diff options
author | Jonathan Brandmeyer <jbrandmeyer@chromium.org> | 2018-08-09 12:58:03 -0600 |
---|---|---|
committer | chrome-bot <chrome-bot@chromium.org> | 2018-08-16 13:14:49 -0700 |
commit | 39ab7a039676ef3738171ed00dc3c7f61802e7b3 (patch) | |
tree | a32fdd410d4d41f2092dbaf283c44b528d4b03c4 /common/usb_i2c.c | |
parent | 94d06cd5f92e163fca391529509f85909ed65ac6 (diff) | |
download | chrome-ec-39ab7a039676ef3738171ed00dc3c7f61802e7b3.tar.gz |
i2c: Split i2c_xfer into locked/unlocked versions.
Rename i2c_xfer to i2c_xfer_unlocked. Audit all users of i2c_xfer to
see if they can use simple locking semantics or require their own
locking. Since locked accesses are only safe for I2C_XFER_SINGLE
transactions, remove the flags parameter from i2c_xfer. This also makes
the audit a bit easier.
Some remaining applications hold the bus locked across several
transactions that would otherwise be atomic, and a few others implement
complex I2C transactions in multiple commands.
Add a (nondeterministic) verification on the I2C port locking
correctness. This will catch all statically incorrect locking patterns,
although dynamically incorrect locking patterns may not be caught.
Related: Revise the i2c_port_active_count to be a bitmap of the active
ports instead of a count of the active ports. The EC's mutex does not
provide an is_locked() primitive, and we would prefer not to have one.
- board/glados: Custom locking for battery reset
- board/mchpevb1: Custom locking for battery reset
- board/oak: Custom locking for battery reset
- board/samus: Custom locking for lightbar reset
- board/sweetberry: simple locking
- board/servo_micro: Custom locking for funky i2c transfers
- capsense: simple locking
- host_command_master: multi-command transactions
- lb_common: manual locking to support samus power sequence
- smbus: multi-command transactions
- usb_i2c: simple locking
- driver/tcpm/fusb302: simple locking and multi-command transactions
- driver/tcpm/tcpi: Forward _unlocked and implicitly locked interface to
fusb302
- driver/touchpad_elan: simple locking
BUG=chromium:871851
BRANCH=none
TEST=buildall; very careful audit
TEST=grunt clamshell and Coral clamshell, test boot, battery charging,
PD communication, and TCPC port low-power mode.
Signed-off-by: Jonathan Brandmeyer <jbrandmeyer@chromium.org>
Change-Id: Ieabf22bcab42780bdb994fca3ced5d8c62519d56
Reviewed-on: https://chromium-review.googlesource.com/1169913
Commit-Ready: Jonathan Brandmeyer <jbrandmeyer@chromium.org>
Tested-by: Jonathan Brandmeyer <jbrandmeyer@chromium.org>
Reviewed-by: Randall Spangler <rspangler@chromium.org>
Reviewed-by: Jett Rink <jettrink@chromium.org>
Diffstat (limited to 'common/usb_i2c.c')
-rw-r--r-- | common/usb_i2c.c | 9 |
1 files changed, 3 insertions, 6 deletions
diff --git a/common/usb_i2c.c b/common/usb_i2c.c index 4ceb4fe031..a92c411f1b 100644 --- a/common/usb_i2c.c +++ b/common/usb_i2c.c @@ -85,7 +85,6 @@ void usb_i2c_execute(struct usb_i2c_config const *config) int write_count = (config->buffer[1] >> 0) & 0xff; int read_count = (config->buffer[1] >> 8) & 0xff; int offset = 0; /* Offset for extended reading header. */ - int port; config->buffer[0] = 0; config->buffer[1] = 0; @@ -117,14 +116,12 @@ void usb_i2c_execute(struct usb_i2c_config const *config) * knows about. It should behave closer to * EC_CMD_I2C_PASSTHRU, which can protect ports and ranges. */ - port = i2c_ports[portindex].port; - i2c_lock(port, 1); - ret = i2c_xfer(port, slave_addr, + ret = i2c_xfer(i2c_ports[portindex].port, + slave_addr, (uint8_t *)(config->buffer + 2) + offset, write_count, (uint8_t *)(config->buffer + 2), - read_count, I2C_XFER_SINGLE); - i2c_lock(port, 0); + read_count); config->buffer[0] = usb_i2c_map_error(ret); } usb_i2c_write_packet(config, read_count + 4); |