+ return IRQ_WAKE_THREAD;
+}
+
+void irq_polling_work(struct work_struct *work)
+{
+ struct dma_bridge_chan *dbc = container_of(work, struct dma_bridge_chan, poll_work);
+ unsigned long flags;
+ int rcu_id;
+ u32 head;
+ u32 tail;
+
+ rcu_id = srcu_read_lock(&dbc->ch_lock);
+
+ while (1) {
+ if (dbc->qdev->in_reset) {
+ srcu_read_unlock(&dbc->ch_lock, rcu_id);
+ return;
+ }
+ if (!dbc->usr) {
+ srcu_read_unlock(&dbc->ch_lock, rcu_id);
+ return;
+ }
+ spin_lock_irqsave(&dbc->xfer_lock, flags);
+ if (list_empty(&dbc->xfer_list)) {
+ spin_unlock_irqrestore(&dbc->xfer_lock, flags);
+ srcu_read_unlock(&dbc->ch_lock, rcu_id);
+ return;
+ }
+ spin_unlock_irqrestore(&dbc->xfer_lock, flags);
+
+ head = readl(dbc->dbc_base + RSPHP_OFF);
+ if (head == U32_MAX) { /* PCI link error */
+ srcu_read_unlock(&dbc->ch_lock, rcu_id);
+ return;
+ }
+
+ tail = readl(dbc->dbc_base + RSPTP_OFF);
+ if (tail == U32_MAX) { /* PCI link error */
+ srcu_read_unlock(&dbc->ch_lock, rcu_id);
+ return;
+ }
+
+ if (head != tail) {
+ irq_wake_thread(dbc->irq, dbc);
+ srcu_read_unlock(&dbc->ch_lock, rcu_id);
+ return;
+ }
+
+ cond_resched();
+ usleep_range(datapath_poll_interval_us, 2 * datapath_poll_interval_us);
+ }
+}
+
+irqreturn_t dbc_irq_threaded_fn(int irq, void *data)
+{
+ struct dma_bridge_chan *dbc = data;
+ int event_count = NUM_EVENTS;
+ int delay_count = NUM_DELAYS;
+ struct qaic_device *qdev;
+ struct qaic_bo *bo, *i;
+ struct dbc_rsp *rsp;
+ unsigned long flags;
+ int rcu_id;
+ u16 status;
+ u16 req_id;
+ u32 head;
+ u32 tail;
+
+ rcu_id = srcu_read_lock(&dbc->ch_lock);
+
+ head = readl(dbc->dbc_base + RSPHP_OFF);
+ if (head == U32_MAX) /* PCI link error */
+ goto error_out;
+
+ qdev = dbc->qdev;
+read_fifo:
+
+ if (!event_count) {
+ event_count = NUM_EVENTS;
+ cond_resched();
+ }
+
+ /*
+ * if this channel isn't assigned or gets unassigned during processing
+ * we have nothing further to do
+ */
+ if (!dbc->usr)
+ goto error_out;
+
+ tail = readl(dbc->dbc_base + RSPTP_OFF);
+ if (tail == U32_MAX) /* PCI link error */
+ goto error_out;
+
+ if (head == tail) { /* queue empty */
+ if (delay_count) {
+ --delay_count;
+ usleep_range(100, 200);
+ goto read_fifo; /* check for a new event */
+ }
+ goto normal_out;
+ }
+
+ delay_count = NUM_DELAYS;
+ while (head != tail) {
+ if (!event_count)
+ break;
+ --event_count;
+ rsp = dbc->rsp_q_base + head * sizeof(*rsp);
+ req_id = le16_to_cpu(rsp->req_id);
+ status = le16_to_cpu(rsp->status);
+ if (status)
+ pci_dbg(qdev->pdev, "req_id %d failed with status %d\n", req_id, status);
+ spin_lock_irqsave(&dbc->xfer_lock, flags);
+ /*
+ * A BO can receive multiple interrupts, since a BO can be
+ * divided into multiple slices and a buffer receives as many
+ * interrupts as slices. So until it receives interrupts for
+ * all the slices we cannot mark that buffer complete.
+ */
+ list_for_each_entry_safe(bo, i, &dbc->xfer_list, xfer_list) {
+ if (bo->req_id == req_id)
+ bo->nr_slice_xfer_done++;
+ else
+ continue;
+
+ if (bo->nr_slice_xfer_done < bo->nr_slice)
+ break;
+
+ /*
+ * At this point we have received all the interrupts for
+ * BO, which means BO execution is complete.
+ */
+ dma_sync_sgtable_for_cpu(&qdev->pdev->dev, bo->sgt, bo->dir);
+ bo->nr_slice_xfer_done = 0;
+ bo->queued = false;
+ list_del(&bo->xfer_list);
+ bo->perf_stats.req_processed_ts = ktime_get_ns();
+ complete_all(&bo->xfer_done);
+ drm_gem_object_put(&bo->base);
+ break;
+ }
+ spin_unlock_irqrestore(&dbc->xfer_lock, flags);
+ head = (head + 1) % dbc->nelem;
+ }
+
+ /*
+ * Update the head pointer of response queue and let the device know
+ * that we have consumed elements from the queue.
+ */
+ writel(head, dbc->dbc_base + RSPHP_OFF);
+
+ /* elements might have been put in the queue while we were processing */
+ goto read_fifo;
+
+normal_out:
+ if (likely(!datapath_polling))
+ enable_irq(irq);
+ else
+ schedule_work(&dbc->poll_work);
+ /* checking the fifo and enabling irqs is a race, missed event check */
+ tail = readl(dbc->dbc_base + RSPTP_OFF);
+ if (tail != U32_MAX && head != tail) {
+ if (likely(!datapath_polling))
+ disable_irq_nosync(irq);
+ goto read_fifo;
+ }
+ srcu_read_unlock(&dbc->ch_lock, rcu_id);
+ return IRQ_HANDLED;
+
+error_out:
+ srcu_read_unlock(&dbc->ch_lock, rcu_id);
+ if (likely(!datapath_polling))
+ enable_irq(irq);
+ else
+ schedule_work(&dbc->poll_work);
+
+ return IRQ_HANDLED;
+}
+
+int qaic_wait_bo_ioctl(struct drm_device *dev, void *data, struct drm_file *file_priv)
+{
+ struct qaic_wait *args = data;
+ int usr_rcu_id, qdev_rcu_id;
+ struct dma_bridge_chan *dbc;
+ struct drm_gem_object *obj;
+ struct qaic_device *qdev;
+ unsigned long timeout;
+ struct qaic_user *usr;
+ struct qaic_bo *bo;
+ int rcu_id;
+ int ret;
+
+ usr = file_priv->driver_priv;
+ usr_rcu_id = srcu_read_lock(&usr->qddev_lock);
+ if (!usr->qddev) {
+ ret = -ENODEV;
+ goto unlock_usr_srcu;
+ }
+
+ qdev = usr->qddev->qdev;
+ qdev_rcu_id = srcu_read_lock(&qdev->dev_lock);
+ if (qdev->in_reset) {
+ ret = -ENODEV;
+ goto unlock_dev_srcu;
+ }
+
+ if (args->pad != 0) {
+ ret = -EINVAL;
+ goto unlock_dev_srcu;
+ }
+
+ if (args->dbc_id >= qdev->num_dbc) {
+ ret = -EINVAL;
+ goto unlock_dev_srcu;
+ }
+
+ dbc = &qdev->dbc[args->dbc_id];
+
+ rcu_id = srcu_read_lock(&dbc->ch_lock);
+ if (dbc->usr != usr) {
+ ret = -EPERM;
+ goto unlock_ch_srcu;
+ }
+
+ obj = drm_gem_object_lookup(file_priv, args->handle);
+ if (!obj) {
+ ret = -ENOENT;
+ goto unlock_ch_srcu;
+ }
+
+ bo = to_qaic_bo(obj);
+ timeout = args->timeout ? args->timeout : wait_exec_default_timeout_ms;
+ timeout = msecs_to_jiffies(timeout);
+ ret = wait_for_completion_interruptible_timeout(&bo->xfer_done, timeout);
+ if (!ret) {
+ ret = -ETIMEDOUT;
+ goto put_obj;
+ }
+ if (ret > 0)
+ ret = 0;
+
+ if (!dbc->usr)
+ ret = -EPERM;
+
+put_obj:
+ drm_gem_object_put(obj);
+unlock_ch_srcu:
+ srcu_read_unlock(&dbc->ch_lock, rcu_id);
+unlock_dev_srcu:
+ srcu_read_unlock(&qdev->dev_lock, qdev_rcu_id);
+unlock_usr_srcu:
+ srcu_read_unlock(&usr->qddev_lock, usr_rcu_id);
+ return ret;
+}
+
+int qaic_perf_stats_bo_ioctl(struct drm_device *dev, void *data, struct drm_file *file_priv)
+{
+ struct qaic_perf_stats_entry *ent = NULL;
+ struct qaic_perf_stats *args = data;
+ int usr_rcu_id, qdev_rcu_id;
+ struct drm_gem_object *obj;
+ struct qaic_device *qdev;
+ struct qaic_user *usr;
+ struct qaic_bo *bo;
+ int ret, i;
+
+ usr = file_priv->driver_priv;
+ usr_rcu_id = srcu_read_lock(&usr->qddev_lock);
+ if (!usr->qddev) {
+ ret = -ENODEV;
+ goto unlock_usr_srcu;
+ }
+
+ qdev = usr->qddev->qdev;
+ qdev_rcu_id = srcu_read_lock(&qdev->dev_lock);
+ if (qdev->in_reset) {
+ ret = -ENODEV;
+ goto unlock_dev_srcu;
+ }
+
+ if (args->hdr.dbc_id >= qdev->num_dbc) {
+ ret = -EINVAL;
+ goto unlock_dev_srcu;
+ }
+
+ ent = kcalloc(args->hdr.count, sizeof(*ent), GFP_KERNEL);
+ if (!ent) {
+ ret = -EINVAL;
+ goto unlock_dev_srcu;
+ }
+
+ ret = copy_from_user(ent, u64_to_user_ptr(args->data), args->hdr.count * sizeof(*ent));
+ if (ret) {
+ ret = -EFAULT;
+ goto free_ent;
+ }
+
+ for (i = 0; i < args->hdr.count; i++) {
+ obj = drm_gem_object_lookup(file_priv, ent[i].handle);
+ if (!obj) {
+ ret = -ENOENT;
+ goto free_ent;
+ }
+ bo = to_qaic_bo(obj);
+ /*
+ * perf stats ioctl is called before wait ioctl is complete then
+ * the latency information is invalid.
