linux-watchdog.vger.kernel.org archive mirror
 help / color / mirror / Atom feed
From: Guenter Roeck <linux@roeck-us.net>
To: Srinivas Neeli <srinivas.neeli@xilinx.com>,
	shubhraj@xilinx.com, sgoud@xilinx.com, michal.simek@xilinx.com
Cc: wim@linux-watchdog.org, linux-watchdog@vger.kernel.org,
	linux-kernel@vger.kernel.org, git@xilinx.com
Subject: Re: [PATCH 2/3] watchdog: xilinx_wwdt: Add Versal Window watchdog mode
Date: Thu, 27 May 2021 07:39:21 -0700	[thread overview]
Message-ID: <fb83065e-7870-c830-7dde-efb3af8cc25c@roeck-us.net> (raw)
In-Reply-To: <20210527095709.12841-3-srinivas.neeli@xilinx.com>

The subject is confusing. This patch does not add "Versal Window watchdog mode",
it adds a driver which supports the Versal Window watchdog. The subject should
be something like "watchdog: Add Versal Window watchdog driver".

On 5/27/21 2:57 AM, Srinivas Neeli wrote:
> Versal watchdog driver uses Window watchdog mode. Window watchdog
> timer(WWDT) contains closed(first) and open(second) window with
> 32 bit width. WWDT will generate an interrupt after the first window
> timeout and reset signal after the second window timeout. Timeout
> and Pre-timeout configuration, Stop and Refresh trigger only in open
> window.

Does that mean that userspace can not send a keepalive before the pretimeout
expires ? That is not how this is supposed to work. If that is the case,
this is not a pretimeout. Pretimeout is supposed to be used for emergency
handling, such as triggering a crash dump. The above description as well
as the code suggests that an interrupt is all but mandatory.

Assuming that is all the case, ie that keepalive is only possible
after the interrupt was received, and that timeout updates are also
only possible after an interrupt was received, one possible handling would
to record that a keepalive was received, and to record that a request to
update the timeout was received if that happens while the window
is closed. Then, when the interrupt is received, those actions can be
applied. Otherwise, if there was no keepalive request prior to the interrupt,
the interrupt handler would call watchdog_notify_pretimeout().

Of course, this is all just guesswork.

More comments below.

