2000-10-28 06:59:48 +00:00
|
|
|
/*-
|
Rewrite the EC driver event model. The main goal is to avoid
polling/interrupt-driven fallback and instead use polling only during
boot and pure interrupt-driven mode after boot. Polled mode could be
relegated completely to a legacy role if we could enable interrupts
during boot. Polled mode can be forced after boot by setting
debug.acpi.ec.polled="1", i.e. if there are timeouts.
- Use polling only during boot, shutdown, or if requested by the user.
Otherwise, use a generation count of GPEs, incremented atomically. This
prevents an old status value from being used if the EC is really slow
and the same condition (i.e. multiple IBEs for a write transaction) is
being checked.
- Check for and run the query handler directly if the SCI bit is set in
the status register during boot. Previously, the query handler wouldn't
run until interrupts were finally enabled late in boot.
- During boot and after starting a command, check if the event appears
to already have occurred before we even start waiting. If so, it's
possible the EC is very slow and we might accept an old status value.
Print a warning in this case. Once we've booted, interrupt-driven mode
should work just fine but polled mode could be unreliable. There's not
much more we can do about this until interrupts are enabled during boot.
- In the above case, we also do one final check if the interrupt-driven
mode gets a timeout. If the status is complete, it will force the
system back into polled mode since interrupt mode doesn't work. For
polled mode during boot, if the status appears to be already complete
before beginning the check loop, it waits 10 us before actually checking
the status, just in case the EC is really slow and hasn't gotten to work
on the new request yet.
- Use upper-case hex for the _Qxx method
- Use device_printf for errors, don't hide them under verbose
- Increase default total timeout to 750 ms and decrease polling interval
to 5 us.
- Don't pass the status value via the softc. Just read it directly.
- Remove the mutex. We use the sx lock for transaction serialization
with the query handler.
- Remove the Intel copyright notice as no code of theirs was ever
present in this file (verified against rev 1.1)
- Allow KTR module-only builds for ease of testing
Thanks to jkim and Alexey Starikovskiy for helpful discussions and testing.
Approved by: re
MFC after: 2 weeks
2007-09-24 16:59:06 +00:00
|
|
|
* Copyright (c) 2003-2007 Nate Lawson
|
2000-10-28 06:59:48 +00:00
|
|
|
* Copyright (c) 2000 Michael Smith
|
|
|
|
* Copyright (c) 2000 BSDi
|
|
|
|
* All rights reserved.
|
|
|
|
*
|
|
|
|
* Redistribution and use in source and binary forms, with or without
|
|
|
|
* modification, are permitted provided that the following conditions
|
|
|
|
* are met:
|
|
|
|
* 1. Redistributions of source code must retain the above copyright
|
|
|
|
* notice, this list of conditions and the following disclaimer.
|
|
|
|
* 2. Redistributions in binary form must reproduce the above copyright
|
|
|
|
* notice, this list of conditions and the following disclaimer in the
|
|
|
|
* documentation and/or other materials provided with the distribution.
|
|
|
|
*
|
|
|
|
* THIS SOFTWARE IS PROVIDED BY THE AUTHOR AND CONTRIBUTORS ``AS IS'' AND
|
|
|
|
* ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
|
|
|
|
* IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
|
|
|
|
* ARE DISCLAIMED. IN NO EVENT SHALL THE AUTHOR OR CONTRIBUTORS BE LIABLE
|
|
|
|
* FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
|
|
|
|
* DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
|
|
|
|
* OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
|
|
|
|
* HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
|
|
|
|
* LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
|
|
|
|
* OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
|
|
|
|
* SUCH DAMAGE.
|
|
|
|
*/
|
|
|
|
|
2003-08-24 17:55:58 +00:00
|
|
|
#include <sys/cdefs.h>
|
|
|
|
__FBSDID("$FreeBSD$");
|
|
|
|
|
2000-10-28 06:59:48 +00:00
|
|
|
#include "opt_acpi.h"
|
|
|
|
#include <sys/param.h>
|
|
|
|
#include <sys/kernel.h>
|
|
|
|
#include <sys/bus.h>
|
2007-03-31 23:23:42 +00:00
|
|
|
#include <sys/lock.h>
|
2004-05-14 04:17:56 +00:00
|
|
|
#include <sys/malloc.h>
|
2004-07-01 00:51:31 +00:00
|
|
|
#include <sys/module.h>
|
|
|
|
#include <sys/sx.h>
|
2000-10-28 06:59:48 +00:00
|
|
|
|
|
|
|
#include <machine/bus.h>
|
|
|
|
#include <machine/resource.h>
|
|
|
|
#include <sys/rman.h>
|
|
|
|
|
2005-09-11 18:39:03 +00:00
|
|
|
#include <contrib/dev/acpica/acpi.h>
|
2000-10-28 06:59:48 +00:00
|
|
|
#include <dev/acpica/acpivar.h>
|
|
|
|
|
2004-08-13 06:21:53 +00:00
|
|
|
/* Hooks for the ACPI CA debugging infrastructure */
|
2001-06-29 20:31:37 +00:00
|
|
|
#define _COMPONENT ACPI_EC
|
2002-02-23 05:26:31 +00:00
|
|
|
ACPI_MODULE_NAME("EC")
|
2000-12-08 09:16:20 +00:00
|
|
|
|
2001-07-21 04:06:44 +00:00
|
|
|
/*
|
|
|
|
* EC_COMMAND:
|
|
|
|
* -----------
|
|
|
|
*/
|
|
|
|
typedef UINT8 EC_COMMAND;
|
|
|
|
|
|
|
|
#define EC_COMMAND_UNKNOWN ((EC_COMMAND) 0x00)
|
|
|
|
#define EC_COMMAND_READ ((EC_COMMAND) 0x80)
|
|
|
|
#define EC_COMMAND_WRITE ((EC_COMMAND) 0x81)
|
|
|
|
#define EC_COMMAND_BURST_ENABLE ((EC_COMMAND) 0x82)
|
|
|
|
#define EC_COMMAND_BURST_DISABLE ((EC_COMMAND) 0x83)
|
|
|
|
#define EC_COMMAND_QUERY ((EC_COMMAND) 0x84)
|
|
|
|
|
Rewrite the EC driver event model. The main goal is to avoid
polling/interrupt-driven fallback and instead use polling only during
boot and pure interrupt-driven mode after boot. Polled mode could be
relegated completely to a legacy role if we could enable interrupts
during boot. Polled mode can be forced after boot by setting
debug.acpi.ec.polled="1", i.e. if there are timeouts.
- Use polling only during boot, shutdown, or if requested by the user.
Otherwise, use a generation count of GPEs, incremented atomically. This
prevents an old status value from being used if the EC is really slow
and the same condition (i.e. multiple IBEs for a write transaction) is
being checked.
- Check for and run the query handler directly if the SCI bit is set in
the status register during boot. Previously, the query handler wouldn't
run until interrupts were finally enabled late in boot.
- During boot and after starting a command, check if the event appears
to already have occurred before we even start waiting. If so, it's
possible the EC is very slow and we might accept an old status value.
Print a warning in this case. Once we've booted, interrupt-driven mode
should work just fine but polled mode could be unreliable. There's not
much more we can do about this until interrupts are enabled during boot.
- In the above case, we also do one final check if the interrupt-driven
mode gets a timeout. If the status is complete, it will force the
system back into polled mode since interrupt mode doesn't work. For
polled mode during boot, if the status appears to be already complete
before beginning the check loop, it waits 10 us before actually checking
the status, just in case the EC is really slow and hasn't gotten to work
on the new request yet.
- Use upper-case hex for the _Qxx method
- Use device_printf for errors, don't hide them under verbose
- Increase default total timeout to 750 ms and decrease polling interval
to 5 us.
- Don't pass the status value via the softc. Just read it directly.
- Remove the mutex. We use the sx lock for transaction serialization
with the query handler.
- Remove the Intel copyright notice as no code of theirs was ever
present in this file (verified against rev 1.1)
- Allow KTR module-only builds for ease of testing
Thanks to jkim and Alexey Starikovskiy for helpful discussions and testing.
Approved by: re
MFC after: 2 weeks
2007-09-24 16:59:06 +00:00
|
|
|
/*
|
2001-07-21 04:06:44 +00:00
|
|
|
* EC_STATUS:
|
|
|
|
* ----------
|
|
|
|
* The encoding of the EC status register is illustrated below.
|
|
|
|
* Note that a set bit (1) indicates the property is TRUE
|
|
|
|
* (e.g. if bit 0 is set then the output buffer is full).
|
|
|
|
* +-+-+-+-+-+-+-+-+
|
2004-12-27 05:36:47 +00:00
|
|
|
* |7|6|5|4|3|2|1|0|
|
2001-07-21 04:06:44 +00:00
|
|
|
* +-+-+-+-+-+-+-+-+
|
|
|
|
* | | | | | | | |
|
|
|
|
* | | | | | | | +- Output Buffer Full?
|
|
|
|
* | | | | | | +--- Input Buffer Full?
|
|
|
|
* | | | | | +----- <reserved>
|
|
|
|
* | | | | +------- Data Register is Command Byte?
|
|
|
|
* | | | +--------- Burst Mode Enabled?
|
|
|
|
* | | +----------- SCI Event?
|
|
|
|
* | +------------- SMI Event?
|
2007-02-27 00:14:20 +00:00
|
|
|
* +--------------- <reserved>
|
2001-07-21 04:06:44 +00:00
|
|
|
*
|
|
|
|
*/
|
|
|
|
typedef UINT8 EC_STATUS;
|
|
|
|
|
|
|
|
#define EC_FLAG_OUTPUT_BUFFER ((EC_STATUS) 0x01)
|
|
|
|
#define EC_FLAG_INPUT_BUFFER ((EC_STATUS) 0x02)
|
2007-02-27 00:14:20 +00:00
|
|
|
#define EC_FLAG_DATA_IS_CMD ((EC_STATUS) 0x08)
|
2001-07-21 04:06:44 +00:00
|
|
|
#define EC_FLAG_BURST_MODE ((EC_STATUS) 0x10)
|
|
|
|
|
|
|
|
/*
|
|
|
|
* EC_EVENT:
|
|
|
|
* ---------
|
|
|
|
*/
|
|
|
|
typedef UINT8 EC_EVENT;
|
|
|
|
|
|
|
|
#define EC_EVENT_UNKNOWN ((EC_EVENT) 0x00)
|
|
|
|
#define EC_EVENT_OUTPUT_BUFFER_FULL ((EC_EVENT) 0x01)
|
|
|
|
#define EC_EVENT_INPUT_BUFFER_EMPTY ((EC_EVENT) 0x02)
|
|
|
|
#define EC_EVENT_SCI ((EC_EVENT) 0x20)
|
2007-02-27 00:14:20 +00:00
|
|
|
#define EC_EVENT_SMI ((EC_EVENT) 0x40)
|
|
|
|
|
|
|
|
/* Data byte returned after burst enable indicating it was successful. */
|
|
|
|
#define EC_BURST_ACK 0x90
|
2001-07-21 04:06:44 +00:00
|
|
|
|
|
|
|
/*
|
|
|
|
* Register access primitives
|
|
|
|
*/
|
|
|
|
#define EC_GET_DATA(sc) \
|
|
|
|
bus_space_read_1((sc)->ec_data_tag, (sc)->ec_data_handle, 0)
|
|
|
|
|
|
|
|
#define EC_SET_DATA(sc, v) \
|
|
|
|
bus_space_write_1((sc)->ec_data_tag, (sc)->ec_data_handle, 0, (v))
|
|
|
|
|
|
|
|
#define EC_GET_CSR(sc) \
|
|
|
|
bus_space_read_1((sc)->ec_csr_tag, (sc)->ec_csr_handle, 0)
|
|
|
|
|
|
|
|
#define EC_SET_CSR(sc, v) \
|
|
|
|
bus_space_write_1((sc)->ec_csr_tag, (sc)->ec_csr_handle, 0, (v))
|
|
|
|
|
2004-05-14 04:17:56 +00:00
|
|
|
/* Additional params to pass from the probe routine */
|
|
|
|
struct acpi_ec_params {
|
|
|
|
int glk;
|
|
|
|
int gpe_bit;
|
|
|
|
ACPI_HANDLE gpe_handle;
|
|
|
|
int uid;
|
|
|
|
};
|
2003-07-20 00:48:38 +00:00
|
|
|
|
2004-05-14 04:17:56 +00:00
|
|
|
/* Indicate that this device has already been probed via ECDT. */
|
2007-06-15 18:02:34 +00:00
|
|
|
#define DEV_ECDT(x) (acpi_get_magic(x) == (uintptr_t)&acpi_ec_devclass)
|
2003-07-20 00:48:38 +00:00
|
|
|
|
2001-07-21 04:06:44 +00:00
|
|
|
/*
|
|
|
|
* Driver softc.
|
|
|
|
*/
|
2000-10-28 06:59:48 +00:00
|
|
|
struct acpi_ec_softc {
|
|
|
|
device_t ec_dev;
|
|
|
|
ACPI_HANDLE ec_handle;
|
2004-05-14 04:17:56 +00:00
|
|
|
int ec_uid;
|
|
|
|
ACPI_HANDLE ec_gpehandle;
|
2003-07-20 21:11:32 +00:00
|
|
|
UINT8 ec_gpebit;
|
Rewrite the EC driver event model. The main goal is to avoid
polling/interrupt-driven fallback and instead use polling only during
boot and pure interrupt-driven mode after boot. Polled mode could be
relegated completely to a legacy role if we could enable interrupts
during boot. Polled mode can be forced after boot by setting
debug.acpi.ec.polled="1", i.e. if there are timeouts.
- Use polling only during boot, shutdown, or if requested by the user.
Otherwise, use a generation count of GPEs, incremented atomically. This
prevents an old status value from being used if the EC is really slow
and the same condition (i.e. multiple IBEs for a write transaction) is
being checked.
- Check for and run the query handler directly if the SCI bit is set in
the status register during boot. Previously, the query handler wouldn't
run until interrupts were finally enabled late in boot.
- During boot and after starting a command, check if the event appears
to already have occurred before we even start waiting. If so, it's
possible the EC is very slow and we might accept an old status value.
Print a warning in this case. Once we've booted, interrupt-driven mode
should work just fine but polled mode could be unreliable. There's not
much more we can do about this until interrupts are enabled during boot.
- In the above case, we also do one final check if the interrupt-driven
mode gets a timeout. If the status is complete, it will force the
system back into polled mode since interrupt mode doesn't work. For
polled mode during boot, if the status appears to be already complete
before beginning the check loop, it waits 10 us before actually checking
the status, just in case the EC is really slow and hasn't gotten to work
on the new request yet.
- Use upper-case hex for the _Qxx method
- Use device_printf for errors, don't hide them under verbose
- Increase default total timeout to 750 ms and decrease polling interval
to 5 us.
- Don't pass the status value via the softc. Just read it directly.
- Remove the mutex. We use the sx lock for transaction serialization
with the query handler.
- Remove the Intel copyright notice as no code of theirs was ever
present in this file (verified against rev 1.1)
- Allow KTR module-only builds for ease of testing
Thanks to jkim and Alexey Starikovskiy for helpful discussions and testing.
Approved by: re
MFC after: 2 weeks
2007-09-24 16:59:06 +00:00
|
|
|
|
2000-10-28 06:59:48 +00:00
|
|
|
int ec_data_rid;
|
|
|
|
struct resource *ec_data_res;
|
|
|
|
bus_space_tag_t ec_data_tag;
|
|
|
|
bus_space_handle_t ec_data_handle;
|
|
|
|
|
|
|
|
int ec_csr_rid;
|
|
|
|
struct resource *ec_csr_res;
|
|
|
|
bus_space_tag_t ec_csr_tag;
|
|
|
|
bus_space_handle_t ec_csr_handle;
|
|
|
|
|
2003-07-10 17:22:46 +00:00
|
|
|
int ec_glk;
|
|
|
|
int ec_glkhandle;
|
2007-02-27 00:14:20 +00:00
|
|
|
int ec_burstactive;
|
|
|
|
int ec_sci_pend;
|
Rewrite the EC driver event model. The main goal is to avoid
polling/interrupt-driven fallback and instead use polling only during
boot and pure interrupt-driven mode after boot. Polled mode could be
relegated completely to a legacy role if we could enable interrupts
during boot. Polled mode can be forced after boot by setting
debug.acpi.ec.polled="1", i.e. if there are timeouts.
- Use polling only during boot, shutdown, or if requested by the user.
Otherwise, use a generation count of GPEs, incremented atomically. This
prevents an old status value from being used if the EC is really slow
and the same condition (i.e. multiple IBEs for a write transaction) is
being checked.
- Check for and run the query handler directly if the SCI bit is set in
the status register during boot. Previously, the query handler wouldn't
run until interrupts were finally enabled late in boot.
- During boot and after starting a command, check if the event appears
to already have occurred before we even start waiting. If so, it's
possible the EC is very slow and we might accept an old status value.
Print a warning in this case. Once we've booted, interrupt-driven mode
should work just fine but polled mode could be unreliable. There's not
much more we can do about this until interrupts are enabled during boot.
- In the above case, we also do one final check if the interrupt-driven
mode gets a timeout. If the status is complete, it will force the
system back into polled mode since interrupt mode doesn't work. For
polled mode during boot, if the status appears to be already complete
before beginning the check loop, it waits 10 us before actually checking
the status, just in case the EC is really slow and hasn't gotten to work
on the new request yet.
- Use upper-case hex for the _Qxx method
- Use device_printf for errors, don't hide them under verbose
- Increase default total timeout to 750 ms and decrease polling interval
to 5 us.
- Don't pass the status value via the softc. Just read it directly.
- Remove the mutex. We use the sx lock for transaction serialization
with the query handler.
- Remove the Intel copyright notice as no code of theirs was ever
present in this file (verified against rev 1.1)
- Allow KTR module-only builds for ease of testing
Thanks to jkim and Alexey Starikovskiy for helpful discussions and testing.
Approved by: re
MFC after: 2 weeks
2007-09-24 16:59:06 +00:00
|
|
|
u_int ec_gencount;
|
2007-10-25 16:49:07 +00:00
|
|
|
int ec_suspending;
|
2000-10-28 06:59:48 +00:00
|
|
|
};
|
|
|
|
|
2003-07-10 17:22:46 +00:00
|
|
|
/*
|
2004-08-13 06:21:53 +00:00
|
|
|
* XXX njl
|
2003-07-10 17:22:46 +00:00
|
|
|
* I couldn't find it in the spec but other implementations also use a
|
|
|
|
* value of 1 ms for the time to acquire global lock.
|
|
|
|
*/
|
|
|
|
#define EC_LOCK_TIMEOUT 1000
|
|
|
|
|
2007-02-27 00:14:20 +00:00
|
|
|
/* Default delay in microseconds between each run of the status polling loop. */
|
Rewrite the EC driver event model. The main goal is to avoid
polling/interrupt-driven fallback and instead use polling only during
boot and pure interrupt-driven mode after boot. Polled mode could be
relegated completely to a legacy role if we could enable interrupts
during boot. Polled mode can be forced after boot by setting
debug.acpi.ec.polled="1", i.e. if there are timeouts.
- Use polling only during boot, shutdown, or if requested by the user.
Otherwise, use a generation count of GPEs, incremented atomically. This
prevents an old status value from being used if the EC is really slow
and the same condition (i.e. multiple IBEs for a write transaction) is
being checked.
- Check for and run the query handler directly if the SCI bit is set in
the status register during boot. Previously, the query handler wouldn't
run until interrupts were finally enabled late in boot.
- During boot and after starting a command, check if the event appears
to already have occurred before we even start waiting. If so, it's
possible the EC is very slow and we might accept an old status value.
Print a warning in this case. Once we've booted, interrupt-driven mode
should work just fine but polled mode could be unreliable. There's not
much more we can do about this until interrupts are enabled during boot.
- In the above case, we also do one final check if the interrupt-driven
mode gets a timeout. If the status is complete, it will force the
system back into polled mode since interrupt mode doesn't work. For
polled mode during boot, if the status appears to be already complete
before beginning the check loop, it waits 10 us before actually checking
the status, just in case the EC is really slow and hasn't gotten to work
on the new request yet.
- Use upper-case hex for the _Qxx method
- Use device_printf for errors, don't hide them under verbose
- Increase default total timeout to 750 ms and decrease polling interval
to 5 us.
- Don't pass the status value via the softc. Just read it directly.
- Remove the mutex. We use the sx lock for transaction serialization
with the query handler.
- Remove the Intel copyright notice as no code of theirs was ever
present in this file (verified against rev 1.1)
- Allow KTR module-only builds for ease of testing
Thanks to jkim and Alexey Starikovskiy for helpful discussions and testing.
Approved by: re
MFC after: 2 weeks
2007-09-24 16:59:06 +00:00
|
|
|
#define EC_POLL_DELAY 5
|
2007-02-27 00:14:20 +00:00
|
|
|
|
|
|
|
/* Total time in ms spent waiting for a response from EC. */
|
Rewrite the EC driver event model. The main goal is to avoid
polling/interrupt-driven fallback and instead use polling only during
boot and pure interrupt-driven mode after boot. Polled mode could be
relegated completely to a legacy role if we could enable interrupts
during boot. Polled mode can be forced after boot by setting
debug.acpi.ec.polled="1", i.e. if there are timeouts.
- Use polling only during boot, shutdown, or if requested by the user.
Otherwise, use a generation count of GPEs, incremented atomically. This
prevents an old status value from being used if the EC is really slow
and the same condition (i.e. multiple IBEs for a write transaction) is
being checked.
- Check for and run the query handler directly if the SCI bit is set in
the status register during boot. Previously, the query handler wouldn't
run until interrupts were finally enabled late in boot.
- During boot and after starting a command, check if the event appears
to already have occurred before we even start waiting. If so, it's
possible the EC is very slow and we might accept an old status value.
Print a warning in this case. Once we've booted, interrupt-driven mode
should work just fine but polled mode could be unreliable. There's not
much more we can do about this until interrupts are enabled during boot.
- In the above case, we also do one final check if the interrupt-driven
mode gets a timeout. If the status is complete, it will force the
system back into polled mode since interrupt mode doesn't work. For
polled mode during boot, if the status appears to be already complete
before beginning the check loop, it waits 10 us before actually checking
the status, just in case the EC is really slow and hasn't gotten to work
on the new request yet.
- Use upper-case hex for the _Qxx method
- Use device_printf for errors, don't hide them under verbose
- Increase default total timeout to 750 ms and decrease polling interval
to 5 us.
- Don't pass the status value via the softc. Just read it directly.
- Remove the mutex. We use the sx lock for transaction serialization
with the query handler.
- Remove the Intel copyright notice as no code of theirs was ever
present in this file (verified against rev 1.1)
- Allow KTR module-only builds for ease of testing
Thanks to jkim and Alexey Starikovskiy for helpful discussions and testing.
Approved by: re
MFC after: 2 weeks
2007-09-24 16:59:06 +00:00
|
|
|
#define EC_TIMEOUT 750
|
2000-10-28 06:59:48 +00:00
|
|
|
|
2003-07-15 19:24:36 +00:00
|
|
|
#define EVENT_READY(event, status) \
|
|
|
|
(((event) == EC_EVENT_OUTPUT_BUFFER_FULL && \
|
|
|
|
((status) & EC_FLAG_OUTPUT_BUFFER) != 0) || \
|
|
|
|
((event) == EC_EVENT_INPUT_BUFFER_EMPTY && \
|
|
|
|
((status) & EC_FLAG_INPUT_BUFFER) == 0))
|
|
|
|
|
2004-08-13 06:21:53 +00:00
|
|
|
ACPI_SERIAL_DECL(ec, "ACPI embedded controller");
|
|
|
|
|
2007-02-27 00:14:20 +00:00
|
|
|
SYSCTL_DECL(_debug_acpi);
|
|
|
|
SYSCTL_NODE(_debug_acpi, OID_AUTO, ec, CTLFLAG_RD, NULL, "EC debugging");
|
|
|
|
|
2007-03-18 01:03:03 +00:00
|
|
|
static int ec_burst_mode;
|
2007-02-27 00:14:20 +00:00
|
|
|
TUNABLE_INT("debug.acpi.ec.burst", &ec_burst_mode);
|
2007-03-18 01:03:03 +00:00
|
|
|
SYSCTL_INT(_debug_acpi_ec, OID_AUTO, burst, CTLFLAG_RW, &ec_burst_mode, 0,
|
2007-02-27 00:14:20 +00:00
|
|
|
"Enable use of burst mode (faster for nearly all systems)");
|
Rewrite the EC driver event model. The main goal is to avoid
polling/interrupt-driven fallback and instead use polling only during
boot and pure interrupt-driven mode after boot. Polled mode could be
relegated completely to a legacy role if we could enable interrupts
during boot. Polled mode can be forced after boot by setting
debug.acpi.ec.polled="1", i.e. if there are timeouts.
- Use polling only during boot, shutdown, or if requested by the user.
Otherwise, use a generation count of GPEs, incremented atomically. This
prevents an old status value from being used if the EC is really slow
and the same condition (i.e. multiple IBEs for a write transaction) is
being checked.
- Check for and run the query handler directly if the SCI bit is set in
the status register during boot. Previously, the query handler wouldn't
run until interrupts were finally enabled late in boot.
- During boot and after starting a command, check if the event appears
to already have occurred before we even start waiting. If so, it's
possible the EC is very slow and we might accept an old status value.
Print a warning in this case. Once we've booted, interrupt-driven mode
should work just fine but polled mode could be unreliable. There's not
much more we can do about this until interrupts are enabled during boot.
- In the above case, we also do one final check if the interrupt-driven
mode gets a timeout. If the status is complete, it will force the
system back into polled mode since interrupt mode doesn't work. For
polled mode during boot, if the status appears to be already complete
before beginning the check loop, it waits 10 us before actually checking
the status, just in case the EC is really slow and hasn't gotten to work
on the new request yet.
