1 .. SPDX-License-Identifier: GPL-2.0-only
7 The QAIC driver is the Kernel Mode Driver (KMD) for the AIC100 family of AI
13 While the AIC100 DMA Bridge hardware implements an IRQ storm mitigation
14 mechanism, it is still possible for an IRQ storm to occur. A storm can happen
15 if the workload is particularly quick, and the host is responsive. If the host
16 can drain the response FIFO as quickly as the device can insert elements into
17 it, then the device will frequently transition the response FIFO from empty to
18 non-empty and generate MSIs at a rate equivalent to the speed of the
19 workload's ability to process inputs. The lprnet (license plate reader network)
20 workload is known to trigger this condition, and can generate in excess of 100k
21 MSIs per second. It has been observed that most systems cannot tolerate this
22 for long, and will crash due to some form of watchdog due to the overhead of
23 the interrupt controller interrupting the host CPU.
25 To mitigate this issue, the QAIC driver implements specific IRQ handling. When
26 QAIC receives an IRQ, it disables that line. This prevents the interrupt
27 controller from interrupting the CPU. Then AIC drains the FIFO. Once the FIFO
28 is drained, QAIC implements a "last chance" polling algorithm where QAIC will
29 sleep for a time to see if the workload will generate more activity. The IRQ
30 line remains disabled during this time. If no activity is detected, QAIC exits
31 polling mode and reenables the IRQ line.
33 This mitigation in QAIC is very effective. The same lprnet usecase that
34 generates 100k IRQs per second (per /proc/interrupts) is reduced to roughly 64
35 IRQs over 5 minutes while keeping the host system stable, and having the same
36 workload throughput performance (within run to run noise variation).
39 Neural Network Control (NNC) Protocol
40 =====================================
42 The implementation of NNC is split between the KMD (QAIC) and UMD. In general
43 QAIC understands how to encode/decode NNC wire protocol, and elements of the
44 protocol which require kernel space knowledge to process (for example, mapping
45 host memory to device IOVAs). QAIC understands the structure of a message, and
46 all of the transactions. QAIC does not understand commands (the payload of a
47 passthrough transaction).
49 QAIC handles and enforces the required little endianness and 64-bit alignment,
50 to the degree that it can. Since QAIC does not know the contents of a
51 passthrough transaction, it relies on the UMD to satisfy the requirements.
53 The terminate transaction is of particular use to QAIC. QAIC is not aware of
54 the resources that are loaded onto a device since the majority of that activity
55 occurs within NNC commands. As a result, QAIC does not have the means to
56 roll back userspace activity. To ensure that a userspace client's resources
57 are fully released in the case of a process crash, or a bug, QAIC uses the
58 terminate command to let QSM know when a user has gone away, and the resources
61 QSM can report a version number of the NNC protocol it supports. This is in the
62 form of a Major number and a Minor number.
64 Major number updates indicate changes to the NNC protocol which impact the
65 message format, or transactions (impacts QAIC).
67 Minor number updates indicate changes to the NNC protocol which impact the
68 commands (does not impact QAIC).
73 QAIC defines a number of driver specific IOCTLs as part of the userspace API.
74 This section describes those APIs.
77 This IOCTL allows userspace to send a NNC request to the QSM. The call will
78 block until a response is received, or the request has timed out.
80 DRM_IOCTL_QAIC_CREATE_BO
81 This IOCTL allows userspace to allocate a buffer object (BO) which can send
82 or receive data from a workload. The call will return a GEM handle that
83 represents the allocated buffer. The BO is not usable until it has been
84 sliced (see DRM_IOCTL_QAIC_ATTACH_SLICE_BO).
86 DRM_IOCTL_QAIC_MMAP_BO
87 This IOCTL allows userspace to prepare an allocated BO to be mmap'd into the
90 DRM_IOCTL_QAIC_ATTACH_SLICE_BO
91 This IOCTL allows userspace to slice a BO in preparation for sending the BO
92 to the device. Slicing is the operation of describing what portions of a BO
93 get sent where to a workload. This requires a set of DMA transfers for the
94 DMA Bridge, and as such, locks the BO to a specific DBC.
96 DRM_IOCTL_QAIC_EXECUTE_BO
97 This IOCTL allows userspace to submit a set of sliced BOs to the device. The
98 call is non-blocking. Success only indicates that the BOs have been queued
99 to the device, but does not guarantee they have been executed.