Guenter

> 
> Signed-off-by: Srinivas Neeli <srinivas.neeli@xilinx.com>
> ---
>   drivers/watchdog/Kconfig       |  16 ++
>   drivers/watchdog/Makefile      |   1 +
>   drivers/watchdog/xilinx_wwdt.c | 382 +++++++++++++++++++++++++++++++++
>   3 files changed, 399 insertions(+)
>   create mode 100644 drivers/watchdog/xilinx_wwdt.c
> 
> diff --git a/drivers/watchdog/Kconfig b/drivers/watchdog/Kconfig
> index 355100dad60a..2ed17f278783 100644
> --- a/drivers/watchdog/Kconfig
> +++ b/drivers/watchdog/Kconfig
> @@ -307,6 +307,22 @@ config XILINX_WATCHDOG
>   	  To compile this driver as a module, choose M here: the
>   	  module will be called of_xilinx_wdt.
>   
> +config XILINX_WINDOW_WATCHDOG
> +	tristate "Xilinx Window Watchdog timer"
> +	depends on HAS_IOMEM
> +	select WATCHDOG_CORE
> +	help
> +	  Window Watchdog driver for the versal_wwdt ip core.
> +	  Window watchdog timer(WWDT) contains closed(first) and
> +	  open(second) window with 32 bit width. WWDT will generate
> +	  an interrupt after the first window timeout and reset
> +	  signal after the second window timeout. Timeout and
> +	  Pre-timeout configuration, Stop and Refresh trigger
> +	  only in open window.
> +
> +	  To compile this driver as a module, choose M here: the
> +	  module will be called xilinx_wwdt.
> +
>   config ZIIRAVE_WATCHDOG
>   	tristate "Zodiac RAVE Watchdog Timer"
>   	depends on I2C
> diff --git a/drivers/watchdog/Makefile b/drivers/watchdog/Makefile
> index a7eade8b4d45..97d10c3d39c7 100644
> --- a/drivers/watchdog/Makefile
> +++ b/drivers/watchdog/Makefile
> @@ -148,6 +148,7 @@ obj-$(CONFIG_M54xx_WATCHDOG) += m54xx_wdt.o
>   
>   # MicroBlaze Architecture
>   obj-$(CONFIG_XILINX_WATCHDOG) += of_xilinx_wdt.o
> +obj-$(CONFIG_XILINX_WINDOW_WATCHDOG) += xilinx_wwdt.o
>   
>   # MIPS Architecture
>   obj-$(CONFIG_ATH79_WDT) += ath79_wdt.o
> diff --git a/drivers/watchdog/xilinx_wwdt.c b/drivers/watchdog/xilinx_wwdt.c
> new file mode 100644
> index 000000000000..8fa2cb2225e8
> --- /dev/null
> +++ b/drivers/watchdog/xilinx_wwdt.c
> @@ -0,0 +1,382 @@
> +// SPDX-License-Identifier: GPL-2.0
> +/*
> + * Window Watchdog Device Driver for Xilinx Versal WWDT
> + *
> + * (C) Copyright 2021 Xilinx, Inc.
> + */
> +
> +#include <linux/clk.h>
> +#include <linux/interrupt.h>
> +#include <linux/io.h>
> +#include <linux/ioport.h>
> +#include <linux/of_device.h>
> +#include <linux/of_address.h>
> +#include <linux/module.h>
> +#include <linux/watchdog.h>
> +
> +#define XWWDT_DEFAULT_TIMEOUT	40
> +#define XWWDT_MIN_TIMEOUT	1
> +#define XWWDT_MAX_TIMEOUT	42
> +
> +/* Register offsets for the WWdt device */
> +#define XWWDT_MWR_OFFSET	0x00
> +#define XWWDT_ESR_OFFSET	0x04
> +#define XWWDT_FCR_OFFSET	0x08
> +#define XWWDT_FWR_OFFSET	0x0c
> +#define XWWDT_SWR_OFFSET	0x10
> +
> +/* Master Write Control Register Masks */
> +#define XWWDT_MWR_MASK		BIT(0)
> +
> +/* Enable and Status Register Masks */
> +#define XWWDT_ESR_WINT_MASK	BIT(16)
> +#define XWWDT_ESR_WSW_MASK	BIT(8)
> +#define XWWDT_ESR_WEN_MASK	BIT(0)
> +
> +/* Function control Register Masks */
> +#define XWWDT_SBC_MASK		0xFF00
> +#define XWWDT_SBC_SHIFT		16
> +#define XWWDT_BSS_MASK		0xC0
> +
> +static int wwdt_timeout;
> +
> +module_param(wwdt_timeout, int, 0644);
> +MODULE_PARM_DESC(wwdt_timeout,
> +		 "Watchdog time in seconds. (default="
> +		 __MODULE_STRING(XWWDT_DEFAULT_TIMEOUT) ")");
> +
> +struct xwwdt_device {
> +	void __iomem *base;
> +	spinlock_t spinlock; /* spinlock for register handling */
> +	struct watchdog_device xilinx_wwdt_wdd;
> +	struct clk *clk;
> +	int irq;
> +};
> +
> +static int is_wwdt_in_closed_window(struct watchdog_device *wdd)

should return bool

> +{
> +	u32 control_status_reg;
> +	struct xwwdt_device *xdev = watchdog_get_drvdata(wdd);
> +
> +	spin_lock(&xdev->spinlock);
> +	control_status_reg = ioread32(xdev->base + XWWDT_ESR_OFFSET);
> +	spin_unlock(&xdev->spinlock);
> +	if (control_status_reg & XWWDT_ESR_WEN_MASK)
> +		if (!(control_status_reg & XWWDT_ESR_WSW_MASK))
> +			return 0;
> +
> +	return 1;

Can be written as expresion without if statement

> +}
> +
> +static int xilinx_wwdt_start(struct watchdog_device *wdd)
> +{
> +	struct xwwdt_device *xdev = watchdog_get_drvdata(wdd);
> +	struct watchdog_device *xilinx_wwdt_wdd = &xdev->xilinx_wwdt_wdd;
> +	u64 time_out, pre_timeout, count;
> +	u32 control_status_reg, fcr;
> +	int ret;
> +
> +	count = clk_get_rate(xdev->clk);
> +	if (!count)
> +		return -EINVAL;
> +
Is there a reason to believe that the clock rate can change ?
If not I would suggest to read it once and store it in struct xwwdt_device.

> +	/* Calculate timeout count */
> +	pre_timeout = count * wdd->pretimeout;
> +	time_out = count * wdd->timeout;
> +	if (!watchdog_active(xilinx_wwdt_wdd)) {
> +		ret  = clk_enable(xdev->clk);

This is at the very least confusing.
The last action in the probe function is to call clk_disable_unprepare(),
meaning the clock should not be prepared here. How can it be enabled
without being prepared ?

> +		if (ret) {
> +			dev_err(wdd->parent, "Failed to enable clock\n");
> +			return ret;
> +		}
> +	}
> +
I think it would be better to separate the functions to start/stop the clock
from the function to update/enable the watchdog. The use of watchdog_active()
combined with clk_enable/clk_disable/clk_prepare_enable/clk_disable_unprepare
is confusing and makes it all but impossible to review the driver.