- Use upper-case hex for the _Qxx method
- Use device_printf for errors, don't hide them under verbose
- Increase default total timeout to 750 ms and decrease polling interval
to 5 us.
- Don't pass the status value via the softc. Just read it directly.
- Remove the mutex. We use the sx lock for transaction serialization
with the query handler.
- Remove the Intel copyright notice as no code of theirs was ever
present in this file (verified against rev 1.1)
- Allow KTR module-only builds for ease of testing
Thanks to jkim and Alexey Starikovskiy for helpful discussions and testing.
Approved by: re
MFC after: 2 weeks
2007-09-24 16:59:06 +00:00
|
|
|
static int ec_polled_mode;
|
|
|
|
TUNABLE_INT("debug.acpi.ec.polled", &ec_polled_mode);
|
|
|
|
SYSCTL_INT(_debug_acpi_ec, OID_AUTO, polled, CTLFLAG_RW, &ec_polled_mode, 0,
|
|
|
|
"Force use of polled mode (only if interrupt mode doesn't work)");
|
2007-02-27 00:14:20 +00:00
|
|
|
static int ec_timeout = EC_TIMEOUT;
|
|
|
|
TUNABLE_INT("debug.acpi.ec.timeout", &ec_timeout);
|
|
|
|
SYSCTL_INT(_debug_acpi_ec, OID_AUTO, timeout, CTLFLAG_RW, &ec_timeout,
|
|
|
|
EC_TIMEOUT, "Total time spent waiting for a response (poll+sleep)");
|
|
|
|
|
Rewrite the EC driver event model. The main goal is to avoid
polling/interrupt-driven fallback and instead use polling only during
boot and pure interrupt-driven mode after boot. Polled mode could be
relegated completely to a legacy role if we could enable interrupts
during boot. Polled mode can be forced after boot by setting
debug.acpi.ec.polled="1", i.e. if there are timeouts.
- Use polling only during boot, shutdown, or if requested by the user.
Otherwise, use a generation count of GPEs, incremented atomically. This
prevents an old status value from being used if the EC is really slow
and the same condition (i.e. multiple IBEs for a write transaction) is
being checked.
- Check for and run the query handler directly if the SCI bit is set in
the status register during boot. Previously, the query handler wouldn't
run until interrupts were finally enabled late in boot.
- During boot and after starting a command, check if the event appears
to already have occurred before we even start waiting. If so, it's
possible the EC is very slow and we might accept an old status value.
Print a warning in this case. Once we've booted, interrupt-driven mode
should work just fine but polled mode could be unreliable. There's not
much more we can do about this until interrupts are enabled during boot.
- In the above case, we also do one final check if the interrupt-driven
mode gets a timeout. If the status is complete, it will force the
system back into polled mode since interrupt mode doesn't work. For
polled mode during boot, if the status appears to be already complete
before beginning the check loop, it waits 10 us before actually checking
the status, just in case the EC is really slow and hasn't gotten to work
on the new request yet.
- Use upper-case hex for the _Qxx method
- Use device_printf for errors, don't hide them under verbose
- Increase default total timeout to 750 ms and decrease polling interval
to 5 us.
- Don't pass the status value via the softc. Just read it directly.
- Remove the mutex. We use the sx lock for transaction serialization
with the query handler.
- Remove the Intel copyright notice as no code of theirs was ever
present in this file (verified against rev 1.1)
- Allow KTR module-only builds for ease of testing
Thanks to jkim and Alexey Starikovskiy for helpful discussions and testing.
Approved by: re
MFC after: 2 weeks
2007-09-24 16:59:06 +00:00
|
|
|
static ACPI_STATUS
|
|
|
|
EcLock(struct acpi_ec_softc *sc)
|
2000-10-28 06:59:48 +00:00
|
|
|
{
|
2004-08-13 06:21:53 +00:00
|
|
|
ACPI_STATUS status;
|
2000-10-28 06:59:48 +00:00
|
|
|
|
2007-06-02 19:21:40 +00:00
|
|
|
/* If _GLK is non-zero, acquire the global lock. */
|
|
|
|
status = AE_OK;
|
|
|
|
if (sc->ec_glk) {
|
|
|
|
status = AcpiAcquireGlobalLock(EC_LOCK_TIMEOUT, &sc->ec_glkhandle);
|
|
|
|
if (ACPI_FAILURE(status))
|
|
|
|
return (status);
|
|
|
|
}
|
Rewrite the EC driver event model. The main goal is to avoid
polling/interrupt-driven fallback and instead use polling only during
boot and pure interrupt-driven mode after boot. Polled mode could be
relegated completely to a legacy role if we could enable interrupts
during boot. Polled mode can be forced after boot by setting
debug.acpi.ec.polled="1", i.e. if there are timeouts.
- Use polling only during boot, shutdown, or if requested by the user.
Otherwise, use a generation count of GPEs, incremented atomically. This
prevents an old status value from being used if the EC is really slow
and the same condition (i.e. multiple IBEs for a write transaction) is
being checked.
- Check for and run the query handler directly if the SCI bit is set in
the status register during boot. Previously, the query handler wouldn't
run until interrupts were finally enabled late in boot.
- During boot and after starting a command, check if the event appears
to already have occurred before we even start waiting. If so, it's
possible the EC is very slow and we might accept an old status value.
Print a warning in this case. Once we've booted, interrupt-driven mode
should work just fine but polled mode could be unreliable. There's not
much more we can do about this until interrupts are enabled during boot.
- In the above case, we also do one final check if the interrupt-driven
mode gets a timeout. If the status is complete, it will force the
system back into polled mode since interrupt mode doesn't work. For
polled mode during boot, if the status appears to be already complete
before beginning the check loop, it waits 10 us before actually checking
the status, just in case the EC is really slow and hasn't gotten to work
on the new request yet.
- Use upper-case hex for the _Qxx method
- Use device_printf for errors, don't hide them under verbose
- Increase default total timeout to 750 ms and decrease polling interval
to 5 us.
- Don't pass the status value via the softc. Just read it directly.
- Remove the mutex. We use the sx lock for transaction serialization
with the query handler.
- Remove the Intel copyright notice as no code of theirs was ever
present in this file (verified against rev 1.1)
- Allow KTR module-only builds for ease of testing
Thanks to jkim and Alexey Starikovskiy for helpful discussions and testing.
Approved by: re
MFC after: 2 weeks
2007-09-24 16:59:06 +00:00
|
|
|
ACPI_SERIAL_BEGIN(ec);
|
2003-06-27 21:57:42 +00:00
|
|
|
return (status);
|
2000-10-28 06:59:48 +00:00
|
|
|
}
|
|
|
|
|
Rewrite the EC driver event model. The main goal is to avoid
polling/interrupt-driven fallback and instead use polling only during
boot and pure interrupt-driven mode after boot. Polled mode could be
relegated completely to a legacy role if we could enable interrupts
during boot. Polled mode can be forced after boot by setting
debug.acpi.ec.polled="1", i.e. if there are timeouts.
- Use polling only during boot, shutdown, or if requested by the user.
Otherwise, use a generation count of GPEs, incremented atomically. This
prevents an old status value from being used if the EC is really slow
and the same condition (i.e. multiple IBEs for a write transaction) is
being checked.
- Check for and run the query handler directly if the SCI bit is set in
the status register during boot. Previously, the query handler wouldn't
run until interrupts were finally enabled late in boot.
- During boot and after starting a command, check if the event appears
to already have occurred before we even start waiting. If so, it's
possible the EC is very slow and we might accept an old status value.
Print a warning in this case. Once we've booted, interrupt-driven mode
should work just fine but polled mode could be unreliable. There's not
much more we can do about this until interrupts are enabled during boot.
- In the above case, we also do one final check if the interrupt-driven
mode gets a timeout. If the status is complete, it will force the
system back into polled mode since interrupt mode doesn't work. For
polled mode during boot, if the status appears to be already complete
before beginning the check loop, it waits 10 us before actually checking
the status, just in case the EC is really slow and hasn't gotten to work
on the new request yet.
- Use upper-case hex for the _Qxx method
- Use device_printf for errors, don't hide them under verbose
- Increase default total timeout to 750 ms and decrease polling interval
to 5 us.
- Don't pass the status value via the softc. Just read it directly.
- Remove the mutex. We use the sx lock for transaction serialization
with the query handler.
- Remove the Intel copyright notice as no code of theirs was ever
present in this file (verified against rev 1.1)
- Allow KTR module-only builds for ease of testing
Thanks to jkim and Alexey Starikovskiy for helpful discussions and testing.
Approved by: re
MFC after: 2 weeks
2007-09-24 16:59:06 +00:00
|
|
|
static void
|
2000-10-28 06:59:48 +00:00
|
|
|
EcUnlock(struct acpi_ec_softc *sc)
|
|
|
|
{
|
Rewrite the EC driver event model. The main goal is to avoid
polling/interrupt-driven fallback and instead use polling only during
boot and pure interrupt-driven mode after boot. Polled mode could be
relegated completely to a legacy role if we could enable interrupts
during boot. Polled mode can be forced after boot by setting
debug.acpi.ec.polled="1", i.e. if there are timeouts.
- Use polling only during boot, shutdown, or if requested by the user.
Otherwise, use a generation count of GPEs, incremented atomically. This
prevents an old status value from being used if the EC is really slow
and the same condition (i.e. multiple IBEs for a write transaction) is
being checked.
- Check for and run the query handler directly if the SCI bit is set in
the status register during boot. Previously, the query handler wouldn't
run until interrupts were finally enabled late in boot.
- During boot and after starting a command, check if the event appears
to already have occurred before we even start waiting. If so, it's
possible the EC is very slow and we might accept an old status value.
Print a warning in this case. Once we've booted, interrupt-driven mode
should work just fine but polled mode could be unreliable. There's not
much more we can do about this until interrupts are enabled during boot.
- In the above case, we also do one final check if the interrupt-driven
mode gets a timeout. If the status is complete, it will force the
system back into polled mode since interrupt mode doesn't work. For
polled mode during boot, if the status appears to be already complete
before beginning the check loop, it waits 10 us before actually checking
the status, just in case the EC is really slow and hasn't gotten to work
on the new request yet.
- Use upper-case hex for the _Qxx method
- Use device_printf for errors, don't hide them under verbose
- Increase default total timeout to 750 ms and decrease polling interval
to 5 us.
- Don't pass the status value via the softc. Just read it directly.
- Remove the mutex. We use the sx lock for transaction serialization
with the query handler.
- Remove the Intel copyright notice as no code of theirs was ever
present in this file (verified against rev 1.1)
- Allow KTR module-only builds for ease of testing
Thanks to jkim and Alexey Starikovskiy for helpful discussions and testing.
Approved by: re
MFC after: 2 weeks
2007-09-24 16:59:06 +00:00
|
|
|
ACPI_SERIAL_END(ec);
|
2007-06-02 19:21:40 +00:00
|
|
|
if (sc->ec_glk)
|
|
|
|
AcpiReleaseGlobalLock(sc->ec_glkhandle);
|
2000-10-28 06:59:48 +00:00
|
|
|
}
|
|
|
|
|
2004-05-25 02:47:35 +00:00
|
|
|
static uint32_t EcGpeHandler(void *Context);
|
Rewrite the EC driver event model. The main goal is to avoid
polling/interrupt-driven fallback and instead use polling only during
boot and pure interrupt-driven mode after boot. Polled mode could be
relegated completely to a legacy role if we could enable interrupts
during boot. Polled mode can be forced after boot by setting
debug.acpi.ec.polled="1", i.e. if there are timeouts.
- Use polling only during boot, shutdown, or if requested by the user.
Otherwise, use a generation count of GPEs, incremented atomically. This
prevents an old status value from being used if the EC is really slow
and the same condition (i.e. multiple IBEs for a write transaction) is
being checked.
- Check for and run the query handler directly if the SCI bit is set in
the status register during boot. Previously, the query handler wouldn't
run until interrupts were finally enabled late in boot.
- During boot and after starting a command, check if the event appears
to already have occurred before we even start waiting. If so, it's
possible the EC is very slow and we might accept an old status value.
Print a warning in this case. Once we've booted, interrupt-driven mode
should work just fine but polled mode could be unreliable. There's not
much more we can do about this until interrupts are enabled during boot.
- In the above case, we also do one final check if the interrupt-driven
mode gets a timeout. If the status is complete, it will force the
system back into polled mode since interrupt mode doesn't work. For
polled mode during boot, if the status appears to be already complete
before beginning the check loop, it waits 10 us before actually checking
the status, just in case the EC is really slow and hasn't gotten to work
on the new request yet.
- Use upper-case hex for the _Qxx method
- Use device_printf for errors, don't hide them under verbose
- Increase default total timeout to 750 ms and decrease polling interval
to 5 us.
- Don't pass the status value via the softc. Just read it directly.
- Remove the mutex. We use the sx lock for transaction serialization
with the query handler.
- Remove the Intel copyright notice as no code of theirs was ever
present in this file (verified against rev 1.1)
- Allow KTR module-only builds for ease of testing
Thanks to jkim and Alexey Starikovskiy for helpful discussions and testing.
Approved by: re
MFC after: 2 weeks
2007-09-24 16:59:06 +00:00
|
|
|
static ACPI_STATUS EcSpaceSetup(ACPI_HANDLE Region, UINT32 Function,
|
2003-06-27 21:57:42 +00:00
|
|
|
void *Context, void **return_Context);
|
|
|
|
static ACPI_STATUS EcSpaceHandler(UINT32 Function,
|
|
|
|
ACPI_PHYSICAL_ADDRESS Address,
|
|
|
|
UINT32 width, ACPI_INTEGER *Value,
|
|
|
|
void *Context, void *RegionContext);
|
Rewrite the EC driver event model. The main goal is to avoid
polling/interrupt-driven fallback and instead use polling only during
boot and pure interrupt-driven mode after boot. Polled mode could be
relegated completely to a legacy role if we could enable interrupts
during boot. Polled mode can be forced after boot by setting
debug.acpi.ec.polled="1", i.e. if there are timeouts.
- Use polling only during boot, shutdown, or if requested by the user.
Otherwise, use a generation count of GPEs, incremented atomically. This
prevents an old status value from being used if the EC is really slow
and the same condition (i.e. multiple IBEs for a write transaction) is
being checked.
- Check for and run the query handler directly if the SCI bit is set in
the status register during boot. Previously, the query handler wouldn't
run until interrupts were finally enabled late in boot.
- During boot and after starting a command, check if the event appears
to already have occurred before we even start waiting. If so, it's
possible the EC is very slow and we might accept an old status value.
Print a warning in this case. Once we've booted, interrupt-driven mode
should work just fine but polled mode could be unreliable. There's not
much more we can do about this until interrupts are enabled during boot.
- In the above case, we also do one final check if the interrupt-driven
mode gets a timeout. If the status is complete, it will force the
system back into polled mode since interrupt mode doesn't work. For
polled mode during boot, if the status appears to be already complete
before beginning the check loop, it waits 10 us before actually checking
the status, just in case the EC is really slow and hasn't gotten to work
on the new request yet.
- Use upper-case hex for the _Qxx method
- Use device_printf for errors, don't hide them under verbose
- Increase default total timeout to 750 ms and decrease polling interval
to 5 us.
- Don't pass the status value via the softc. Just read it directly.
- Remove the mutex. We use the sx lock for transaction serialization
with the query handler.
- Remove the Intel copyright notice as no code of theirs was ever
present in this file (verified against rev 1.1)
- Allow KTR module-only builds for ease of testing
Thanks to jkim and Alexey Starikovskiy for helpful discussions and testing.
Approved by: re
MFC after: 2 weeks
2007-09-24 16:59:06 +00:00
|
|
|
static ACPI_STATUS EcWaitEvent(struct acpi_ec_softc *sc, EC_EVENT Event,
|
|
|
|
u_int gen_count);
|
2003-07-10 17:22:46 +00:00
|
|
|
static ACPI_STATUS EcCommand(struct acpi_ec_softc *sc, EC_COMMAND cmd);
|
2003-06-27 21:57:42 +00:00
|
|
|
static ACPI_STATUS EcRead(struct acpi_ec_softc *sc, UINT8 Address,
|
|
|
|
UINT8 *Data);
|
|
|
|
static ACPI_STATUS EcWrite(struct acpi_ec_softc *sc, UINT8 Address,
|
|
|
|
UINT8 *Data);
|
2000-10-28 06:59:48 +00:00
|
|
|
static int acpi_ec_probe(device_t dev);
|
|
|
|
static int acpi_ec_attach(device_t dev);
|
2007-10-25 16:49:07 +00:00
|
|
|
static int acpi_ec_suspend(device_t dev);
|
|
|
|
static int acpi_ec_resume(device_t dev);
|
2004-06-30 16:00:20 +00:00
|
|
|
static int acpi_ec_shutdown(device_t dev);
|
2005-03-20 01:27:27 +00:00
|
|
|
static int acpi_ec_read_method(device_t dev, u_int addr,
|
|
|
|
ACPI_INTEGER *val, int width);
|
|
|
|
static int acpi_ec_write_method(device_t dev, u_int addr,
|
|
|
|
ACPI_INTEGER val, int width);
|
2000-10-28 06:59:48 +00:00
|
|
|
|
|
|
|
static device_method_t acpi_ec_methods[] = {
|
|
|
|
/* Device interface */
|
|
|
|
DEVMETHOD(device_probe, acpi_ec_probe),
|
|
|
|
DEVMETHOD(device_attach, acpi_ec_attach),
|
2007-10-25 16:49:07 +00:00
|
|
|
DEVMETHOD(device_suspend, acpi_ec_suspend),
|
|
|
|
DEVMETHOD(device_resume, acpi_ec_resume),
|
2004-06-30 16:00:20 +00:00
|
|
|
DEVMETHOD(device_shutdown, acpi_ec_shutdown),
|
2000-10-28 06:59:48 +00:00
|
|
|
|
2005-03-20 01:27:27 +00:00
|
|
|
/* Embedded controller interface */
|
|
|
|
DEVMETHOD(acpi_ec_read, acpi_ec_read_method),
|
|
|
|
DEVMETHOD(acpi_ec_write, acpi_ec_write_method),
|
|
|
|
|
2000-10-28 06:59:48 +00:00
|
|
|
{0, 0}
|
|
|
|
};
|
|
|
|
|
|
|
|
static driver_t acpi_ec_driver = {
|
|
|
|
"acpi_ec",
|
|
|
|
acpi_ec_methods,
|
|
|
|
sizeof(struct acpi_ec_softc),
|
|
|
|
};
|
|
|
|
|
2002-01-08 06:46:01 +00:00
|
|
|
static devclass_t acpi_ec_devclass;
|
2000-10-28 06:59:48 +00:00
|
|
|
DRIVER_MODULE(acpi_ec, acpi, acpi_ec_driver, acpi_ec_devclass, 0, 0);
|
2004-04-09 18:14:32 +00:00
|
|
|
MODULE_DEPEND(acpi_ec, acpi, 1, 1, 1);
|
2000-10-28 06:59:48 +00:00
|
|
|
|
|
|
|
/*
|
Rewrite the EC driver event model. The main goal is to avoid
polling/interrupt-driven fallback and instead use polling only during
boot and pure interrupt-driven mode after boot. Polled mode could be
relegated completely to a legacy role if we could enable interrupts
during boot. Polled mode can be forced after boot by setting
debug.acpi.ec.polled="1", i.e. if there are timeouts.
- Use polling only during boot, shutdown, or if requested by the user.
Otherwise, use a generation count of GPEs, incremented atomically. This
prevents an old status value from being used if the EC is really slow
and the same condition (i.e. multiple IBEs for a write transaction) is
being checked.
- Check for and run the query handler directly if the SCI bit is set in
the status register during boot. Previously, the query handler wouldn't
run until interrupts were finally enabled late in boot.
- During boot and after starting a command, check if the event appears
to already have occurred before we even start waiting. If so, it's
possible the EC is very slow and we might accept an old status value.
Print a warning in this case. Once we've booted, interrupt-driven mode
should work just fine but polled mode could be unreliable. There's not
much more we can do about this until interrupts are enabled during boot.
- In the above case, we also do one final check if the interrupt-driven
mode gets a timeout. If the status is complete, it will force the
system back into polled mode since interrupt mode doesn't work. For
polled mode during boot, if the status appears to be already complete
before beginning the check loop, it waits 10 us before actually checking
the status, just in case the EC is really slow and hasn't gotten to work
on the new request yet.
- Use upper-case hex for the _Qxx method
- Use device_printf for errors, don't hide them under verbose
- Increase default total timeout to 750 ms and decrease polling interval
to 5 us.
- Don't pass the status value via the softc. Just read it directly.
- Remove the mutex. We use the sx lock for transaction serialization
with the query handler.
- Remove the Intel copyright notice as no code of theirs was ever
present in this file (verified against rev 1.1)
- Allow KTR module-only builds for ease of testing
Thanks to jkim and Alexey Starikovskiy for helpful discussions and testing.
Approved by: re
MFC after: 2 weeks
2007-09-24 16:59:06 +00:00
|
|
|
* Look for an ECDT and if we find one, set up default GPE and
|
2003-07-20 00:48:38 +00:00
|
|
|
* space handlers to catch attempts to access EC space before
|
2000-10-28 06:59:48 +00:00
|
|
|
* we have a real driver instance in place.
|
Rewrite the EC driver event model. The main goal is to avoid
polling/interrupt-driven fallback and instead use polling only during
boot and pure interrupt-driven mode after boot. Polled mode could be
relegated completely to a legacy role if we could enable interrupts
during boot. Polled mode can be forced after boot by setting
debug.acpi.ec.polled="1", i.e. if there are timeouts.
- Use polling only during boot, shutdown, or if requested by the user.
Otherwise, use a generation count of GPEs, incremented atomically. This
prevents an old status value from being used if the EC is really slow
and the same condition (i.e. multiple IBEs for a write transaction) is
being checked.
- Check for and run the query handler directly if the SCI bit is set in
the status register during boot. Previously, the query handler wouldn't
run until interrupts were finally enabled late in boot.
- During boot and after starting a command, check if the event appears
to already have occurred before we even start waiting. If so, it's
possible the EC is very slow and we might accept an old status value.
Print a warning in this case. Once we've booted, interrupt-driven mode
should work just fine but polled mode could be unreliable. There's not
much more we can do about this until interrupts are enabled during boot.
- In the above case, we also do one final check if the interrupt-driven
mode gets a timeout. If the status is complete, it will force the
system back into polled mode since interrupt mode doesn't work. For
polled mode during boot, if the status appears to be already complete
before beginning the check loop, it waits 10 us before actually checking
the status, just in case the EC is really slow and hasn't gotten to work
on the new request yet.
- Use upper-case hex for the _Qxx method
- Use device_printf for errors, don't hide them under verbose
- Increase default total timeout to 750 ms and decrease polling interval
to 5 us.
- Don't pass the status value via the softc. Just read it directly.
- Remove the mutex. We use the sx lock for transaction serialization
with the query handler.
- Remove the Intel copyright notice as no code of theirs was ever
present in this file (verified against rev 1.1)
- Allow KTR module-only builds for ease of testing
Thanks to jkim and Alexey Starikovskiy for helpful discussions and testing.
