Commit | Line | Data |
---|---|---|
a1bd3bae AH |
1 | # NTB Drivers |
2 | ||
3 | NTB (Non-Transparent Bridge) is a type of PCI-Express bridge chip that connects | |
cdcca896 SS |
4 | the separate memory systems of two or more computers to the same PCI-Express |
5 | fabric. Existing NTB hardware supports a common feature set: doorbell | |
6 | registers and memory translation windows, as well as non common features like | |
7 | scratchpad and message registers. Scratchpad registers are read-and-writable | |
8 | registers that are accessible from either side of the device, so that peers can | |
9 | exchange a small amount of information at a fixed address. Message registers can | |
10 | be utilized for the same purpose. Additionally they are provided with with | |
11 | special status bits to make sure the information isn't rewritten by another | |
12 | peer. Doorbell registers provide a way for peers to send interrupt events. | |
13 | Memory windows allow translated read and write access to the peer memory. | |
a1bd3bae AH |
14 | |
15 | ## NTB Core Driver (ntb) | |
16 | ||
17 | The NTB core driver defines an api wrapping the common feature set, and allows | |
18 | clients interested in NTB features to discover NTB the devices supported by | |
19 | hardware drivers. The term "client" is used here to mean an upper layer | |
20 | component making use of the NTB api. The term "driver," or "hardware driver," | |
21 | is used here to mean a driver for a specific vendor and model of NTB hardware. | |
22 | ||
23 | ## NTB Client Drivers | |
24 | ||
25 | NTB client drivers should register with the NTB core driver. After | |
26 | registering, the client probe and remove functions will be called appropriately | |
27 | as ntb hardware, or hardware drivers, are inserted and removed. The | |
28 | registration uses the Linux Device framework, so it should feel familiar to | |
29 | anyone who has written a pci driver. | |
30 | ||
cdcca896 SS |
31 | ### NTB Typical client driver implementation |
32 | ||
33 | Primary purpose of NTB is to share some peace of memory between at least two | |
34 | systems. So the NTB device features like Scratchpad/Message registers are | |
35 | mainly used to perform the proper memory window initialization. Typically | |
36 | there are two types of memory window interfaces supported by the NTB API: | |
37 | inbound translation configured on the local ntb port and outbound translation | |
38 | configured by the peer, on the peer ntb port. The first type is | |
39 | depicted on the next figure | |
40 | ||
41 | Inbound translation: | |
42 | Memory: Local NTB Port: Peer NTB Port: Peer MMIO: | |
43 | ____________ | |
44 | | dma-mapped |-ntb_mw_set_trans(addr) | | |
45 | | memory | _v____________ | ______________ | |
46 | | (addr) |<======| MW xlat addr |<====| MW base addr |<== memory-mapped IO | |
47 | |------------| |--------------| | |--------------| | |
48 | ||
49 | So typical scenario of the first type memory window initialization looks: | |
50 | 1) allocate a memory region, 2) put translated address to NTB config, | |
51 | 3) somehow notify a peer device of performed initialization, 4) peer device | |
52 | maps corresponding outbound memory window so to have access to the shared | |
53 | memory region. | |
54 | ||
55 | The second type of interface, that implies the shared windows being | |
56 | initialized by a peer device, is depicted on the figure: | |
57 | ||
58 | Outbound translation: | |
59 | Memory: Local NTB Port: Peer NTB Port: Peer MMIO: | |
60 | ____________ ______________ | |
61 | | dma-mapped | | | MW base addr |<== memory-mapped IO | |
62 | | memory | | |--------------| | |
63 | | (addr) |<===================| MW xlat addr |<-ntb_peer_mw_set_trans(addr) | |
64 | |------------| | |--------------| | |
65 | ||
66 | Typical scenario of the second type interface initialization would be: | |
67 | 1) allocate a memory region, 2) somehow deliver a translated address to a peer | |
68 | device, 3) peer puts the translated address to NTB config, 4) peer device maps | |
69 | outbound memory window so to have access to the shared memory region. | |
70 | ||