> +	spin_lock(&xdev->spinlock);
> +	iowrite32(XWWDT_MWR_MASK, xdev->base + XWWDT_MWR_OFFSET);
> +	iowrite32(~(u32)XWWDT_ESR_WEN_MASK,
> +		  xdev->base + XWWDT_ESR_OFFSET);
> +
> +	if (pre_timeout) {
> +		iowrite32((u32)(time_out - pre_timeout),
> +			  xdev->base + XWWDT_FWR_OFFSET);
> +		iowrite32((u32)pre_timeout, xdev->base + XWWDT_SWR_OFFSET);
> +		fcr = ioread32(xdev->base + XWWDT_SWR_OFFSET);
> +		fcr = (fcr >> XWWDT_SBC_SHIFT) & XWWDT_SBC_MASK;
> +		fcr = fcr | XWWDT_BSS_MASK;
> +		iowrite32(fcr, xdev->base + XWWDT_FCR_OFFSET);
> +	} else {
> +		iowrite32((u32)pre_timeout,
> +			  xdev->base + XWWDT_FWR_OFFSET);
> +		iowrite32((u32)time_out, xdev->base + XWWDT_SWR_OFFSET);
> +		iowrite32(0x0, xdev->base + XWWDT_FCR_OFFSET);
> +	}
> +
> +	/* Enable the window watchdog timer */
> +	control_status_reg = ioread32(xdev->base + XWWDT_ESR_OFFSET);
> +	control_status_reg |= XWWDT_ESR_WEN_MASK;
> +	iowrite32(control_status_reg, xdev->base + XWWDT_ESR_OFFSET);
> +
> +	spin_unlock(&xdev->spinlock);
> +
> +	dev_dbg(xilinx_wwdt_wdd->parent, "Watchdog Started!\n");
> +
> +	return 0;
> +}
> +
> +static int xilinx_wwdt_stop(struct watchdog_device *wdd)
> +{
> +	struct xwwdt_device *xdev = watchdog_get_drvdata(wdd);
> +	struct watchdog_device *xilinx_wwdt_wdd = &xdev->xilinx_wwdt_wdd;
> +
> +	if (!is_wwdt_in_closed_window(wdd)) {
> +		dev_warn(xilinx_wwdt_wdd->parent, "timer in closed window");
> +		return -EINVAL;
> +	}
> +
> +	spin_lock(&xdev->spinlock);
> +
> +	iowrite32(XWWDT_MWR_MASK, xdev->base + XWWDT_MWR_OFFSET);
> +	/* Disable the Window watchdog timer */
> +	iowrite32(~(u32)XWWDT_ESR_WEN_MASK,
> +		  xdev->base + XWWDT_ESR_OFFSET);
> +
> +	spin_unlock(&xdev->spinlock);
> +
> +	if (watchdog_active(xilinx_wwdt_wdd))
> +		clk_disable(xdev->clk);
> +

The stop function won't be called if the watchdog is not running,
ie if watchdog_active() returns false.
I don't undeerstand the logic of clock handling in this driver.
More on that below.

> +	dev_dbg(xilinx_wwdt_wdd->parent, "Watchdog Stopped!\n");
> +
> +	return 0;
> +}
> +
> +static int xilinx_wwdt_keepalive(struct watchdog_device *wdd)
> +{
> +	u32 control_status_reg;
> +	struct xwwdt_device *xdev = watchdog_get_drvdata(wdd);
> +
> +	/* Refresh in open window is ignored */
> +	if (!is_wwdt_in_closed_window(wdd))
> +		return 0;
> +
This is dangerous. There is no reason for userspace to adhere to these restrictions.

> +	spin_lock(&xdev->spinlock);
> +
> +	iowrite32(XWWDT_MWR_MASK, xdev->base + XWWDT_MWR_OFFSET);
> +	control_status_reg = ioread32(xdev->base + XWWDT_ESR_OFFSET);
> +	control_status_reg |= XWWDT_ESR_WINT_MASK;
> +	control_status_reg &= ~XWWDT_ESR_WSW_MASK;
> +	iowrite32(control_status_reg, xdev->base + XWWDT_ESR_OFFSET);
> +	control_status_reg = ioread32(xdev->base + XWWDT_ESR_OFFSET);
> +	control_status_reg |= XWWDT_ESR_WSW_MASK;
> +	iowrite32(control_status_reg, xdev->base + XWWDT_ESR_OFFSET);
> +
> +	spin_unlock(&xdev->spinlock);
> +
> +	return 0;
> +}
> +
> +static int xilinx_wwdt_set_timeout(struct watchdog_device *wdd,
> +				   unsigned int new_time)
> +{
> +	u32 ret = 0;
> +	struct xwwdt_device *xdev = watchdog_get_drvdata(wdd);
> +	struct watchdog_device *xilinx_wwdt_wdd = &xdev->xilinx_wwdt_wdd;
> +
> +	if (!is_wwdt_in_closed_window(wdd)) {
> +		dev_warn(xilinx_wwdt_wdd->parent, "timer in closed window");
> +		return -EINVAL;
> +	}
> +

So userspace can clog the log by continuously trying to set the
timeout ?