+ */
+ if (bo->perf_stats.req_processed_ts < bo->perf_stats.req_submit_ts) {
+ ent[i].device_latency_us = 0;
+ } else {
+ ent[i].device_latency_us = div_u64((bo->perf_stats.req_processed_ts -
+ bo->perf_stats.req_submit_ts), 1000);
+ }
+ ent[i].submit_latency_us = div_u64((bo->perf_stats.req_submit_ts -
+ bo->perf_stats.req_received_ts), 1000);
+ ent[i].queue_level_before = bo->perf_stats.queue_level_before;
+ ent[i].num_queue_element = bo->total_slice_nents;
+ drm_gem_object_put(obj);
+ }
+
+ if (copy_to_user(u64_to_user_ptr(args->data), ent, args->hdr.count * sizeof(*ent)))
+ ret = -EFAULT;
+
+free_ent:
+ kfree(ent);
+unlock_dev_srcu:
+ srcu_read_unlock(&qdev->dev_lock, qdev_rcu_id);
+unlock_usr_srcu:
+ srcu_read_unlock(&usr->qddev_lock, usr_rcu_id);
+ return ret;
+}
+
+static void empty_xfer_list(struct qaic_device *qdev, struct dma_bridge_chan *dbc)
+{
+ unsigned long flags;
+ struct qaic_bo *bo;
+
+ spin_lock_irqsave(&dbc->xfer_lock, flags);
+ while (!list_empty(&dbc->xfer_list)) {
+ bo = list_first_entry(&dbc->xfer_list, typeof(*bo), xfer_list);
+ bo->queued = false;
+ list_del(&bo->xfer_list);
+ spin_unlock_irqrestore(&dbc->xfer_lock, flags);
+ dma_sync_sgtable_for_cpu(&qdev->pdev->dev, bo->sgt, bo->dir);
+ complete_all(&bo->xfer_done);
+ drm_gem_object_put(&bo->base);
+ spin_lock_irqsave(&dbc->xfer_lock, flags);
+ }
+ spin_unlock_irqrestore(&dbc->xfer_lock, flags);
+}
+
+int disable_dbc(struct qaic_device *qdev, u32 dbc_id, struct qaic_user *usr)
+{
+ if (!qdev->dbc[dbc_id].usr || qdev->dbc[dbc_id].usr->handle != usr->handle)
+ return -EPERM;
+
+ qdev->dbc[dbc_id].usr = NULL;
+ synchronize_srcu(&qdev->dbc[dbc_id].ch_lock);
+ return 0;
+}
+
+/**
+ * enable_dbc - Enable the DBC. DBCs are disabled by removing the context of
+ * user. Add user context back to DBC to enable it. This function trusts the
+ * DBC ID passed and expects the DBC to be disabled.
+ * @qdev: Qranium device handle
+ * @dbc_id: ID of the DBC
+ * @usr: User context
+ */
+void enable_dbc(struct qaic_device *qdev, u32 dbc_id, struct qaic_user *usr)
+{
+ qdev->dbc[dbc_id].usr = usr;
+}
+
+void wakeup_dbc(struct qaic_device *qdev, u32 dbc_id)
+{
+ struct dma_bridge_chan *dbc = &qdev->dbc[dbc_id];
+
+ dbc->usr = NULL;
+ empty_xfer_list(qdev, dbc);
+ synchronize_srcu(&dbc->ch_lock);
+}
+
+void release_dbc(struct qaic_device *qdev, u32 dbc_id)
+{
+ struct bo_slice *slice, *slice_temp;
+ struct qaic_bo *bo, *bo_temp;
+ struct dma_bridge_chan *dbc;
+
+ dbc = &qdev->dbc[dbc_id];
+ if (!dbc->in_use)
+ return;
+
+ wakeup_dbc(qdev, dbc_id);
+
+ dma_free_coherent(&qdev->pdev->dev, dbc->total_size, dbc->req_q_base, dbc->dma_addr);
+ dbc->total_size = 0;
+ dbc->req_q_base = NULL;
+ dbc->dma_addr = 0;
+ dbc->nelem = 0;
+ dbc->usr = NULL;
+
+ list_for_each_entry_safe(bo, bo_temp, &dbc->bo_lists, bo_list) {
+ list_for_each_entry_safe(slice, slice_temp, &bo->slices, slice)
+ kref_put(&slice->ref_count, free_slice);
+ bo->sliced = false;
+ INIT_LIST_HEAD(&bo->slices);
+ bo->total_slice_nents = 0;
+ bo->dir = 0;
+ bo->dbc = NULL;
+ bo->nr_slice = 0;
+ bo->nr_slice_xfer_done = 0;
+ bo->queued = false;
+ bo->req_id = 0;
+ init_completion(&bo->xfer_done);
+ complete_all(&bo->xfer_done);
+ list_del(&bo->bo_list);
+ bo->perf_stats.req_received_ts = 0;
+ bo->perf_stats.req_submit_ts = 0;
+ bo->perf_stats.req_processed_ts = 0;
+ bo->perf_stats.queue_level_before = 0;
+ }
+
+ dbc->in_use = false;
+ wake_up(&dbc->dbc_release);
+}
--
2.7.4
^ permalink raw reply related [flat|nested] 16+ messages in thread* [PATCH v5 6/8] accel/qaic: Add mhi_qaic_cntl
2023-03-27 15:54 [PATCH v5 0/8] QAIC accel driver Jeffrey Hugo
` (4 preceding siblings ...)
2023-03-27 15:54 ` [PATCH v5 5/8] accel/qaic: Add datapath Jeffrey Hugo
@ 2023-03-27 15:54 ` Jeffrey Hugo
2023-03-27 15:54 ` [PATCH v5 7/8] accel/qaic: Add qaic driver to the build system Jeffrey Hugo
` (2 subsequent siblings)
8 siblings, 0 replies; 16+ messages in thread
From: Jeffrey Hugo @ 2023-03-27 15:54 UTC (permalink / raw)
To: ogabbay, airlied, daniel
Cc: jacek.lawrynowicz, stanislaw.gruszka, dri-devel, quic_pkanojiy,
quic_carlv, quic_ajitpals, linux-doc, linux-arm-msm, bagasdotme,
mani, Jeffrey Hugo
From: Pranjal Ramajor Asha Kanojiya <quic_pkanojiy@quicinc.com>
Some of the MHI channels for an AIC100 device need to be routed to
userspace so that userspace can communicate directly with QSM. The MHI
bus does not support this, and while the WWAN subsystem does (for the same
reasons), AIC100 is not a WWAN device. Also, MHI is not something that
other accelerators are expected to share, thus an accel subsystem function
that meets this usecase is unlikely.
