| lists.openwall.net | lists / announce owl-users owl-dev john-users john-dev passwdqc-users yescrypt popa3d-users / oss-security kernel-hardening musl sabotage tlsify passwords / crypt-dev xvendor / Bugtraq Full-Disclosure linux-kernel linux-netdev linux-ext4 linux-hardening linux-cve-announce PHC | |
|
Open Source and information security mailing list archives
| ||
|
Message-Id: <20230519123801.e1123cab701b6cb0ffbf367f@hugovil.com>
Date: Fri, 19 May 2023 12:38:01 -0400
From: Hugo Villeneuve <hugo@...ovil.com>
To: Lech Perczak <lech.perczak@...lingroup.com>
Cc: Greg Kroah-Hartman <gregkh@...uxfoundation.org>,
Jiri Slaby <jirislaby@...nel.org>,
Lech Perczak <l.perczak@...lintechnologies.com>,
Tomasz Moń <tomasz.mon@...lingroup.com>,
linux-gpio@...r.kernel.org,
Hugo Villeneuve <hvilleneuve@...onoff.com>,
linux-serial@...r.kernel.org, linux-kernel@...r.kernel.org
Subject: Re: [RFC PATCH v2 2/2] serial: sc16is7xx: fix regression with GPIO
configuration
On Fri, 19 May 2023 18:00:54 +0200
Lech Perczak <lech.perczak@...lingroup.com> wrote:
>
> W dniu 19.05.2023 o 15:30, Hugo Villeneuve pisze:
> > On Fri, 19 May 2023 15:00:58 +0200
> > Lech Perczak <lech.perczak@...lingroup.com> wrote:
> >
> >> Hello Hugo,
> >>
> >> A couple of remarks inline.
> >>
> >> W dniu 18.05.2023 o 15:29, Hugo Villeneuve pisze:
> >>> From: Hugo Villeneuve <hvilleneuve@...onoff.com>
> >>>
> >>> Commit 679875d1d880 ("sc16is7xx: Separate GPIOs from modem control lines")
> >>> and commit 21144bab4f11 ("sc16is7xx: Handle modem status lines")
> >>> changed the function of the GPIOs pins to act as modem control
> >>> lines without any possibility of selecting GPIO function.
> >>>
> >>> As a consequence, applications that depends on GPIO lines configured
> >>> by default as GPIO pins no longer work as expected.
> >>>
> >>> Also, the change to select modem control lines function was done only
> >>> for channel A of dual UART variants (752/762). This was not documented
> >>> in the log message.
> >>>
> >>> This new patch allows to specify GPIO or modem control line function
> >>> in the device tree, and for each of the ports (A or B).
> >>>
> >>> This is done by using the new device-tree property named
> >>> "modem-control-line-ports" (property added in separate patch).
> >>>
> >>> Boards that need to have GPIOS configured as modem control lines
> >>> should add that property to their device tree. Here is a list of
> >>> boards using the sc16is7xx driver in their device tree and that may
> >>> need to be modified:
> >>> arm64/boot/dts/freescale/fsl-ls1012a-frdm.dts
> >>> mips/boot/dts/ingenic/cu1830-neo.dts
> >>> mips/boot/dts/ingenic/cu1000-neo.dts
> >>>
> >>> Fixes: 679875d1d880 ("sc16is7xx: Separate GPIOs from modem control lines")
> >>> Fixes: 21144bab4f11 ("sc16is7xx: Handle modem status lines")
> >>> Signed-off-by: Hugo Villeneuve <hvilleneuve@...onoff.com>
> >>> ---
> >>> v1 -> v2: Rebase because original patch did not apply properly
> >>> Add git base-commit
> >>>
> >>> drivers/tty/serial/sc16is7xx.c | 42 ++++++++++++++++++++--------------
> >>> 1 file changed, 25 insertions(+), 17 deletions(-)
> >>>
> >>> diff --git a/drivers/tty/serial/sc16is7xx.c b/drivers/tty/serial/sc16is7xx.c
> >>> index abad091baeea..4e3d2325ef6e 100644
> >>> --- a/drivers/tty/serial/sc16is7xx.c
> >>> +++ b/drivers/tty/serial/sc16is7xx.c
> >>> @@ -236,7 +236,8 @@
> >>>
> >>> /* IOControl register bits (Only 750/760) */
> >>> #define SC16IS7XX_IOCONTROL_LATCH_BIT (1 << 0) /* Enable input latching */
> >>> -#define SC16IS7XX_IOCONTROL_MODEM_BIT (1 << 1) /* Enable GPIO[7:4] as modem pins */
> >>> +#define SC16IS7XX_IOCONTROL_MODEM_A_BIT (1 << 1) /* Enable GPIO[7:4] as modem A pins */
> >>> +#define SC16IS7XX_IOCONTROL_MODEM_B_BIT (1 << 2) /* Enable GPIO[3:0] as modem B pins */
> >>> #define SC16IS7XX_IOCONTROL_SRESET_BIT (1 << 3) /* Software Reset */
> >>>
> >>> /* EFCR register bits */
> >>> @@ -306,7 +307,6 @@ struct sc16is7xx_devtype {
> >>> char name[10];
> >>> int nr_gpio;
> >>> int nr_uart;
> >>> - int has_mctrl;
> >>> };
> >>>
> >>> #define SC16IS7XX_RECONF_MD (1 << 0)
> >>> @@ -447,35 +447,30 @@ static const struct sc16is7xx_devtype sc16is74x_devtype = {
> >>> .name = "SC16IS74X",
> >>> .nr_gpio = 0,
> >>> .nr_uart = 1,
> >>> - .has_mctrl = 0,
> >>> };
> >>>
> >>> static const struct sc16is7xx_devtype sc16is750_devtype = {
> >>> .name = "SC16IS750",
> >>> - .nr_gpio = 4,
> >>> + .nr_gpio = 8,
> >>> .nr_uart = 1,
> >>> - .has_mctrl = 1,
> >>> };
> >>>