Approved by: re
MFC after: 2 weeks
2007-09-24 16:59:06 +00:00
|
|
|
*
|
|
|
|
* TODO: Some old Gateway laptops need us to fake up an ECDT or
|
|
|
|
* otherwise attach early so that _REG methods can run.
|
2000-10-28 06:59:48 +00:00
|
|
|
*/
|
2003-07-20 00:48:38 +00:00
|
|
|
void
|
|
|
|
acpi_ec_ecdt_probe(device_t parent)
|
2000-10-28 06:59:48 +00:00
|
|
|
{
|
2003-07-20 00:48:38 +00:00
|
|
|
ACPI_TABLE_ECDT *ecdt;
|
|
|
|
ACPI_STATUS status;
|
|
|
|
device_t child;
|
|
|
|
ACPI_HANDLE h;
|
2004-05-14 04:17:56 +00:00
|
|
|
struct acpi_ec_params *params;
|
2003-07-20 00:48:38 +00:00
|
|
|
|
2002-05-19 06:16:47 +00:00
|
|
|
ACPI_FUNCTION_TRACE((char *)(uintptr_t)__func__);
|
2000-10-28 06:59:48 +00:00
|
|
|
|
2003-07-20 00:48:38 +00:00
|
|
|
/* Find and validate the ECDT. */
|
2007-03-22 18:16:43 +00:00
|
|
|
status = AcpiGetTable(ACPI_SIG_ECDT, 1, (ACPI_TABLE_HEADER **)&ecdt);
|
2003-07-20 00:48:38 +00:00
|
|
|
if (ACPI_FAILURE(status) ||
|
2007-03-22 18:16:43 +00:00
|
|
|
ecdt->Control.BitWidth != 8 ||
|
|
|
|
ecdt->Data.BitWidth != 8) {
|
2003-07-20 00:48:38 +00:00
|
|
|
return;
|
|
|
|
}
|
|
|
|
|
|
|
|
/* Create the child device with the given unit number. */
|
2007-03-22 18:16:43 +00:00
|
|
|
child = BUS_ADD_CHILD(parent, 0, "acpi_ec", ecdt->Uid);
|
2003-07-20 00:48:38 +00:00
|
|
|
if (child == NULL) {
|
2004-05-14 04:17:56 +00:00
|
|
|
printf("%s: can't add child\n", __func__);
|
2003-07-20 00:48:38 +00:00
|
|
|
return;
|
|
|
|
}
|
|
|
|
|
|
|
|
/* Find and save the ACPI handle for this device. */
|
2007-03-22 18:16:43 +00:00
|
|
|
status = AcpiGetHandle(NULL, ecdt->Id, &h);
|
2003-07-20 00:48:38 +00:00
|
|
|
if (ACPI_FAILURE(status)) {
|
|
|
|
device_delete_child(parent, child);
|
2004-05-14 04:17:56 +00:00
|
|
|
printf("%s: can't get handle\n", __func__);
|
2003-07-20 00:48:38 +00:00
|
|
|
return;
|
|
|
|
}
|
|
|
|
acpi_set_handle(child, h);
|
|
|
|
|
|
|
|
/* Set the data and CSR register addresses. */
|
2007-03-22 18:16:43 +00:00
|
|
|
bus_set_resource(child, SYS_RES_IOPORT, 0, ecdt->Data.Address,
|
2003-07-20 00:48:38 +00:00
|
|
|
/*count*/1);
|
2007-03-22 18:16:43 +00:00
|
|
|
bus_set_resource(child, SYS_RES_IOPORT, 1, ecdt->Control.Address,
|
2003-07-20 00:48:38 +00:00
|
|
|
/*count*/1);
|
|
|
|
|
|
|
|
/*
|
|
|
|
* Store values for the probe/attach routines to use. Store the
|
2004-05-13 03:17:08 +00:00
|
|
|
* ECDT GPE bit and set the global lock flag according to _GLK.
|
2004-05-14 04:17:56 +00:00
|
|
|
* Note that it is not perfectly correct to be evaluating a method
|
|
|
|
* before initializing devices, but in practice this function
|
|
|
|
* should be safe to call at this point.
|
2003-07-20 00:48:38 +00:00
|
|
|
*/
|
2004-05-14 04:17:56 +00:00
|
|
|
params = malloc(sizeof(struct acpi_ec_params), M_TEMP, M_WAITOK | M_ZERO);
|
|
|
|
params->gpe_handle = NULL;
|
2007-03-22 18:16:43 +00:00
|
|
|
params->gpe_bit = ecdt->Gpe;
|
|
|
|
params->uid = ecdt->Uid;
|
2004-05-14 04:17:56 +00:00
|
|
|
acpi_GetInteger(h, "_GLK", ¶ms->glk);
|
|
|
|
acpi_set_private(child, params);
|
2007-06-15 18:02:34 +00:00
|
|
|
acpi_set_magic(child, (uintptr_t)&acpi_ec_devclass);
|
2003-07-20 00:48:38 +00:00
|
|
|
|
|
|
|
/* Finish the attach process. */
|
|
|
|
if (device_probe_and_attach(child) != 0)
|
|
|
|
device_delete_child(parent, child);
|
2000-10-28 06:59:48 +00:00
|
|
|
}
|
|
|
|
|
|
|
|
static int
|
|
|
|
acpi_ec_probe(device_t dev)
|
|
|
|
{
|
2004-05-14 04:17:56 +00:00
|
|
|
ACPI_BUFFER buf;
|
2003-07-20 00:48:38 +00:00
|
|
|
ACPI_HANDLE h;
|
2004-05-14 04:17:56 +00:00
|
|
|
ACPI_OBJECT *obj;
|
2003-07-20 00:48:38 +00:00
|
|
|
ACPI_STATUS status;
|
|
|
|
device_t peer;
|
|
|
|
char desc[64];
|
2004-05-14 04:17:56 +00:00
|
|
|
int ret;
|
|
|
|
struct acpi_ec_params *params;
|
2004-06-29 19:02:27 +00:00
|
|
|
static char *ec_ids[] = { "PNP0C09", NULL };
|
2003-07-20 00:48:38 +00:00
|
|
|
|
2003-07-20 21:12:21 +00:00
|
|
|
/* Check that this is a device and that EC is not disabled. */
|
|
|
|
if (acpi_get_type(dev) != ACPI_TYPE_DEVICE || acpi_disabled("ec"))
|
2003-07-20 00:48:38 +00:00
|
|
|
return (ENXIO);
|
2000-12-08 09:16:20 +00:00
|
|
|
|
2003-07-20 00:48:38 +00:00
|
|
|
/*
|
|
|
|
* If probed via ECDT, set description and continue. Otherwise,
|
|
|
|
* we can access the namespace and make sure this is not a
|
|
|
|
* duplicate probe.
|
|
|
|
*/
|
2004-05-14 04:17:56 +00:00
|
|
|
ret = ENXIO;
|
|
|
|
params = NULL;
|
|
|
|
buf.Pointer = NULL;
|
|
|
|
buf.Length = ACPI_ALLOCATE_BUFFER;
|
|
|
|
if (DEV_ECDT(dev)) {
|
|
|
|
params = acpi_get_private(dev);
|
2003-07-20 00:48:38 +00:00
|
|
|
ret = 0;
|
2004-06-29 19:02:27 +00:00
|
|
|
} else if (!acpi_disabled("ec") &&
|
|
|
|
ACPI_ID_PROBE(device_get_parent(dev), dev, ec_ids)) {
|
2004-05-14 04:17:56 +00:00
|
|
|
params = malloc(sizeof(struct acpi_ec_params), M_TEMP,
|
|
|
|
M_WAITOK | M_ZERO);
|
2003-07-20 00:48:38 +00:00
|
|
|
h = acpi_get_handle(dev);
|
2000-10-28 06:59:48 +00:00
|
|
|
|
|
|
|
/*
|
2003-07-20 00:48:38 +00:00
|
|
|
* Read the unit ID to check for duplicate attach and the
|
|
|
|
* global lock value to see if we should acquire it when
|
|
|
|
* accessing the EC.
|
2000-10-28 06:59:48 +00:00
|
|
|
*/
|
2004-05-14 04:17:56 +00:00
|
|
|
status = acpi_GetInteger(h, "_UID", ¶ms->uid);
|
2003-07-20 00:48:38 +00:00
|
|
|
if (ACPI_FAILURE(status))
|
2004-05-14 04:17:56 +00:00
|
|
|
params->uid = 0;
|
|
|
|
status = acpi_GetInteger(h, "_GLK", ¶ms->glk);
|
2003-07-20 00:48:38 +00:00
|
|
|
if (ACPI_FAILURE(status))
|
2004-05-14 04:17:56 +00:00
|
|
|
params->glk = 0;
|
2003-07-20 00:48:38 +00:00
|
|
|
|
|
|
|
/*
|
|
|
|
* Evaluate the _GPE method to find the GPE bit used by the EC to
|
2004-05-14 04:17:56 +00:00
|
|
|
* signal status (SCI). If it's a package, it contains a reference
|
|
|
|
* and GPE bit, similar to _PRW.
|
2003-07-20 00:48:38 +00:00
|
|
|
*/
|
2004-05-14 04:17:56 +00:00
|
|
|
status = AcpiEvaluateObject(h, "_GPE", NULL, &buf);
|
2003-07-20 00:48:38 +00:00
|
|
|
if (ACPI_FAILURE(status)) {
|
|
|
|
device_printf(dev, "can't evaluate _GPE - %s\n",
|
|
|
|
AcpiFormatException(status));
|
2005-05-12 22:35:08 +00:00
|
|
|
goto out;
|
2003-07-20 00:48:38 +00:00
|
|
|
}
|
2004-05-14 04:17:56 +00:00
|
|
|
obj = (ACPI_OBJECT *)buf.Pointer;
|
|
|
|
if (obj == NULL)
|
2005-05-12 22:35:08 +00:00
|
|
|
goto out;
|
2004-05-14 04:17:56 +00:00
|
|
|
|
|
|
|
switch (obj->Type) {
|
|
|
|
case ACPI_TYPE_INTEGER:
|
|
|
|
params->gpe_handle = NULL;
|
|
|
|
params->gpe_bit = obj->Integer.Value;
|
|
|
|
break;
|
|
|
|
case ACPI_TYPE_PACKAGE:
|
|
|
|
if (!ACPI_PKG_VALID(obj, 2))
|
|
|
|
goto out;
|
|
|
|
params->gpe_handle =
|
|
|
|
acpi_GetReference(NULL, &obj->Package.Elements[0]);
|
|
|
|
if (params->gpe_handle == NULL ||
|
|
|
|
acpi_PkgInt32(obj, 1, ¶ms->gpe_bit) != 0)
|
|
|
|
goto out;
|
|
|
|
break;
|
|
|
|
default:
|
|
|
|
device_printf(dev, "_GPE has invalid type %d\n", obj->Type);
|
|
|
|
goto out;
|
|
|
|
}
|
2003-07-20 00:48:38 +00:00
|
|
|
|
|
|
|
/* Store the values we got from the namespace for attach. */
|
2004-05-14 04:17:56 +00:00
|
|
|
acpi_set_private(dev, params);
|
2003-07-20 00:48:38 +00:00
|
|
|
|
|
|
|
/*
|
|
|
|
* Check for a duplicate probe. This can happen when a probe
|
2004-05-13 03:17:08 +00:00
|
|
|
* via ECDT succeeded already. If this is a duplicate, disable
|
|
|
|
* this device.
|
2003-07-20 00:48:38 +00:00
|
|
|
*/
|
2004-05-14 04:17:56 +00:00
|
|
|
peer = devclass_get_device(acpi_ec_devclass, params->uid);
|
2004-05-13 03:17:08 +00:00
|
|
|
if (peer == NULL || !device_is_alive(peer))
|
2003-07-20 00:48:38 +00:00
|
|
|
ret = 0;
|
2004-05-13 03:17:08 +00:00
|
|
|
else
|
|
|
|
device_disable(dev);
|
|
|
|
}
|
|
|
|
|
2004-05-14 04:17:56 +00:00
|
|
|
out:
|
2004-05-13 03:17:08 +00:00
|
|
|
if (ret == 0) {
|
|
|
|
snprintf(desc, sizeof(desc), "Embedded Controller: GPE %#x%s%s",
|
2004-05-14 04:17:56 +00:00
|
|
|
params->gpe_bit, (params->glk) ? ", GLK" : "",
|
2004-05-13 03:17:08 +00:00
|
|
|
DEV_ECDT(dev) ? ", ECDT" : "");
|
|
|
|
device_set_desc_copy(dev, desc);
|
2000-10-28 06:59:48 +00:00
|
|
|
}
|
2003-07-10 17:22:46 +00:00
|
|
|
|
2004-05-14 04:17:56 +00:00
|
|
|
if (ret > 0 && params)
|
|
|
|
free(params, M_TEMP);
|
|
|
|
if (buf.Pointer)
|
|
|
|
AcpiOsFree(buf.Pointer);
|
2003-07-10 17:22:46 +00:00
|
|
|
return (ret);
|
2000-10-28 06:59:48 +00:00
|
|
|
}
|
|
|
|
|
|
|
|
static int
|
|
|
|
acpi_ec_attach(device_t dev)
|
|
|
|
{
|
|
|
|
struct acpi_ec_softc *sc;
|
2004-05-14 04:17:56 +00:00
|
|
|
struct acpi_ec_params *params;
|
2000-10-28 06:59:48 +00:00
|
|
|
ACPI_STATUS Status;
|
2002-02-23 05:26:31 +00:00
|
|
|
|
2002-05-19 06:16:47 +00:00
|
|
|
ACPI_FUNCTION_TRACE((char *)(uintptr_t)__func__);
|
2000-12-08 09:16:20 +00:00
|
|
|
|
2003-07-10 17:22:46 +00:00
|
|
|
/* Fetch/initialize softc (assumes softc is pre-zeroed). */
|
2000-10-28 06:59:48 +00:00
|
|
|
sc = device_get_softc(dev);
|
2004-05-14 04:17:56 +00:00
|
|
|
params = acpi_get_private(dev);
|
2000-10-28 06:59:48 +00:00
|
|
|
sc->ec_dev = dev;
|
|
|
|
sc->ec_handle = acpi_get_handle(dev);
|
|
|
|
|
2003-07-20 00:48:38 +00:00
|
|
|
/* Retrieve previously probed values via device ivars. */
|
2004-05-14 04:17:56 +00:00
|
|
|
sc->ec_glk = params->glk;
|
|
|
|
sc->ec_gpebit = params->gpe_bit;
|
|
|
|
sc->ec_gpehandle = params->gpe_handle;
|
|
|
|
sc->ec_uid = params->uid;
|
2007-10-25 20:02:38 +00:00
|
|
|
sc->ec_suspending = FALSE;
|
2004-05-14 04:17:56 +00:00
|
|
|
free(params, M_TEMP);
|
2003-07-20 00:48:38 +00:00
|
|
|
|
2003-07-10 17:22:46 +00:00
|
|
|
/* Attach bus resources for data and command/status ports. */
|
2000-10-28 06:59:48 +00:00
|
|
|
sc->ec_data_rid = 0;
|
2004-03-17 17:50:55 +00:00
|
|
|
sc->ec_data_res = bus_alloc_resource_any(sc->ec_dev, SYS_RES_IOPORT,
|
|
|
|
&sc->ec_data_rid, RF_ACTIVE);
|
2003-06-27 21:57:42 +00:00
|
|
|
if (sc->ec_data_res == NULL) {
|
2000-10-28 06:59:48 +00:00
|
|
|
device_printf(dev, "can't allocate data port\n");
|
2004-05-25 02:47:35 +00:00
|
|
|
goto error;
|
2000-10-28 06:59:48 +00:00
|
|
|
}
|
|
|
|
sc->ec_data_tag = rman_get_bustag(sc->ec_data_res);
|
|
|
|
sc->ec_data_handle = rman_get_bushandle(sc->ec_data_res);
|
|
|
|
|
|
|
|
sc->ec_csr_rid = 1;
|
2004-03-17 17:50:55 +00:00
|
|
|
sc->ec_csr_res = bus_alloc_resource_any(sc->ec_dev, SYS_RES_IOPORT,
|
|
|
|
&sc->ec_csr_rid, RF_ACTIVE);
|
2003-06-27 21:57:42 +00:00
|
|
|
if (sc->ec_csr_res == NULL) {
|
2000-10-28 06:59:48 +00:00
|
|
|
device_printf(dev, "can't allocate command/status port\n");
|
2004-05-25 02:47:35 +00:00
|
|
|
goto error;
|
2000-10-28 06:59:48 +00:00
|
|
|
}
|
|
|
|
sc->ec_csr_tag = rman_get_bustag(sc->ec_csr_res);
|
|
|
|
sc->ec_csr_handle = rman_get_bushandle(sc->ec_csr_res);
|
|
|
|
|
|
|
|
/*
|
2003-07-10 17:22:46 +00:00
|
|
|
* Install a handler for this EC's GPE bit. We want edge-triggered
|
|
|
|
* behavior.
|
2000-10-28 06:59:48 +00:00
|
|
|
*/
|
2003-07-10 17:22:46 +00:00
|
|
|
ACPI_DEBUG_PRINT((ACPI_DB_RESOURCES, "attaching GPE handler\n"));
|
2004-05-14 04:17:56 +00:00
|
|
|
Status = AcpiInstallGpeHandler(sc->ec_gpehandle, sc->ec_gpebit,
|
2004-04-14 02:20:35 +00:00
|
|
|
ACPI_GPE_EDGE_TRIGGERED, &EcGpeHandler, sc);
|
2003-06-27 21:57:42 +00:00
|
|
|
if (ACPI_FAILURE(Status)) {
|
2001-07-05 07:21:12 +00:00
|
|
|
device_printf(dev, "can't install GPE handler for %s - %s\n",
|
2001-07-21 10:24:37 +00:00
|
|
|
acpi_name(sc->ec_handle), AcpiFormatException(Status));
|
2004-05-25 02:47:35 +00:00
|
|
|
goto error;
|
2000-10-28 06:59:48 +00:00
|
|
|
}
|
|
|
|
|
Rewrite the EC driver event model. The main goal is to avoid
polling/interrupt-driven fallback and instead use polling only during
boot and pure interrupt-driven mode after boot. Polled mode could be
relegated completely to a legacy role if we could enable interrupts
during boot. Polled mode can be forced after boot by setting
debug.acpi.ec.polled="1", i.e. if there are timeouts.
- Use polling only during boot, shutdown, or if requested by the user.
Otherwise, use a generation count of GPEs, incremented atomically. This
prevents an old status value from being used if the EC is really slow
and the same condition (i.e. multiple IBEs for a write transaction) is
being checked.
- Check for and run the query handler directly if the SCI bit is set in
the status register during boot. Previously, the query handler wouldn't
run until interrupts were finally enabled late in boot.
- During boot and after starting a command, check if the event appears
to already have occurred before we even start waiting. If so, it's
possible the EC is very slow and we might accept an old status value.
Print a warning in this case. Once we've booted, interrupt-driven mode
should work just fine but polled mode could be unreliable. There's not
much more we can do about this until interrupts are enabled during boot.
- In the above case, we also do one final check if the interrupt-driven
mode gets a timeout. If the status is complete, it will force the
system back into polled mode since interrupt mode doesn't work. For
polled mode during boot, if the status appears to be already complete
before beginning the check loop, it waits 10 us before actually checking
the status, just in case the EC is really slow and hasn't gotten to work
on the new request yet.
- Use upper-case hex for the _Qxx method
- Use device_printf for errors, don't hide them under verbose
- Increase default total timeout to 750 ms and decrease polling interval
to 5 us.
- Don't pass the status value via the softc. Just read it directly.
- Remove the mutex. We use the sx lock for transaction serialization
with the query handler.
- Remove the Intel copyright notice as no code of theirs was ever
present in this file (verified against rev 1.1)
- Allow KTR module-only builds for ease of testing
Thanks to jkim and Alexey Starikovskiy for helpful discussions and testing.
Approved by: re
MFC after: 2 weeks
2007-09-24 16:59:06 +00:00
|
|
|
/*
|
2000-10-28 06:59:48 +00:00
|
|
|
* Install address space handler
|
|
|
|
*/
|
2001-08-26 22:50:15 +00:00
|
|
|
ACPI_DEBUG_PRINT((ACPI_DB_RESOURCES, "attaching address space handler\n"));
|
2003-06-27 21:57:42 +00:00
|
|
|
Status = AcpiInstallAddressSpaceHandler(sc->ec_handle, ACPI_ADR_SPACE_EC,
|
2003-07-10 17:22:46 +00:00
|
|
|
&EcSpaceHandler, &EcSpaceSetup, sc);
|
2003-06-27 21:57:42 +00:00
|
|
|
if (ACPI_FAILURE(Status)) {
|
2001-07-05 07:21:12 +00:00
|
|
|
device_printf(dev, "can't install address space handler for %s - %s\n",
|
2001-07-21 10:24:37 +00:00
|
|
|
acpi_name(sc->ec_handle), AcpiFormatException(Status));
|
2004-05-25 02:47:35 +00:00
|
|
|
goto error;
|
|
|
|
}
|
|
|
|
|
|
|
|
/* Enable runtime GPEs for the handler. */
|
|
|
|
Status = AcpiSetGpeType(sc->ec_gpehandle, sc->ec_gpebit,
|
|
|
|
ACPI_GPE_TYPE_RUNTIME);
|
|
|
|
if (ACPI_FAILURE(Status)) {
|
|
|
|
device_printf(dev, "AcpiSetGpeType failed: %s\n",
|
|
|
|
AcpiFormatException(Status));
|
|
|
|
goto error;
|
|
|
|
}
|
|
|
|
Status = AcpiEnableGpe(sc->ec_gpehandle, sc->ec_gpebit, ACPI_NOT_ISR);
|
|
|
|
if (ACPI_FAILURE(Status)) {
|
|
|
|
device_printf(dev, "AcpiEnableGpe failed: %s\n",
|
|
|
|
AcpiFormatException(Status));
|
|
|
|
goto error;
|
2000-10-28 06:59:48 +00:00
|
|
|
}
|
2003-07-10 17:22:46 +00:00
|
|
|
|
|
|
|
ACPI_DEBUG_PRINT((ACPI_DB_RESOURCES, "acpi_ec_attach complete\n"));
|
|
|
|
return (0);
|
2003-06-27 21:57:42 +00:00
|
|
|
|
2004-05-25 02:47:35 +00:00
|
|
|
error:
|
|
|
|
AcpiRemoveGpeHandler(sc->ec_gpehandle, sc->ec_gpebit, &EcGpeHandler);
|
|
|
|
AcpiRemoveAddressSpaceHandler(sc->ec_handle, ACPI_ADR_SPACE_EC,
|
|
|
|
EcSpaceHandler);
|
2003-06-27 21:57:42 +00:00
|
|
|
if (sc->ec_csr_res)
|
Rewrite the EC driver event model. The main goal is to avoid
polling/interrupt-driven fallback and instead use polling only during
boot and pure interrupt-driven mode after boot. Polled mode could be
relegated completely to a legacy role if we could enable interrupts
during boot. Polled mode can be forced after boot by setting
debug.acpi.ec.polled="1", i.e. if there are timeouts.
- Use polling only during boot, shutdown, or if requested by the user.
Otherwise, use a generation count of GPEs, incremented atomically. This
prevents an old status value from being used if the EC is really slow
and the same condition (i.e. multiple IBEs for a write transaction) is
being checked.
- Check for and run the query handler directly if the SCI bit is set in
the status register during boot. Previously, the query handler wouldn't
run until interrupts were finally enabled late in boot.
- During boot and after starting a command, check if the event appears
to already have occurred before we even start waiting. If so, it's
possible the EC is very slow and we might accept an old status value.
Print a warning in this case. Once we've booted, interrupt-driven mode
should work just fine but polled mode could be unreliable. There's not
much more we can do about this until interrupts are enabled during boot.
- In the above case, we also do one final check if the interrupt-driven
mode gets a timeout. If the status is complete, it will force the
system back into polled mode since interrupt mode doesn't work. For
polled mode during boot, if the status appears to be already complete
before beginning the check loop, it waits 10 us before actually checking
the status, just in case the EC is really slow and hasn't gotten to work
on the new request yet.
- Use upper-case hex for the _Qxx method
- Use device_printf for errors, don't hide them under verbose
- Increase default total timeout to 750 ms and decrease polling interval
to 5 us.
- Don't pass the status value via the softc. Just read it directly.
- Remove the mutex. We use the sx lock for transaction serialization
with the query handler.
- Remove the Intel copyright notice as no code of theirs was ever
present in this file (verified against rev 1.1)
- Allow KTR module-only builds for ease of testing
Thanks to jkim and Alexey Starikovskiy for helpful discussions and testing.
Approved by: re
MFC after: 2 weeks
2007-09-24 16:59:06 +00:00
|
|
|
bus_release_resource(sc->ec_dev, SYS_RES_IOPORT, sc->ec_csr_rid,
|
2002-01-31 01:23:22 +00:00
|
|
|
sc->ec_csr_res);
|
2003-06-27 21:57:42 +00:00
|
|
|
if (sc->ec_data_res)
|
2003-07-10 17:22:46 +00:00
|
|
|
bus_release_resource(sc->ec_dev, SYS_RES_IOPORT, sc->ec_data_rid,
|
2002-01-31 01:23:22 +00:00
|
|
|
sc->ec_data_res);
|
2004-05-25 02:47:35 +00:00
|
|
|
return (ENXIO);
|
2000-10-28 06:59:48 +00:00
|
|
|
}
|
|
|
|
|
2007-10-25 16:49:07 +00:00
|
|
|
static int
|
|
|
|
acpi_ec_suspend(device_t dev)
|
|
|
|
{
|
|
|
|
struct acpi_ec_softc *sc;
|
|
|
|
|
|
|
|
sc = device_get_softc(dev);
|
2007-10-25 20:02:38 +00:00
|
|
|
sc->ec_suspending = TRUE;
|
2007-10-25 16:49:07 +00:00
|
|
|
return (0);
|
|
|
|
}
|
|
|
|
|
|
|
|
static int
|
|
|
|
acpi_ec_resume(device_t dev)
|
|
|
|
{
|
|
|
|
struct acpi_ec_softc *sc;
|
|
|
|
|
|
|
|
sc = device_get_softc(dev);
|
2007-10-25 20:02:38 +00:00
|
|
|
sc->ec_suspending = FALSE;
|
2007-10-25 16:49:07 +00:00
|
|
|
return (0);
|
|
|
|
}
|
|
|
|
|
2004-06-30 16:00:20 +00:00
|
|
|
static int
|
|
|
|
acpi_ec_shutdown(device_t dev)
|
|
|
|
{
|
|
|
|
struct acpi_ec_softc *sc;
|
|
|
|
|
|
|
|
/* Disable the GPE so we don't get EC events during shutdown. */
|
|
|
|
sc = device_get_softc(dev);
|
|
|
|
AcpiDisableGpe(sc->ec_gpehandle, sc->ec_gpebit, ACPI_NOT_ISR);
|
|
|
|
return (0);
|
|
|
|
}
|
|
|
|
|
2005-03-20 01:27:27 +00:00
|
|
|
/* Methods to allow other devices (e.g., smbat) to read/write EC space. */
|
|
|
|
static int
|
|
|
|
acpi_ec_read_method(device_t dev, u_int addr, ACPI_INTEGER *val, int width)
|
|
|
|
{
|
|
|
|
struct acpi_ec_softc *sc;
|
|
|
|
ACPI_STATUS status;
|
|
|
|
|
|
|
|
sc = device_get_softc(dev);
|
|
|
|
status = EcSpaceHandler(ACPI_READ, addr, width * 8, val, sc, NULL);
|
|
|
|
if (ACPI_FAILURE(status))
|
|
|
|
return (ENXIO);
|
|
|
|
return (0);
|
|
|
|
}
|
|
|
|
|
|
|
|
static int
|
|
|
|
acpi_ec_write_method(device_t dev, u_int addr, ACPI_INTEGER val, int width)
|
|
|
|
{
|
|
|
|
struct acpi_ec_softc *sc;
|
|
|
|
ACPI_STATUS status;
|
|
|
|
|
|
|
|
sc = device_get_softc(dev);
|
|
|
|
status = EcSpaceHandler(ACPI_WRITE, addr, width * 8, &val, sc, NULL);
|
|
|
|
if (ACPI_FAILURE(status))
|
|
|
|
return (ENXIO);
|
|
|
|
return (0);
|
|
|
|
}
|
|
|
|
|
2000-10-28 06:59:48 +00:00
|
|
|
static void
|
2000-12-14 12:43:22 +00:00
|
|
|
EcGpeQueryHandler(void *Context)
|
2000-10-28 06:59:48 +00:00
|
|
|
{
|
|
|
|
struct acpi_ec_softc *sc = (struct acpi_ec_softc *)Context;
|
|
|
|
UINT8 Data;
|
|
|
|
ACPI_STATUS Status;
|
|
|
|
char qxx[5];
|
|
|
|
|
2002-05-19 06:16:47 +00:00
|
|
|
ACPI_FUNCTION_TRACE((char *)(uintptr_t)__func__);
|
2003-07-10 17:22:46 +00:00
|
|
|
KASSERT(Context != NULL, ("EcGpeQueryHandler called with NULL"));
|
2000-12-08 09:16:20 +00:00
|
|
|
|
2007-02-27 00:14:20 +00:00
|
|
|
/* Serialize user access with EcSpaceHandler(). */
|
Rewrite the EC driver event model. The main goal is to avoid
polling/interrupt-driven fallback and instead use polling only during
boot and pure interrupt-driven mode after boot. Polled mode could be
relegated completely to a legacy role if we could enable interrupts
during boot. Polled mode can be forced after boot by setting
debug.acpi.ec.polled="1", i.e. if there are timeouts.