Create a QAIC specific MHI userspace shim that exposes these channels.
Start with QAIC_SAHARA which is required to boot AIC100 and is consumed by
the kickstart application as documented in aic100.rst
Each AIC100 instance (currently, up to 16) in a system will create a
chardev for QAIC_SAHARA. This chardev will be found as
/dev/<mhi instance>_QAIC_SAHARA
For example - /dev/mhi0_QAIC_SAHARA
Signed-off-by: Pranjal Ramajor Asha Kanojiya <quic_pkanojiy@quicinc.com>
Signed-off-by: Jeffrey Hugo <quic_jhugo@quicinc.com>
Reviewed-by: Carl Vanderlip <quic_carlv@quicinc.com>
Reviewed-by: Stanislaw Gruszka <stanislaw.gruszka@linux.intel.com>
Reviewed-by: Jacek Lawrynowicz <jacek.lawrynowicz@linux.intel.com>
---
drivers/accel/qaic/mhi_qaic_ctrl.c | 570 +++++++++++++++++++++++++++++++++++++
drivers/accel/qaic/mhi_qaic_ctrl.h | 12 +
2 files changed, 582 insertions(+)
create mode 100644 drivers/accel/qaic/mhi_qaic_ctrl.c
create mode 100644 drivers/accel/qaic/mhi_qaic_ctrl.h
diff --git a/drivers/accel/qaic/mhi_qaic_ctrl.c b/drivers/accel/qaic/mhi_qaic_ctrl.c
new file mode 100644
index 0000000..21bbf3d
--- /dev/null
+++ b/drivers/accel/qaic/mhi_qaic_ctrl.c
@@ -0,0 +1,570 @@
+// SPDX-License-Identifier: GPL-2.0-only
+/* Copyright (c) 2022-2023 Qualcomm Innovation Center, Inc. All rights reserved. */
+
+#include <linux/kernel.h>
+#include <linux/mhi.h>
+#include <linux/mod_devicetable.h>
+#include <linux/module.h>
+#include <linux/poll.h>
+#include <linux/xarray.h>
+#include <uapi/linux/eventpoll.h>
+
+#include "mhi_qaic_ctrl.h"
+#include "qaic.h"
+
+#define MHI_QAIC_CTRL_DRIVER_NAME "mhi_qaic_ctrl"
+#define MHI_QAIC_CTRL_MAX_MINORS 128
+#define MHI_MAX_MTU 0xffff
+static DEFINE_XARRAY_ALLOC(mqc_xa);
+static struct class *mqc_dev_class;
+static int mqc_dev_major;
+
+/**
+ * struct mqc_buf - Buffer structure used to receive data from device
+ * @data: Address of data to read from
+ * @odata: Original address returned from *alloc() API. Used to free this buf.
+ * @len: Length of data in byte
+ * @node: This buffer will be part of list managed in struct mqc_dev
+ */
+struct mqc_buf {
+ void *data;
+ void *odata;
+ size_t len;
+ struct list_head node;
+};
+
+/**
+ * struct mqc_dev - MHI QAIC Control Device
+ * @minor: MQC device node minor number
+ * @mhi_dev: Associated mhi device object
+ * @mtu: Max TRE buffer length
+ * @enabled: Flag to track the state of the MQC device
+ * @lock: Mutex lock to serialize access to open_count
+ * @read_lock: Mutex lock to serialize readers
+ * @write_lock: Mutex lock to serialize writers
+ * @ul_wq: Wait queue for writers
+ * @dl_wq: Wait queue for readers
+ * @dl_queue_lock: Spin lock to serialize access to download queue
+ * @dl_queue: Queue of downloaded buffers
+ * @open_count: Track open counts
+ * @ref_count: Reference count for this structure
+ */
+struct mqc_dev {
+ uint32_t minor;
+ struct mhi_device *mhi_dev;
+ size_t mtu;
+ bool enabled;
+ struct mutex lock;
+ struct mutex read_lock;
+ struct mutex write_lock;
+ wait_queue_head_t ul_wq;
+ wait_queue_head_t dl_wq;
+ spinlock_t dl_queue_lock;
+ struct list_head dl_queue;
+ unsigned int open_count;
+ struct kref ref_count;
+};
+
+static void mqc_dev_release(struct kref *ref)
+{
+ struct mqc_dev *mqcdev = container_of(ref, struct mqc_dev, ref_count);
+
+ mutex_destroy(&mqcdev->read_lock);
+ mutex_destroy(&mqcdev->write_lock);
+ mutex_destroy(&mqcdev->lock);
+ kfree(mqcdev);
+}
+
+static int mhi_qaic_ctrl_fill_dl_queue(struct mqc_dev *mqcdev)
+{
+ struct mhi_device *mhi_dev = mqcdev->mhi_dev;
+ struct mqc_buf *ctrlbuf;
+ int rx_budget;
+ int ret = 0;
+ void *data;
+
+ rx_budget = mhi_get_free_desc_count(mhi_dev, DMA_FROM_DEVICE);
+ if (rx_budget < 0)
+ return -EIO;
+
+ while (rx_budget--) {
+ data = kzalloc(mqcdev->mtu + sizeof(*ctrlbuf), GFP_KERNEL);
+ if (!data)
+ return -ENOMEM;
+
+ ctrlbuf = data + mqcdev->mtu;
+ ctrlbuf->odata = data;
+
+ ret = mhi_queue_buf(mhi_dev, DMA_FROM_DEVICE, data, mqcdev->mtu, MHI_EOT);
+ if (ret) {
+ kfree(data);
+ dev_err(&mhi_dev->dev, "Failed to queue buffer\n");
+ return ret;
+ }
+ }
+
+ return ret;
+}
+
+static int mhi_qaic_ctrl_dev_start_chan(struct mqc_dev *mqcdev)
+{
+ struct device *dev = &mqcdev->mhi_dev->dev;
+ int ret = 0;
+
+ ret = mutex_lock_interruptible(&mqcdev->lock);
+ if (ret)
+ return ret;
+ if (!mqcdev->enabled) {
+ ret = -ENODEV;
+ goto release_dev_lock;
+ }
+ if (!mqcdev->open_count) {
+ ret = mhi_prepare_for_transfer(mqcdev->mhi_dev);
+ if (ret) {
+ dev_err(dev, "Error starting transfer channels\n");
+ goto release_dev_lock;
+ }
+
+ ret = mhi_qaic_ctrl_fill_dl_queue(mqcdev);
+ if (ret) {
+ dev_err(dev, "Error filling download queue.\n");
+ goto mhi_unprepare;
+ }
+ }
+ mqcdev->open_count++;
+ mutex_unlock(&mqcdev->lock);
+
+ return 0;
+
+mhi_unprepare:
+ mhi_unprepare_from_transfer(mqcdev->mhi_dev);
+release_dev_lock:
+ mutex_unlock(&mqcdev->lock);
+ return ret;
+}
+
+static struct mqc_dev *mqc_dev_get_by_minor(unsigned int minor)
+{
+ struct mqc_dev *mqcdev;
+
+ xa_lock(&mqc_xa);
+ mqcdev = xa_load(&mqc_xa, minor);
+ if (mqcdev)
+ kref_get(&mqcdev->ref_count);
+ xa_unlock(&mqc_xa);
+
+ return mqcdev;
+}
+
+static int mhi_qaic_ctrl_open(struct inode *inode, struct file *filp)
+{
+ struct mqc_dev *mqcdev;
+ int ret;
+
+ mqcdev = mqc_dev_get_by_minor(iminor(inode));
+ if (!mqcdev) {
+ pr_debug("mqc: minor %d not found\n", iminor(inode));
+ return -EINVAL;
+ }
+
+ ret = mhi_qaic_ctrl_dev_start_chan(mqcdev);
+ if (ret) {
+ kref_put(&mqcdev->ref_count, mqc_dev_release);
+ return ret;
+ }
+
+ filp->private_data = mqcdev;
+
+ return 0;
+}
+
+static void mhi_qaic_ctrl_buf_free(struct mqc_buf *ctrlbuf)
+{
+ list_del(&ctrlbuf->node);
+ kfree(ctrlbuf->odata);
+}
+
+static void __mhi_qaic_ctrl_release(struct mqc_dev *mqcdev)
+{
+ struct mqc_buf *ctrlbuf, *tmp;
+
+ mhi_unprepare_from_transfer(mqcdev->mhi_dev);
+ wake_up_interruptible(&mqcdev->ul_wq);
+ wake_up_interruptible(&mqcdev->dl_wq);
+ /*
+ * Free the dl_queue. As we have already unprepared mhi transfers, we
+ * do not expect any callback functions that update dl_queue hence no need
+ * to grab dl_queue lock.
