1. Pass the correct conn_id to protocomm_req_handle
In transport_simple_ble_write(), passing param->exec_write.conn_id would
be invalid. Instead param->write.conn_id should be passed
Similar change in transport_simple_ble_exec_write() to use
param->exec_write.conn_id
2. simple_ble_start() assumes that the mode is BLE only and enables
Bluetooth controller accordingly. For, cases having BT + BLE like Alexa
(Provisioning over BLE + Audio over classic BT), this assumption should
be removed.
Anurag Kar [Mon, 1 Apr 2019 09:22:04 +0000 (14:52 +0530)]
esp_http_server : Only accept new connections if server has capacity to handle more
This fix prevents HTTP server from accepting new connections when the total count of connected
sockets has reached the max_open_sockets limit set during configuration. The pending connections
are kept in backlog until atleast one of the connected sockets is closed. The maximum number of
connection requests that can kept in backlog is specified as backlog_conn configuration option.
Note that this modification has no effect when LRU purge is enabled.
Also added sanity check on setting for max_open_sockets during configuration.
Solution suggested by jimparis https://github.com/espressif/esp-idf/issues/3183#issue-421234265
spi: fix the bug of connecting SPI peripheral to read-only pins
The requirements of pin capabilites is different for spi master and
slave. The master needs CS, SCLK, MOSI to be output-able, while slave
needs MISO to be output-able.
Previous code is for master only.
This commit allows to place other 3 pins than MISO on input-only pins
for slaves. Refactoring for spi_common is also included.
Angus Gratton [Tue, 19 Mar 2019 23:42:29 +0000 (10:42 +1100)]
cmake: Set COMPONENT_PATH during early expansion of dependencies
Works around bug where components/soc/CMakeLists.txt was testing "EXISTS
${COMPONENT_PATH}/${SOC_NAME}" and this test could pass during early
expansion if COMPONENT_PATH was empty and a directory /esp32 exists
on Windows.
krzychb [Tue, 19 Mar 2019 15:47:39 +0000 (16:47 +0100)]
docs: Upgraded sphinx package version to prevent build failures on readthedocs site. Upgraded versions of other packages that are used during documentation build. (backport v3.2)
Ivan Grokhotkov [Wed, 13 Mar 2019 12:19:55 +0000 (20:19 +0800)]
examples: don't enable buffering on stdout in console examples
newlib uses significantly more stack space when printing to an
unbuffered stream. To reduce the amount of stack space required to
use the console, don’t disable buffering. linenoise should support
unbuffered stdout instead.
David Cermak [Mon, 11 Mar 2019 10:49:29 +0000 (11:49 +0100)]
mdns: fix possible crash when probing on particular interface with duplicated service instances due to naming conflicts on network
Issue: MDNS server initially sends probing packets to resolve naming confilicts with already registered service instances. In case of a conflict, instance name is altered and probing restarts. Original instance however wasnnot removed from the structure and upon service removal only one entry was removed and a dangling service might have been kept in the structure to bring about a crash.
Resolution: Keep only one instance of a service in the probing structure.
David Cermak [Mon, 4 Mar 2019 11:32:10 +0000 (12:32 +0100)]
mdns: enable pcbs before starting service thread to avoid updating pcb's internal variables from concurent tasks
possible race condition: user task runs mdns_init, which enables pcbs while mdns-task already created could execute enable/disable of the same pcbs if an appropriate system event received
David Cermak [Mon, 4 Mar 2019 10:13:52 +0000 (11:13 +0100)]
mdns: fix possible deadlock on mdns deinit calling mdns_free()
mnds_free() initiates stop and delete timer tasks, which after locking the mutex could lead to a dead lock in case timer task executed before deleting the task, as it would wait indefinitelly for unlocking the mutex. This condition is fixed by calling _mdns_stop_timer without locking the mutex, because there's no need to protect any data when stopping and deleting the timer task
David Cermak [Fri, 1 Mar 2019 15:59:38 +0000 (16:59 +0100)]
mdsn: fix race condition in updating packet data from user task when failed to allocate or queue a new service
Issue: mdns_service_add API allocates and queues an action to be processed in mdns task context; when allocation or queueing fails, allocated structure needs to be freed. Function _mdns_free_service did not only fee all the structures, but also updates packet data.
Resolution: Moved removal of packet data outside of _mdns_free_service function.
David Cermak [Mon, 25 Feb 2019 13:29:39 +0000 (14:29 +0100)]
mdns: fix possible crash when packet scheduled to transmit contained service which might have been already removed
packets scheduled to transmit are pushed to action queue and removed from tx_queue_head structure, which is searched for all remaining services and while service is removed, then service questions/asnwers are also removed from this structure. This update fixes possible crash when packet is pushed to action queue, and when service is removed, its answers are removed from tx_queue_head, but not from action queue. this could lead to a crash when the packet is poped from action queue containing questions/answers to already removed (freed) service
Ivan Grokhotkov [Mon, 4 Mar 2019 08:59:09 +0000 (16:59 +0800)]
mdns: use binary semaphore instead of mutex when searching
mdns_search_once_t::lock is used to synchronize tasks (taken by one
task and given by the other) so it should not be a mutex.
Convert to semaphore, and rename to indicate its purpose.