- Use polling only during boot, shutdown, or if requested by the user.
Otherwise, use a generation count of GPEs, incremented atomically. This
prevents an old status value from being used if the EC is really slow
and the same condition (i.e. multiple IBEs for a write transaction) is
being checked.
- Check for and run the query handler directly if the SCI bit is set in
the status register during boot. Previously, the query handler wouldn't
run until interrupts were finally enabled late in boot.
- During boot and after starting a command, check if the event appears
to already have occurred before we even start waiting. If so, it's
possible the EC is very slow and we might accept an old status value.
Print a warning in this case. Once we've booted, interrupt-driven mode
should work just fine but polled mode could be unreliable. There's not
much more we can do about this until interrupts are enabled during boot.
- In the above case, we also do one final check if the interrupt-driven
mode gets a timeout. If the status is complete, it will force the
system back into polled mode since interrupt mode doesn't work. For
polled mode during boot, if the status appears to be already complete
before beginning the check loop, it waits 10 us before actually checking
the status, just in case the EC is really slow and hasn't gotten to work
on the new request yet.
- Use upper-case hex for the _Qxx method
- Use device_printf for errors, don't hide them under verbose
- Increase default total timeout to 750 ms and decrease polling interval
to 5 us.
- Don't pass the status value via the softc. Just read it directly.
- Remove the mutex. We use the sx lock for transaction serialization
with the query handler.
- Remove the Intel copyright notice as no code of theirs was ever
present in this file (verified against rev 1.1)
- Allow KTR module-only builds for ease of testing
Thanks to jkim and Alexey Starikovskiy for helpful discussions and testing.
Approved by: re
MFC after: 2 weeks
2007-09-24 16:59:06 +00:00
|
|
|
Status = EcLock(sc);
|
2003-07-20 21:11:32 +00:00
|
|
|
if (ACPI_FAILURE(Status)) {
|
Rewrite the EC driver event model. The main goal is to avoid
polling/interrupt-driven fallback and instead use polling only during
boot and pure interrupt-driven mode after boot. Polled mode could be
relegated completely to a legacy role if we could enable interrupts
during boot. Polled mode can be forced after boot by setting
debug.acpi.ec.polled="1", i.e. if there are timeouts.
- Use polling only during boot, shutdown, or if requested by the user.
Otherwise, use a generation count of GPEs, incremented atomically. This
prevents an old status value from being used if the EC is really slow
and the same condition (i.e. multiple IBEs for a write transaction) is
being checked.
- Check for and run the query handler directly if the SCI bit is set in
the status register during boot. Previously, the query handler wouldn't
run until interrupts were finally enabled late in boot.
- During boot and after starting a command, check if the event appears
to already have occurred before we even start waiting. If so, it's
possible the EC is very slow and we might accept an old status value.
Print a warning in this case. Once we've booted, interrupt-driven mode
should work just fine but polled mode could be unreliable. There's not
much more we can do about this until interrupts are enabled during boot.
- In the above case, we also do one final check if the interrupt-driven
mode gets a timeout. If the status is complete, it will force the
system back into polled mode since interrupt mode doesn't work. For
polled mode during boot, if the status appears to be already complete
before beginning the check loop, it waits 10 us before actually checking
the status, just in case the EC is really slow and hasn't gotten to work
on the new request yet.
- Use upper-case hex for the _Qxx method
- Use device_printf for errors, don't hide them under verbose
- Increase default total timeout to 750 ms and decrease polling interval
to 5 us.
- Don't pass the status value via the softc. Just read it directly.
- Remove the mutex. We use the sx lock for transaction serialization
with the query handler.
- Remove the Intel copyright notice as no code of theirs was ever
present in this file (verified against rev 1.1)
- Allow KTR module-only builds for ease of testing
Thanks to jkim and Alexey Starikovskiy for helpful discussions and testing.
Approved by: re
MFC after: 2 weeks
2007-09-24 16:59:06 +00:00
|
|
|
device_printf(sc->ec_dev, "GpeQuery lock error: %s\n",
|
|
|
|
AcpiFormatException(Status));
|
2003-07-20 21:11:32 +00:00
|
|
|
return;
|
|
|
|
}
|
|
|
|
|
|
|
|
/*
|
|
|
|
* Send a query command to the EC to find out which _Qxx call it
|
|
|
|
* wants to make. This command clears the SCI bit and also the
|
Rewrite the EC driver event model. The main goal is to avoid
polling/interrupt-driven fallback and instead use polling only during
boot and pure interrupt-driven mode after boot. Polled mode could be
relegated completely to a legacy role if we could enable interrupts
during boot. Polled mode can be forced after boot by setting
debug.acpi.ec.polled="1", i.e. if there are timeouts.
- Use polling only during boot, shutdown, or if requested by the user.
Otherwise, use a generation count of GPEs, incremented atomically. This
prevents an old status value from being used if the EC is really slow
and the same condition (i.e. multiple IBEs for a write transaction) is
being checked.
- Check for and run the query handler directly if the SCI bit is set in
the status register during boot. Previously, the query handler wouldn't
run until interrupts were finally enabled late in boot.
- During boot and after starting a command, check if the event appears
to already have occurred before we even start waiting. If so, it's
possible the EC is very slow and we might accept an old status value.
Print a warning in this case. Once we've booted, interrupt-driven mode
should work just fine but polled mode could be unreliable. There's not
much more we can do about this until interrupts are enabled during boot.
- In the above case, we also do one final check if the interrupt-driven
mode gets a timeout. If the status is complete, it will force the
system back into polled mode since interrupt mode doesn't work. For
polled mode during boot, if the status appears to be already complete
before beginning the check loop, it waits 10 us before actually checking
the status, just in case the EC is really slow and hasn't gotten to work
on the new request yet.
- Use upper-case hex for the _Qxx method
- Use device_printf for errors, don't hide them under verbose
- Increase default total timeout to 750 ms and decrease polling interval
to 5 us.
- Don't pass the status value via the softc. Just read it directly.
- Remove the mutex. We use the sx lock for transaction serialization
with the query handler.
- Remove the Intel copyright notice as no code of theirs was ever
present in this file (verified against rev 1.1)
- Allow KTR module-only builds for ease of testing
Thanks to jkim and Alexey Starikovskiy for helpful discussions and testing.
Approved by: re
MFC after: 2 weeks
2007-09-24 16:59:06 +00:00
|
|
|
* interrupt source since we are edge-triggered. To prevent the GPE
|
|
|
|
* that may arise from running the query from causing another query
|
|
|
|
* to be queued, we clear the pending flag only after running it.
|
2003-07-20 21:11:32 +00:00
|
|
|
*/
|
|
|
|
Status = EcCommand(sc, EC_COMMAND_QUERY);
|
Rewrite the EC driver event model. The main goal is to avoid
polling/interrupt-driven fallback and instead use polling only during
boot and pure interrupt-driven mode after boot. Polled mode could be
relegated completely to a legacy role if we could enable interrupts
during boot. Polled mode can be forced after boot by setting
debug.acpi.ec.polled="1", i.e. if there are timeouts.
- Use polling only during boot, shutdown, or if requested by the user.
Otherwise, use a generation count of GPEs, incremented atomically. This
prevents an old status value from being used if the EC is really slow
and the same condition (i.e. multiple IBEs for a write transaction) is
being checked.
- Check for and run the query handler directly if the SCI bit is set in
the status register during boot. Previously, the query handler wouldn't
run until interrupts were finally enabled late in boot.
- During boot and after starting a command, check if the event appears
to already have occurred before we even start waiting. If so, it's
possible the EC is very slow and we might accept an old status value.
Print a warning in this case. Once we've booted, interrupt-driven mode
should work just fine but polled mode could be unreliable. There's not
much more we can do about this until interrupts are enabled during boot.
- In the above case, we also do one final check if the interrupt-driven
mode gets a timeout. If the status is complete, it will force the
system back into polled mode since interrupt mode doesn't work. For
polled mode during boot, if the status appears to be already complete
before beginning the check loop, it waits 10 us before actually checking
the status, just in case the EC is really slow and hasn't gotten to work
on the new request yet.
- Use upper-case hex for the _Qxx method
- Use device_printf for errors, don't hide them under verbose
- Increase default total timeout to 750 ms and decrease polling interval
to 5 us.
- Don't pass the status value via the softc. Just read it directly.
- Remove the mutex. We use the sx lock for transaction serialization
with the query handler.
- Remove the Intel copyright notice as no code of theirs was ever
present in this file (verified against rev 1.1)
- Allow KTR module-only builds for ease of testing
Thanks to jkim and Alexey Starikovskiy for helpful discussions and testing.
Approved by: re
MFC after: 2 weeks
2007-09-24 16:59:06 +00:00
|
|
|
sc->ec_sci_pend = FALSE;
|
2003-07-10 17:22:46 +00:00
|
|
|
if (ACPI_FAILURE(Status)) {
|
2003-07-20 21:11:32 +00:00
|
|
|
EcUnlock(sc);
|
Rewrite the EC driver event model. The main goal is to avoid
polling/interrupt-driven fallback and instead use polling only during
boot and pure interrupt-driven mode after boot. Polled mode could be
relegated completely to a legacy role if we could enable interrupts
during boot. Polled mode can be forced after boot by setting
debug.acpi.ec.polled="1", i.e. if there are timeouts.
- Use polling only during boot, shutdown, or if requested by the user.
Otherwise, use a generation count of GPEs, incremented atomically. This
prevents an old status value from being used if the EC is really slow
and the same condition (i.e. multiple IBEs for a write transaction) is
being checked.
- Check for and run the query handler directly if the SCI bit is set in
the status register during boot. Previously, the query handler wouldn't
run until interrupts were finally enabled late in boot.
- During boot and after starting a command, check if the event appears
to already have occurred before we even start waiting. If so, it's
possible the EC is very slow and we might accept an old status value.
Print a warning in this case. Once we've booted, interrupt-driven mode
should work just fine but polled mode could be unreliable. There's not
much more we can do about this until interrupts are enabled during boot.
- In the above case, we also do one final check if the interrupt-driven
mode gets a timeout. If the status is complete, it will force the
system back into polled mode since interrupt mode doesn't work. For
polled mode during boot, if the status appears to be already complete
before beginning the check loop, it waits 10 us before actually checking
the status, just in case the EC is really slow and hasn't gotten to work
on the new request yet.
- Use upper-case hex for the _Qxx method
- Use device_printf for errors, don't hide them under verbose
- Increase default total timeout to 750 ms and decrease polling interval
to 5 us.
- Don't pass the status value via the softc. Just read it directly.
- Remove the mutex. We use the sx lock for transaction serialization
with the query handler.
- Remove the Intel copyright notice as no code of theirs was ever
present in this file (verified against rev 1.1)
- Allow KTR module-only builds for ease of testing
Thanks to jkim and Alexey Starikovskiy for helpful discussions and testing.
Approved by: re
MFC after: 2 weeks
2007-09-24 16:59:06 +00:00
|
|
|
device_printf(sc->ec_dev, "GPE query failed: %s\n",
|
|
|
|
AcpiFormatException(Status));
|
|
|
|
return;
|
2003-07-10 17:22:46 +00:00
|
|
|
}
|
2003-07-20 21:11:32 +00:00
|
|
|
Data = EC_GET_DATA(sc);
|
2007-02-27 00:14:20 +00:00
|
|
|
|
Rewrite the EC driver event model. The main goal is to avoid
polling/interrupt-driven fallback and instead use polling only during
boot and pure interrupt-driven mode after boot. Polled mode could be
relegated completely to a legacy role if we could enable interrupts
during boot. Polled mode can be forced after boot by setting
debug.acpi.ec.polled="1", i.e. if there are timeouts.
- Use polling only during boot, shutdown, or if requested by the user.
Otherwise, use a generation count of GPEs, incremented atomically. This
prevents an old status value from being used if the EC is really slow
and the same condition (i.e. multiple IBEs for a write transaction) is
being checked.
- Check for and run the query handler directly if the SCI bit is set in
the status register during boot. Previously, the query handler wouldn't
run until interrupts were finally enabled late in boot.
- During boot and after starting a command, check if the event appears
to already have occurred before we even start waiting. If so, it's
possible the EC is very slow and we might accept an old status value.
Print a warning in this case. Once we've booted, interrupt-driven mode
should work just fine but polled mode could be unreliable. There's not
much more we can do about this until interrupts are enabled during boot.
- In the above case, we also do one final check if the interrupt-driven
mode gets a timeout. If the status is complete, it will force the
system back into polled mode since interrupt mode doesn't work. For
polled mode during boot, if the status appears to be already complete
before beginning the check loop, it waits 10 us before actually checking
the status, just in case the EC is really slow and hasn't gotten to work
on the new request yet.
- Use upper-case hex for the _Qxx method
- Use device_printf for errors, don't hide them under verbose
- Increase default total timeout to 750 ms and decrease polling interval
to 5 us.
- Don't pass the status value via the softc. Just read it directly.
- Remove the mutex. We use the sx lock for transaction serialization
with the query handler.
- Remove the Intel copyright notice as no code of theirs was ever
present in this file (verified against rev 1.1)
- Allow KTR module-only builds for ease of testing
Thanks to jkim and Alexey Starikovskiy for helpful discussions and testing.
Approved by: re
MFC after: 2 weeks
2007-09-24 16:59:06 +00:00
|
|
|
/*
|
|
|
|
* We have to unlock before running the _Qxx method below since that
|
|
|
|
* method may attempt to read/write from EC address space, causing
|
|
|
|
* recursive acquisition of the lock.
|
|
|
|
*/
|
2003-07-20 21:11:32 +00:00
|
|
|
EcUnlock(sc);
|
2000-10-28 06:59:48 +00:00
|
|
|
|
2003-07-10 17:22:46 +00:00
|
|
|
/* Ignore the value for "no outstanding event". (13.3.5) */
|
Rewrite the EC driver event model. The main goal is to avoid
polling/interrupt-driven fallback and instead use polling only during
boot and pure interrupt-driven mode after boot. Polled mode could be
relegated completely to a legacy role if we could enable interrupts
during boot. Polled mode can be forced after boot by setting
debug.acpi.ec.polled="1", i.e. if there are timeouts.
- Use polling only during boot, shutdown, or if requested by the user.
Otherwise, use a generation count of GPEs, incremented atomically. This
prevents an old status value from being used if the EC is really slow
and the same condition (i.e. multiple IBEs for a write transaction) is
being checked.
- Check for and run the query handler directly if the SCI bit is set in
the status register during boot. Previously, the query handler wouldn't
run until interrupts were finally enabled late in boot.
- During boot and after starting a command, check if the event appears
to already have occurred before we even start waiting. If so, it's
possible the EC is very slow and we might accept an old status value.
Print a warning in this case. Once we've booted, interrupt-driven mode
should work just fine but polled mode could be unreliable. There's not
much more we can do about this until interrupts are enabled during boot.
- In the above case, we also do one final check if the interrupt-driven
mode gets a timeout. If the status is complete, it will force the
system back into polled mode since interrupt mode doesn't work. For
polled mode during boot, if the status appears to be already complete
before beginning the check loop, it waits 10 us before actually checking
the status, just in case the EC is really slow and hasn't gotten to work
on the new request yet.
- Use upper-case hex for the _Qxx method
- Use device_printf for errors, don't hide them under verbose
- Increase default total timeout to 750 ms and decrease polling interval
to 5 us.
- Don't pass the status value via the softc. Just read it directly.
- Remove the mutex. We use the sx lock for transaction serialization
with the query handler.
- Remove the Intel copyright notice as no code of theirs was ever
present in this file (verified against rev 1.1)
- Allow KTR module-only builds for ease of testing
Thanks to jkim and Alexey Starikovskiy for helpful discussions and testing.
Approved by: re
MFC after: 2 weeks
2007-09-24 16:59:06 +00:00
|
|
|
CTR2(KTR_ACPI, "ec query ok,%s running _Q%02X", Data ? "" : " not", Data);
|
2003-07-10 17:22:46 +00:00
|
|
|
if (Data == 0)
|
Rewrite the EC driver event model. The main goal is to avoid
polling/interrupt-driven fallback and instead use polling only during
boot and pure interrupt-driven mode after boot. Polled mode could be
relegated completely to a legacy role if we could enable interrupts
during boot. Polled mode can be forced after boot by setting
debug.acpi.ec.polled="1", i.e. if there are timeouts.
- Use polling only during boot, shutdown, or if requested by the user.
Otherwise, use a generation count of GPEs, incremented atomically. This
prevents an old status value from being used if the EC is really slow
and the same condition (i.e. multiple IBEs for a write transaction) is
being checked.
- Check for and run the query handler directly if the SCI bit is set in
the status register during boot. Previously, the query handler wouldn't
run until interrupts were finally enabled late in boot.
- During boot and after starting a command, check if the event appears
to already have occurred before we even start waiting. If so, it's
possible the EC is very slow and we might accept an old status value.
Print a warning in this case. Once we've booted, interrupt-driven mode
should work just fine but polled mode could be unreliable. There's not
much more we can do about this until interrupts are enabled during boot.
- In the above case, we also do one final check if the interrupt-driven
mode gets a timeout. If the status is complete, it will force the
system back into polled mode since interrupt mode doesn't work. For
polled mode during boot, if the status appears to be already complete
before beginning the check loop, it waits 10 us before actually checking
the status, just in case the EC is really slow and hasn't gotten to work
on the new request yet.
- Use upper-case hex for the _Qxx method
- Use device_printf for errors, don't hide them under verbose
- Increase default total timeout to 750 ms and decrease polling interval
to 5 us.
- Don't pass the status value via the softc. Just read it directly.
- Remove the mutex. We use the sx lock for transaction serialization
with the query handler.
- Remove the Intel copyright notice as no code of theirs was ever
present in this file (verified against rev 1.1)
- Allow KTR module-only builds for ease of testing
Thanks to jkim and Alexey Starikovskiy for helpful discussions and testing.
Approved by: re
MFC after: 2 weeks
2007-09-24 16:59:06 +00:00
|
|
|
return;
|
2000-10-28 06:59:48 +00:00
|
|
|
|
2003-07-10 17:22:46 +00:00
|
|
|
/* Evaluate _Qxx to respond to the controller. */
|
Rewrite the EC driver event model. The main goal is to avoid
polling/interrupt-driven fallback and instead use polling only during
boot and pure interrupt-driven mode after boot. Polled mode could be
relegated completely to a legacy role if we could enable interrupts
during boot. Polled mode can be forced after boot by setting
debug.acpi.ec.polled="1", i.e. if there are timeouts.
- Use polling only during boot, shutdown, or if requested by the user.
Otherwise, use a generation count of GPEs, incremented atomically. This
prevents an old status value from being used if the EC is really slow
and the same condition (i.e. multiple IBEs for a write transaction) is
being checked.
- Check for and run the query handler directly if the SCI bit is set in
the status register during boot. Previously, the query handler wouldn't
run until interrupts were finally enabled late in boot.
- During boot and after starting a command, check if the event appears
to already have occurred before we even start waiting. If so, it's
possible the EC is very slow and we might accept an old status value.
Print a warning in this case. Once we've booted, interrupt-driven mode
should work just fine but polled mode could be unreliable. There's not
much more we can do about this until interrupts are enabled during boot.
- In the above case, we also do one final check if the interrupt-driven
mode gets a timeout. If the status is complete, it will force the
system back into polled mode since interrupt mode doesn't work. For
polled mode during boot, if the status appears to be already complete
before beginning the check loop, it waits 10 us before actually checking
the status, just in case the EC is really slow and hasn't gotten to work
on the new request yet.
- Use upper-case hex for the _Qxx method
- Use device_printf for errors, don't hide them under verbose
- Increase default total timeout to 750 ms and decrease polling interval
to 5 us.
- Don't pass the status value via the softc. Just read it directly.
- Remove the mutex. We use the sx lock for transaction serialization
with the query handler.
- Remove the Intel copyright notice as no code of theirs was ever
present in this file (verified against rev 1.1)
- Allow KTR module-only builds for ease of testing
Thanks to jkim and Alexey Starikovskiy for helpful discussions and testing.
Approved by: re
MFC after: 2 weeks
2007-09-24 16:59:06 +00:00
|
|
|
snprintf(qxx, sizeof(qxx), "_Q%02X", Data);
|
2005-04-14 06:50:13 +00:00
|
|
|
AcpiUtStrupr(qxx);
|
2003-07-10 17:22:46 +00:00
|
|
|
Status = AcpiEvaluateObject(sc->ec_handle, qxx, NULL, NULL);
|
|
|
|
if (ACPI_FAILURE(Status) && Status != AE_NOT_FOUND) {
|
Rewrite the EC driver event model. The main goal is to avoid
polling/interrupt-driven fallback and instead use polling only during
boot and pure interrupt-driven mode after boot. Polled mode could be
relegated completely to a legacy role if we could enable interrupts
during boot. Polled mode can be forced after boot by setting
debug.acpi.ec.polled="1", i.e. if there are timeouts.
- Use polling only during boot, shutdown, or if requested by the user.
Otherwise, use a generation count of GPEs, incremented atomically. This
prevents an old status value from being used if the EC is really slow
and the same condition (i.e. multiple IBEs for a write transaction) is
being checked.
- Check for and run the query handler directly if the SCI bit is set in
the status register during boot. Previously, the query handler wouldn't
run until interrupts were finally enabled late in boot.
- During boot and after starting a command, check if the event appears
to already have occurred before we even start waiting. If so, it's
possible the EC is very slow and we might accept an old status value.
Print a warning in this case. Once we've booted, interrupt-driven mode
should work just fine but polled mode could be unreliable. There's not
much more we can do about this until interrupts are enabled during boot.
- In the above case, we also do one final check if the interrupt-driven
mode gets a timeout. If the status is complete, it will force the
system back into polled mode since interrupt mode doesn't work. For
polled mode during boot, if the status appears to be already complete
before beginning the check loop, it waits 10 us before actually checking
the status, just in case the EC is really slow and hasn't gotten to work
on the new request yet.
- Use upper-case hex for the _Qxx method
- Use device_printf for errors, don't hide them under verbose
- Increase default total timeout to 750 ms and decrease polling interval
to 5 us.
- Don't pass the status value via the softc. Just read it directly.
- Remove the mutex. We use the sx lock for transaction serialization
with the query handler.
- Remove the Intel copyright notice as no code of theirs was ever
present in this file (verified against rev 1.1)
- Allow KTR module-only builds for ease of testing
Thanks to jkim and Alexey Starikovskiy for helpful discussions and testing.
Approved by: re
MFC after: 2 weeks
2007-09-24 16:59:06 +00:00
|
|
|
device_printf(sc->ec_dev, "evaluation of query method %s failed: %s\n",
|
|
|
|
qxx, AcpiFormatException(Status));
|
2000-10-28 06:59:48 +00:00
|
|
|
}
|
2000-12-14 12:43:22 +00:00
|
|
|
}
|
2001-01-31 09:32:44 +00:00
|
|
|
|
2001-06-29 20:31:37 +00:00
|
|
|
/*
|
Rewrite the EC driver event model. The main goal is to avoid
polling/interrupt-driven fallback and instead use polling only during
boot and pure interrupt-driven mode after boot. Polled mode could be
relegated completely to a legacy role if we could enable interrupts
during boot. Polled mode can be forced after boot by setting
debug.acpi.ec.polled="1", i.e. if there are timeouts.
- Use polling only during boot, shutdown, or if requested by the user.
Otherwise, use a generation count of GPEs, incremented atomically. This
prevents an old status value from being used if the EC is really slow
and the same condition (i.e. multiple IBEs for a write transaction) is
being checked.
- Check for and run the query handler directly if the SCI bit is set in
the status register during boot. Previously, the query handler wouldn't
run until interrupts were finally enabled late in boot.
- During boot and after starting a command, check if the event appears
to already have occurred before we even start waiting. If so, it's
possible the EC is very slow and we might accept an old status value.
Print a warning in this case. Once we've booted, interrupt-driven mode
should work just fine but polled mode could be unreliable. There's not
much more we can do about this until interrupts are enabled during boot.
- In the above case, we also do one final check if the interrupt-driven
mode gets a timeout. If the status is complete, it will force the
system back into polled mode since interrupt mode doesn't work. For
polled mode during boot, if the status appears to be already complete
before beginning the check loop, it waits 10 us before actually checking
the status, just in case the EC is really slow and hasn't gotten to work
on the new request yet.
- Use upper-case hex for the _Qxx method
- Use device_printf for errors, don't hide them under verbose
- Increase default total timeout to 750 ms and decrease polling interval
to 5 us.
- Don't pass the status value via the softc. Just read it directly.
- Remove the mutex. We use the sx lock for transaction serialization
with the query handler.
- Remove the Intel copyright notice as no code of theirs was ever
present in this file (verified against rev 1.1)
- Allow KTR module-only builds for ease of testing
Thanks to jkim and Alexey Starikovskiy for helpful discussions and testing.