+ */
+ mutex_lock(&mqcdev->read_lock);
+ list_for_each_entry_safe(ctrlbuf, tmp, &mqcdev->dl_queue, node)
+ mhi_qaic_ctrl_buf_free(ctrlbuf);
+ mutex_unlock(&mqcdev->read_lock);
+}
+
+static int mhi_qaic_ctrl_release(struct inode *inode, struct file *file)
+{
+ struct mqc_dev *mqcdev = file->private_data;
+
+ mutex_lock(&mqcdev->lock);
+ mqcdev->open_count--;
+ if (!mqcdev->open_count && mqcdev->enabled)
+ __mhi_qaic_ctrl_release(mqcdev);
+ mutex_unlock(&mqcdev->lock);
+
+ kref_put(&mqcdev->ref_count, mqc_dev_release);
+
+ return 0;
+}
+
+static __poll_t mhi_qaic_ctrl_poll(struct file *file, poll_table *wait)
+{
+ struct mqc_dev *mqcdev = file->private_data;
+ struct mhi_device *mhi_dev;
+ __poll_t mask = 0;
+
+ mhi_dev = mqcdev->mhi_dev;
+
+ poll_wait(file, &mqcdev->ul_wq, wait);
+ poll_wait(file, &mqcdev->dl_wq, wait);
+
+ mutex_lock(&mqcdev->lock);
+ if (!mqcdev->enabled) {
+ mutex_unlock(&mqcdev->lock);
+ return EPOLLERR;
+ }
+
+ spin_lock_bh(&mqcdev->dl_queue_lock);
+ if (!list_empty(&mqcdev->dl_queue))
+ mask |= EPOLLIN | EPOLLRDNORM;
+ spin_unlock_bh(&mqcdev->dl_queue_lock);
+
+ if (mutex_lock_interruptible(&mqcdev->write_lock)) {
+ mutex_unlock(&mqcdev->lock);
+ return EPOLLERR;
+ }
+ if (mhi_get_free_desc_count(mhi_dev, DMA_TO_DEVICE) > 0)
+ mask |= EPOLLOUT | EPOLLWRNORM;
+ mutex_unlock(&mqcdev->write_lock);
+ mutex_unlock(&mqcdev->lock);
+
+ dev_dbg(&mhi_dev->dev, "Client attempted to poll, returning mask 0x%x\n", mask);
+
+ return mask;
+}
+
+static int mhi_qaic_ctrl_tx(struct mqc_dev *mqcdev)
+{
+ int ret;
+
+ ret = wait_event_interruptible(mqcdev->ul_wq, (!mqcdev->enabled ||
+ mhi_get_free_desc_count(mqcdev->mhi_dev,
+ DMA_TO_DEVICE) > 0));
+
+ if (!mqcdev->enabled)
+ return -ENODEV;
+
+ return ret;
+}
+
+static ssize_t mhi_qaic_ctrl_write(struct file *file, const char __user *buf, size_t count,
+ loff_t *offp)
+{
+ struct mqc_dev *mqcdev = file->private_data;
+ struct mhi_device *mhi_dev;
+ size_t bytes_xfered = 0;
+ struct device *dev;
+ int ret, nr_desc;
+
+ mhi_dev = mqcdev->mhi_dev;
+ dev = &mhi_dev->dev;
+
+ if (!mhi_dev->ul_chan)
+ return -EOPNOTSUPP;
+
+ if (!buf || !count)
+ return -EINVAL;
+
+ dev_dbg(dev, "Request to transfer %zu bytes\n", count);
+
+ ret = mhi_qaic_ctrl_tx(mqcdev);
+ if (ret)
+ return ret;
+
+ if (mutex_lock_interruptible(&mqcdev->write_lock))
+ return -EINTR;
+
+ nr_desc = mhi_get_free_desc_count(mhi_dev, DMA_TO_DEVICE);
+ if (nr_desc * mqcdev->mtu < count) {
+ ret = -EMSGSIZE;
+ dev_dbg(dev, "Buffer too big to transfer\n");
+ goto unlock_mutex;
+ }
+
+ while (count != bytes_xfered) {
+ enum mhi_flags flags;
+ size_t to_copy;
+ void *kbuf;
+
+ to_copy = min_t(size_t, count - bytes_xfered, mqcdev->mtu);
+ kbuf = kmalloc(to_copy, GFP_KERNEL);
+ if (!kbuf) {
+ ret = -ENOMEM;
+ goto unlock_mutex;
+ }
+
+ ret = copy_from_user(kbuf, buf + bytes_xfered, to_copy);
+ if (ret) {
+ kfree(kbuf);
+ ret = -EFAULT;
+ goto unlock_mutex;
+ }
+
+ if (bytes_xfered + to_copy == count)
+ flags = MHI_EOT;
+ else
+ flags = MHI_CHAIN;
+
+ ret = mhi_queue_buf(mhi_dev, DMA_TO_DEVICE, kbuf, to_copy, flags);
+ if (ret) {
+ kfree(kbuf);
+ dev_err(dev, "Failed to queue buf of size %zu\n", to_copy);
+ goto unlock_mutex;
+ }
+
+ bytes_xfered += to_copy;
+ }
+
+ mutex_unlock(&mqcdev->write_lock);
+ dev_dbg(dev, "bytes xferred: %zu\n", bytes_xfered);
+
+ return bytes_xfered;
+
+unlock_mutex:
+ mutex_unlock(&mqcdev->write_lock);
+ return ret;
+}
+
+static int mhi_qaic_ctrl_rx(struct mqc_dev *mqcdev)
+{
+ int ret;
+
+ ret = wait_event_interruptible(mqcdev->dl_wq, (!mqcdev->enabled ||
+ !list_empty(&mqcdev->dl_queue)));
+
+ if (!mqcdev->enabled)
+ return -ENODEV;
+
+ return ret;
+}
+
+static ssize_t mhi_qaic_ctrl_read(struct file *file, char __user *buf, size_t count, loff_t *ppos)
+{
+ struct mqc_dev *mqcdev = file->private_data;
+ struct mqc_buf *ctrlbuf;
+ size_t to_copy;
+ int ret;
+
+ if (!mqcdev->mhi_dev->dl_chan)
+ return -EOPNOTSUPP;
+
+ ret = mhi_qaic_ctrl_rx(mqcdev);
+ if (ret)
+ return ret;
+
+ if (mutex_lock_interruptible(&mqcdev->read_lock))
+ return -EINTR;
+
+ ctrlbuf = list_first_entry_or_null(&mqcdev->dl_queue, struct mqc_buf, node);
+ if (!ctrlbuf) {
+ mutex_unlock(&mqcdev->read_lock);
+ ret = -ENODEV;
+ goto error_out;
+ }
+
+ to_copy = min_t(size_t, count, ctrlbuf->len);
+ if (copy_to_user(buf, ctrlbuf->data, to_copy)) {
+ mutex_unlock(&mqcdev->read_lock);
+ dev_dbg(&mqcdev->mhi_dev->dev, "Failed to copy data to user buffer\n");
+ ret = -EFAULT;
+ goto error_out;
+ }
+
+ ctrlbuf->len -= to_copy;
+ ctrlbuf->data += to_copy;
+
+ if (!ctrlbuf->len) {
+ spin_lock_bh(&mqcdev->dl_queue_lock);
+ mhi_qaic_ctrl_buf_free(ctrlbuf);
+ spin_unlock_bh(&mqcdev->dl_queue_lock);
+ mhi_qaic_ctrl_fill_dl_queue(mqcdev);
+ dev_dbg(&mqcdev->mhi_dev->dev, "Read buf freed\n");
+ }
+
+ mutex_unlock(&mqcdev->read_lock);
+ return to_copy;
+
+error_out:
+ mutex_unlock(&mqcdev->read_lock);
+ return ret;
+}
+
+static const struct file_operations mhidev_fops = {
+ .owner = THIS_MODULE,