71 | As one can see the described scenarios can be combined in one portable | |
72 | algorithm. | |
73 | Local device: | |
74 | 1) Allocate memory for a shared window | |
75 | 2) Initialize memory window by translated address of the allocated region | |
76 | (it may fail if local memory window initialization is unsupported) | |
77 | 3) Send the translated address and memory window index to a peer device | |
78 | Peer device: | |
79 | 1) Initialize memory window with retrieved address of the allocated | |
80 | by another device memory region (it may fail if peer memory window | |
81 | initialization is unsupported) | |
82 | 2) Map outbound memory window | |
83 | ||
84 | In accordance with this scenario, the NTB Memory Window API can be used as | |
85 | follows: | |
86 | Local device: | |
87 | 1) ntb_mw_count(pidx) - retrieve number of memory ranges, which can | |
88 | be allocated for memory windows between local device and peer device | |
89 | of port with specified index. | |
90 | 2) ntb_get_align(pidx, midx) - retrieve parameters restricting the | |
91 | shared memory region alignment and size. Then memory can be properly | |
92 | allocated. | |
93 | 3) Allocate physically contiguous memory region in compliance with | |
94 | restrictions retrieved in 2). | |
95 | 4) ntb_mw_set_trans(pidx, midx) - try to set translation address of | |
96 | the memory window with specified index for the defined peer device | |
97 | (it may fail if local translated address setting is not supported) | |
98 | 5) Send translated base address (usually together with memory window | |
99 | number) to the peer device using, for instance, scratchpad or message | |
100 | registers. | |
101 | Peer device: | |
102 | 1) ntb_peer_mw_set_trans(pidx, midx) - try to set received from other | |
103 | device (related to pidx) translated address for specified memory | |
104 | window. It may fail if retrieved address, for instance, exceeds | |
105 | maximum possible address or isn't properly aligned. | |
106 | 2) ntb_peer_mw_get_addr(widx) - retrieve MMIO address to map the memory | |
107 | window so to have an access to the shared memory. | |
108 | ||
109 | Also it is worth to note, that method ntb_mw_count(pidx) should return the | |
110 | same value as ntb_peer_mw_count() on the peer with port index - pidx. | |
111 | ||
e26a5843 AH |
112 | ### NTB Transport Client (ntb\_transport) and NTB Netdev (ntb\_netdev) |
113 | ||
114 | The primary client for NTB is the Transport client, used in tandem with NTB | |
115 | Netdev. These drivers function together to create a logical link to the peer, | |
116 | across the ntb, to exchange packets of network data. The Transport client | |
117 | establishes a logical link to the peer, and creates queue pairs to exchange | |
118 | messages and data. The NTB Netdev then creates an ethernet device using a | |
119 | Transport queue pair. Network data is copied between socket buffers and the | |
120 | Transport queue pair buffer. The Transport client may be used for other things | |
121 | besides Netdev, however no other applications have yet been written. | |
122 | ||
963de473 AH |
123 | ### NTB Ping Pong Test Client (ntb\_pingpong) |
124 | ||
125 | The Ping Pong test client serves as a demonstration to exercise the doorbell | |
126 | and scratchpad registers of NTB hardware, and as an example simple NTB client. | |
127 | Ping Pong enables the link when started, waits for the NTB link to come up, and | |
128 | then proceeds to read and write the doorbell scratchpad registers of the NTB. | |
129 | The peers interrupt each other using a bit mask of doorbell bits, which is | |
130 | shifted by one in each round, to test the behavior of multiple doorbell bits | |
131 | and interrupt vectors. The Ping Pong driver also reads the first local | |
132 | scratchpad, and writes the value plus one to the first peer scratchpad, each | |
133 | round before writing the peer doorbell register. | |
134 | ||
135 | Module Parameters: | |
136 | ||
137 | * unsafe - Some hardware has known issues with scratchpad and doorbell | |
138 | registers. By default, Ping Pong will not attempt to exercise such | |