Angus Gratton [Fri, 1 Mar 2019 04:38:23 +0000 (15:38 +1100)]
confgen: Fix bug with JSON metadata conditional range generation
When generating JSON metadata for ranges where there are conditional ranges (ie different allowed range
depending on another config setting), the JSON metadata would always have the last named range as
the expression was not evaluated properly.
Thanks to ulfalizer on GitHub for pointing this out.
Piyush Shah [Mon, 4 Mar 2019 18:51:03 +0000 (00:21 +0530)]
esp_http_server: Provide apps an option to let http_server ignore sess_ctx changes
By default, if a URI handler changes the http session context,
the webserver internally clears the older context after the handler
returns. However, if applications want to change this behavior and
manage the allocation/de-allocation/freeing themselves and let the
server handle only the "socket close" case, this commit provides such
an option.
Darian Leung [Tue, 26 Feb 2019 12:58:02 +0000 (20:58 +0800)]
freertos: Make xPortGetCoreID() volatile
When xPortGetCoreID() is called twice within a function,
it might only be called once after compilation. This
commit makes the inline assembly of the function volatile.
Ivan Grokhotkov [Fri, 22 Feb 2019 09:28:43 +0000 (17:28 +0800)]
nvs: do eager cleanup of HashListBlocks
Previously when HashList was removing items, HashListBlocks were
removed lazily. This resulted in empty HashListBlocks dangling around
in full pages, even when all items have been erased. These blocks
would only be deleted when NVS was re-initialized
(nvs_flash_deinit/nvs_flash_init).
This change does eager cleanup instead, based on the code from
@negativekelvin offered in
https://github.com/espressif/esp-idf/issues/1642#issuecomment-367227994.
Ivan Grokhotkov [Fri, 22 Feb 2019 13:27:43 +0000 (21:27 +0800)]
esp_timer: fix occasional failures in "monotonic values" test
1. ref_clock used in unit tests occasionally produces time off by ~100
microseconds shortly after being started. Add a delay to let
ref_clock stabilise, until the cause is found.
2. Reduce roundoff error accumulation which would occasionally cause
the test to fail, by choosing an overflow value which can be divided
by APB frequency.
3. Move time sampling part of the test into an IRAM function to
reduce variations due to cache behavior.
4. Remove calculation of "standard deviation" in the test, as what was
calculated was not actually standard deviation, and it did not add any
useful information.
Ivan Grokhotkov [Thu, 29 Nov 2018 07:18:11 +0000 (15:18 +0800)]
bootloader: check previously used clock frequency at run time
In the situation when bootloader was compiled for 240MHz, and app was
compiled for 160MHz, and the chip is a revision 0 chip, the
bootloader will assume that the application has also been running at
240MHz. This will cause the chip to lock up later. Modify this to use
a run time check of DPORT_CPUPERIOD_SEL, which indicates which of the
PLL frequencies was used.
Ivan Grokhotkov [Wed, 13 Feb 2019 03:40:48 +0000 (11:40 +0800)]
make: fix issues related to EXTRA_COMPONENT_DIRS
1. When one of the COMPONENT_DIRS points to a component directory
(i.e. a directory containing component.mk, not a directory of multiple
components), and there is a subdirectory in it which also contains
a component, the subdirectory was mistakenly added to the list of
components and compiled.
2. When one of the COMPONENT_DIRS points to a component directory, and
the parent directory contained a directory with the same name as
another component, that directory would be mistakenly added to the
COMPONENT_PATHS.
and Makefile sets EXTRA_COMPONENT_DIRS=$(realpath ../mycomponent),
then "esp32" directory which is at the same level as mycomponent
was added to COMPONENT_PATHS.
3. If EXTRA_COMPONENT_DIRS pointed to a directory with a list of
components, and one of the subdirectories was not a component, but
had the same name as another component, than that directory would be
mistakenly added to COMPONENT_PATHS instead of the real esp32
component directory.
For example:
my_components/
my_component/
component.mk
my_component.c
esp32/
(some random stuff)
and EXTRA_COMPONENT_DIRS would point to my_components/, then "esp32"
directory would be added to COMPONENT_PATHS instead of the real esp32
component directory.
krzychb [Sat, 16 Feb 2019 19:46:49 +0000 (20:46 +0100)]
docs: Preparation for upgrade of Sphinx and Breathe packages to versions 1.8.4 and 4.11.1. Updated check of Sphinx warnings to account for the filename instead of the whole path, that for the same errors may be different between English and Chinese docs. (backport v3.2)
Angus Gratton [Fri, 22 Feb 2019 01:20:11 +0000 (12:20 +1100)]
cmake: Fix psram workaround compiler flag application
Previously, this compiler flag was not being applied
regardless of CONFIG_SPIRAM_CACHE_WORKAROUND setting.
Explanation: add_compile_options() only applies to
source files added after the function is run, or in
subdirectories added after the function is run. In
this case, no new source files were being added after
this function was run.
Paul Abbott [Mon, 24 Sep 2018 19:23:26 +0000 (15:23 -0400)]
driver(rmt): Fixed some RMT related issues.
1. Add missing get function to RMT API: rmt_get_idle_level().
2. Suppress error message from rmt_wait_tx_done() if wait_time==0 (allows for polling).
3. Remove the blank lines between parameter lists.
projects / esp-idf / log