+ .open = mhi_qaic_ctrl_open,
+ .release = mhi_qaic_ctrl_release,
+ .read = mhi_qaic_ctrl_read,
+ .write = mhi_qaic_ctrl_write,
+ .poll = mhi_qaic_ctrl_poll,
+};
+
+static void mhi_qaic_ctrl_ul_xfer_cb(struct mhi_device *mhi_dev, struct mhi_result *mhi_result)
+{
+ struct mqc_dev *mqcdev = dev_get_drvdata(&mhi_dev->dev);
+
+ dev_dbg(&mhi_dev->dev, "%s: status: %d xfer_len: %zu\n", __func__,
+ mhi_result->transaction_status, mhi_result->bytes_xferd);
+
+ kfree(mhi_result->buf_addr);
+
+ if (!mhi_result->transaction_status)
+ wake_up_interruptible(&mqcdev->ul_wq);
+}
+
+static void mhi_qaic_ctrl_dl_xfer_cb(struct mhi_device *mhi_dev, struct mhi_result *mhi_result)
+{
+ struct mqc_dev *mqcdev = dev_get_drvdata(&mhi_dev->dev);
+ struct mqc_buf *ctrlbuf;
+
+ dev_dbg(&mhi_dev->dev, "%s: status: %d receive_len: %zu\n", __func__,
+ mhi_result->transaction_status, mhi_result->bytes_xferd);
+
+ if (mhi_result->transaction_status &&
+ mhi_result->transaction_status != -EOVERFLOW) {
+ kfree(mhi_result->buf_addr);
+ return;
+ }
+
+ ctrlbuf = mhi_result->buf_addr + mqcdev->mtu;
+ ctrlbuf->data = mhi_result->buf_addr;
+ ctrlbuf->len = mhi_result->bytes_xferd;
+ spin_lock_bh(&mqcdev->dl_queue_lock);
+ list_add_tail(&ctrlbuf->node, &mqcdev->dl_queue);
+ spin_unlock_bh(&mqcdev->dl_queue_lock);
+
+ wake_up_interruptible(&mqcdev->dl_wq);
+}
+
+static int mhi_qaic_ctrl_probe(struct mhi_device *mhi_dev, const struct mhi_device_id *id)
+{
+ struct mqc_dev *mqcdev;
+ struct device *dev;
+ int ret;
+
+ mqcdev = kzalloc(sizeof(*mqcdev), GFP_KERNEL);
+ if (!mqcdev)
+ return -ENOMEM;
+
+ kref_init(&mqcdev->ref_count);
+ mutex_init(&mqcdev->lock);
+ mqcdev->mhi_dev = mhi_dev;
+
+ ret = xa_alloc(&mqc_xa, &mqcdev->minor, mqcdev, XA_LIMIT(0, MHI_QAIC_CTRL_MAX_MINORS),
+ GFP_KERNEL);
+ if (ret) {
+ kfree(mqcdev);
+ return ret;
+ }
+
+ init_waitqueue_head(&mqcdev->ul_wq);
+ init_waitqueue_head(&mqcdev->dl_wq);
+ mutex_init(&mqcdev->read_lock);
+ mutex_init(&mqcdev->write_lock);
+ spin_lock_init(&mqcdev->dl_queue_lock);
+ INIT_LIST_HEAD(&mqcdev->dl_queue);
+ mqcdev->mtu = min_t(size_t, id->driver_data, MHI_MAX_MTU);
+ mqcdev->enabled = true;
+ mqcdev->open_count = 0;
+ dev_set_drvdata(&mhi_dev->dev, mqcdev);
+
+ dev = device_create(mqc_dev_class, &mhi_dev->dev, MKDEV(mqc_dev_major, mqcdev->minor),
+ mqcdev, "%s", dev_name(&mhi_dev->dev));
+ if (IS_ERR(dev)) {
+ xa_erase(&mqc_xa, mqcdev->minor);
+ dev_set_drvdata(&mhi_dev->dev, NULL);
+ kfree(mqcdev);
+ return PTR_ERR(dev);
+ }
+
+ return 0;
+};
+
+static void mhi_qaic_ctrl_remove(struct mhi_device *mhi_dev)
+{
+ struct mqc_dev *mqcdev = dev_get_drvdata(&mhi_dev->dev);
+
+ device_destroy(mqc_dev_class, MKDEV(mqc_dev_major, mqcdev->minor));
+
+ mutex_lock(&mqcdev->lock);
+ mqcdev->enabled = false;
+ if (mqcdev->open_count)
+ __mhi_qaic_ctrl_release(mqcdev);
+ mutex_unlock(&mqcdev->lock);
+
+ xa_erase(&mqc_xa, mqcdev->minor);
+ kref_put(&mqcdev->ref_count, mqc_dev_release);
+}
+
+/* .driver_data stores max mtu */
+static const struct mhi_device_id mhi_qaic_ctrl_match_table[] = {
+ { .chan = "QAIC_SAHARA", .driver_data = SZ_32K},
+ {},
+};
+MODULE_DEVICE_TABLE(mhi, mhi_qaic_ctrl_match_table);
+
+static struct mhi_driver mhi_qaic_ctrl_driver = {
+ .id_table = mhi_qaic_ctrl_match_table,
+ .remove = mhi_qaic_ctrl_remove,
+ .probe = mhi_qaic_ctrl_probe,
+ .ul_xfer_cb = mhi_qaic_ctrl_ul_xfer_cb,
+ .dl_xfer_cb = mhi_qaic_ctrl_dl_xfer_cb,
+ .driver = {
+ .name = MHI_QAIC_CTRL_DRIVER_NAME,
+ },
+};
+
+int mhi_qaic_ctrl_init(void)
+{
+ int ret;
+
+ ret = register_chrdev(0, MHI_QAIC_CTRL_DRIVER_NAME, &mhidev_fops);
+ if (ret < 0)
+ return ret;
+
+ mqc_dev_major = ret;
+ mqc_dev_class = class_create(THIS_MODULE, MHI_QAIC_CTRL_DRIVER_NAME);
+ if (IS_ERR(mqc_dev_class)) {
+ ret = PTR_ERR(mqc_dev_class);
+ goto unregister_chrdev;
+ }
+
+ ret = mhi_driver_register(&mhi_qaic_ctrl_driver);
+ if (ret)
+ goto destroy_class;
+
+ return 0;
+
+destroy_class:
+ class_destroy(mqc_dev_class);
+unregister_chrdev:
+ unregister_chrdev(mqc_dev_major, MHI_QAIC_CTRL_DRIVER_NAME);
+ return ret;
+}
+
+void mhi_qaic_ctrl_deinit(void)
+{
+ mhi_driver_unregister(&mhi_qaic_ctrl_driver);
+ class_destroy(mqc_dev_class);
+ unregister_chrdev(mqc_dev_major, MHI_QAIC_CTRL_DRIVER_NAME);
+ xa_destroy(&mqc_xa);
+}
diff --git a/drivers/accel/qaic/mhi_qaic_ctrl.h b/drivers/accel/qaic/mhi_qaic_ctrl.h
new file mode 100644
index 0000000..930b3ac
--- /dev/null
+++ b/drivers/accel/qaic/mhi_qaic_ctrl.h
@@ -0,0 +1,12 @@
+/* SPDX-License-Identifier: GPL-2.0-only
+ *
+ * Copyright (c) 2022 Qualcomm Innovation Center, Inc. All rights reserved.
+ */
+
+#ifndef __MHI_QAIC_CTRL_H__
+#define __MHI_QAIC_CTRL_H__
+
+int mhi_qaic_ctrl_init(void);
+void mhi_qaic_ctrl_deinit(void);
+
+#endif /* __MHI_QAIC_CTRL_H__ */
--
2.7.4
^ permalink raw reply related [flat|nested] 16+ messages in thread* [PATCH v5 7/8] accel/qaic: Add qaic driver to the build system
2023-03-27 15:54 [PATCH v5 0/8] QAIC accel driver Jeffrey Hugo
` (5 preceding siblings ...)
