tegra: add kernel 4.19 support
[openwrt/openwrt.git] / target / linux / tegra / patches-4.19 / 100-serial8250-on-tegra-hsuart-recover-from-spurious-interrupts-due-to-tegra2-silicon-bug.patch
1 From patchwork Fri Jul 13 11:32:42 2018
2 Content-Type: text/plain; charset="utf-8"
3 MIME-Version: 1.0
4 Content-Transfer-Encoding: 7bit
5 Subject: serial8250 on tegra hsuart: recover from spurious interrupts due to
6 tegra2 silicon bug
7 X-Patchwork-Submitter: "David R. Piegdon" <lkml@p23q.org>
8 X-Patchwork-Id: 943440
9 Message-Id: <4676ea34-69ce-5422-1ded-94218b89f7d9@p23q.org>
10 To: linux-tegra@vger.kernel.org
11 Date: Fri, 13 Jul 2018 11:32:42 +0000
12 From: "David R. Piegdon" <lkml@p23q.org>
13 List-Id: <linux-tegra.vger.kernel.org>
14
15 Hi,
16 a while back I sent a few mails regarding spurious interrupts in the
17 UARTA (hsuart) block of the Tegra2 SoC, when using the 8250 driver for
18 it instead of the hsuart driver. After going down a pretty deep
19 debugging/testing hole, I think I found a patch that fixes the issue. So
20 far testing in a reboot-cycle suggests that the error frequency dropped
21 from >3% of all reboots to at least <0.05% of all reboots. Tests
22 continue to run over the weekend.
23
24 The patch below already is a second iteration; the first did not reset
25 the MCR or contain the lines below '// clear interrupts'. This resulted
26 in no more spurious interrupts, but in a few % of spurious interrupts
27 that were recovered the UART block did not receive any characters any
28 more. So further resetting was required to fully reacquire operational
29 state of the UART block.
30
31 I'd love any comments/suggestions on this!
32
33 Cheers,
34
35 David
36
37 --- a/drivers/tty/serial/8250/8250_core.c
38 +++ b/drivers/tty/serial/8250/8250_core.c
39 @@ -136,6 +136,38 @@ static irqreturn_t serial8250_interrupt(
40 "serial8250: too much work for irq%d\n", irq);
41 break;
42 }
43 +
44 +#ifdef CONFIG_ARCH_TEGRA_2x_SOC
45 + if (!handled && (port->type == PORT_TEGRA)) {
46 + /*
47 + * Fix Tegra 2 CPU silicon bug where sometimes
48 + * "TX holding register empty" interrupts result in a
49 + * bad (metastable?) state in Tegras HSUART IP core.
50 + * Only way to recover seems to be to reset all
51 + * interrupts as well as the TX queue and the MCR.
52 + * But we don't want to loose any outgoing characters,
53 + * so only do it if the RX and TX queues are empty.
54 + */
55 + unsigned char lsr = port->serial_in(port, UART_LSR);
56 + const unsigned char fifo_empty_mask =
57 + (UART_LSR_TEMT | UART_LSR_THRE);
58 + if (((lsr & (UART_LSR_DR | fifo_empty_mask)) ==
59 + fifo_empty_mask)) {
60 + port->serial_out(port, UART_IER, 0);
61 + port->serial_out(port, UART_MCR, 0);
62 + serial8250_clear_and_reinit_fifos(up);
63 + port->serial_out(port, UART_MCR, up->mcr);
64 + port->serial_out(port, UART_IER, up->ier);
65 + // clear interrupts
66 + serial_port_in(port, UART_LSR);
67 + serial_port_in(port, UART_RX);
68 + serial_port_in(port, UART_IIR);
69 + serial_port_in(port, UART_MSR);
70 + up->lsr_saved_flags = 0;
71 + up->msr_saved_flags = 0;
72 + }
73 + }
74 +#endif
75 } while (l != end);
76
77 spin_unlock(&i->lock);