Approved by: re
MFC after: 2 weeks
2007-09-24 16:59:06 +00:00
|
|
|
* The GPE handler is called when IBE/OBF or SCI events occur. We are
|
|
|
|
* called from an unknown lock context.
|
2001-06-29 20:31:37 +00:00
|
|
|
*/
|
2004-05-25 02:47:35 +00:00
|
|
|
static uint32_t
|
2001-06-29 20:31:37 +00:00
|
|
|
EcGpeHandler(void *Context)
|
2000-12-14 12:43:22 +00:00
|
|
|
{
|
2001-06-29 20:31:37 +00:00
|
|
|
struct acpi_ec_softc *sc = Context;
|
2003-07-10 17:22:46 +00:00
|
|
|
ACPI_STATUS Status;
|
2007-02-27 00:14:20 +00:00
|
|
|
EC_STATUS EcStatus;
|
2001-06-29 20:31:37 +00:00
|
|
|
|
2003-07-10 17:22:46 +00:00
|
|
|
KASSERT(Context != NULL, ("EcGpeHandler called with NULL"));
|
Rewrite the EC driver event model. The main goal is to avoid
polling/interrupt-driven fallback and instead use polling only during
boot and pure interrupt-driven mode after boot. Polled mode could be
relegated completely to a legacy role if we could enable interrupts
during boot. Polled mode can be forced after boot by setting
debug.acpi.ec.polled="1", i.e. if there are timeouts.
- Use polling only during boot, shutdown, or if requested by the user.
Otherwise, use a generation count of GPEs, incremented atomically. This
prevents an old status value from being used if the EC is really slow
and the same condition (i.e. multiple IBEs for a write transaction) is
being checked.
- Check for and run the query handler directly if the SCI bit is set in
the status register during boot. Previously, the query handler wouldn't
run until interrupts were finally enabled late in boot.
- During boot and after starting a command, check if the event appears
to already have occurred before we even start waiting. If so, it's
possible the EC is very slow and we might accept an old status value.
Print a warning in this case. Once we've booted, interrupt-driven mode
should work just fine but polled mode could be unreliable. There's not
much more we can do about this until interrupts are enabled during boot.
- In the above case, we also do one final check if the interrupt-driven
mode gets a timeout. If the status is complete, it will force the
system back into polled mode since interrupt mode doesn't work. For
polled mode during boot, if the status appears to be already complete
before beginning the check loop, it waits 10 us before actually checking
the status, just in case the EC is really slow and hasn't gotten to work
on the new request yet.
- Use upper-case hex for the _Qxx method
- Use device_printf for errors, don't hide them under verbose
- Increase default total timeout to 750 ms and decrease polling interval
to 5 us.
- Don't pass the status value via the softc. Just read it directly.
- Remove the mutex. We use the sx lock for transaction serialization
with the query handler.
- Remove the Intel copyright notice as no code of theirs was ever
present in this file (verified against rev 1.1)
- Allow KTR module-only builds for ease of testing
Thanks to jkim and Alexey Starikovskiy for helpful discussions and testing.
Approved by: re
MFC after: 2 weeks
2007-09-24 16:59:06 +00:00
|
|
|
CTR0(KTR_ACPI, "ec gpe handler start");
|
2003-07-10 17:22:46 +00:00
|
|
|
|
2005-02-21 23:38:41 +00:00
|
|
|
/*
|
Rewrite the EC driver event model. The main goal is to avoid
polling/interrupt-driven fallback and instead use polling only during
boot and pure interrupt-driven mode after boot. Polled mode could be
relegated completely to a legacy role if we could enable interrupts
during boot. Polled mode can be forced after boot by setting
debug.acpi.ec.polled="1", i.e. if there are timeouts.
- Use polling only during boot, shutdown, or if requested by the user.
Otherwise, use a generation count of GPEs, incremented atomically. This
prevents an old status value from being used if the EC is really slow
and the same condition (i.e. multiple IBEs for a write transaction) is
being checked.
- Check for and run the query handler directly if the SCI bit is set in
the status register during boot. Previously, the query handler wouldn't
run until interrupts were finally enabled late in boot.
- During boot and after starting a command, check if the event appears
to already have occurred before we even start waiting. If so, it's
possible the EC is very slow and we might accept an old status value.
Print a warning in this case. Once we've booted, interrupt-driven mode
should work just fine but polled mode could be unreliable. There's not
much more we can do about this until interrupts are enabled during boot.
- In the above case, we also do one final check if the interrupt-driven
mode gets a timeout. If the status is complete, it will force the
system back into polled mode since interrupt mode doesn't work. For
polled mode during boot, if the status appears to be already complete
before beginning the check loop, it waits 10 us before actually checking
the status, just in case the EC is really slow and hasn't gotten to work
on the new request yet.
- Use upper-case hex for the _Qxx method
- Use device_printf for errors, don't hide them under verbose
- Increase default total timeout to 750 ms and decrease polling interval
to 5 us.
- Don't pass the status value via the softc. Just read it directly.
- Remove the mutex. We use the sx lock for transaction serialization
with the query handler.
- Remove the Intel copyright notice as no code of theirs was ever
present in this file (verified against rev 1.1)
- Allow KTR module-only builds for ease of testing
Thanks to jkim and Alexey Starikovskiy for helpful discussions and testing.
Approved by: re
MFC after: 2 weeks
2007-09-24 16:59:06 +00:00
|
|
|
* Notify EcWaitEvent() that the status register is now fresh. If we
|
|
|
|
* didn't do this, it wouldn't be possible to distinguish an old IBE
|
|
|
|
* from a new one, for example when doing a write transaction (writing
|
|
|
|
* address and then data values.)
|
2005-02-21 23:38:41 +00:00
|
|
|
*/
|
Rewrite the EC driver event model. The main goal is to avoid
polling/interrupt-driven fallback and instead use polling only during
boot and pure interrupt-driven mode after boot. Polled mode could be
relegated completely to a legacy role if we could enable interrupts
during boot. Polled mode can be forced after boot by setting
debug.acpi.ec.polled="1", i.e. if there are timeouts.
- Use polling only during boot, shutdown, or if requested by the user.
Otherwise, use a generation count of GPEs, incremented atomically. This
prevents an old status value from being used if the EC is really slow
and the same condition (i.e. multiple IBEs for a write transaction) is
being checked.
- Check for and run the query handler directly if the SCI bit is set in
the status register during boot. Previously, the query handler wouldn't
run until interrupts were finally enabled late in boot.
- During boot and after starting a command, check if the event appears
to already have occurred before we even start waiting. If so, it's
possible the EC is very slow and we might accept an old status value.
Print a warning in this case. Once we've booted, interrupt-driven mode
should work just fine but polled mode could be unreliable. There's not
much more we can do about this until interrupts are enabled during boot.
- In the above case, we also do one final check if the interrupt-driven
mode gets a timeout. If the status is complete, it will force the
system back into polled mode since interrupt mode doesn't work. For
polled mode during boot, if the status appears to be already complete
before beginning the check loop, it waits 10 us before actually checking
the status, just in case the EC is really slow and hasn't gotten to work
on the new request yet.
- Use upper-case hex for the _Qxx method
- Use device_printf for errors, don't hide them under verbose
- Increase default total timeout to 750 ms and decrease polling interval
to 5 us.
- Don't pass the status value via the softc. Just read it directly.
- Remove the mutex. We use the sx lock for transaction serialization
with the query handler.
- Remove the Intel copyright notice as no code of theirs was ever
present in this file (verified against rev 1.1)
- Allow KTR module-only builds for ease of testing
Thanks to jkim and Alexey Starikovskiy for helpful discussions and testing.
Approved by: re
MFC after: 2 weeks
2007-09-24 16:59:06 +00:00
|
|
|
atomic_add_int(&sc->ec_gencount, 1);
|
|
|
|
wakeup(&sc->ec_gencount);
|
2004-05-25 02:47:35 +00:00
|
|
|
|
2007-02-27 00:14:20 +00:00
|
|
|
/*
|
Rewrite the EC driver event model. The main goal is to avoid
polling/interrupt-driven fallback and instead use polling only during
boot and pure interrupt-driven mode after boot. Polled mode could be
relegated completely to a legacy role if we could enable interrupts
during boot. Polled mode can be forced after boot by setting
debug.acpi.ec.polled="1", i.e. if there are timeouts.
- Use polling only during boot, shutdown, or if requested by the user.
Otherwise, use a generation count of GPEs, incremented atomically. This
prevents an old status value from being used if the EC is really slow
and the same condition (i.e. multiple IBEs for a write transaction) is
being checked.
- Check for and run the query handler directly if the SCI bit is set in
the status register during boot. Previously, the query handler wouldn't
run until interrupts were finally enabled late in boot.
- During boot and after starting a command, check if the event appears
to already have occurred before we even start waiting. If so, it's
possible the EC is very slow and we might accept an old status value.
Print a warning in this case. Once we've booted, interrupt-driven mode
should work just fine but polled mode could be unreliable. There's not
much more we can do about this until interrupts are enabled during boot.
- In the above case, we also do one final check if the interrupt-driven
mode gets a timeout. If the status is complete, it will force the
system back into polled mode since interrupt mode doesn't work. For
polled mode during boot, if the status appears to be already complete
before beginning the check loop, it waits 10 us before actually checking
the status, just in case the EC is really slow and hasn't gotten to work
on the new request yet.
- Use upper-case hex for the _Qxx method
- Use device_printf for errors, don't hide them under verbose
- Increase default total timeout to 750 ms and decrease polling interval
to 5 us.
- Don't pass the status value via the softc. Just read it directly.
- Remove the mutex. We use the sx lock for transaction serialization
with the query handler.
- Remove the Intel copyright notice as no code of theirs was ever
present in this file (verified against rev 1.1)
- Allow KTR module-only builds for ease of testing
Thanks to jkim and Alexey Starikovskiy for helpful discussions and testing.
Approved by: re
MFC after: 2 weeks
2007-09-24 16:59:06 +00:00
|
|
|
* If the EC_SCI bit of the status register is set, queue a query handler.
|
|
|
|
* It will run the query and _Qxx method later, under the lock.
|
2007-02-27 00:14:20 +00:00
|
|
|
*/
|
|
|
|
EcStatus = EC_GET_CSR(sc);
|
Rewrite the EC driver event model. The main goal is to avoid
polling/interrupt-driven fallback and instead use polling only during
boot and pure interrupt-driven mode after boot. Polled mode could be
relegated completely to a legacy role if we could enable interrupts
during boot. Polled mode can be forced after boot by setting
debug.acpi.ec.polled="1", i.e. if there are timeouts.
- Use polling only during boot, shutdown, or if requested by the user.
Otherwise, use a generation count of GPEs, incremented atomically. This
prevents an old status value from being used if the EC is really slow
and the same condition (i.e. multiple IBEs for a write transaction) is
being checked.
- Check for and run the query handler directly if the SCI bit is set in
the status register during boot. Previously, the query handler wouldn't
run until interrupts were finally enabled late in boot.
- During boot and after starting a command, check if the event appears
to already have occurred before we even start waiting. If so, it's
possible the EC is very slow and we might accept an old status value.
Print a warning in this case. Once we've booted, interrupt-driven mode
should work just fine but polled mode could be unreliable. There's not
much more we can do about this until interrupts are enabled during boot.
- In the above case, we also do one final check if the interrupt-driven
mode gets a timeout. If the status is complete, it will force the
system back into polled mode since interrupt mode doesn't work. For
polled mode during boot, if the status appears to be already complete
before beginning the check loop, it waits 10 us before actually checking
the status, just in case the EC is really slow and hasn't gotten to work
on the new request yet.
- Use upper-case hex for the _Qxx method
- Use device_printf for errors, don't hide them under verbose
- Increase default total timeout to 750 ms and decrease polling interval
to 5 us.
- Don't pass the status value via the softc. Just read it directly.
- Remove the mutex. We use the sx lock for transaction serialization
with the query handler.
- Remove the Intel copyright notice as no code of theirs was ever
present in this file (verified against rev 1.1)
- Allow KTR module-only builds for ease of testing
Thanks to jkim and Alexey Starikovskiy for helpful discussions and testing.
Approved by: re
MFC after: 2 weeks
2007-09-24 16:59:06 +00:00
|
|
|
if ((EcStatus & EC_EVENT_SCI) && !sc->ec_sci_pend) {
|
|
|
|
CTR0(KTR_ACPI, "ec gpe queueing query handler");
|
2007-03-22 18:16:43 +00:00
|
|
|
Status = AcpiOsExecute(OSL_GPE_HANDLER, EcGpeQueryHandler, Context);
|
Rewrite the EC driver event model. The main goal is to avoid
polling/interrupt-driven fallback and instead use polling only during
boot and pure interrupt-driven mode after boot. Polled mode could be
relegated completely to a legacy role if we could enable interrupts
during boot. Polled mode can be forced after boot by setting
debug.acpi.ec.polled="1", i.e. if there are timeouts.
- Use polling only during boot, shutdown, or if requested by the user.
Otherwise, use a generation count of GPEs, incremented atomically. This
prevents an old status value from being used if the EC is really slow
and the same condition (i.e. multiple IBEs for a write transaction) is
being checked.
- Check for and run the query handler directly if the SCI bit is set in
the status register during boot. Previously, the query handler wouldn't
run until interrupts were finally enabled late in boot.
- During boot and after starting a command, check if the event appears
to already have occurred before we even start waiting. If so, it's
possible the EC is very slow and we might accept an old status value.
Print a warning in this case. Once we've booted, interrupt-driven mode
should work just fine but polled mode could be unreliable. There's not
much more we can do about this until interrupts are enabled during boot.
- In the above case, we also do one final check if the interrupt-driven
mode gets a timeout. If the status is complete, it will force the
system back into polled mode since interrupt mode doesn't work. For
polled mode during boot, if the status appears to be already complete
before beginning the check loop, it waits 10 us before actually checking
the status, just in case the EC is really slow and hasn't gotten to work
on the new request yet.
- Use upper-case hex for the _Qxx method
- Use device_printf for errors, don't hide them under verbose
- Increase default total timeout to 750 ms and decrease polling interval
to 5 us.
- Don't pass the status value via the softc. Just read it directly.
- Remove the mutex. We use the sx lock for transaction serialization
with the query handler.
- Remove the Intel copyright notice as no code of theirs was ever
present in this file (verified against rev 1.1)
- Allow KTR module-only builds for ease of testing
Thanks to jkim and Alexey Starikovskiy for helpful discussions and testing.
Approved by: re
MFC after: 2 weeks
2007-09-24 16:59:06 +00:00
|
|
|
if (ACPI_SUCCESS(Status))
|
2007-02-27 00:14:20 +00:00
|
|
|
sc->ec_sci_pend = TRUE;
|
Rewrite the EC driver event model. The main goal is to avoid
polling/interrupt-driven fallback and instead use polling only during
boot and pure interrupt-driven mode after boot. Polled mode could be
relegated completely to a legacy role if we could enable interrupts
during boot. Polled mode can be forced after boot by setting
debug.acpi.ec.polled="1", i.e. if there are timeouts.
- Use polling only during boot, shutdown, or if requested by the user.
Otherwise, use a generation count of GPEs, incremented atomically. This
prevents an old status value from being used if the EC is really slow
and the same condition (i.e. multiple IBEs for a write transaction) is
being checked.
- Check for and run the query handler directly if the SCI bit is set in
the status register during boot. Previously, the query handler wouldn't
run until interrupts were finally enabled late in boot.
- During boot and after starting a command, check if the event appears
to already have occurred before we even start waiting. If so, it's
possible the EC is very slow and we might accept an old status value.
Print a warning in this case. Once we've booted, interrupt-driven mode
should work just fine but polled mode could be unreliable. There's not
much more we can do about this until interrupts are enabled during boot.
- In the above case, we also do one final check if the interrupt-driven
mode gets a timeout. If the status is complete, it will force the
system back into polled mode since interrupt mode doesn't work. For
polled mode during boot, if the status appears to be already complete
before beginning the check loop, it waits 10 us before actually checking
the status, just in case the EC is really slow and hasn't gotten to work
on the new request yet.
- Use upper-case hex for the _Qxx method
- Use device_printf for errors, don't hide them under verbose
- Increase default total timeout to 750 ms and decrease polling interval
to 5 us.
- Don't pass the status value via the softc. Just read it directly.
- Remove the mutex. We use the sx lock for transaction serialization
with the query handler.
- Remove the Intel copyright notice as no code of theirs was ever
present in this file (verified against rev 1.1)
- Allow KTR module-only builds for ease of testing
Thanks to jkim and Alexey Starikovskiy for helpful discussions and testing.
Approved by: re
MFC after: 2 weeks
2007-09-24 16:59:06 +00:00
|
|
|
else
|
|
|
|
printf("EcGpeHandler: queuing GPE query handler failed\n");
|
2007-02-27 00:14:20 +00:00
|
|
|
}
|
2004-05-25 02:47:35 +00:00
|
|
|
return (0);
|
2000-10-28 06:59:48 +00:00
|
|
|
}
|
|
|
|
|
|
|
|
static ACPI_STATUS
|
2003-06-27 21:57:42 +00:00
|
|
|
EcSpaceSetup(ACPI_HANDLE Region, UINT32 Function, void *Context,
|
|
|
|
void **RegionContext)
|
2000-10-28 06:59:48 +00:00
|
|
|
{
|
2001-01-31 09:32:44 +00:00
|
|
|
|
2002-05-19 06:16:47 +00:00
|
|
|
ACPI_FUNCTION_TRACE((char *)(uintptr_t)__func__);
|
2001-01-31 09:32:44 +00:00
|
|
|
|
2000-10-28 06:59:48 +00:00
|
|
|
/*
|
2004-01-03 02:01:39 +00:00
|
|
|
* If deactivating a region, always set the output to NULL. Otherwise,
|
|
|
|
* just pass the context through.
|
2000-10-28 06:59:48 +00:00
|
|
|
*/
|
2004-01-03 02:01:39 +00:00
|
|
|
if (Function == ACPI_REGION_DEACTIVATE)
|
|
|
|
*RegionContext = NULL;
|
|
|
|
else
|
|
|
|
*RegionContext = Context;
|
2000-10-28 06:59:48 +00:00
|
|
|
|
2003-06-27 21:57:42 +00:00
|
|
|
return_ACPI_STATUS (AE_OK);
|
2000-10-28 06:59:48 +00:00
|
|
|
}
|
|
|
|
|
|
|
|
static ACPI_STATUS
|
2003-06-27 21:57:42 +00:00
|
|
|
EcSpaceHandler(UINT32 Function, ACPI_PHYSICAL_ADDRESS Address, UINT32 width,
|
|
|
|
ACPI_INTEGER *Value, void *Context, void *RegionContext)
|
2000-10-28 06:59:48 +00:00
|
|
|
{
|
|
|
|
struct acpi_ec_softc *sc = (struct acpi_ec_softc *)Context;
|
2004-03-04 05:58:50 +00:00
|
|
|
ACPI_STATUS Status;
|
2003-07-10 17:22:46 +00:00
|
|
|
UINT8 EcAddr, EcData;
|
2001-02-26 20:36:56 +00:00
|
|
|
int i;
|
2000-10-28 06:59:48 +00:00
|
|
|
|
2002-05-19 06:16:47 +00:00
|
|
|
ACPI_FUNCTION_TRACE_U32((char *)(uintptr_t)__func__, (UINT32)Address);
|
2000-12-08 09:16:20 +00:00
|
|
|
|
2004-03-04 05:58:50 +00:00
|
|
|
if (width % 8 != 0 || Value == NULL || Context == NULL)
|
2003-07-10 17:22:46 +00:00
|
|
|
return_ACPI_STATUS (AE_BAD_PARAMETER);
|
2004-03-04 05:58:50 +00:00
|
|
|
if (Address + (width / 8) - 1 > 0xFF)
|
|
|
|
return_ACPI_STATUS (AE_BAD_ADDRESS);
|
2000-10-28 06:59:48 +00:00
|
|
|
|
2004-03-04 05:58:50 +00:00
|
|
|
if (Function == ACPI_READ)
|
|
|
|
*Value = 0;
|
2003-07-10 17:22:46 +00:00
|
|
|
EcAddr = Address;
|
2004-03-04 05:58:50 +00:00
|
|
|
Status = AE_ERROR;
|
|
|
|
|
Rewrite the EC driver event model. The main goal is to avoid
polling/interrupt-driven fallback and instead use polling only during
boot and pure interrupt-driven mode after boot. Polled mode could be
relegated completely to a legacy role if we could enable interrupts
during boot. Polled mode can be forced after boot by setting
debug.acpi.ec.polled="1", i.e. if there are timeouts.
- Use polling only during boot, shutdown, or if requested by the user.
Otherwise, use a generation count of GPEs, incremented atomically. This
prevents an old status value from being used if the EC is really slow
and the same condition (i.e. multiple IBEs for a write transaction) is
being checked.
- Check for and run the query handler directly if the SCI bit is set in
the status register during boot. Previously, the query handler wouldn't
run until interrupts were finally enabled late in boot.
- During boot and after starting a command, check if the event appears
to already have occurred before we even start waiting. If so, it's
possible the EC is very slow and we might accept an old status value.
Print a warning in this case. Once we've booted, interrupt-driven mode
should work just fine but polled mode could be unreliable. There's not
much more we can do about this until interrupts are enabled during boot.
- In the above case, we also do one final check if the interrupt-driven
mode gets a timeout. If the status is complete, it will force the
system back into polled mode since interrupt mode doesn't work. For
polled mode during boot, if the status appears to be already complete
before beginning the check loop, it waits 10 us before actually checking
the status, just in case the EC is really slow and hasn't gotten to work
on the new request yet.
- Use upper-case hex for the _Qxx method
- Use device_printf for errors, don't hide them under verbose
- Increase default total timeout to 750 ms and decrease polling interval
to 5 us.
- Don't pass the status value via the softc. Just read it directly.
- Remove the mutex. We use the sx lock for transaction serialization
with the query handler.
- Remove the Intel copyright notice as no code of theirs was ever
present in this file (verified against rev 1.1)
- Allow KTR module-only builds for ease of testing
Thanks to jkim and Alexey Starikovskiy for helpful discussions and testing.
Approved by: re
MFC after: 2 weeks
2007-09-24 16:59:06 +00:00
|
|
|
/*
|
|
|
|
* If booting, check if we need to run the query handler. If so, we
|
|
|
|
* we call it directly here since our thread taskq is not active yet.
|
|
|
|
*/
|
2009-03-17 00:48:11 +00:00
|
|
|
if (cold || rebooting || sc->ec_suspending) {
|
Rewrite the EC driver event model. The main goal is to avoid
polling/interrupt-driven fallback and instead use polling only during
boot and pure interrupt-driven mode after boot. Polled mode could be
relegated completely to a legacy role if we could enable interrupts
during boot. Polled mode can be forced after boot by setting
debug.acpi.ec.polled="1", i.e. if there are timeouts.
- Use polling only during boot, shutdown, or if requested by the user.
Otherwise, use a generation count of GPEs, incremented atomically. This
prevents an old status value from being used if the EC is really slow
and the same condition (i.e. multiple IBEs for a write transaction) is
being checked.
- Check for and run the query handler directly if the SCI bit is set in
the status register during boot. Previously, the query handler wouldn't
run until interrupts were finally enabled late in boot.
- During boot and after starting a command, check if the event appears
to already have occurred before we even start waiting. If so, it's
possible the EC is very slow and we might accept an old status value.
Print a warning in this case. Once we've booted, interrupt-driven mode
should work just fine but polled mode could be unreliable. There's not
much more we can do about this until interrupts are enabled during boot.
- In the above case, we also do one final check if the interrupt-driven
mode gets a timeout. If the status is complete, it will force the
system back into polled mode since interrupt mode doesn't work. For
polled mode during boot, if the status appears to be already complete
before beginning the check loop, it waits 10 us before actually checking
the status, just in case the EC is really slow and hasn't gotten to work
on the new request yet.
- Use upper-case hex for the _Qxx method
- Use device_printf for errors, don't hide them under verbose
- Increase default total timeout to 750 ms and decrease polling interval
to 5 us.
- Don't pass the status value via the softc. Just read it directly.
- Remove the mutex. We use the sx lock for transaction serialization
with the query handler.
- Remove the Intel copyright notice as no code of theirs was ever
present in this file (verified against rev 1.1)
- Allow KTR module-only builds for ease of testing
Thanks to jkim and Alexey Starikovskiy for helpful discussions and testing.
Approved by: re
MFC after: 2 weeks
2007-09-24 16:59:06 +00:00
|
|
|
if ((EC_GET_CSR(sc) & EC_EVENT_SCI)) {
|
|
|
|
CTR0(KTR_ACPI, "ec running gpe handler directly");
|
|
|
|
EcGpeQueryHandler(sc);
|
|
|
|
}
|
|
|
|
}
|
|
|
|
|
|
|
|
/* Serialize with EcGpeQueryHandler() at transaction granularity. */
|
|
|
|
Status = EcLock(sc);
|
2004-08-13 06:21:53 +00:00
|
|
|
if (ACPI_FAILURE(Status))
|
|
|
|
return_ACPI_STATUS (Status);
|
|
|
|
|
2004-03-04 05:58:50 +00:00
|
|
|
/* Perform the transaction(s), based on width. */
|
|
|
|
for (i = 0; i < width; i += 8, EcAddr++) {
|
2003-07-10 17:22:46 +00:00
|
|
|
switch (Function) {
|
|
|
|
case ACPI_READ:
|
|
|
|
Status = EcRead(sc, EcAddr, &EcData);
|
2004-03-04 05:58:50 +00:00
|
|
|
if (ACPI_SUCCESS(Status))
|
|
|
|
*Value |= ((ACPI_INTEGER)EcData) << i;
|
2003-07-10 17:22:46 +00:00
|
|
|
break;
|
|
|
|
case ACPI_WRITE:
|
|
|
|
EcData = (UINT8)((*Value) >> i);
|
|
|
|
Status = EcWrite(sc, EcAddr, &EcData);
|
|
|
|
break;
|
|
|
|
default:
|
|
|
|
device_printf(sc->ec_dev, "invalid EcSpaceHandler function %d\n",
|
|
|
|
Function);
|
|
|
|
Status = AE_BAD_PARAMETER;
|
|
|
|
break;
|
|
|
|
}
|
|
|
|
if (ACPI_FAILURE(Status))
|
2004-03-04 05:58:50 +00:00
|
|
|
break;
|
2000-12-14 12:43:22 +00:00
|
|
|
}
|
2004-03-04 05:58:50 +00:00
|
|
|
|
2004-08-13 06:21:53 +00:00
|
|
|
EcUnlock(sc);
|
2003-07-10 17:22:46 +00:00
|
|
|
return_ACPI_STATUS (Status);
|
2000-12-14 12:43:22 +00:00
|
|
|
}
|
2000-10-28 06:59:48 +00:00
|
|
|
|
|
|
|
static ACPI_STATUS
|
Rewrite the EC driver event model. The main goal is to avoid
polling/interrupt-driven fallback and instead use polling only during
boot and pure interrupt-driven mode after boot. Polled mode could be
relegated completely to a legacy role if we could enable interrupts
during boot. Polled mode can be forced after boot by setting
debug.acpi.ec.polled="1", i.e. if there are timeouts.