2023-03-27 15:54 ` [PATCH v5 6/8] accel/qaic: Add mhi_qaic_cntl Jeffrey Hugo
@ 2023-03-27 15:54 ` Jeffrey Hugo
2023-03-27 15:54 ` [PATCH v5 8/8] MAINTAINERS: Add entry for QAIC driver Jeffrey Hugo
2023-04-03 17:22 ` [PATCH v5 0/8] QAIC accel driver Jeffrey Hugo
8 siblings, 0 replies; 16+ messages in thread
From: Jeffrey Hugo @ 2023-03-27 15:54 UTC (permalink / raw)
To: ogabbay, airlied, daniel
Cc: jacek.lawrynowicz, stanislaw.gruszka, dri-devel, quic_pkanojiy,
quic_carlv, quic_ajitpals, linux-doc, linux-arm-msm, bagasdotme,
mani, Jeffrey Hugo
Now that we have all the components of a minimum QAIC which can boot and
run an AIC100 device, add the infrastructure that allows the QAIC driver
to be built.
Signed-off-by: Jeffrey Hugo <quic_jhugo@quicinc.com>
Reviewed-by: Carl Vanderlip <quic_carlv@quicinc.com>
Reviewed-by: Pranjal Ramajor Asha Kanojiya <quic_pkanojiy@quicinc.com>
Reviewed-by: Stanislaw Gruszka <stanislaw.gruszka@linux.intel.com>
Reviewed-by: Jacek Lawrynowicz <jacek.lawrynowicz@linux.intel.com>
---
drivers/accel/Kconfig | 1 +
drivers/accel/Makefile | 1 +
drivers/accel/qaic/Kconfig | 23 +++++++++++++++++++++++
drivers/accel/qaic/Makefile | 13 +++++++++++++
4 files changed, 38 insertions(+)
create mode 100644 drivers/accel/qaic/Kconfig
create mode 100644 drivers/accel/qaic/Makefile
diff --git a/drivers/accel/Kconfig b/drivers/accel/Kconfig
index c437206..64065fb 100644
--- a/drivers/accel/Kconfig
+++ b/drivers/accel/Kconfig
@@ -26,5 +26,6 @@ menuconfig DRM_ACCEL
source "drivers/accel/habanalabs/Kconfig"
source "drivers/accel/ivpu/Kconfig"
+source "drivers/accel/qaic/Kconfig"
endif
diff --git a/drivers/accel/Makefile b/drivers/accel/Makefile
index f22fd44..ab3df93 100644
--- a/drivers/accel/Makefile
+++ b/drivers/accel/Makefile
@@ -2,3 +2,4 @@
obj-$(CONFIG_DRM_ACCEL_HABANALABS) += habanalabs/
obj-$(CONFIG_DRM_ACCEL_IVPU) += ivpu/
+obj-$(CONFIG_DRM_ACCEL_QAIC) += qaic/
diff --git a/drivers/accel/qaic/Kconfig b/drivers/accel/qaic/Kconfig
new file mode 100644
index 0000000..a9f8662
--- /dev/null
+++ b/drivers/accel/qaic/Kconfig
@@ -0,0 +1,23 @@
+# SPDX-License-Identifier: GPL-2.0-only
+#
+# Qualcomm Cloud AI accelerators driver
+#
+
+config DRM_ACCEL_QAIC
+ tristate "Qualcomm Cloud AI accelerators"
+ depends on DRM_ACCEL
+ depends on PCI && HAS_IOMEM
+ depends on MHI_BUS
+ depends on MMU
+ select CRC32
+ help
+ Enables driver for Qualcomm's Cloud AI accelerator PCIe cards that are
+ designed to accelerate Deep Learning inference workloads.
+
+ The driver manages the PCIe devices and provides an IOCTL interface
+ for users to submit workloads to the devices.
+
+ If unsure, say N.
+
+ To compile this driver as a module, choose M here: the
+ module will be called qaic.
diff --git a/drivers/accel/qaic/Makefile b/drivers/accel/qaic/Makefile
new file mode 100644
index 0000000..d5f4952
--- /dev/null
+++ b/drivers/accel/qaic/Makefile
@@ -0,0 +1,13 @@
+# SPDX-License-Identifier: GPL-2.0-only
+#
+# Makefile for Qualcomm Cloud AI accelerators driver
+#
+
+obj-$(CONFIG_DRM_ACCEL_QAIC) := qaic.o
+
+qaic-y := \
+ mhi_controller.o \
+ mhi_qaic_ctrl.o \
+ qaic_control.o \
+ qaic_data.o \
+ qaic_drv.o
--
2.7.4
^ permalink raw reply related [flat|nested] 16+ messages in thread* [PATCH v5 8/8] MAINTAINERS: Add entry for QAIC driver
2023-03-27 15:54 [PATCH v5 0/8] QAIC accel driver Jeffrey Hugo
` (6 preceding siblings ...)
2023-03-27 15:54 ` [PATCH v5 7/8] accel/qaic: Add qaic driver to the build system Jeffrey Hugo
@ 2023-03-27 15:54 ` Jeffrey Hugo
2023-04-03 17:22 ` [PATCH v5 0/8] QAIC accel driver Jeffrey Hugo
8 siblings, 0 replies; 16+ messages in thread
From: Jeffrey Hugo @ 2023-03-27 15:54 UTC (permalink / raw)
To: ogabbay, airlied, daniel
Cc: jacek.lawrynowicz, stanislaw.gruszka, dri-devel, quic_pkanojiy,
quic_carlv, quic_ajitpals, linux-doc, linux-arm-msm, bagasdotme,
mani, Jeffrey Hugo
Add MAINTAINERS entry for the Qualcomm Cloud AI 100 driver.
Signed-off-by: Jeffrey Hugo <quic_jhugo@quicinc.com>
Reviewed-by: Carl Vanderlip <quic_carlv@quicinc.com>
Reviewed-by: Pranjal Ramajor Asha Kanojiya <quic_pkanojiy@quicinc.com>
Reviewed-by: Stanislaw Gruszka <stanislaw.gruszka@linux.intel.com>
Reviewed-by: Jacek Lawrynowicz <jacek.lawrynowicz@linux.intel.com>
---
MAINTAINERS | 10 ++++++++++
1 file changed, 10 insertions(+)
diff --git a/MAINTAINERS b/MAINTAINERS
index d8ebab5..6cc18f4 100644
--- a/MAINTAINERS
+++ b/MAINTAINERS
@@ -17248,6 +17248,16 @@ F: Documentation/devicetree/bindings/clock/qcom,*
F: drivers/clk/qcom/
F: include/dt-bindings/clock/qcom,*
+QUALCOMM CLOUD AI (QAIC) DRIVER
+M: Jeffrey Hugo <quic_jhugo@quicinc.com>
+L: linux-arm-msm@vger.kernel.org
+L: dri-devel@lists.freedesktop.org
+S: Supported
+T: git git://anongit.freedesktop.org/drm/drm-misc
+F: Documentation/accel/qaic/
+F: drivers/accel/qaic/
+F: include/uapi/drm/qaic_accel.h
+
QUALCOMM CORE POWER REDUCTION (CPR) AVS DRIVER
M: Bjorn Andersson <andersson@kernel.org>
M: Konrad Dybcio <konrad.dybcio@linaro.org>
--
2.7.4
^ permalink raw reply related [flat|nested] 16+ messages in thread* Re: [PATCH v5 0/8] QAIC accel driver
2023-03-27 15:54 [PATCH v5 0/8] QAIC accel driver Jeffrey Hugo
` (7 preceding siblings ...)
2023-03-27 15:54 ` [PATCH v5 8/8] MAINTAINERS: Add entry for QAIC driver Jeffrey Hugo
@ 2023-04-03 17:22 ` Jeffrey Hugo
2023-04-05 11:26 ` Jacek Lawrynowicz
8 siblings, 1 reply; 16+ messages in thread
From: Jeffrey Hugo @ 2023-04-03 17:22 UTC (permalink / raw)
To: ogabbay, airlied, daniel
Cc: jacek.lawrynowicz, stanislaw.gruszka, dri-devel, quic_pkanojiy,
quic_carlv, quic_ajitpals, linux-doc, linux-arm-msm, bagasdotme,
mani
On 3/27/2023 9:54 AM, Jeffrey Hugo wrote:
> This series introduces a driver under the accel subsystem (QAIC -
> Qualcomm AIC) for the Qualcomm Cloud AI 100 product (AIC100). AIC100 is
> a PCIe adapter card that hosts a dedicated machine learning inference
> accelerator.
>
> The previous version (v4) can be found at:
> https://lore.kernel.org/all/1679325074-5494-1-git-send-email-quic_jhugo@quicinc.com/
Looks like things have been silent on this revision and we have a number
of review tags already. Seems like this series is ready for merge.