> >>> static const struct sc16is7xx_devtype sc16is752_devtype = {
> >>> .name = "SC16IS752",
> >>> - .nr_gpio = 0,
> >>> + .nr_gpio = 8,
> >>> .nr_uart = 2,
> >>> - .has_mctrl = 1,
> >>> };
> >>>
> >>> static const struct sc16is7xx_devtype sc16is760_devtype = {
> >>> .name = "SC16IS760",
> >>> - .nr_gpio = 4,
> >>> + .nr_gpio = 8,
> >>> .nr_uart = 1,
> >>> - .has_mctrl = 1,
> >>> };
> >>>
> >>> static const struct sc16is7xx_devtype sc16is762_devtype = {
> >>> .name = "SC16IS762",
> >>> - .nr_gpio = 0,
> >>> + .nr_gpio = 8,
> >>> .nr_uart = 2,
> >>> - .has_mctrl = 1,
> >>> };
> >>>
> >>> static bool sc16is7xx_regmap_volatile(struct device *dev, unsigned int reg)
> >>> @@ -1456,12 +1451,6 @@ static int sc16is7xx_probe(struct device *dev,
> >>> SC16IS7XX_EFCR_RXDISABLE_BIT |
> >>> SC16IS7XX_EFCR_TXDISABLE_BIT);
> >>>
> >>> - /* Use GPIO lines as modem status registers */
> >>> - if (devtype->has_mctrl)
> >>> - sc16is7xx_port_write(&s->p[i].port,
> >>> - SC16IS7XX_IOCONTROL_REG,
> >>> - SC16IS7XX_IOCONTROL_MODEM_BIT);
> >>> -
> >>> /* Initialize kthread work structs */
> >>> kthread_init_work(&s->p[i].tx_work, sc16is7xx_tx_proc);
> >>> kthread_init_work(&s->p[i].reg_work, sc16is7xx_reg_proc);
> >>> @@ -1497,6 +1486,25 @@ static int sc16is7xx_probe(struct device *dev,
> >>> prop, p, u)
> >>> if (u < devtype->nr_uart)
> >>> s->p[u].irda_mode = true;
> >>> +
> >>> + val = 0;
> >>> +
> >>> + of_property_for_each_u32(dev->of_node, "nxp,modem-control-line-ports",
> >>> + prop, p, u)
> >> The general idea looks solid to me.
> >> I think, that with this property set, we should also reduce number of GPIOs exported by gpiochip.
> > OK, I will look into that.
> >
> >>> + if (u < devtype->nr_uart) {
> >>> + /* Use GPIO lines as modem control lines */
> >>> + if (u == 0)
> >>> + val |= SC16IS7XX_IOCONTROL_MODEM_A_BIT;
> >>> + else if (u == 1)
> >>> + val |= SC16IS7XX_IOCONTROL_MODEM_B_BIT;
> >>> + }
> >>> +
> >>> + if (val)
> >>> + regmap_update_bits(
> >>> + s->regmap,
> >>> + SC16IS7XX_IOCONTROL_REG << SC16IS7XX_REG_SHIFT,
> >>> + SC16IS7XX_IOCONTROL_MODEM_A_BIT |
> >>> + SC16IS7XX_IOCONTROL_MODEM_B_BIT, val);
> >> I just had a chance to test this on my hardware (with SC16IS760), and something still doesn't work.
> >> When looking into /sys/kernel/debug/regmap/<device>/registers,
> >> it seems like the proper bits are set, also taking the regmap shift into account.
> >> For reference - I did my test basing on v6.4-rc2.
> >>
> >> I used a loopback plug and statserial to perform the test.
> >> Without your patch, when DTR is set, all other pins (DCD, DSR, RI) are set. With it - they are cleared, according to statserial.
> > It is strange, because the only conceptual change is the moment in time when we set the IOControl register. Before, IOControl was set within the for/loop for initializing each port. Now it is done after.
> >
> > Maybe you could try to put the section that reads the DT property just after resetting the device.
> >
> > While suggesting this, I just noticed that when the device is reset, we do not wait before initiating communication with the device. Thefore, I would suggest the following change (that I will submit as a separate patch):
> >
> > regmap_write(s->regmap, SC16IS7XX_IOCONTROL_REG << SC16IS7XX_REG_SHIFT,
> > SC16IS7XX_IOCONTROL_SRESET_BIT);
> >
> > + /* After reset, the host must wait at least 3us before initializing a
> > + * communication with the device: */
> > + usleep_range(3, 5);
> >
> > Then like I said, insert the code that read the modem-control-line-port DT property right after that to see if it helps.
> >
> > Hugo.
>
> Hello,
>
> I tried those suggestions, together with the patch you mentioned in the other email - all to no avail.
> I added a debug print to confirm, that the property is detected, and regmap access is executed - and it is. This is so strange.
> Maybe the offsets in regmap are still wrong, for some reason?
> I also double checked if it was possible to alter and check GPIO states through GPIO subsystem - it is.
Then it seems that the remaining difference is the number of GPIOs that are configured, which for the moment is always the value set in constant nr_gpio (for each device).
I already implemented and tested your suggestion of lowering the number of GPIOs exported when using some pins as modem status lines. It seems to be working good, altough I have only tested that the correct number of GPIOs are now exported.
I will send you the test patch and let me know if it helps.
Hugo.
Powered by blists - more mailing lists