kernel/linux.git/include/linux/serial_core.h, branch v4.9.319

serial: core: Allow processing sysrq at port unlock time

2019-12-21T09:40:50+00:00

[ Upstream commit d6e1935819db0c91ce4a5af82466f3ab50d17346 ] Right now serial drivers process sysrq keys deep in their character receiving code. This means that they've already grabbed their port->lock spinlock. This can end up getting in the way if we've go to do serial stuff (especially kgdb) in response to the sysrq. Serial drivers have various hacks in them to handle this. Looking at '8250_port.c' you can see that the console_write() skips locking if we're in the sysrq handler. Looking at 'msm_serial.c' you can see that the port lock is dropped around uart_handle_sysrq_char(). It turns out that these hacks aren't exactly perfect. If you have lockdep turned on and use something like the 8250_port hack you'll get a splat that looks like: WARNING: possible circular locking dependency detected [...] is trying to acquire lock: ... (console_owner){-.-.}, at: console_unlock+0x2e0/0x5e4 but task is already holding lock: ... (&port_lock_key){-.-.}, at: serial8250_handle_irq+0x30/0xe4 which lock already depends on the new lock. the existing dependency chain (in reverse order) is: -> #1 (&port_lock_key){-.-.}: _raw_spin_lock_irqsave+0x58/0x70 serial8250_console_write+0xa8/0x250 univ8250_console_write+0x40/0x4c console_unlock+0x528/0x5e4 register_console+0x2c4/0x3b0 uart_add_one_port+0x350/0x478 serial8250_register_8250_port+0x350/0x3a8 dw8250_probe+0x67c/0x754 platform_drv_probe+0x58/0xa4 really_probe+0x150/0x294 driver_probe_device+0xac/0xe8 __driver_attach+0x98/0xd0 bus_for_each_dev+0x84/0xc8 driver_attach+0x2c/0x34 bus_add_driver+0xf0/0x1ec driver_register+0xb4/0x100 __platform_driver_register+0x60/0x6c dw8250_platform_driver_init+0x20/0x28 ... -> #0 (console_owner){-.-.}: lock_acquire+0x1e8/0x214 console_unlock+0x35c/0x5e4 vprintk_emit+0x230/0x274 vprintk_default+0x7c/0x84 vprintk_func+0x190/0x1bc printk+0x80/0xa0 __handle_sysrq+0x104/0x21c handle_sysrq+0x30/0x3c serial8250_read_char+0x15c/0x18c serial8250_rx_chars+0x34/0x74 serial8250_handle_irq+0x9c/0xe4 dw8250_handle_irq+0x98/0xcc serial8250_interrupt+0x50/0xe8 ... other info that might help us debug this: Possible unsafe locking scenario: CPU0 CPU1 ---- ---- lock(&port_lock_key); lock(console_owner); lock(&port_lock_key); lock(console_owner); *** DEADLOCK *** The hack used in 'msm_serial.c' doesn't cause the above splats but it seems a bit ugly to unlock / lock our spinlock deep in our irq handler. It seems like we could defer processing the sysrq until the end of the interrupt handler right after we've unlocked the port. With this scheme if a whole batch of sysrq characters comes in one irq then we won't handle them all, but that seems like it should be a fine compromise. Signed-off-by: Douglas Anderson Signed-off-by: Greg Kroah-Hartman Signed-off-by: Sasha Levin

serial: core: Make sure compiler barfs for 16-byte earlycon names

2018-08-03T05:55:25+00:00

[ Upstream commit c1c734cb1f54b062f7e67ffc9656d82f5b412b9c ] As part of bringup I ended up wanting to call an earlycon driver by a name that was exactly 16-bytes big, specifically "qcom_geni_serial". Unfortunately, when I tried this I found that things compiled just fine. They just didn't work. Specifically the compiler felt perfectly justified in initting the ".name" field of "struct earlycon_id" with the full 16-bytes and just skipping the '\0'. Needless to say, that behavior didn't seem ideal, but I guess someone must have allowed it for a reason. One way to fix this is to shorten the name field to 15 bytes and then add an extra byte after that nobody touches. This should always be initted to 0 and we're golden. There are, of course, other ways to fix this too. We could audit all the users of the "name" field and make them stop at both null termination or at 16 bytes. We could also just make the name field much bigger so that we're not likely to run into this. ...but both seem like we'll just hit the bug again. Signed-off-by: Douglas Anderson Signed-off-by: Greg Kroah-Hartman Signed-off-by: Sasha Levin Signed-off-by: Greg Kroah-Hartman

