Copyright (c) 2004, Sun Microsystems, Inc., All Rights Reserved
The contents of this file are subject to the terms of the Common Development and Distribution License (the "License"). You may not use this file except in compliance with the License.
You can obtain a copy of the license at usr/src/OPENSOLARIS.LICENSE or http://www.opensolaris.org/os/licensing. See the License for the specific language governing permissions and limitations under the License.
When distributing Covered Code, include this CDDL HEADER in each file and include the License file at usr/src/OPENSOLARIS.LICENSE. If applicable, add the following below this CDDL HEADER, with the fields enclosed by brackets "[]" replaced with your own identifying information: Portions Copyright [yyyy] [name of copyright owner]
#include <sys/usb/usba.h>
Solaris DDI specific (Solaris DDI)
If the USB framework detects an error during a request execution, it calls the client driver's exception callback handler to report what happened. Callback flags (which are set prior to calling the exception callback handler) detail errors discovered during the exception recovery process, and summarize recovery actions taken by the USBA framework.
Information from the callback flags supplements information from the original transport error. For transfers, the original transport error status is returned to the callback handler through the original request (whose completion reason field contains any transport error indication). For command completion callbacks, the callback's rval argument contains the transport error status. A completion reason of USB_CR_OK means the transfer completed with no errors detected.
The usb_cb_flags_t enumerated type contains the following definitions: USB_CB_NO_INFO
No additional errors discovered or recovery actions taken.
A functional stall occurred during the transfer. A functional stall is usually caused by a hardware error, and must be explicitly cleared. A functional stall is fatal if it cannot be cleared. The default control pipe never shows a functional stall.
A functional stall has been cleared by the USBA framework. This can happen if USB_ATTRS_AUTOCLEARING is set in the request's xxxx_attributes field.
A protocol stall has occurred during the transfer. A protocol stall is caused usually by an invalid or misunderstood command. It is cleared automatically when the device is given its next command. The USBA framework treats stalls detected on default pipe transfers as protocol stalls.
A pipe with a stall has been reset automatically via autoclearing, or via an explicit call to usb_pipe_reset(9F). Resetting a pipe consists of stopping all transactions on a pipe, setting the pipe to the idle state, and if the pipe is not the default pipe, flushing all pending requests. The request which has the error, plus all pending requests which are flushed, show USB_CB_RESET_PIPE set in the usb_cb_flags_t when their exception callback is called.
Resources could not be allocated to process callbacks asynchronously. Callbacks receiving this flag must not block, since those callbacks are executing in a context which holds resources shared by the rest of the system. Note that exception callbacks with USB_CB_ASYNC_REQ_FAILED set may execute out of order from the requests which preceded them. Normal callbacks may be already queued when an exception hits that the USBA is unable to queue.
A queued request was submitted to the host controller driver and was rejected. The usb_completion_reason shows why the request was rejected by the host controller.
Insufficient resources were available for recovery to proceed.
Callback is executing in interrupt context and should not block.
The usb_cb_flags_t enumerated type defines a bitmask. Multiple bits can be set, reporting back multiple statuses to the exception callback handler.
The USBA framework supports callback handling as a way of asynchronous client driver notification. There are three kinds of callbacks: Normal completion transfer callback, exception (error) completion transfer callback, and command completion callback, each described below.
Callback handlers are called whenever they are specified in a request or command, regardless of whether or not that request or command specifies the USB_FLAGS_SLEEP flag. (USB_FLAGS_SLEEP tells the request or command to block until completed.) Callback handlers must be specified whenever an asynchronous transfer is requested.
Each pipe is associated with a pool of threads that are used to run callbacks associated with requests on that pipe. All transfer completion callbacks for a particular pipe are run serially by a single thread.
Pipes taking requests with callbacks which can block must have their pipe policy properly initialized. If a callback blocks on a condition that is only met by another thread associated with the same pipe, there must be sufficient threads available. Otherwise that callback thread will block forever. Similarly, problems will ensue when callbacks overlap and there are not enough threads to handle the number of overlapping callbacks.