- Use polling only during boot, shutdown, or if requested by the user.
Otherwise, use a generation count of GPEs, incremented atomically. This
prevents an old status value from being used if the EC is really slow
and the same condition (i.e. multiple IBEs for a write transaction) is
being checked.
- Check for and run the query handler directly if the SCI bit is set in
the status register during boot. Previously, the query handler wouldn't
run until interrupts were finally enabled late in boot.
- During boot and after starting a command, check if the event appears
to already have occurred before we even start waiting. If so, it's
possible the EC is very slow and we might accept an old status value.
Print a warning in this case. Once we've booted, interrupt-driven mode
should work just fine but polled mode could be unreliable. There's not
much more we can do about this until interrupts are enabled during boot.
- In the above case, we also do one final check if the interrupt-driven
mode gets a timeout. If the status is complete, it will force the
system back into polled mode since interrupt mode doesn't work. For
polled mode during boot, if the status appears to be already complete
before beginning the check loop, it waits 10 us before actually checking
the status, just in case the EC is really slow and hasn't gotten to work
on the new request yet.
- Use upper-case hex for the _Qxx method
- Use device_printf for errors, don't hide them under verbose
- Increase default total timeout to 750 ms and decrease polling interval
to 5 us.
- Don't pass the status value via the softc. Just read it directly.
- Remove the mutex. We use the sx lock for transaction serialization
with the query handler.
- Remove the Intel copyright notice as no code of theirs was ever
present in this file (verified against rev 1.1)
- Allow KTR module-only builds for ease of testing
Thanks to jkim and Alexey Starikovskiy for helpful discussions and testing.
Approved by: re
MFC after: 2 weeks
2007-09-24 16:59:06 +00:00
|
|
|
EcCheckStatus(struct acpi_ec_softc *sc, const char *msg, EC_EVENT event)
|
|
|
|
{
|
|
|
|
ACPI_STATUS status;
|
|
|
|
EC_STATUS ec_status;
|
|
|
|
|
|
|
|
status = AE_NO_HARDWARE_RESPONSE;
|
|
|
|
ec_status = EC_GET_CSR(sc);
|
|
|
|
if (sc->ec_burstactive && !(ec_status & EC_FLAG_BURST_MODE)) {
|
|
|
|
CTR1(KTR_ACPI, "ec burst disabled in waitevent (%s)", msg);
|
|
|
|
sc->ec_burstactive = FALSE;
|
|
|
|
}
|
|
|
|
if (EVENT_READY(event, ec_status)) {
|
|
|
|
CTR2(KTR_ACPI, "ec %s wait ready, status %#x", msg, ec_status);
|
|
|
|
status = AE_OK;
|
|
|
|
}
|
|
|
|
return (status);
|
|
|
|
}
|
|
|
|
|
|
|
|
static ACPI_STATUS
|
|
|
|
EcWaitEvent(struct acpi_ec_softc *sc, EC_EVENT Event, u_int gen_count)
|
2000-10-28 06:59:48 +00:00
|
|
|
{
|
2003-07-10 17:22:46 +00:00
|
|
|
ACPI_STATUS Status;
|
Rewrite the EC driver event model. The main goal is to avoid
polling/interrupt-driven fallback and instead use polling only during
boot and pure interrupt-driven mode after boot. Polled mode could be
relegated completely to a legacy role if we could enable interrupts
during boot. Polled mode can be forced after boot by setting
debug.acpi.ec.polled="1", i.e. if there are timeouts.
- Use polling only during boot, shutdown, or if requested by the user.
Otherwise, use a generation count of GPEs, incremented atomically. This
prevents an old status value from being used if the EC is really slow
and the same condition (i.e. multiple IBEs for a write transaction) is
being checked.
- Check for and run the query handler directly if the SCI bit is set in
the status register during boot. Previously, the query handler wouldn't
run until interrupts were finally enabled late in boot.
- During boot and after starting a command, check if the event appears
to already have occurred before we even start waiting. If so, it's
possible the EC is very slow and we might accept an old status value.
Print a warning in this case. Once we've booted, interrupt-driven mode
should work just fine but polled mode could be unreliable. There's not
much more we can do about this until interrupts are enabled during boot.
- In the above case, we also do one final check if the interrupt-driven
mode gets a timeout. If the status is complete, it will force the
system back into polled mode since interrupt mode doesn't work. For
polled mode during boot, if the status appears to be already complete
before beginning the check loop, it waits 10 us before actually checking
the status, just in case the EC is really slow and hasn't gotten to work
on the new request yet.
- Use upper-case hex for the _Qxx method
- Use device_printf for errors, don't hide them under verbose
- Increase default total timeout to 750 ms and decrease polling interval
to 5 us.
- Don't pass the status value via the softc. Just read it directly.
- Remove the mutex. We use the sx lock for transaction serialization
with the query handler.
- Remove the Intel copyright notice as no code of theirs was ever
present in this file (verified against rev 1.1)
- Allow KTR module-only builds for ease of testing
Thanks to jkim and Alexey Starikovskiy for helpful discussions and testing.
Approved by: re
MFC after: 2 weeks
2007-09-24 16:59:06 +00:00
|
|
|
int count, i, slp_ival;
|
2000-10-28 06:59:48 +00:00
|
|
|
|
2004-08-13 06:21:53 +00:00
|
|
|
ACPI_SERIAL_ASSERT(ec);
|
2003-07-10 17:22:46 +00:00
|
|
|
Status = AE_NO_HARDWARE_RESPONSE;
|
2007-10-25 17:03:07 +00:00
|
|
|
int need_poll = cold || rebooting || ec_polled_mode || sc->ec_suspending;
|
2004-06-28 20:41:30 +00:00
|
|
|
/*
|
Rewrite the EC driver event model. The main goal is to avoid
polling/interrupt-driven fallback and instead use polling only during
boot and pure interrupt-driven mode after boot. Polled mode could be
relegated completely to a legacy role if we could enable interrupts
during boot. Polled mode can be forced after boot by setting
debug.acpi.ec.polled="1", i.e. if there are timeouts.
- Use polling only during boot, shutdown, or if requested by the user.
Otherwise, use a generation count of GPEs, incremented atomically. This
prevents an old status value from being used if the EC is really slow
and the same condition (i.e. multiple IBEs for a write transaction) is
being checked.
- Check for and run the query handler directly if the SCI bit is set in
the status register during boot. Previously, the query handler wouldn't
run until interrupts were finally enabled late in boot.
- During boot and after starting a command, check if the event appears
to already have occurred before we even start waiting. If so, it's
possible the EC is very slow and we might accept an old status value.
Print a warning in this case. Once we've booted, interrupt-driven mode
should work just fine but polled mode could be unreliable. There's not
much more we can do about this until interrupts are enabled during boot.
- In the above case, we also do one final check if the interrupt-driven
mode gets a timeout. If the status is complete, it will force the
system back into polled mode since interrupt mode doesn't work. For
polled mode during boot, if the status appears to be already complete
before beginning the check loop, it waits 10 us before actually checking
the status, just in case the EC is really slow and hasn't gotten to work
on the new request yet.
- Use upper-case hex for the _Qxx method
- Use device_printf for errors, don't hide them under verbose
- Increase default total timeout to 750 ms and decrease polling interval
to 5 us.
- Don't pass the status value via the softc. Just read it directly.
- Remove the mutex. We use the sx lock for transaction serialization
with the query handler.
- Remove the Intel copyright notice as no code of theirs was ever
present in this file (verified against rev 1.1)
- Allow KTR module-only builds for ease of testing
Thanks to jkim and Alexey Starikovskiy for helpful discussions and testing.
Approved by: re
MFC after: 2 weeks
2007-09-24 16:59:06 +00:00
|
|
|
* The main CPU should be much faster than the EC. So the status should
|
|
|
|
* be "not ready" when we start waiting. But if the main CPU is really
|
|
|
|
* slow, it's possible we see the current "ready" response. Since that
|
|
|
|
* can't be distinguished from the previous response in polled mode,
|
|
|
|
* this is a potential issue. We really should have interrupts enabled
|
|
|
|
* during boot so there is no ambiguity in polled mode.
|
|
|
|
*
|
|
|
|
* If this occurs, we add an additional delay before actually entering
|
|
|
|
* the status checking loop, hopefully to allow the EC to go to work
|
|
|
|
* and produce a non-stale status.
|
2000-10-28 06:59:48 +00:00
|
|
|
*/
|
2007-10-25 17:03:07 +00:00
|
|
|
if (need_poll) {
|
Rewrite the EC driver event model. The main goal is to avoid
polling/interrupt-driven fallback and instead use polling only during
boot and pure interrupt-driven mode after boot. Polled mode could be
relegated completely to a legacy role if we could enable interrupts
during boot. Polled mode can be forced after boot by setting
debug.acpi.ec.polled="1", i.e. if there are timeouts.
- Use polling only during boot, shutdown, or if requested by the user.
Otherwise, use a generation count of GPEs, incremented atomically. This
prevents an old status value from being used if the EC is really slow
and the same condition (i.e. multiple IBEs for a write transaction) is
being checked.
- Check for and run the query handler directly if the SCI bit is set in
the status register during boot. Previously, the query handler wouldn't
run until interrupts were finally enabled late in boot.
- During boot and after starting a command, check if the event appears
to already have occurred before we even start waiting. If so, it's
possible the EC is very slow and we might accept an old status value.
Print a warning in this case. Once we've booted, interrupt-driven mode
should work just fine but polled mode could be unreliable. There's not
much more we can do about this until interrupts are enabled during boot.
- In the above case, we also do one final check if the interrupt-driven
mode gets a timeout. If the status is complete, it will force the
system back into polled mode since interrupt mode doesn't work. For
polled mode during boot, if the status appears to be already complete
before beginning the check loop, it waits 10 us before actually checking
the status, just in case the EC is really slow and hasn't gotten to work
on the new request yet.
- Use upper-case hex for the _Qxx method
- Use device_printf for errors, don't hide them under verbose
- Increase default total timeout to 750 ms and decrease polling interval
to 5 us.
- Don't pass the status value via the softc. Just read it directly.
- Remove the mutex. We use the sx lock for transaction serialization
with the query handler.
- Remove the Intel copyright notice as no code of theirs was ever
present in this file (verified against rev 1.1)
- Allow KTR module-only builds for ease of testing
Thanks to jkim and Alexey Starikovskiy for helpful discussions and testing.
Approved by: re
MFC after: 2 weeks
2007-09-24 16:59:06 +00:00
|
|
|
static int once;
|
|
|
|
|
|
|
|
if (EcCheckStatus(sc, "pre-check", Event) == AE_OK) {
|
|
|
|
if (!once) {
|
|
|
|
device_printf(sc->ec_dev,
|
|
|
|
"warning: EC done before starting event wait\n");
|
|
|
|
once = 1;
|
|
|
|
}
|
|
|
|
AcpiOsStall(10);
|
2003-07-10 17:22:46 +00:00
|
|
|
}
|
2000-10-28 06:59:48 +00:00
|
|
|
}
|
2003-07-15 19:24:36 +00:00
|
|
|
|
Rewrite the EC driver event model. The main goal is to avoid
polling/interrupt-driven fallback and instead use polling only during
boot and pure interrupt-driven mode after boot. Polled mode could be
relegated completely to a legacy role if we could enable interrupts
during boot. Polled mode can be forced after boot by setting
debug.acpi.ec.polled="1", i.e. if there are timeouts.
- Use polling only during boot, shutdown, or if requested by the user.
Otherwise, use a generation count of GPEs, incremented atomically. This
prevents an old status value from being used if the EC is really slow
and the same condition (i.e. multiple IBEs for a write transaction) is
being checked.
- Check for and run the query handler directly if the SCI bit is set in
the status register during boot. Previously, the query handler wouldn't
run until interrupts were finally enabled late in boot.
- During boot and after starting a command, check if the event appears
to already have occurred before we even start waiting. If so, it's
possible the EC is very slow and we might accept an old status value.
Print a warning in this case. Once we've booted, interrupt-driven mode
should work just fine but polled mode could be unreliable. There's not
much more we can do about this until interrupts are enabled during boot.
- In the above case, we also do one final check if the interrupt-driven
mode gets a timeout. If the status is complete, it will force the
system back into polled mode since interrupt mode doesn't work. For
polled mode during boot, if the status appears to be already complete
before beginning the check loop, it waits 10 us before actually checking
the status, just in case the EC is really slow and hasn't gotten to work
on the new request yet.
- Use upper-case hex for the _Qxx method
- Use device_printf for errors, don't hide them under verbose
- Increase default total timeout to 750 ms and decrease polling interval
to 5 us.
- Don't pass the status value via the softc. Just read it directly.
- Remove the mutex. We use the sx lock for transaction serialization
with the query handler.
- Remove the Intel copyright notice as no code of theirs was ever
present in this file (verified against rev 1.1)
- Allow KTR module-only builds for ease of testing
Thanks to jkim and Alexey Starikovskiy for helpful discussions and testing.
Approved by: re
MFC after: 2 weeks
2007-09-24 16:59:06 +00:00
|
|
|
/* Wait for event by polling or GPE (interrupt). */
|
2007-10-25 17:03:07 +00:00
|
|
|
if (need_poll) {
|
Rewrite the EC driver event model. The main goal is to avoid
polling/interrupt-driven fallback and instead use polling only during
boot and pure interrupt-driven mode after boot. Polled mode could be
relegated completely to a legacy role if we could enable interrupts
during boot. Polled mode can be forced after boot by setting
debug.acpi.ec.polled="1", i.e. if there are timeouts.
- Use polling only during boot, shutdown, or if requested by the user.
Otherwise, use a generation count of GPEs, incremented atomically. This
prevents an old status value from being used if the EC is really slow
and the same condition (i.e. multiple IBEs for a write transaction) is
being checked.
- Check for and run the query handler directly if the SCI bit is set in
the status register during boot. Previously, the query handler wouldn't
run until interrupts were finally enabled late in boot.
- During boot and after starting a command, check if the event appears
to already have occurred before we even start waiting. If so, it's
possible the EC is very slow and we might accept an old status value.
Print a warning in this case. Once we've booted, interrupt-driven mode
should work just fine but polled mode could be unreliable. There's not
much more we can do about this until interrupts are enabled during boot.
- In the above case, we also do one final check if the interrupt-driven
mode gets a timeout. If the status is complete, it will force the
system back into polled mode since interrupt mode doesn't work. For
polled mode during boot, if the status appears to be already complete
before beginning the check loop, it waits 10 us before actually checking
the status, just in case the EC is really slow and hasn't gotten to work
on the new request yet.
- Use upper-case hex for the _Qxx method
- Use device_printf for errors, don't hide them under verbose
- Increase default total timeout to 750 ms and decrease polling interval
to 5 us.
- Don't pass the status value via the softc. Just read it directly.
- Remove the mutex. We use the sx lock for transaction serialization
with the query handler.
- Remove the Intel copyright notice as no code of theirs was ever
present in this file (verified against rev 1.1)
- Allow KTR module-only builds for ease of testing
Thanks to jkim and Alexey Starikovskiy for helpful discussions and testing.
Approved by: re
MFC after: 2 weeks
2007-09-24 16:59:06 +00:00
|
|
|
count = (ec_timeout * 1000) / EC_POLL_DELAY;
|
|
|
|
if (count == 0)
|
|
|
|
count = 1;
|
2004-07-01 00:51:31 +00:00
|
|
|
for (i = 0; i < count; i++) {
|
Rewrite the EC driver event model. The main goal is to avoid
polling/interrupt-driven fallback and instead use polling only during
boot and pure interrupt-driven mode after boot. Polled mode could be
relegated completely to a legacy role if we could enable interrupts
during boot. Polled mode can be forced after boot by setting
debug.acpi.ec.polled="1", i.e. if there are timeouts.
- Use polling only during boot, shutdown, or if requested by the user.
Otherwise, use a generation count of GPEs, incremented atomically. This
prevents an old status value from being used if the EC is really slow
and the same condition (i.e. multiple IBEs for a write transaction) is
being checked.
- Check for and run the query handler directly if the SCI bit is set in
the status register during boot. Previously, the query handler wouldn't
run until interrupts were finally enabled late in boot.
- During boot and after starting a command, check if the event appears
to already have occurred before we even start waiting. If so, it's
possible the EC is very slow and we might accept an old status value.
Print a warning in this case. Once we've booted, interrupt-driven mode
should work just fine but polled mode could be unreliable. There's not
much more we can do about this until interrupts are enabled during boot.
- In the above case, we also do one final check if the interrupt-driven
mode gets a timeout. If the status is complete, it will force the
system back into polled mode since interrupt mode doesn't work. For
polled mode during boot, if the status appears to be already complete
before beginning the check loop, it waits 10 us before actually checking
the status, just in case the EC is really slow and hasn't gotten to work
on the new request yet.
- Use upper-case hex for the _Qxx method
- Use device_printf for errors, don't hide them under verbose
- Increase default total timeout to 750 ms and decrease polling interval
to 5 us.
- Don't pass the status value via the softc. Just read it directly.
- Remove the mutex. We use the sx lock for transaction serialization
with the query handler.
- Remove the Intel copyright notice as no code of theirs was ever
present in this file (verified against rev 1.1)
- Allow KTR module-only builds for ease of testing
Thanks to jkim and Alexey Starikovskiy for helpful discussions and testing.
Approved by: re
MFC after: 2 weeks
2007-09-24 16:59:06 +00:00
|
|
|
Status = EcCheckStatus(sc, "poll", Event);
|
|
|
|
if (Status == AE_OK)
|
2003-07-15 19:24:36 +00:00
|
|
|
break;
|
Rewrite the EC driver event model. The main goal is to avoid
polling/interrupt-driven fallback and instead use polling only during
boot and pure interrupt-driven mode after boot. Polled mode could be
relegated completely to a legacy role if we could enable interrupts
during boot. Polled mode can be forced after boot by setting
debug.acpi.ec.polled="1", i.e. if there are timeouts.
- Use polling only during boot, shutdown, or if requested by the user.
Otherwise, use a generation count of GPEs, incremented atomically. This
prevents an old status value from being used if the EC is really slow
and the same condition (i.e. multiple IBEs for a write transaction) is
being checked.
- Check for and run the query handler directly if the SCI bit is set in
the status register during boot. Previously, the query handler wouldn't
run until interrupts were finally enabled late in boot.
- During boot and after starting a command, check if the event appears
to already have occurred before we even start waiting. If so, it's
possible the EC is very slow and we might accept an old status value.
Print a warning in this case. Once we've booted, interrupt-driven mode
should work just fine but polled mode could be unreliable. There's not
much more we can do about this until interrupts are enabled during boot.
- In the above case, we also do one final check if the interrupt-driven
mode gets a timeout. If the status is complete, it will force the
system back into polled mode since interrupt mode doesn't work. For
polled mode during boot, if the status appears to be already complete
before beginning the check loop, it waits 10 us before actually checking
the status, just in case the EC is really slow and hasn't gotten to work
on the new request yet.
- Use upper-case hex for the _Qxx method
- Use device_printf for errors, don't hide them under verbose
- Increase default total timeout to 750 ms and decrease polling interval
to 5 us.
- Don't pass the status value via the softc. Just read it directly.
- Remove the mutex. We use the sx lock for transaction serialization
with the query handler.
- Remove the Intel copyright notice as no code of theirs was ever
present in this file (verified against rev 1.1)
- Allow KTR module-only builds for ease of testing
Thanks to jkim and Alexey Starikovskiy for helpful discussions and testing.
Approved by: re
MFC after: 2 weeks
2007-09-24 16:59:06 +00:00
|
|
|
AcpiOsStall(EC_POLL_DELAY);
|
|
|
|
}
|
|
|
|
} else {
|
|
|
|
slp_ival = hz / 1000;
|
|
|
|
if (slp_ival != 0) {
|
|
|
|
count = ec_timeout;
|
|
|
|
} else {
|
|
|
|
/* hz has less than 1 ms resolution so scale timeout. */
|
|
|
|
slp_ival = 1;
|
|
|
|
count = ec_timeout / (1000 / hz);
|
|
|
|
}
|
|
|
|
|
|
|
|
/*
|
|
|
|
* Wait for the GPE to signal the status changed, checking the
|
|
|
|
* status register each time we get one. It's possible to get a
|
|
|
|
* GPE for an event we're not interested in here (i.e., SCI for
|
|
|
|
* EC query).
|
|
|
|
*/
|
|
|
|
for (i = 0; i < count; i++) {
|
|
|
|
if (gen_count != sc->ec_gencount) {
|
|
|
|
/*
|
|
|
|
* Record new generation count. It's possible the GPE was
|
|
|
|
* just to notify us that a query is needed and we need to
|
|
|
|
* wait for a second GPE to signal the completion of the
|
|
|
|
* event we are actually waiting for.
|
|
|
|
*/
|
|
|
|
gen_count = sc->ec_gencount;
|
|
|
|
Status = EcCheckStatus(sc, "sleep", Event);
|
|
|
|
if (Status == AE_OK)
|
|
|
|
break;
|
2003-07-15 19:24:36 +00:00
|
|
|
}
|
Rewrite the EC driver event model. The main goal is to avoid
polling/interrupt-driven fallback and instead use polling only during
boot and pure interrupt-driven mode after boot. Polled mode could be
relegated completely to a legacy role if we could enable interrupts
during boot. Polled mode can be forced after boot by setting
debug.acpi.ec.polled="1", i.e. if there are timeouts.
- Use polling only during boot, shutdown, or if requested by the user.
Otherwise, use a generation count of GPEs, incremented atomically. This
prevents an old status value from being used if the EC is really slow
and the same condition (i.e. multiple IBEs for a write transaction) is
being checked.
- Check for and run the query handler directly if the SCI bit is set in
the status register during boot. Previously, the query handler wouldn't
run until interrupts were finally enabled late in boot.
- During boot and after starting a command, check if the event appears
to already have occurred before we even start waiting. If so, it's
possible the EC is very slow and we might accept an old status value.
Print a warning in this case. Once we've booted, interrupt-driven mode
should work just fine but polled mode could be unreliable. There's not
much more we can do about this until interrupts are enabled during boot.
- In the above case, we also do one final check if the interrupt-driven
mode gets a timeout. If the status is complete, it will force the
system back into polled mode since interrupt mode doesn't work. For
polled mode during boot, if the status appears to be already complete
before beginning the check loop, it waits 10 us before actually checking
the status, just in case the EC is really slow and hasn't gotten to work
on the new request yet.
- Use upper-case hex for the _Qxx method
- Use device_printf for errors, don't hide them under verbose
- Increase default total timeout to 750 ms and decrease polling interval
to 5 us.
- Don't pass the status value via the softc. Just read it directly.
- Remove the mutex. We use the sx lock for transaction serialization
with the query handler.
- Remove the Intel copyright notice as no code of theirs was ever
present in this file (verified against rev 1.1)
- Allow KTR module-only builds for ease of testing
Thanks to jkim and Alexey Starikovskiy for helpful discussions and testing.
Approved by: re
MFC after: 2 weeks
2007-09-24 16:59:06 +00:00
|
|
|
tsleep(&sc->ec_gencount, PZERO, "ecgpe", slp_ival);
|
2003-07-15 19:24:36 +00:00
|
|
|
}
|
2003-07-10 17:22:46 +00:00
|
|
|
|
Rewrite the EC driver event model. The main goal is to avoid
polling/interrupt-driven fallback and instead use polling only during
boot and pure interrupt-driven mode after boot. Polled mode could be
relegated completely to a legacy role if we could enable interrupts
during boot. Polled mode can be forced after boot by setting
debug.acpi.ec.polled="1", i.e. if there are timeouts.
- Use polling only during boot, shutdown, or if requested by the user.
Otherwise, use a generation count of GPEs, incremented atomically. This
prevents an old status value from being used if the EC is really slow
and the same condition (i.e. multiple IBEs for a write transaction) is
being checked.
- Check for and run the query handler directly if the SCI bit is set in
the status register during boot. Previously, the query handler wouldn't
run until interrupts were finally enabled late in boot.
- During boot and after starting a command, check if the event appears
to already have occurred before we even start waiting. If so, it's
possible the EC is very slow and we might accept an old status value.
Print a warning in this case. Once we've booted, interrupt-driven mode
should work just fine but polled mode could be unreliable. There's not
much more we can do about this until interrupts are enabled during boot.
- In the above case, we also do one final check if the interrupt-driven
mode gets a timeout. If the status is complete, it will force the
system back into polled mode since interrupt mode doesn't work. For
polled mode during boot, if the status appears to be already complete
before beginning the check loop, it waits 10 us before actually checking
the status, just in case the EC is really slow and hasn't gotten to work
on the new request yet.
- Use upper-case hex for the _Qxx method
- Use device_printf for errors, don't hide them under verbose
- Increase default total timeout to 750 ms and decrease polling interval
to 5 us.