139 | hardware. You may override this behavior at your own risk by setting | |
140 | unsafe=1. | |
141 | * delay\_ms - Specify the delay between receiving a doorbell | |
142 | interrupt event and setting the peer doorbell register for the next | |
143 | round. | |
144 | * init\_db - Specify the doorbell bits to start new series of rounds. A new | |
145 | series begins once all the doorbell bits have been shifted out of | |
146 | range. | |
147 | * dyndbg - It is suggested to specify dyndbg=+p when loading this module, and | |
148 | then to observe debugging output on the console. | |
149 | ||
578b881b AH |
150 | ### NTB Tool Test Client (ntb\_tool) |
151 | ||
152 | The Tool test client serves for debugging, primarily, ntb hardware and drivers. | |
153 | The Tool provides access through debugfs for reading, setting, and clearing the | |
154 | NTB doorbell, and reading and writing scratchpads. | |
155 | ||
156 | The Tool does not currently have any module parameters. | |
157 | ||
158 | Debugfs Files: | |
159 | ||
160 | * *debugfs*/ntb\_tool/*hw*/ - A directory in debugfs will be created for each | |
161 | NTB device probed by the tool. This directory is shortened to *hw* | |
162 | below. | |
163 | * *hw*/db - This file is used to read, set, and clear the local doorbell. Not | |
164 | all operations may be supported by all hardware. To read the doorbell, | |
165 | read the file. To set the doorbell, write `s` followed by the bits to | |
166 | set (eg: `echo 's 0x0101' > db`). To clear the doorbell, write `c` | |
167 | followed by the bits to clear. | |
168 | * *hw*/mask - This file is used to read, set, and clear the local doorbell mask. | |
169 | See *db* for details. | |
170 | * *hw*/peer\_db - This file is used to read, set, and clear the peer doorbell. | |
171 | See *db* for details. | |
172 | * *hw*/peer\_mask - This file is used to read, set, and clear the peer doorbell | |
173 | mask. See *db* for details. | |
174 | * *hw*/spad - This file is used to read and write local scratchpads. To read | |
175 | the values of all scratchpads, read the file. To write values, write a | |
176 | series of pairs of scratchpad number and value | |
177 | (eg: `echo '4 0x123 7 0xabc' > spad` | |
178 | # to set scratchpads `4` and `7` to `0x123` and `0xabc`, respectively). | |
179 | * *hw*/peer\_spad - This file is used to read and write peer scratchpads. See | |
180 | *spad* for details. | |
181 | ||
a1bd3bae AH |
182 | ## NTB Hardware Drivers |
183 | ||
184 | NTB hardware drivers should register devices with the NTB core driver. After | |
185 | registering, clients probe and remove functions will be called. | |
e26a5843 AH |
186 | |
187 | ### NTB Intel Hardware Driver (ntb\_hw\_intel) | |
188 | ||
189 | The Intel hardware driver supports NTB on Xeon and Atom CPUs. | |
190 | ||
191 | Module Parameters: | |
192 | ||
193 | * b2b\_mw\_idx - If the peer ntb is to be accessed via a memory window, then use | |
194 | this memory window to access the peer ntb. A value of zero or positive | |
195 | starts from the first mw idx, and a negative value starts from the last | |
196 | mw idx. Both sides MUST set the same value here! The default value is | |
197 | `-1`. | |
198 | * b2b\_mw\_share - If the peer ntb is to be accessed via a memory window, and if | |
199 | the memory window is large enough, still allow the client to use the | |
200 | second half of the memory window for address translation to the peer. | |
2f887b9a DJ |
201 | * xeon\_b2b\_usd\_bar2\_addr64 - If using B2B topology on Xeon hardware, use |
202 | this 64 bit address on the bus between the NTB devices for the window | |
203 | at BAR2, on the upstream side of the link. | |
204 | * xeon\_b2b\_usd\_bar4\_addr64 - See *xeon\_b2b\_bar2\_addr64*. | |
205 | * xeon\_b2b\_usd\_bar4\_addr32 - See *xeon\_b2b\_bar2\_addr64*. | |
206 | * xeon\_b2b\_usd\_bar5\_addr32 - See *xeon\_b2b\_bar2\_addr64*. | |
207 | * xeon\_b2b\_dsd\_bar2\_addr64 - See *xeon\_b2b\_bar2\_addr64*. | |
208 | * xeon\_b2b\_dsd\_bar4\_addr64 - See *xeon\_b2b\_bar2\_addr64*. | |
209 | * xeon\_b2b\_dsd\_bar4\_addr32 - See *xeon\_b2b\_bar2\_addr64*. | |
210 | * xeon\_b2b\_dsd\_bar5\_addr32 - See *xeon\_b2b\_bar2\_addr64*. |