The pp_max_async_reqs field of the pipe_policy provides a hint of how many threads to allocate for asynchronous processing of request callbacks on a pipe. Set this value high enough per pipe to accommodate all of the pipe's possible asynchronous conditions. The pipe_policy is passed to usb_pipe_open(9F).
Transfer completion callbacks (normal completion and exception):
Most transfer completion callbacks are allowed to block, but only under certain conditions:
1. No callback is allowed to block if the callback flags show USB_CB_INTR_CONTEXT set, since that flag indicates that the callback is running in interrupt context instead of kernel context. Isochronous normal completion callbacks, plus those with USB_CB_ASYNC_REQ_FAILED set, execute in interrupt context.
2. Any callback except for isochronous normal completion can block for resources (for example to allocate memory).
3. No callback can block for synchronous completion of a command (for example, a call to usb_pipe_close(9F) with the USB_FLAGS_SLEEP flag passed) done on the same pipe. The command could wait for all callbacks to complete, including the callback which issued that command, causing all operations on the pipe to deadlock. Note that asynchronous commands can start from a callback, providing that the pipe's policy pp_max_async_reqs field is initialized to accommodate them.
4. Avoid callbacks that block for synchronous completion of commands done on other pipes. Such conditions can cause complex dependencies and unpredictable results.
5. No callback can block waiting for a synchronous transfer request to complete. (Note that making an asynchronous request to start a new transfer or start polling does not block, and is OK.)
6. No callback can block waiting for another callback to complete. (This is because all callbacks are done by a single thread.)
7. Note that if a callback blocks, other callbacks awaiting processing can backup behind it, impacting system resources.
A transfer request can specify a non-null normal-completion callback. Such requests conclude by calling the normal-completion callback when the transfer completes normally. Similarly, a transfer request can specify a non-null exception callback. Such requests conclude by calling the exception callback when the transfer completes abnormally. Note that the same callback can be used for both normal completion and exception callback handling. A completion reason of USB_CR_OK defines normal completion.
All request-callbacks take as arguments a usb_pipe_handle_t and a pointer to the request:
xxxx_cb(usb_pipe_handle_t ph, struct usb_ctrl_req *req);
Such callbacks can retrieve saved state or other information from the private area of the pipe handle. (See usb_pipe_set_private(9F).) Handlers also have access to the completion reason (usb_cr_t) and callback flags (usb_cb_flags_t) through the request argument they are passed.
Request information follows. In the data below, xxxx below represents the type of request (ctrl, intr, isoc or bulk.)
Request structure name is usb_xxxx_req_t. Normal completion callback handler field is xxxx_cb. Exception callback handler field is xxxx_exc_cb. Completion reason field is xxxx_completion_reason. Callback flags field is xxxx_cb_flags.
Calls to some non-transfer functions can be set up for callback notification. These include usb_pipe_close(9F), usb_pipe_reset(9F), usb_pipe_drain_reqs(9F), usb_set_cfg(9F), usb_set_alt_if(9F) and usb_clr_feature(9F).
The signature of a command completion callback is as follows:
command_cb( usb_pipe_handle_t cb_pipe_handle, usb_opaque_t arg, int rval, usb_cb_flags_t flags);
As with transfer completion callbacks, command completion callbacks take a usb_pipe_handle_t to retrieve saved state or other information from the pipe's private area. Also, command completion callbacks are provided with an additional user-definable argument (usb_opaque_t arg), the return status of the executed command (int rval), and the callback flags (usb_cb_flags_t flags).
The rval argument is roughly equivalent to the completion reason of a transfer callback, indicating the overall status. See the return values of the relevant function for possible rval values which can be passed to the callback.
The callback flags can be checked when rval indicates failure status. Just as for transfer completion callbacks, callback flags return additional information on execution events.
See attributes(5) for descriptions of the following attributes:
| ATTRIBUTE TYPE ATTRIBUTE VALUE |
| Architecture PCI-based systems |
| Interface stability Committed |
usb_alloc_request(9F), usb_pipe_bulk_xfer(9F), usb_pipe_ctrl_xfer(9F), usb_pipe_intr_xfer(9F), usb_pipe_isoc_xfer(9F), usb_bulk_request(9S), usb_ctrl_request(9S), usb_intr_request(9S), usb_isoc_request(9S)