pppd & n_gsm problem

[pavel havlik]

Hello

i'd like to ask to questions related to serial communication.

1. Is there any reason why DMA is disabled by default in beagle bone black serial_omap driver (CONFIG_SERIAL_8250_DMA=n) ?

    Is it safe to recompile it with CONFIG_SERIAL_8250_DMA=y?

2. I encountered board freeze (most probably deadlock but I'm not sure) when using pppd and n_gsm to communicate with GSM modem over serial line.

    When I connected to the board from internet and issued some command wich output a lot of data, board got stuck but when output little data

    board behaved correctly. After digging into n_gsm driver I discovered that board gets stuck when 8250 driver calls gsmld_write_wakeu() indicating upper layer (n_gsm)

    that is has room again for new data. n_gsm in return sends more data to 8250 driver directly in the gsmld_write_wakeup() , which is called from interrupt. I thought this could be the source of the deadlog but I didn't discover anything wrong. However the problem got treated when I instead of directly sending data in  gsmld_write_wakeup()  scheduled tasklet and only inside it I sent the data.

Any hint what could be wrong with the original code? The behaviour was the same both in 4.19.94-ti-r37 and 4.14.49-ti-r54.

Tank you for any help

regards

Pavel

Instead of

static void gsmld_write_wakeup(struct tty_struct *tty)
{
    struct gsm_mux *gsm = tty->disc_data;
    unsigned long flags;

    /* Queue poll */
    clear_bit(TTY_DO_WRITE_WAKEUP, &tty->flags);

    spin_lock_irqsave(&gsm->tx_lock, flags);
    gsm_data_kick(gsm);
    if (gsm->tx_bytes < TX_THRESH_LO) {
        gsm_dlci_data_sweep(gsm);
    }
    spin_unlock_irqrestore(&gsm->tx_lock, flags);
}

this

static void gsmld_write_wakeup(struct tty_struct *tty)
{
    struct gsm_mux *gsm = tty->disc_data;
 

    /* Queue poll */
    clear_bit(TTY_DO_WRITE_WAKEUP, &tty->flags);
    tasklet_schedule(&gsm->wkup_tasklet);

}

static void wkup_tasklet_handler(unsigned long data)
{
    struct gsm_mux *gsm = (struct gsm_mux *)data;
    unsigned long flags;

    spin_lock_irqsave(&gsm->tx_lock, flags);
    gsm_data_kick(gsm);
    if (gsm->tx_bytes < TX_THRESH_LO) {
        gsm_dlci_data_sweep(gsm);
    }
    spin_unlock_irqrestore(&gsm->tx_lock, flags);
}

this is call stack trace from omap driver up to n_gsm driver resulting in freez

[  361.652243] [<bf309b98>] (gsmld_write_wakeup [n_gsm]) from [<c07f7e5c>] (tty_wakeup+0x60/0x6c)
[  361.652255] [<c07f7e5c>] (tty_wakeup) from [<c0802ac4>] (tty_port_default_wakeup+0x24/0x30)
[  361.652266] [<c0802ac4>] (tty_port_default_wakeup) from [<c0802614>] (tty_port_tty_wakeup+0x20/0x24)
[  361.652277] [<c0802614>] (tty_port_tty_wakeup) from [<c081c074>] (uart_write_wakeup+0x24/0x2c)
[  361.652288] [<c081c074>] (uart_write_wakeup) from [<c0820460>] (serial8250_tx_chars+0x104/0x208)
[  361.652300] [<c0820460>] (serial8250_tx_chars) from [<c0821280>] (serial8250_handle_irq.part.11+0xd8/0xdc)
[  361.652312] [<c0821280>] (serial8250_handle_irq.part.11) from [<c08212ac>] (serial8250_handle_irq+0x28/0x2c)
[  361.652323] [<c08212ac>] (serial8250_handle_irq) from [<c082359c>] (omap8250_irq+0x3c/0x54)
[  361.652336] [<c082359c>] (omap8250_irq) from [<c01af844>] (__handle_irq_event_percpu+0x84/0x2d0)
[  361.652349] [<c01af844>] (__handle_irq_event_percpu) from [<c01afacc>] (handle_irq_event_percpu+0x3c/0x90)
[  361.652361] [<c01afacc>] (handle_irq_event_percpu) from [<c01afb68>] (handle_irq_event+0x48/0x6c)
[  361.652373] [<c01afb68>] (handle_irq_event) from [<c01b3aa8>] (handle_level_irq+0xdc/0x158)
[  361.652384] [<c01b3aa8>] (handle_level_irq) from [<c01ae8d4>] (generic_handle_irq+0x34/0x44)
[  361.652396] [<c01ae8d4>] (generic_handle_irq) from [<c01aefd4>] (__handle_domain_irq+0x8c/0xfc)
[  361.652408] [<c01aefd4>] (__handle_domain_irq) from [<c01022c0>] (omap_intc_handle_irq+0x44/0x9c)
[  361.652419] [<c01022c0>] (omap_intc_handle_irq) from [<c0101a0c>] (__irq_svc+0x6c/0xa8)