I'd like to see this queued for 6.4 if possible. Given that we are at
6.3-rc5, it seems like this would need to be merged now(ish) to make 6.4.
Jacek, since you have commit permissions in drm-misc and are an active
Accel maintainer, I wonder if it would be appropriate for you to merge
this series to drm-misc. Thoughts?
-Jeff
^ permalink raw reply [flat|nested] 16+ messages in thread* Re: [PATCH v5 0/8] QAIC accel driver
2023-04-03 17:22 ` [PATCH v5 0/8] QAIC accel driver Jeffrey Hugo
@ 2023-04-05 11:26 ` Jacek Lawrynowicz
2023-04-05 12:35 ` Oded Gabbay
0 siblings, 1 reply; 16+ messages in thread
From: Jacek Lawrynowicz @ 2023-04-05 11:26 UTC (permalink / raw)
To: Jeffrey Hugo, ogabbay, airlied, daniel
Cc: linux-doc, linux-arm-msm, mani, dri-devel, quic_ajitpals,
quic_pkanojiy, stanislaw.gruszka, quic_carlv, bagasdotme
Hi,
On 03.04.2023 19:22, Jeffrey Hugo wrote:
> On 3/27/2023 9:54 AM, Jeffrey Hugo wrote:
>> This series introduces a driver under the accel subsystem (QAIC -
>> Qualcomm AIC) for the Qualcomm Cloud AI 100 product (AIC100). AIC100 is
>> a PCIe adapter card that hosts a dedicated machine learning inference
>> accelerator.
>>
>> The previous version (v4) can be found at:
>> https://lore.kernel.org/all/1679325074-5494-1-git-send-email-quic_jhugo@quicinc.com/
>
> Looks like things have been silent on this revision and we have a number of review tags already. Seems like this series is ready for merge.
>
> I'd like to see this queued for 6.4 if possible. Given that we are at 6.3-rc5, it seems like this would need to be merged now(ish) to make 6.4.
>
> Jacek, since you have commit permissions in drm-misc and are an active Accel maintainer, I wonder if it would be appropriate for you to merge this series to drm-misc. Thoughts?
I'm would be happy to merge it but I think it needs to be acked by Oded first.
Regards,
Jacek
^ permalink raw reply [flat|nested] 16+ messages in thread
* Re: [PATCH v5 0/8] QAIC accel driver
2023-04-05 11:26 ` Jacek Lawrynowicz
@ 2023-04-05 12:35 ` Oded Gabbay
2023-04-05 13:27 ` Daniel Vetter
0 siblings, 1 reply; 16+ messages in thread
From: Oded Gabbay @ 2023-04-05 12:35 UTC (permalink / raw)
To: Jacek Lawrynowicz
Cc: Jeffrey Hugo, airlied, daniel, mani, linux-arm-msm, linux-doc,
dri-devel, quic_ajitpals, quic_pkanojiy, stanislaw.gruszka,
quic_carlv, bagasdotme
On Wed, Apr 5, 2023 at 2:26 PM Jacek Lawrynowicz
<jacek.lawrynowicz@linux.intel.com> wrote:
>
> Hi,
>
> On 03.04.2023 19:22, Jeffrey Hugo wrote:
> > On 3/27/2023 9:54 AM, Jeffrey Hugo wrote:
> >> This series introduces a driver under the accel subsystem (QAIC -
> >> Qualcomm AIC) for the Qualcomm Cloud AI 100 product (AIC100). AIC100 is
> >> a PCIe adapter card that hosts a dedicated machine learning inference
> >> accelerator.
> >>
> >> The previous version (v4) can be found at:
> >> https://lore.kernel.org/all/1679325074-5494-1-git-send-email-quic_jhugo@quicinc.com/
> >
> > Looks like things have been silent on this revision and we have a number of review tags already. Seems like this series is ready for merge.
> >
> > I'd like to see this queued for 6.4 if possible. Given that we are at 6.3-rc5, it seems like this would need to be merged now(ish) to make 6.4.
> >
> > Jacek, since you have commit permissions in drm-misc and are an active Accel maintainer, I wonder if it would be appropriate for you to merge this series to drm-misc. Thoughts?
>
> I'm would be happy to merge it but I think it needs to be acked by Oded first.
>
> Regards,
> Jacek
Hi,
Entire patch-set is:
Acked-by: Oded Gabbay <ogabbay@kernel.org>
Thanks,
Oded
^ permalink raw reply [flat|nested] 16+ messages in thread
* Re: [PATCH v5 0/8] QAIC accel driver
2023-04-05 12:35 ` Oded Gabbay
@ 2023-04-05 13:27 ` Daniel Vetter
2023-04-05 14:28 ` Jeffrey Hugo
0 siblings, 1 reply; 16+ messages in thread
From: Daniel Vetter @ 2023-04-05 13:27 UTC (permalink / raw)
To: Oded Gabbay
Cc: Jacek Lawrynowicz, Jeffrey Hugo, airlied, daniel, mani,
linux-arm-msm, linux-doc, dri-devel, quic_ajitpals, quic_pkanojiy,
stanislaw.gruszka, quic_carlv, bagasdotme
On Wed, Apr 05, 2023 at 03:35:19PM +0300, Oded Gabbay wrote:
> On Wed, Apr 5, 2023 at 2:26 PM Jacek Lawrynowicz
> <jacek.lawrynowicz@linux.intel.com> wrote:
> >
> > Hi,
> >
> > On 03.04.2023 19:22, Jeffrey Hugo wrote:
> > > On 3/27/2023 9:54 AM, Jeffrey Hugo wrote:
> > >> This series introduces a driver under the accel subsystem (QAIC -
> > >> Qualcomm AIC) for the Qualcomm Cloud AI 100 product (AIC100). AIC100 is
> > >> a PCIe adapter card that hosts a dedicated machine learning inference
> > >> accelerator.
> > >>
> > >> The previous version (v4) can be found at:
> > >> https://lore.kernel.org/all/1679325074-5494-1-git-send-email-quic_jhugo@quicinc.com/
> > >
> > > Looks like things have been silent on this revision and we have a number of review tags already. Seems like this series is ready for merge.
> > >
> > > I'd like to see this queued for 6.4 if possible. Given that we are at 6.3-rc5, it seems like this would need to be merged now(ish) to make 6.4.
> > >
> > > Jacek, since you have commit permissions in drm-misc and are an active Accel maintainer, I wonder if it would be appropriate for you to merge this series to drm-misc. Thoughts?
> >
> > I'm would be happy to merge it but I think it needs to be acked by Oded first.
> >
> > Regards,
> > Jacek
>
> Hi,
> Entire patch-set is:
> Acked-by: Oded Gabbay <ogabbay@kernel.org>
Once Jacke has pushed this I htink it would also be good to get Jeffrey
commit rights for drm-misc, so that in the future bugfixes for the qaic
driver can be pushed directly by the qaic team. Still with acks/r-b
requirements as per usual, and I guess for anything bigger/new uapi an ack
from oded is needed.
https://drm.pages.freedesktop.org/maintainer-tools/commit-access.html#drm-misc
Cheers, Daniel
--
Daniel Vetter
Software Engineer, Intel Corporation
http://blog.ffwll.ch
^ permalink raw reply [flat|nested] 16+ messages in thread
* Re: [PATCH v5 0/8] QAIC accel driver
2023-04-05 13:27 ` Daniel Vetter
@ 2023-04-05 14:28 ` Jeffrey Hugo
2023-04-06 6:34 ` Jacek Lawrynowicz
0 siblings, 1 reply; 16+ messages in thread
From: Jeffrey Hugo @ 2023-04-05 14:28 UTC (permalink / raw)
To: Daniel Vetter, Oded Gabbay
Cc: Jacek Lawrynowicz, airlied, mani, linux-arm-msm, linux-doc,
dri-devel, quic_ajitpals, quic_pkanojiy, stanislaw.gruszka,
quic_carlv, bagasdotme
On 4/5/2023 7:27 AM, Daniel Vetter wrote:
> On Wed, Apr 05, 2023 at 03:35:19PM +0300, Oded Gabbay wrote:
>> On Wed, Apr 5, 2023 at 2:26 PM Jacek Lawrynowicz
>> <jacek.lawrynowicz@linux.intel.com> wrote:
>>>
>>> Hi,
>>>
>>> On 03.04.2023 19:22, Jeffrey Hugo wrote:
>>>> On 3/27/2023 9:54 AM, Jeffrey Hugo wrote:
>>>>> This series introduces a driver under the accel subsystem (QAIC -
>>>>> Qualcomm AIC) for the Qualcomm Cloud AI 100 product (AIC100). AIC100 is
>>>>> a PCIe adapter card that hosts a dedicated machine learning inference
>>>>> accelerator.