101 DRM_IOCTL_QAIC_PARTIAL_EXECUTE_BO
102 This IOCTL operates like DRM_IOCTL_QAIC_EXECUTE_BO, but it allows userspace
103 to shrink the BOs sent to the device for this specific call. If a BO
104 typically has N inputs, but only a subset of those is available, this IOCTL
105 allows userspace to indicate that only the first M bytes of the BO should be
106 sent to the device to minimize data transfer overhead. This IOCTL dynamically
107 recomputes the slicing, and therefore has some processing overhead before the
108 BOs can be queued to the device.
110 DRM_IOCTL_QAIC_WAIT_BO
111 This IOCTL allows userspace to determine when a particular BO has been
112 processed by the device. The call will block until either the BO has been
113 processed and can be re-queued to the device, or a timeout occurs.
115 DRM_IOCTL_QAIC_PERF_STATS_BO
116 This IOCTL allows userspace to collect performance statistics on the most
117 recent execution of a BO. This allows userspace to construct an end to end
118 timeline of the BO processing for a performance analysis.
120 DRM_IOCTL_QAIC_PART_DEV
121 This IOCTL allows userspace to request a duplicate "shadow device". This extra
122 accelN device is associated with a specific partition of resources on the
123 AIC100 device and can be used for limiting a process to some subset of
126 DRM_IOCTL_QAIC_DETACH_SLICE_BO
127 This IOCTL allows userspace to remove the slicing information from a BO that
128 was originally provided by a call to DRM_IOCTL_QAIC_ATTACH_SLICE_BO. This
129 is the inverse of DRM_IOCTL_QAIC_ATTACH_SLICE_BO. The BO must be idle for
130 DRM_IOCTL_QAIC_DETACH_SLICE_BO to be called. After a successful detach slice
131 operation the BO may have new slicing information attached with a new call
132 to DRM_IOCTL_QAIC_ATTACH_SLICE_BO. After detach slice, the BO cannot be
133 executed until after a new attach slice operation. Combining attach slice
134 and detach slice calls allows userspace to use a BO with multiple workloads.
136 Userspace Client Isolation
137 ==========================
139 AIC100 supports multiple clients. Multiple DBCs can be consumed by a single
140 client, and multiple clients can each consume one or more DBCs. Workloads
141 may contain sensitive information therefore only the client that owns the
142 workload should be allowed to interface with the DBC.
144 Clients are identified by the instance associated with their open(). A client
145 may only use memory they allocate, and DBCs that are assigned to their
146 workloads. Attempts to access resources assigned to other clients will be
152 QAIC supports the following module parameters:
154 **datapath_polling (bool)**
156 Configures QAIC to use a polling thread for datapath events instead of relying
157 on the device interrupts. Useful for platforms with broken multiMSI. Must be
158 set at QAIC driver initialization. Default is 0 (off).
160 **mhi_timeout_ms (unsigned int)**
162 Sets the timeout value for MHI operations in milliseconds (ms). Must be set
163 at the time the driver detects a device. Default is 2000 (2 seconds).
165 **control_resp_timeout_s (unsigned int)**
167 Sets the timeout value for QSM responses to NNC messages in seconds (s). Must
168 be set at the time the driver is sending a request to QSM. Default is 60 (one
171 **wait_exec_default_timeout_ms (unsigned int)**
173 Sets the default timeout for the wait_exec ioctl in milliseconds (ms). Must be
174 set prior to the waic_exec ioctl call. A value specified in the ioctl call
175 overrides this for that call. Default is 5000 (5 seconds).
177 **datapath_poll_interval_us (unsigned int)**
179 Sets the polling interval in microseconds (us) when datapath polling is active.
180 Takes effect at the next polling interval. Default is 100 (100 us).