- Don't pass the status value via the softc. Just read it directly.
- Remove the mutex. We use the sx lock for transaction serialization
with the query handler.
- Remove the Intel copyright notice as no code of theirs was ever
present in this file (verified against rev 1.1)
- Allow KTR module-only builds for ease of testing
Thanks to jkim and Alexey Starikovskiy for helpful discussions and testing.
Approved by: re
MFC after: 2 weeks
2007-09-24 16:59:06 +00:00
|
|
|
/*
|
|
|
|
* We finished waiting for the GPE and it never arrived. Try to
|
|
|
|
* read the register once and trust whatever value we got. This is
|
|
|
|
* the best we can do at this point. Then, force polled mode on
|
|
|
|
* since this system doesn't appear to generate GPEs.
|
|
|
|
*/
|
|
|
|
if (Status != AE_OK) {
|
|
|
|
Status = EcCheckStatus(sc, "sleep_end", Event);
|
|
|
|
device_printf(sc->ec_dev,
|
|
|
|
"wait timed out (%sresponse), forcing polled mode\n",
|
|
|
|
Status == AE_OK ? "" : "no ");
|
|
|
|
ec_polled_mode = TRUE;
|
|
|
|
}
|
|
|
|
}
|
|
|
|
if (Status != AE_OK)
|
|
|
|
CTR0(KTR_ACPI, "error: ec wait timed out");
|
2003-06-27 21:57:42 +00:00
|
|
|
return (Status);
|
Rewrite the EC driver event model. The main goal is to avoid
polling/interrupt-driven fallback and instead use polling only during
boot and pure interrupt-driven mode after boot. Polled mode could be
relegated completely to a legacy role if we could enable interrupts
during boot. Polled mode can be forced after boot by setting
debug.acpi.ec.polled="1", i.e. if there are timeouts.
- Use polling only during boot, shutdown, or if requested by the user.
Otherwise, use a generation count of GPEs, incremented atomically. This
prevents an old status value from being used if the EC is really slow
and the same condition (i.e. multiple IBEs for a write transaction) is
being checked.
- Check for and run the query handler directly if the SCI bit is set in
the status register during boot. Previously, the query handler wouldn't
run until interrupts were finally enabled late in boot.
- During boot and after starting a command, check if the event appears
to already have occurred before we even start waiting. If so, it's
possible the EC is very slow and we might accept an old status value.
Print a warning in this case. Once we've booted, interrupt-driven mode
should work just fine but polled mode could be unreliable. There's not
much more we can do about this until interrupts are enabled during boot.
- In the above case, we also do one final check if the interrupt-driven
mode gets a timeout. If the status is complete, it will force the
system back into polled mode since interrupt mode doesn't work. For
polled mode during boot, if the status appears to be already complete
before beginning the check loop, it waits 10 us before actually checking
the status, just in case the EC is really slow and hasn't gotten to work
on the new request yet.
- Use upper-case hex for the _Qxx method
- Use device_printf for errors, don't hide them under verbose
- Increase default total timeout to 750 ms and decrease polling interval
to 5 us.
- Don't pass the status value via the softc. Just read it directly.
- Remove the mutex. We use the sx lock for transaction serialization
with the query handler.
- Remove the Intel copyright notice as no code of theirs was ever
present in this file (verified against rev 1.1)
- Allow KTR module-only builds for ease of testing
Thanks to jkim and Alexey Starikovskiy for helpful discussions and testing.
Approved by: re
MFC after: 2 weeks
2007-09-24 16:59:06 +00:00
|
|
|
}
|
2000-10-28 06:59:48 +00:00
|
|
|
|
|
|
|
static ACPI_STATUS
|
2003-07-10 17:22:46 +00:00
|
|
|
EcCommand(struct acpi_ec_softc *sc, EC_COMMAND cmd)
|
2000-10-28 06:59:48 +00:00
|
|
|
{
|
2007-02-27 00:14:20 +00:00
|
|
|
ACPI_STATUS status;
|
|
|
|
EC_EVENT event;
|
|
|
|
EC_STATUS ec_status;
|
Rewrite the EC driver event model. The main goal is to avoid
polling/interrupt-driven fallback and instead use polling only during
boot and pure interrupt-driven mode after boot. Polled mode could be
relegated completely to a legacy role if we could enable interrupts
during boot. Polled mode can be forced after boot by setting
debug.acpi.ec.polled="1", i.e. if there are timeouts.
- Use polling only during boot, shutdown, or if requested by the user.
Otherwise, use a generation count of GPEs, incremented atomically. This
prevents an old status value from being used if the EC is really slow
and the same condition (i.e. multiple IBEs for a write transaction) is
being checked.
- Check for and run the query handler directly if the SCI bit is set in
the status register during boot. Previously, the query handler wouldn't
run until interrupts were finally enabled late in boot.
- During boot and after starting a command, check if the event appears
to already have occurred before we even start waiting. If so, it's
possible the EC is very slow and we might accept an old status value.
Print a warning in this case. Once we've booted, interrupt-driven mode
should work just fine but polled mode could be unreliable. There's not
much more we can do about this until interrupts are enabled during boot.
- In the above case, we also do one final check if the interrupt-driven
mode gets a timeout. If the status is complete, it will force the
system back into polled mode since interrupt mode doesn't work. For
polled mode during boot, if the status appears to be already complete
before beginning the check loop, it waits 10 us before actually checking
the status, just in case the EC is really slow and hasn't gotten to work
on the new request yet.
- Use upper-case hex for the _Qxx method
- Use device_printf for errors, don't hide them under verbose
- Increase default total timeout to 750 ms and decrease polling interval
to 5 us.
- Don't pass the status value via the softc. Just read it directly.
- Remove the mutex. We use the sx lock for transaction serialization
with the query handler.
- Remove the Intel copyright notice as no code of theirs was ever
present in this file (verified against rev 1.1)
- Allow KTR module-only builds for ease of testing
Thanks to jkim and Alexey Starikovskiy for helpful discussions and testing.
Approved by: re
MFC after: 2 weeks
2007-09-24 16:59:06 +00:00
|
|
|
u_int gen_count;
|
2000-10-28 06:59:48 +00:00
|
|
|
|
2004-08-13 06:21:53 +00:00
|
|
|
ACPI_SERIAL_ASSERT(ec);
|
2000-10-28 06:59:48 +00:00
|
|
|
|
2007-02-27 00:14:20 +00:00
|
|
|
/* Don't use burst mode if user disabled it. */
|
|
|
|
if (!ec_burst_mode && cmd == EC_COMMAND_BURST_ENABLE)
|
|
|
|
return (AE_ERROR);
|
|
|
|
|
2003-07-10 17:22:46 +00:00
|
|
|
/* Decide what to wait for based on command type. */
|
|
|
|
switch (cmd) {
|
2000-10-28 06:59:48 +00:00
|
|
|
case EC_COMMAND_READ:
|
|
|
|
case EC_COMMAND_WRITE:
|
2003-07-10 17:22:46 +00:00
|
|
|
case EC_COMMAND_BURST_DISABLE:
|
2007-02-27 00:14:20 +00:00
|
|
|
event = EC_EVENT_INPUT_BUFFER_EMPTY;
|
2000-10-28 06:59:48 +00:00
|
|
|
break;
|
2003-07-10 17:22:46 +00:00
|
|
|
case EC_COMMAND_QUERY:
|
|
|
|
case EC_COMMAND_BURST_ENABLE:
|
2007-02-27 00:14:20 +00:00
|
|
|
event = EC_EVENT_OUTPUT_BUFFER_FULL;
|
2000-10-28 06:59:48 +00:00
|
|
|
break;
|
2003-07-10 17:22:46 +00:00
|
|
|
default:
|
Rewrite the EC driver event model. The main goal is to avoid
polling/interrupt-driven fallback and instead use polling only during
boot and pure interrupt-driven mode after boot. Polled mode could be
relegated completely to a legacy role if we could enable interrupts
during boot. Polled mode can be forced after boot by setting
debug.acpi.ec.polled="1", i.e. if there are timeouts.
- Use polling only during boot, shutdown, or if requested by the user.
Otherwise, use a generation count of GPEs, incremented atomically. This
prevents an old status value from being used if the EC is really slow
and the same condition (i.e. multiple IBEs for a write transaction) is
being checked.
- Check for and run the query handler directly if the SCI bit is set in
the status register during boot. Previously, the query handler wouldn't
run until interrupts were finally enabled late in boot.
- During boot and after starting a command, check if the event appears
to already have occurred before we even start waiting. If so, it's
possible the EC is very slow and we might accept an old status value.
Print a warning in this case. Once we've booted, interrupt-driven mode
should work just fine but polled mode could be unreliable. There's not
much more we can do about this until interrupts are enabled during boot.
- In the above case, we also do one final check if the interrupt-driven
mode gets a timeout. If the status is complete, it will force the
system back into polled mode since interrupt mode doesn't work. For
polled mode during boot, if the status appears to be already complete
before beginning the check loop, it waits 10 us before actually checking
the status, just in case the EC is really slow and hasn't gotten to work
on the new request yet.
- Use upper-case hex for the _Qxx method
- Use device_printf for errors, don't hide them under verbose
- Increase default total timeout to 750 ms and decrease polling interval
to 5 us.
- Don't pass the status value via the softc. Just read it directly.
- Remove the mutex. We use the sx lock for transaction serialization
with the query handler.
- Remove the Intel copyright notice as no code of theirs was ever
present in this file (verified against rev 1.1)
- Allow KTR module-only builds for ease of testing
Thanks to jkim and Alexey Starikovskiy for helpful discussions and testing.
Approved by: re
MFC after: 2 weeks
2007-09-24 16:59:06 +00:00
|
|
|
device_printf(sc->ec_dev, "EcCommand: invalid command %#x\n", cmd);
|
2003-07-10 17:22:46 +00:00
|
|
|
return (AE_BAD_PARAMETER);
|
2003-06-27 21:57:42 +00:00
|
|
|
}
|
2003-07-10 17:22:46 +00:00
|
|
|
|
|
|
|
/* Run the command and wait for the chosen event. */
|
2007-02-27 00:14:20 +00:00
|
|
|
CTR1(KTR_ACPI, "ec running command %#x", cmd);
|
Rewrite the EC driver event model. The main goal is to avoid
polling/interrupt-driven fallback and instead use polling only during
boot and pure interrupt-driven mode after boot. Polled mode could be
relegated completely to a legacy role if we could enable interrupts
during boot. Polled mode can be forced after boot by setting
debug.acpi.ec.polled="1", i.e. if there are timeouts.
- Use polling only during boot, shutdown, or if requested by the user.
Otherwise, use a generation count of GPEs, incremented atomically. This
prevents an old status value from being used if the EC is really slow
and the same condition (i.e. multiple IBEs for a write transaction) is
being checked.
- Check for and run the query handler directly if the SCI bit is set in
the status register during boot. Previously, the query handler wouldn't
run until interrupts were finally enabled late in boot.
- During boot and after starting a command, check if the event appears
to already have occurred before we even start waiting. If so, it's
possible the EC is very slow and we might accept an old status value.
Print a warning in this case. Once we've booted, interrupt-driven mode
should work just fine but polled mode could be unreliable. There's not
much more we can do about this until interrupts are enabled during boot.
- In the above case, we also do one final check if the interrupt-driven
mode gets a timeout. If the status is complete, it will force the
system back into polled mode since interrupt mode doesn't work. For
polled mode during boot, if the status appears to be already complete
before beginning the check loop, it waits 10 us before actually checking
the status, just in case the EC is really slow and hasn't gotten to work
on the new request yet.
- Use upper-case hex for the _Qxx method
- Use device_printf for errors, don't hide them under verbose
- Increase default total timeout to 750 ms and decrease polling interval
to 5 us.
- Don't pass the status value via the softc. Just read it directly.
- Remove the mutex. We use the sx lock for transaction serialization
with the query handler.
- Remove the Intel copyright notice as no code of theirs was ever
present in this file (verified against rev 1.1)
- Allow KTR module-only builds for ease of testing
Thanks to jkim and Alexey Starikovskiy for helpful discussions and testing.
Approved by: re
MFC after: 2 weeks
2007-09-24 16:59:06 +00:00
|
|
|
gen_count = sc->ec_gencount;
|
2003-07-10 17:22:46 +00:00
|
|
|
EC_SET_CSR(sc, cmd);
|
Rewrite the EC driver event model. The main goal is to avoid
polling/interrupt-driven fallback and instead use polling only during
boot and pure interrupt-driven mode after boot. Polled mode could be
relegated completely to a legacy role if we could enable interrupts
during boot. Polled mode can be forced after boot by setting
debug.acpi.ec.polled="1", i.e. if there are timeouts.
- Use polling only during boot, shutdown, or if requested by the user.
Otherwise, use a generation count of GPEs, incremented atomically. This
prevents an old status value from being used if the EC is really slow
and the same condition (i.e. multiple IBEs for a write transaction) is
being checked.
- Check for and run the query handler directly if the SCI bit is set in
the status register during boot. Previously, the query handler wouldn't
run until interrupts were finally enabled late in boot.
- During boot and after starting a command, check if the event appears
to already have occurred before we even start waiting. If so, it's
possible the EC is very slow and we might accept an old status value.
Print a warning in this case. Once we've booted, interrupt-driven mode
should work just fine but polled mode could be unreliable. There's not
much more we can do about this until interrupts are enabled during boot.
- In the above case, we also do one final check if the interrupt-driven
mode gets a timeout. If the status is complete, it will force the
system back into polled mode since interrupt mode doesn't work. For
polled mode during boot, if the status appears to be already complete
before beginning the check loop, it waits 10 us before actually checking
the status, just in case the EC is really slow and hasn't gotten to work
on the new request yet.
- Use upper-case hex for the _Qxx method
- Use device_printf for errors, don't hide them under verbose
- Increase default total timeout to 750 ms and decrease polling interval
to 5 us.
- Don't pass the status value via the softc. Just read it directly.
- Remove the mutex. We use the sx lock for transaction serialization
with the query handler.
- Remove the Intel copyright notice as no code of theirs was ever
present in this file (verified against rev 1.1)
- Allow KTR module-only builds for ease of testing
Thanks to jkim and Alexey Starikovskiy for helpful discussions and testing.
Approved by: re
MFC after: 2 weeks
2007-09-24 16:59:06 +00:00
|
|
|
status = EcWaitEvent(sc, event, gen_count);
|
2007-02-27 00:14:20 +00:00
|
|
|
if (ACPI_SUCCESS(status)) {
|
|
|
|
/* If we succeeded, burst flag should now be present. */
|
|
|
|
if (cmd == EC_COMMAND_BURST_ENABLE) {
|
|
|
|
ec_status = EC_GET_CSR(sc);
|
|
|
|
if ((ec_status & EC_FLAG_BURST_MODE) == 0)
|
|
|
|
status = AE_ERROR;
|
|
|
|
}
|
Rewrite the EC driver event model. The main goal is to avoid
polling/interrupt-driven fallback and instead use polling only during
boot and pure interrupt-driven mode after boot. Polled mode could be
relegated completely to a legacy role if we could enable interrupts
during boot. Polled mode can be forced after boot by setting
debug.acpi.ec.polled="1", i.e. if there are timeouts.
- Use polling only during boot, shutdown, or if requested by the user.
Otherwise, use a generation count of GPEs, incremented atomically. This
prevents an old status value from being used if the EC is really slow
and the same condition (i.e. multiple IBEs for a write transaction) is
being checked.
- Check for and run the query handler directly if the SCI bit is set in
the status register during boot. Previously, the query handler wouldn't
run until interrupts were finally enabled late in boot.
- During boot and after starting a command, check if the event appears
to already have occurred before we even start waiting. If so, it's
possible the EC is very slow and we might accept an old status value.
Print a warning in this case. Once we've booted, interrupt-driven mode
should work just fine but polled mode could be unreliable. There's not
much more we can do about this until interrupts are enabled during boot.
- In the above case, we also do one final check if the interrupt-driven
mode gets a timeout. If the status is complete, it will force the
system back into polled mode since interrupt mode doesn't work. For
polled mode during boot, if the status appears to be already complete
before beginning the check loop, it waits 10 us before actually checking
the status, just in case the EC is really slow and hasn't gotten to work
on the new request yet.
- Use upper-case hex for the _Qxx method
- Use device_printf for errors, don't hide them under verbose
- Increase default total timeout to 750 ms and decrease polling interval
to 5 us.
- Don't pass the status value via the softc. Just read it directly.
- Remove the mutex. We use the sx lock for transaction serialization
with the query handler.
- Remove the Intel copyright notice as no code of theirs was ever
present in this file (verified against rev 1.1)
- Allow KTR module-only builds for ease of testing
Thanks to jkim and Alexey Starikovskiy for helpful discussions and testing.
Approved by: re
MFC after: 2 weeks
2007-09-24 16:59:06 +00:00
|
|
|
} else
|
|
|
|
device_printf(sc->ec_dev, "EcCommand: no response to %#x\n", cmd);
|
2007-02-27 00:14:20 +00:00
|
|
|
return (status);
|
2000-10-28 06:59:48 +00:00
|
|
|
}
|
|
|
|
|
|
|
|
static ACPI_STATUS
|
|
|
|
EcRead(struct acpi_ec_softc *sc, UINT8 Address, UINT8 *Data)
|
|
|
|
{
|
2007-02-27 00:14:20 +00:00
|
|
|
ACPI_STATUS status;
|
|
|
|
UINT8 data;
|
Rewrite the EC driver event model. The main goal is to avoid
polling/interrupt-driven fallback and instead use polling only during
boot and pure interrupt-driven mode after boot. Polled mode could be
relegated completely to a legacy role if we could enable interrupts
during boot. Polled mode can be forced after boot by setting
debug.acpi.ec.polled="1", i.e. if there are timeouts.
- Use polling only during boot, shutdown, or if requested by the user.
Otherwise, use a generation count of GPEs, incremented atomically. This
prevents an old status value from being used if the EC is really slow
and the same condition (i.e. multiple IBEs for a write transaction) is
being checked.
- Check for and run the query handler directly if the SCI bit is set in
the status register during boot. Previously, the query handler wouldn't
run until interrupts were finally enabled late in boot.
- During boot and after starting a command, check if the event appears
to already have occurred before we even start waiting. If so, it's
possible the EC is very slow and we might accept an old status value.
Print a warning in this case. Once we've booted, interrupt-driven mode
should work just fine but polled mode could be unreliable. There's not
much more we can do about this until interrupts are enabled during boot.
- In the above case, we also do one final check if the interrupt-driven
mode gets a timeout. If the status is complete, it will force the
system back into polled mode since interrupt mode doesn't work. For
polled mode during boot, if the status appears to be already complete
before beginning the check loop, it waits 10 us before actually checking
the status, just in case the EC is really slow and hasn't gotten to work
on the new request yet.
- Use upper-case hex for the _Qxx method
- Use device_printf for errors, don't hide them under verbose
- Increase default total timeout to 750 ms and decrease polling interval
to 5 us.
- Don't pass the status value via the softc. Just read it directly.
- Remove the mutex. We use the sx lock for transaction serialization
with the query handler.
- Remove the Intel copyright notice as no code of theirs was ever
present in this file (verified against rev 1.1)
- Allow KTR module-only builds for ease of testing
Thanks to jkim and Alexey Starikovskiy for helpful discussions and testing.
Approved by: re
MFC after: 2 weeks
2007-09-24 16:59:06 +00:00
|
|
|
u_int gen_count;
|
2000-10-28 06:59:48 +00:00
|
|
|
|
2004-08-13 06:21:53 +00:00
|
|
|
ACPI_SERIAL_ASSERT(ec);
|
2005-12-06 14:47:28 +00:00
|
|
|
CTR1(KTR_ACPI, "ec read from %#x", Address);
|
2000-10-28 06:59:48 +00:00
|
|
|
|
2003-07-10 17:22:46 +00:00
|
|
|
/* If we can't start burst mode, continue anyway. */
|
2007-02-27 00:14:20 +00:00
|
|
|
status = EcCommand(sc, EC_COMMAND_BURST_ENABLE);
|
|
|
|
if (status == AE_OK) {
|
|
|
|
data = EC_GET_DATA(sc);
|
|
|
|
if (data == EC_BURST_ACK) {
|
|
|
|
CTR0(KTR_ACPI, "ec burst enabled");
|
|
|
|
sc->ec_burstactive = TRUE;
|
|
|
|
}
|
|
|
|
}
|
2000-10-28 06:59:48 +00:00
|
|
|
|
2007-02-27 00:14:20 +00:00
|
|
|
status = EcCommand(sc, EC_COMMAND_READ);
|
|
|
|
if (ACPI_FAILURE(status))
|
|
|
|
return (status);
|
2000-10-28 06:59:48 +00:00
|
|
|
|
Rewrite the EC driver event model. The main goal is to avoid
polling/interrupt-driven fallback and instead use polling only during
boot and pure interrupt-driven mode after boot. Polled mode could be
relegated completely to a legacy role if we could enable interrupts
during boot. Polled mode can be forced after boot by setting
debug.acpi.ec.polled="1", i.e. if there are timeouts.
- Use polling only during boot, shutdown, or if requested by the user.
Otherwise, use a generation count of GPEs, incremented atomically. This
prevents an old status value from being used if the EC is really slow
and the same condition (i.e. multiple IBEs for a write transaction) is
being checked.
- Check for and run the query handler directly if the SCI bit is set in
the status register during boot. Previously, the query handler wouldn't
run until interrupts were finally enabled late in boot.
- During boot and after starting a command, check if the event appears
to already have occurred before we even start waiting. If so, it's
possible the EC is very slow and we might accept an old status value.
Print a warning in this case. Once we've booted, interrupt-driven mode
should work just fine but polled mode could be unreliable. There's not
much more we can do about this until interrupts are enabled during boot.
- In the above case, we also do one final check if the interrupt-driven
mode gets a timeout. If the status is complete, it will force the
system back into polled mode since interrupt mode doesn't work. For
polled mode during boot, if the status appears to be already complete
before beginning the check loop, it waits 10 us before actually checking
the status, just in case the EC is really slow and hasn't gotten to work
on the new request yet.
- Use upper-case hex for the _Qxx method
- Use device_printf for errors, don't hide them under verbose
- Increase default total timeout to 750 ms and decrease polling interval
to 5 us.
- Don't pass the status value via the softc. Just read it directly.
- Remove the mutex. We use the sx lock for transaction serialization
with the query handler.
- Remove the Intel copyright notice as no code of theirs was ever
present in this file (verified against rev 1.1)
- Allow KTR module-only builds for ease of testing
Thanks to jkim and Alexey Starikovskiy for helpful discussions and testing.
Approved by: re
MFC after: 2 weeks
2007-09-24 16:59:06 +00:00
|
|
|
gen_count = sc->ec_gencount;
|
2000-10-28 06:59:48 +00:00
|
|
|
EC_SET_DATA(sc, Address);
|
Rewrite the EC driver event model. The main goal is to avoid
polling/interrupt-driven fallback and instead use polling only during
boot and pure interrupt-driven mode after boot. Polled mode could be
relegated completely to a legacy role if we could enable interrupts
during boot. Polled mode can be forced after boot by setting
debug.acpi.ec.polled="1", i.e. if there are timeouts.
- Use polling only during boot, shutdown, or if requested by the user.
Otherwise, use a generation count of GPEs, incremented atomically. This
prevents an old status value from being used if the EC is really slow
and the same condition (i.e. multiple IBEs for a write transaction) is
being checked.
- Check for and run the query handler directly if the SCI bit is set in
the status register during boot. Previously, the query handler wouldn't
run until interrupts were finally enabled late in boot.
- During boot and after starting a command, check if the event appears
to already have occurred before we even start waiting. If so, it's
possible the EC is very slow and we might accept an old status value.
Print a warning in this case. Once we've booted, interrupt-driven mode
should work just fine but polled mode could be unreliable. There's not
much more we can do about this until interrupts are enabled during boot.
- In the above case, we also do one final check if the interrupt-driven
mode gets a timeout. If the status is complete, it will force the
system back into polled mode since interrupt mode doesn't work. For
polled mode during boot, if the status appears to be already complete
before beginning the check loop, it waits 10 us before actually checking
the status, just in case the EC is really slow and hasn't gotten to work
on the new request yet.
- Use upper-case hex for the _Qxx method
- Use device_printf for errors, don't hide them under verbose
- Increase default total timeout to 750 ms and decrease polling interval
to 5 us.
- Don't pass the status value via the softc. Just read it directly.
- Remove the mutex. We use the sx lock for transaction serialization
with the query handler.
- Remove the Intel copyright notice as no code of theirs was ever
present in this file (verified against rev 1.1)
- Allow KTR module-only builds for ease of testing
Thanks to jkim and Alexey Starikovskiy for helpful discussions and testing.
Approved by: re
MFC after: 2 weeks
2007-09-24 16:59:06 +00:00
|
|
|
status = EcWaitEvent(sc, EC_EVENT_OUTPUT_BUFFER_FULL, gen_count);
|
2007-02-27 00:14:20 +00:00
|
|
|
if (ACPI_FAILURE(status)) {
|
Rewrite the EC driver event model. The main goal is to avoid
polling/interrupt-driven fallback and instead use polling only during
boot and pure interrupt-driven mode after boot. Polled mode could be
relegated completely to a legacy role if we could enable interrupts
during boot. Polled mode can be forced after boot by setting
debug.acpi.ec.polled="1", i.e. if there are timeouts.
- Use polling only during boot, shutdown, or if requested by the user.
Otherwise, use a generation count of GPEs, incremented atomically. This
prevents an old status value from being used if the EC is really slow
and the same condition (i.e. multiple IBEs for a write transaction) is
being checked.
- Check for and run the query handler directly if the SCI bit is set in
the status register during boot. Previously, the query handler wouldn't
run until interrupts were finally enabled late in boot.
- During boot and after starting a command, check if the event appears
to already have occurred before we even start waiting. If so, it's
possible the EC is very slow and we might accept an old status value.