earlycon: Use a pointer table to fix __earlycon_table stride

2018-05-01T22:13:09+00:00

commit dd709e72cb934eefd44de8d9969097173fbf45dc upstream. Commit 99492c39f39f ("earlycon: Fix __earlycon_table stride") tried to fix __earlycon_table stride by forcing the earlycon_id struct alignment to 32 and asking the linker to 32-byte align the __earlycon_table symbol. This fix was based on commit 07fca0e57fca92 ("tracing: Properly align linker defined symbols") which tried a similar fix for the tracing subsystem. However, this fix doesn't quite work because there is no guarantee that gcc will place structures packed into an array format. In fact, gcc 4.9 chooses to 64-byte align these structs by inserting additional padding between the entries because it has no clue that they are supposed to be in an array. If we are unlucky, the linker will assign symbol "__earlycon_table" to a 32-byte aligned address which does not correspond to the 64-byte aligned contents of section "__earlycon_table". To address this same problem, the fix to the tracing system was subsequently re-implemented using a more robust table of pointers approach by commits: 3d56e331b653 ("tracing: Replace syscall_meta_data struct array with pointer array") 654986462939 ("tracepoints: Fix section alignment using pointer array") e4a9ea5ee7c8 ("tracing: Replace trace_event struct array with pointer array") Let's use this same "array of pointers to structs" approach for EARLYCON_TABLE. Fixes: 99492c39f39f ("earlycon: Fix __earlycon_table stride") Signed-off-by: Daniel Kurtz Suggested-by: Aaron Durbin Reviewed-by: Rob Herring Tested-by: Guenter Roeck Reviewed-by: Guenter Roeck Cc: stable Signed-off-by: Greg Kroah-Hartman

ACPI: parse SPCR and enable matching console

2016-09-28T15:46:46+00:00

'ARM Server Base Boot Requiremets' [1] mentions SPCR (Serial Port Console Redirection Table) [2] as a mandatory ACPI table that specifies the configuration of serial console. Defer initialization of DT earlycon until ACPI/DT decision is made. Parse the ACPI SPCR table, setup earlycon if required, enable specified console. Thanks to Peter Hurley for explaining how this should work. [1] http://infocenter.arm.com/help/index.jsp?topic=/com.arm.doc.den0044a/index.html [2] https://msdn.microsoft.com/en-us/library/windows/hardware/dn639132(v=vs.85).aspx Signed-off-by: Aleksey Makarov Acked-by: Rafael J. Wysocki Reviewed-by: Peter Hurley Tested-by: Kefeng Wang Tested-by: Christopher Covington Signed-off-by: Greg Kroah-Hartman

tty: serial_core: add tty NULL check to uart_tx_stopped

2016-09-15T10:37:09+00:00

Commit 761ed4a94582 ("tty: serial_core: convert uart_close to use tty_port_close") created a case where a port used for a console does not get shutdown on tty closing. Then a call to uart_tx_stopped() segfaults because the tty is NULL. This could be fixed to restore old behavior, but we also want to allow tty_ports to work without a tty attached. So this change to allow a NULL tty_struct is needed either way. Fixes: 761ed4a94582 ("tty: serial_core: convert uart_close to use tty_port_close") Reported-by: kernel test robot Signed-off-by: Rob Herring Signed-off-by: Greg Kroah-Hartman

serial: earlycon: Extend earlycon command line option to support 64-bit addresses

2016-09-02T13:03:35+00:00

earlycon implementation used "unsigned long" internally, but there are systems (ARM with LPAE) where sizeof(unsigned long) == 4 and uart is mapped beyond 4GiB address range. Switch to resource_size_t internally and replace obsoleted simple_strtoul() with kstrtoull(). Signed-off-by: Alexander Sverdlin Signed-off-by: Greg Kroah-Hartman

earlycon: mark earlycon code as __used iif the caller is built-in

2016-06-25T21:31:04+00:00

Keep earlycon related symbols only when CONFIG_SERIAL_EARLYCON is enabled and the driver is built-in. This will be helpful to clean up ifdefs surrounding earlycon code in serial drivers. Signed-off-by: Masahiro Yamada Acked-by: Arnd Bergmann Signed-off-by: Greg Kroah-Hartman

serial: 8250: export get_mctrl function

2016-04-30T16:26:55+00:00

Exposes get_mctrl() function so that it can be overriden with platform specific implementation. Signed-off-by: Wan Ahmad Zainie Reviewed-by: Andy Shevchenko Reviewed-by: Heikki Krogerus Signed-off-by: Greg Kroah-Hartman

serial: core: Prevent unsafe uart port access, part 2

2016-04-30T16:26:55+00:00

For serial core operations not already excluded by holding port->mutex, use reference counting to protect deferencing the state->uart_port. Introduce helper functions, uart_port_ref() and uart_port_deref(), to wrap uart_port access, and helper macros, uart_port_lock() and uart_port_unlock(), to wrap combination uart_port access with uart port lock sections. Port removal in uart_remove_one_port() waits for reference count to drop to zero before detaching the uart port from struct uart_state. For functions only reading the tx circular buffer indexes (where the uart port lock is claimed to prevent concurrent users), a NULL uart port is simply ignored and the operation completes normally. For functions change the tx circular buffer indexes (where the uart port lock is claimed to prevent concurrent users), the operation is aborted if the uart port is NULL (ie., has been detached). Signed-off-by: Peter Hurley Signed-off-by: Greg Kroah-Hartman

of: earlycon: Move address translation to of_setup_earlycon()

2016-02-07T06:07:37+00:00

Cleanup the early DT/earlycon separation; remove the 'addr' parameter from of_setup_earlycon() and get the uart phys addr directly with a new wrapper function, of_flat_dt_translate_addr(). Limit fdt_translate_address() to file scope. Acked-by: Rob Herring Signed-off-by: Peter Hurley Signed-off-by: Greg Kroah-Hartman