>>>>>
>>>>> The previous version (v4) can be found at:
>>>>> https://lore.kernel.org/all/1679325074-5494-1-git-send-email-quic_jhugo@quicinc.com/
>>>>
>>>> Looks like things have been silent on this revision and we have a number of review tags already. Seems like this series is ready for merge.
>>>>
>>>> I'd like to see this queued for 6.4 if possible. Given that we are at 6.3-rc5, it seems like this would need to be merged now(ish) to make 6.4.
>>>>
>>>> Jacek, since you have commit permissions in drm-misc and are an active Accel maintainer, I wonder if it would be appropriate for you to merge this series to drm-misc. Thoughts?
>>>
>>> I'm would be happy to merge it but I think it needs to be acked by Oded first.
>>>
>>> Regards,
>>> Jacek
>>
>> Hi,
>> Entire patch-set is:
>> Acked-by: Oded Gabbay <ogabbay@kernel.org>
>
> Once Jacke has pushed this I htink it would also be good to get Jeffrey
> commit rights for drm-misc, so that in the future bugfixes for the qaic
> driver can be pushed directly by the qaic team. Still with acks/r-b
> requirements as per usual, and I guess for anything bigger/new uapi an ack
> from oded is needed.
>
> https://drm.pages.freedesktop.org/maintainer-tools/commit-access.html#drm-misc
>
> Cheers, Daniel
Yep. This is the plan per my understanding.
-Jeff
^ permalink raw reply [flat|nested] 16+ messages in thread
* Re: [PATCH v5 0/8] QAIC accel driver
2023-04-05 14:28 ` Jeffrey Hugo
@ 2023-04-06 6:34 ` Jacek Lawrynowicz
2023-04-06 13:15 ` Jeffrey Hugo
0 siblings, 1 reply; 16+ messages in thread
From: Jacek Lawrynowicz @ 2023-04-06 6:34 UTC (permalink / raw)
To: Jeffrey Hugo, Daniel Vetter, Oded Gabbay
Cc: linux-doc, linux-arm-msm, mani, dri-devel, quic_ajitpals,
quic_pkanojiy, stanislaw.gruszka, quic_carlv, bagasdotme
I've merged the patchset to drm-misc-next.
Congrats Jeffrey :)
On 05.04.2023 16:28, Jeffrey Hugo wrote:
> On 4/5/2023 7:27 AM, Daniel Vetter wrote:
>> On Wed, Apr 05, 2023 at 03:35:19PM +0300, Oded Gabbay wrote:
>>> On Wed, Apr 5, 2023 at 2:26 PM Jacek Lawrynowicz
>>> <jacek.lawrynowicz@linux.intel.com> wrote:
>>>>
>>>> Hi,
>>>>
>>>> On 03.04.2023 19:22, Jeffrey Hugo wrote:
>>>>> On 3/27/2023 9:54 AM, Jeffrey Hugo wrote:
>>>>>> This series introduces a driver under the accel subsystem (QAIC -
>>>>>> Qualcomm AIC) for the Qualcomm Cloud AI 100 product (AIC100). AIC100 is
>>>>>> a PCIe adapter card that hosts a dedicated machine learning inference
>>>>>> accelerator.
>>>>>>
>>>>>> The previous version (v4) can be found at:
>>>>>> https://lore.kernel.org/all/1679325074-5494-1-git-send-email-quic_jhugo@quicinc.com/
>>>>>
>>>>> Looks like things have been silent on this revision and we have a number of review tags already. Seems like this series is ready for merge.
>>>>>
>>>>> I'd like to see this queued for 6.4 if possible. Given that we are at 6.3-rc5, it seems like this would need to be merged now(ish) to make 6.4.
>>>>>
>>>>> Jacek, since you have commit permissions in drm-misc and are an active Accel maintainer, I wonder if it would be appropriate for you to merge this series to drm-misc. Thoughts?
>>>>
>>>> I'm would be happy to merge it but I think it needs to be acked by Oded first.
>>>>
>>>> Regards,
>>>> Jacek
>>>
>>> Hi,
>>> Entire patch-set is:
>>> Acked-by: Oded Gabbay <ogabbay@kernel.org>
>>
>> Once Jacke has pushed this I htink it would also be good to get Jeffrey
>> commit rights for drm-misc, so that in the future bugfixes for the qaic
>> driver can be pushed directly by the qaic team. Still with acks/r-b
>> requirements as per usual, and I guess for anything bigger/new uapi an ack
>> from oded is needed.
>>
>> https://drm.pages.freedesktop.org/maintainer-tools/commit-access.html#drm-misc
>>
>> Cheers, Daniel
>
> Yep. This is the plan per my understanding.
>
> -Jeff
^ permalink raw reply [flat|nested] 16+ messages in thread
* Re: [PATCH v5 0/8] QAIC accel driver
2023-04-06 6:34 ` Jacek Lawrynowicz
@ 2023-04-06 13:15 ` Jeffrey Hugo
0 siblings, 0 replies; 16+ messages in thread
From: Jeffrey Hugo @ 2023-04-06 13:15 UTC (permalink / raw)
To: Jacek Lawrynowicz, Daniel Vetter, Oded Gabbay
Cc: linux-doc, linux-arm-msm, mani, dri-devel, quic_ajitpals,
quic_pkanojiy, stanislaw.gruszka, quic_carlv, bagasdotme
On 4/6/2023 12:34 AM, Jacek Lawrynowicz wrote:
> I've merged the patchset to drm-misc-next.
> Congrats Jeffrey :)
Thanks!
>
> On 05.04.2023 16:28, Jeffrey Hugo wrote:
>> On 4/5/2023 7:27 AM, Daniel Vetter wrote:
>>> On Wed, Apr 05, 2023 at 03:35:19PM +0300, Oded Gabbay wrote:
>>>> On Wed, Apr 5, 2023 at 2:26 PM Jacek Lawrynowicz
>>>> <jacek.lawrynowicz@linux.intel.com> wrote:
>>>>>
>>>>> Hi,
>>>>>
>>>>> On 03.04.2023 19:22, Jeffrey Hugo wrote:
>>>>>> On 3/27/2023 9:54 AM, Jeffrey Hugo wrote:
>>>>>>> This series introduces a driver under the accel subsystem (QAIC -
>>>>>>> Qualcomm AIC) for the Qualcomm Cloud AI 100 product (AIC100). AIC100 is
>>>>>>> a PCIe adapter card that hosts a dedicated machine learning inference
>>>>>>> accelerator.
>>>>>>>
>>>>>>> The previous version (v4) can be found at:
>>>>>>> https://lore.kernel.org/all/1679325074-5494-1-git-send-email-quic_jhugo@quicinc.com/
>>>>>>
>>>>>> Looks like things have been silent on this revision and we have a number of review tags already. Seems like this series is ready for merge.
>>>>>>
>>>>>> I'd like to see this queued for 6.4 if possible. Given that we are at 6.3-rc5, it seems like this would need to be merged now(ish) to make 6.4.
>>>>>>
>>>>>> Jacek, since you have commit permissions in drm-misc and are an active Accel maintainer, I wonder if it would be appropriate for you to merge this series to drm-misc. Thoughts?
>>>>>
>>>>> I'm would be happy to merge it but I think it needs to be acked by Oded first.
>>>>>
>>>>> Regards,
>>>>> Jacek
>>>>
>>>> Hi,
>>>> Entire patch-set is:
>>>> Acked-by: Oded Gabbay <ogabbay@kernel.org>
>>>
>>> Once Jacke has pushed this I htink it would also be good to get Jeffrey
>>> commit rights for drm-misc, so that in the future bugfixes for the qaic
>>> driver can be pushed directly by the qaic team. Still with acks/r-b
>>> requirements as per usual, and I guess for anything bigger/new uapi an ack
>>> from oded is needed.
>>>
>>> https://drm.pages.freedesktop.org/maintainer-tools/commit-access.html#drm-misc
>>>
>>> Cheers, Daniel
>>
>> Yep. This is the plan per my understanding.
>>
>> -Jeff
^ permalink raw reply [flat|nested] 16+ messages in thread