Print a warning in this case. Once we've booted, interrupt-driven mode
should work just fine but polled mode could be unreliable. There's not
much more we can do about this until interrupts are enabled during boot.
- In the above case, we also do one final check if the interrupt-driven
mode gets a timeout. If the status is complete, it will force the
system back into polled mode since interrupt mode doesn't work. For
polled mode during boot, if the status appears to be already complete
before beginning the check loop, it waits 10 us before actually checking
the status, just in case the EC is really slow and hasn't gotten to work
on the new request yet.
- Use upper-case hex for the _Qxx method
- Use device_printf for errors, don't hide them under verbose
- Increase default total timeout to 750 ms and decrease polling interval
to 5 us.
- Don't pass the status value via the softc. Just read it directly.
- Remove the mutex. We use the sx lock for transaction serialization
with the query handler.
- Remove the Intel copyright notice as no code of theirs was ever
present in this file (verified against rev 1.1)
- Allow KTR module-only builds for ease of testing
Thanks to jkim and Alexey Starikovskiy for helpful discussions and testing.
Approved by: re
MFC after: 2 weeks
2007-09-24 16:59:06 +00:00
|
|
|
device_printf(sc->ec_dev, "EcRead: failed waiting to get data\n");
|
2007-02-27 00:14:20 +00:00
|
|
|
return (status);
|
2000-10-28 06:59:48 +00:00
|
|
|
}
|
2003-06-27 21:57:42 +00:00
|
|
|
*Data = EC_GET_DATA(sc);
|
2000-10-28 06:59:48 +00:00
|
|
|
|
2003-07-10 17:22:46 +00:00
|
|
|
if (sc->ec_burstactive) {
|
Rewrite the EC driver event model. The main goal is to avoid
polling/interrupt-driven fallback and instead use polling only during
boot and pure interrupt-driven mode after boot. Polled mode could be
relegated completely to a legacy role if we could enable interrupts
during boot. Polled mode can be forced after boot by setting
debug.acpi.ec.polled="1", i.e. if there are timeouts.
- Use polling only during boot, shutdown, or if requested by the user.
Otherwise, use a generation count of GPEs, incremented atomically. This
prevents an old status value from being used if the EC is really slow
and the same condition (i.e. multiple IBEs for a write transaction) is
being checked.
- Check for and run the query handler directly if the SCI bit is set in
the status register during boot. Previously, the query handler wouldn't
run until interrupts were finally enabled late in boot.
- During boot and after starting a command, check if the event appears
to already have occurred before we even start waiting. If so, it's
possible the EC is very slow and we might accept an old status value.
Print a warning in this case. Once we've booted, interrupt-driven mode
should work just fine but polled mode could be unreliable. There's not
much more we can do about this until interrupts are enabled during boot.
- In the above case, we also do one final check if the interrupt-driven
mode gets a timeout. If the status is complete, it will force the
system back into polled mode since interrupt mode doesn't work. For
polled mode during boot, if the status appears to be already complete
before beginning the check loop, it waits 10 us before actually checking
the status, just in case the EC is really slow and hasn't gotten to work
on the new request yet.
- Use upper-case hex for the _Qxx method
- Use device_printf for errors, don't hide them under verbose
- Increase default total timeout to 750 ms and decrease polling interval
to 5 us.
- Don't pass the status value via the softc. Just read it directly.
- Remove the mutex. We use the sx lock for transaction serialization
with the query handler.
- Remove the Intel copyright notice as no code of theirs was ever
present in this file (verified against rev 1.1)
- Allow KTR module-only builds for ease of testing
Thanks to jkim and Alexey Starikovskiy for helpful discussions and testing.
Approved by: re
MFC after: 2 weeks
2007-09-24 16:59:06 +00:00
|
|
|
sc->ec_burstactive = FALSE;
|
2007-02-27 00:14:20 +00:00
|
|
|
status = EcCommand(sc, EC_COMMAND_BURST_DISABLE);
|
|
|
|
if (ACPI_FAILURE(status))
|
|
|
|
return (status);
|
|
|
|
CTR0(KTR_ACPI, "ec disabled burst ok");
|
2003-07-10 17:22:46 +00:00
|
|
|
}
|
2000-10-28 06:59:48 +00:00
|
|
|
|
2003-06-27 21:57:42 +00:00
|
|
|
return (AE_OK);
|
Rewrite the EC driver event model. The main goal is to avoid
polling/interrupt-driven fallback and instead use polling only during
boot and pure interrupt-driven mode after boot. Polled mode could be
relegated completely to a legacy role if we could enable interrupts
during boot. Polled mode can be forced after boot by setting
debug.acpi.ec.polled="1", i.e. if there are timeouts.
- Use polling only during boot, shutdown, or if requested by the user.
Otherwise, use a generation count of GPEs, incremented atomically. This
prevents an old status value from being used if the EC is really slow
and the same condition (i.e. multiple IBEs for a write transaction) is
being checked.
- Check for and run the query handler directly if the SCI bit is set in
the status register during boot. Previously, the query handler wouldn't
run until interrupts were finally enabled late in boot.
- During boot and after starting a command, check if the event appears
to already have occurred before we even start waiting. If so, it's
possible the EC is very slow and we might accept an old status value.
Print a warning in this case. Once we've booted, interrupt-driven mode
should work just fine but polled mode could be unreliable. There's not
much more we can do about this until interrupts are enabled during boot.
- In the above case, we also do one final check if the interrupt-driven
mode gets a timeout. If the status is complete, it will force the
system back into polled mode since interrupt mode doesn't work. For
polled mode during boot, if the status appears to be already complete
before beginning the check loop, it waits 10 us before actually checking
the status, just in case the EC is really slow and hasn't gotten to work
on the new request yet.
- Use upper-case hex for the _Qxx method
- Use device_printf for errors, don't hide them under verbose
- Increase default total timeout to 750 ms and decrease polling interval
to 5 us.
- Don't pass the status value via the softc. Just read it directly.
- Remove the mutex. We use the sx lock for transaction serialization
with the query handler.
- Remove the Intel copyright notice as no code of theirs was ever
present in this file (verified against rev 1.1)
- Allow KTR module-only builds for ease of testing
Thanks to jkim and Alexey Starikovskiy for helpful discussions and testing.
Approved by: re
MFC after: 2 weeks
2007-09-24 16:59:06 +00:00
|
|
|
}
|
2000-10-28 06:59:48 +00:00
|
|
|
|
|
|
|
static ACPI_STATUS
|
|
|
|
EcWrite(struct acpi_ec_softc *sc, UINT8 Address, UINT8 *Data)
|
|
|
|
{
|
2007-02-27 00:14:20 +00:00
|
|
|
ACPI_STATUS status;
|
|
|
|
UINT8 data;
|
Rewrite the EC driver event model. The main goal is to avoid
polling/interrupt-driven fallback and instead use polling only during
boot and pure interrupt-driven mode after boot. Polled mode could be
relegated completely to a legacy role if we could enable interrupts
during boot. Polled mode can be forced after boot by setting
debug.acpi.ec.polled="1", i.e. if there are timeouts.
- Use polling only during boot, shutdown, or if requested by the user.
Otherwise, use a generation count of GPEs, incremented atomically. This
prevents an old status value from being used if the EC is really slow
and the same condition (i.e. multiple IBEs for a write transaction) is
being checked.
- Check for and run the query handler directly if the SCI bit is set in
the status register during boot. Previously, the query handler wouldn't
run until interrupts were finally enabled late in boot.
- During boot and after starting a command, check if the event appears
to already have occurred before we even start waiting. If so, it's
possible the EC is very slow and we might accept an old status value.
Print a warning in this case. Once we've booted, interrupt-driven mode
should work just fine but polled mode could be unreliable. There's not
much more we can do about this until interrupts are enabled during boot.
- In the above case, we also do one final check if the interrupt-driven
mode gets a timeout. If the status is complete, it will force the
system back into polled mode since interrupt mode doesn't work. For
polled mode during boot, if the status appears to be already complete
before beginning the check loop, it waits 10 us before actually checking
the status, just in case the EC is really slow and hasn't gotten to work
on the new request yet.
- Use upper-case hex for the _Qxx method
- Use device_printf for errors, don't hide them under verbose
- Increase default total timeout to 750 ms and decrease polling interval
to 5 us.
- Don't pass the status value via the softc. Just read it directly.
- Remove the mutex. We use the sx lock for transaction serialization
with the query handler.
- Remove the Intel copyright notice as no code of theirs was ever
present in this file (verified against rev 1.1)
- Allow KTR module-only builds for ease of testing
Thanks to jkim and Alexey Starikovskiy for helpful discussions and testing.
Approved by: re
MFC after: 2 weeks
2007-09-24 16:59:06 +00:00
|
|
|
u_int gen_count;
|
2000-10-28 06:59:48 +00:00
|
|
|
|
2004-08-13 06:21:53 +00:00
|
|
|
ACPI_SERIAL_ASSERT(ec);
|
2005-12-06 14:47:28 +00:00
|
|
|
CTR2(KTR_ACPI, "ec write to %#x, data %#x", Address, *Data);
|
2000-10-28 06:59:48 +00:00
|
|
|
|
2003-07-10 17:22:46 +00:00
|
|
|
/* If we can't start burst mode, continue anyway. */
|
2007-02-27 00:14:20 +00:00
|
|
|
status = EcCommand(sc, EC_COMMAND_BURST_ENABLE);
|
|
|
|
if (status == AE_OK) {
|
|
|
|
data = EC_GET_DATA(sc);
|
|
|
|
if (data == EC_BURST_ACK) {
|
|
|
|
CTR0(KTR_ACPI, "ec burst enabled");
|
|
|
|
sc->ec_burstactive = TRUE;
|
|
|
|
}
|
|
|
|
}
|
2000-10-28 06:59:48 +00:00
|
|
|
|
2007-02-27 00:14:20 +00:00
|
|
|
status = EcCommand(sc, EC_COMMAND_WRITE);
|
|
|
|
if (ACPI_FAILURE(status))
|
|
|
|
return (status);
|
2000-10-28 06:59:48 +00:00
|
|
|
|
Rewrite the EC driver event model. The main goal is to avoid
polling/interrupt-driven fallback and instead use polling only during
boot and pure interrupt-driven mode after boot. Polled mode could be
relegated completely to a legacy role if we could enable interrupts
during boot. Polled mode can be forced after boot by setting
debug.acpi.ec.polled="1", i.e. if there are timeouts.
- Use polling only during boot, shutdown, or if requested by the user.
Otherwise, use a generation count of GPEs, incremented atomically. This
prevents an old status value from being used if the EC is really slow
and the same condition (i.e. multiple IBEs for a write transaction) is
being checked.
- Check for and run the query handler directly if the SCI bit is set in
the status register during boot. Previously, the query handler wouldn't
run until interrupts were finally enabled late in boot.
- During boot and after starting a command, check if the event appears
to already have occurred before we even start waiting. If so, it's
possible the EC is very slow and we might accept an old status value.
Print a warning in this case. Once we've booted, interrupt-driven mode
should work just fine but polled mode could be unreliable. There's not
much more we can do about this until interrupts are enabled during boot.
- In the above case, we also do one final check if the interrupt-driven
mode gets a timeout. If the status is complete, it will force the
system back into polled mode since interrupt mode doesn't work. For
polled mode during boot, if the status appears to be already complete
before beginning the check loop, it waits 10 us before actually checking
the status, just in case the EC is really slow and hasn't gotten to work
on the new request yet.
- Use upper-case hex for the _Qxx method
- Use device_printf for errors, don't hide them under verbose
- Increase default total timeout to 750 ms and decrease polling interval
to 5 us.
- Don't pass the status value via the softc. Just read it directly.
- Remove the mutex. We use the sx lock for transaction serialization
with the query handler.
- Remove the Intel copyright notice as no code of theirs was ever
present in this file (verified against rev 1.1)
- Allow KTR module-only builds for ease of testing
Thanks to jkim and Alexey Starikovskiy for helpful discussions and testing.
Approved by: re
MFC after: 2 weeks
2007-09-24 16:59:06 +00:00
|
|
|
gen_count = sc->ec_gencount;
|
2000-10-28 06:59:48 +00:00
|
|
|
EC_SET_DATA(sc, Address);
|
Rewrite the EC driver event model. The main goal is to avoid
polling/interrupt-driven fallback and instead use polling only during
boot and pure interrupt-driven mode after boot. Polled mode could be
relegated completely to a legacy role if we could enable interrupts
during boot. Polled mode can be forced after boot by setting
debug.acpi.ec.polled="1", i.e. if there are timeouts.
- Use polling only during boot, shutdown, or if requested by the user.
Otherwise, use a generation count of GPEs, incremented atomically. This
prevents an old status value from being used if the EC is really slow
and the same condition (i.e. multiple IBEs for a write transaction) is
being checked.
- Check for and run the query handler directly if the SCI bit is set in
the status register during boot. Previously, the query handler wouldn't
run until interrupts were finally enabled late in boot.
- During boot and after starting a command, check if the event appears
to already have occurred before we even start waiting. If so, it's
possible the EC is very slow and we might accept an old status value.
Print a warning in this case. Once we've booted, interrupt-driven mode
should work just fine but polled mode could be unreliable. There's not
much more we can do about this until interrupts are enabled during boot.
- In the above case, we also do one final check if the interrupt-driven
mode gets a timeout. If the status is complete, it will force the
system back into polled mode since interrupt mode doesn't work. For
polled mode during boot, if the status appears to be already complete
before beginning the check loop, it waits 10 us before actually checking
the status, just in case the EC is really slow and hasn't gotten to work
on the new request yet.
- Use upper-case hex for the _Qxx method
- Use device_printf for errors, don't hide them under verbose
- Increase default total timeout to 750 ms and decrease polling interval
to 5 us.
- Don't pass the status value via the softc. Just read it directly.
- Remove the mutex. We use the sx lock for transaction serialization
with the query handler.
- Remove the Intel copyright notice as no code of theirs was ever
present in this file (verified against rev 1.1)
- Allow KTR module-only builds for ease of testing
Thanks to jkim and Alexey Starikovskiy for helpful discussions and testing.
Approved by: re
MFC after: 2 weeks
2007-09-24 16:59:06 +00:00
|
|
|
status = EcWaitEvent(sc, EC_EVENT_INPUT_BUFFER_EMPTY, gen_count);
|
2007-02-27 00:14:20 +00:00
|
|
|
if (ACPI_FAILURE(status)) {
|
Rewrite the EC driver event model. The main goal is to avoid
polling/interrupt-driven fallback and instead use polling only during
boot and pure interrupt-driven mode after boot. Polled mode could be
relegated completely to a legacy role if we could enable interrupts
during boot. Polled mode can be forced after boot by setting
debug.acpi.ec.polled="1", i.e. if there are timeouts.
- Use polling only during boot, shutdown, or if requested by the user.
Otherwise, use a generation count of GPEs, incremented atomically. This
prevents an old status value from being used if the EC is really slow
and the same condition (i.e. multiple IBEs for a write transaction) is
being checked.
- Check for and run the query handler directly if the SCI bit is set in
the status register during boot. Previously, the query handler wouldn't
run until interrupts were finally enabled late in boot.
- During boot and after starting a command, check if the event appears
to already have occurred before we even start waiting. If so, it's
possible the EC is very slow and we might accept an old status value.
Print a warning in this case. Once we've booted, interrupt-driven mode
should work just fine but polled mode could be unreliable. There's not
much more we can do about this until interrupts are enabled during boot.
- In the above case, we also do one final check if the interrupt-driven
mode gets a timeout. If the status is complete, it will force the
system back into polled mode since interrupt mode doesn't work. For
polled mode during boot, if the status appears to be already complete
before beginning the check loop, it waits 10 us before actually checking
the status, just in case the EC is really slow and hasn't gotten to work
on the new request yet.
- Use upper-case hex for the _Qxx method
- Use device_printf for errors, don't hide them under verbose
- Increase default total timeout to 750 ms and decrease polling interval
to 5 us.
- Don't pass the status value via the softc. Just read it directly.
- Remove the mutex. We use the sx lock for transaction serialization
with the query handler.
- Remove the Intel copyright notice as no code of theirs was ever
present in this file (verified against rev 1.1)
- Allow KTR module-only builds for ease of testing
Thanks to jkim and Alexey Starikovskiy for helpful discussions and testing.
Approved by: re
MFC after: 2 weeks
2007-09-24 16:59:06 +00:00
|
|
|
device_printf(sc->ec_dev, "EcRead: failed waiting for sent address\n");
|
2007-02-27 00:14:20 +00:00
|
|
|
return (status);
|
2000-10-28 06:59:48 +00:00
|
|
|
}
|
|
|
|
|
Rewrite the EC driver event model. The main goal is to avoid
polling/interrupt-driven fallback and instead use polling only during
boot and pure interrupt-driven mode after boot. Polled mode could be
relegated completely to a legacy role if we could enable interrupts
during boot. Polled mode can be forced after boot by setting
debug.acpi.ec.polled="1", i.e. if there are timeouts.
- Use polling only during boot, shutdown, or if requested by the user.
Otherwise, use a generation count of GPEs, incremented atomically. This
prevents an old status value from being used if the EC is really slow
and the same condition (i.e. multiple IBEs for a write transaction) is
being checked.
- Check for and run the query handler directly if the SCI bit is set in
the status register during boot. Previously, the query handler wouldn't
run until interrupts were finally enabled late in boot.
- During boot and after starting a command, check if the event appears
to already have occurred before we even start waiting. If so, it's
possible the EC is very slow and we might accept an old status value.
Print a warning in this case. Once we've booted, interrupt-driven mode
should work just fine but polled mode could be unreliable. There's not
much more we can do about this until interrupts are enabled during boot.
- In the above case, we also do one final check if the interrupt-driven
mode gets a timeout. If the status is complete, it will force the
system back into polled mode since interrupt mode doesn't work. For
polled mode during boot, if the status appears to be already complete
before beginning the check loop, it waits 10 us before actually checking
the status, just in case the EC is really slow and hasn't gotten to work
on the new request yet.
- Use upper-case hex for the _Qxx method
- Use device_printf for errors, don't hide them under verbose
- Increase default total timeout to 750 ms and decrease polling interval
to 5 us.
- Don't pass the status value via the softc. Just read it directly.
- Remove the mutex. We use the sx lock for transaction serialization
with the query handler.
- Remove the Intel copyright notice as no code of theirs was ever
present in this file (verified against rev 1.1)
- Allow KTR module-only builds for ease of testing
Thanks to jkim and Alexey Starikovskiy for helpful discussions and testing.
Approved by: re
MFC after: 2 weeks
2007-09-24 16:59:06 +00:00
|
|
|
gen_count = sc->ec_gencount;
|
2000-10-28 06:59:48 +00:00
|
|
|
EC_SET_DATA(sc, *Data);
|
Rewrite the EC driver event model. The main goal is to avoid
polling/interrupt-driven fallback and instead use polling only during
boot and pure interrupt-driven mode after boot. Polled mode could be
relegated completely to a legacy role if we could enable interrupts
during boot. Polled mode can be forced after boot by setting
debug.acpi.ec.polled="1", i.e. if there are timeouts.
- Use polling only during boot, shutdown, or if requested by the user.
Otherwise, use a generation count of GPEs, incremented atomically. This
prevents an old status value from being used if the EC is really slow
and the same condition (i.e. multiple IBEs for a write transaction) is
being checked.
- Check for and run the query handler directly if the SCI bit is set in
the status register during boot. Previously, the query handler wouldn't
run until interrupts were finally enabled late in boot.
- During boot and after starting a command, check if the event appears
to already have occurred before we even start waiting. If so, it's
possible the EC is very slow and we might accept an old status value.
Print a warning in this case. Once we've booted, interrupt-driven mode
should work just fine but polled mode could be unreliable. There's not
much more we can do about this until interrupts are enabled during boot.
- In the above case, we also do one final check if the interrupt-driven
mode gets a timeout. If the status is complete, it will force the
system back into polled mode since interrupt mode doesn't work. For
polled mode during boot, if the status appears to be already complete
before beginning the check loop, it waits 10 us before actually checking
the status, just in case the EC is really slow and hasn't gotten to work
on the new request yet.
- Use upper-case hex for the _Qxx method
- Use device_printf for errors, don't hide them under verbose
- Increase default total timeout to 750 ms and decrease polling interval
to 5 us.
- Don't pass the status value via the softc. Just read it directly.
- Remove the mutex. We use the sx lock for transaction serialization
with the query handler.
- Remove the Intel copyright notice as no code of theirs was ever
present in this file (verified against rev 1.1)
- Allow KTR module-only builds for ease of testing
Thanks to jkim and Alexey Starikovskiy for helpful discussions and testing.
Approved by: re
MFC after: 2 weeks
2007-09-24 16:59:06 +00:00
|
|
|
status = EcWaitEvent(sc, EC_EVENT_INPUT_BUFFER_EMPTY, gen_count);
|
2007-02-27 00:14:20 +00:00
|
|
|
if (ACPI_FAILURE(status)) {
|
Rewrite the EC driver event model. The main goal is to avoid
polling/interrupt-driven fallback and instead use polling only during
boot and pure interrupt-driven mode after boot. Polled mode could be
relegated completely to a legacy role if we could enable interrupts
during boot. Polled mode can be forced after boot by setting
debug.acpi.ec.polled="1", i.e. if there are timeouts.
- Use polling only during boot, shutdown, or if requested by the user.
Otherwise, use a generation count of GPEs, incremented atomically. This
prevents an old status value from being used if the EC is really slow
and the same condition (i.e. multiple IBEs for a write transaction) is
being checked.
- Check for and run the query handler directly if the SCI bit is set in
the status register during boot. Previously, the query handler wouldn't
run until interrupts were finally enabled late in boot.
- During boot and after starting a command, check if the event appears
to already have occurred before we even start waiting. If so, it's
possible the EC is very slow and we might accept an old status value.
Print a warning in this case. Once we've booted, interrupt-driven mode
should work just fine but polled mode could be unreliable. There's not
much more we can do about this until interrupts are enabled during boot.
- In the above case, we also do one final check if the interrupt-driven
mode gets a timeout. If the status is complete, it will force the
system back into polled mode since interrupt mode doesn't work. For
polled mode during boot, if the status appears to be already complete
before beginning the check loop, it waits 10 us before actually checking
the status, just in case the EC is really slow and hasn't gotten to work
on the new request yet.
- Use upper-case hex for the _Qxx method
- Use device_printf for errors, don't hide them under verbose
- Increase default total timeout to 750 ms and decrease polling interval
to 5 us.
- Don't pass the status value via the softc. Just read it directly.
- Remove the mutex. We use the sx lock for transaction serialization
with the query handler.
- Remove the Intel copyright notice as no code of theirs was ever
present in this file (verified against rev 1.1)
- Allow KTR module-only builds for ease of testing
Thanks to jkim and Alexey Starikovskiy for helpful discussions and testing.
Approved by: re
MFC after: 2 weeks
2007-09-24 16:59:06 +00:00
|
|
|
device_printf(sc->ec_dev, "EcWrite: failed waiting for sent data\n");
|
2007-02-27 00:14:20 +00:00
|
|
|
return (status);
|
2000-10-28 06:59:48 +00:00
|
|
|
}
|
|
|
|
|
2003-07-10 17:22:46 +00:00
|
|
|
if (sc->ec_burstactive) {
|
Rewrite the EC driver event model. The main goal is to avoid
polling/interrupt-driven fallback and instead use polling only during
boot and pure interrupt-driven mode after boot. Polled mode could be
relegated completely to a legacy role if we could enable interrupts
during boot. Polled mode can be forced after boot by setting
debug.acpi.ec.polled="1", i.e. if there are timeouts.
- Use polling only during boot, shutdown, or if requested by the user.
Otherwise, use a generation count of GPEs, incremented atomically. This
prevents an old status value from being used if the EC is really slow
and the same condition (i.e. multiple IBEs for a write transaction) is
being checked.
- Check for and run the query handler directly if the SCI bit is set in
the status register during boot. Previously, the query handler wouldn't
run until interrupts were finally enabled late in boot.
- During boot and after starting a command, check if the event appears
to already have occurred before we even start waiting. If so, it's
possible the EC is very slow and we might accept an old status value.
Print a warning in this case. Once we've booted, interrupt-driven mode
should work just fine but polled mode could be unreliable. There's not
much more we can do about this until interrupts are enabled during boot.
- In the above case, we also do one final check if the interrupt-driven
mode gets a timeout. If the status is complete, it will force the
system back into polled mode since interrupt mode doesn't work. For
polled mode during boot, if the status appears to be already complete
before beginning the check loop, it waits 10 us before actually checking
the status, just in case the EC is really slow and hasn't gotten to work
on the new request yet.
- Use upper-case hex for the _Qxx method
- Use device_printf for errors, don't hide them under verbose
- Increase default total timeout to 750 ms and decrease polling interval
to 5 us.
- Don't pass the status value via the softc. Just read it directly.
- Remove the mutex. We use the sx lock for transaction serialization
with the query handler.
- Remove the Intel copyright notice as no code of theirs was ever
present in this file (verified against rev 1.1)
- Allow KTR module-only builds for ease of testing
Thanks to jkim and Alexey Starikovskiy for helpful discussions and testing.
Approved by: re
MFC after: 2 weeks
2007-09-24 16:59:06 +00:00
|
|
|
sc->ec_burstactive = FALSE;
|
2007-02-27 00:14:20 +00:00
|
|
|
status = EcCommand(sc, EC_COMMAND_BURST_DISABLE);
|
|
|
|
if (ACPI_FAILURE(status))
|
|
|
|
return (status);
|
|
|
|
CTR0(KTR_ACPI, "ec disabled burst ok");
|
2003-07-10 17:22:46 +00:00
|
|
|
}
|
2000-10-28 06:59:48 +00:00
|
|
|
|
2003-06-27 21:57:42 +00:00
|
|
|
return (AE_OK);
|
2000-12-14 12:43:22 +00:00
|
|
|
}
|