Also, again, this is dangerous. Userspace does not know the window size.


> +	if (new_time < XWWDT_MIN_TIMEOUT ||
> +	    new_time > XWWDT_MAX_TIMEOUT) {
> +		dev_warn(xilinx_wwdt_wdd->parent,
> +			 "timeout value must be %d<=x<=%d, using %d\n",
> +			 XWWDT_MIN_TIMEOUT,
> +			 XWWDT_MAX_TIMEOUT, new_time);
> +		return -EINVAL;
> +	}

This is handled by the watchdog core.

> +
> +	wdd->timeout = new_time;
> +	wdd->pretimeout = 0;
> +
> +	if (watchdog_active(xilinx_wwdt_wdd)) {
> +		ret = xilinx_wwdt_start(wdd);
> +		if (ret)
> +			dev_dbg(xilinx_wwdt_wdd->parent, "timer start failed");
> +	}
> +
> +	return ret;
> +}
> +
> +static int xilinx_wwdt_set_pretimeout(struct watchdog_device *wdd,
> +				      u32 new_pretimeout)
> +{
> +	u32 ret = 0;
> +	struct xwwdt_device *xdev = watchdog_get_drvdata(wdd);
> +	struct watchdog_device *xilinx_wwdt_wdd = &xdev->xilinx_wwdt_wdd;
> +
> +	if (!is_wwdt_in_closed_window(wdd)) {
> +		dev_warn(xilinx_wwdt_wdd->parent, "timer in closed window");
> +		return -EINVAL;
> +	}

This restriction is not only dangerous but confusing, and simply
unacceptable. Userspace will get -EINVAL with no idea what is wrong.
A watchdog daemon will bail out for no good reason.

> +
> +	if (new_pretimeout < wdd->min_timeout ||
> +	    new_pretimeout >= wdd->timeout)
> +		return -EINVAL;
> +
This is handled by the watchdog core.

> +	wdd->pretimeout = new_pretimeout;
> +
> +	if (watchdog_active(xilinx_wwdt_wdd)) {
> +		ret = xilinx_wwdt_start(wdd);
> +		if (ret)
> +			dev_dbg(xilinx_wwdt_wdd->parent, "timer start failed");
> +	}
> +
> +	return ret;
> +}
> +
> +static void xwwdt_clk_disable_unprepare(void *data)
> +{
> +	clk_disable_unprepare(data);
> +}
> +
> +static irqreturn_t xilinx_wwdt_isr(int irq, void *wdog_arg)
> +{
> +	struct xwwdt_device *xdev = wdog_arg;
> +
> +	watchdog_notify_pretimeout(&xdev->xilinx_wwdt_wdd);
> +
> +	return IRQ_HANDLED;
> +}
> +
> +static const struct watchdog_info xilinx_wwdt_ident = {
> +	.options = WDIOF_MAGICCLOSE |
> +		   WDIOF_KEEPALIVEPING |
> +		   WDIOF_SETTIMEOUT,
> +	.firmware_version = 1,
> +	.identity = "xlnx_window watchdog",
> +};
> +
> +static const struct watchdog_info xilinx_wwdt_pretimeout_ident = {
> +	.options = WDIOF_MAGICCLOSE |
> +		   WDIOF_KEEPALIVEPING |
> +		   WDIOF_PRETIMEOUT |
> +		   WDIOF_SETTIMEOUT,
> +	.firmware_version = 1,
> +	.identity = "xlnx_window watchdog",
> +};
> +
> +static const struct watchdog_ops xilinx_wwdt_ops = {
> +	.owner = THIS_MODULE,
> +	.start = xilinx_wwdt_start,
> +	.stop = xilinx_wwdt_stop,
> +	.ping = xilinx_wwdt_keepalive,
> +	.set_timeout = xilinx_wwdt_set_timeout,
> +	.set_pretimeout = xilinx_wwdt_set_pretimeout,
> +};
> +
> +static int xwwdt_probe(struct platform_device *pdev)
> +{
> +	struct device *dev = &pdev->dev;
> +	struct watchdog_device *xilinx_wwdt_wdd;
> +	struct xwwdt_device *xdev;
> +	u32 pre_timeout = 0;
> +	int ret;
> +
> +	xdev = devm_kzalloc(dev, sizeof(*xdev), GFP_KERNEL);
> +	if (!xdev)
> +		return -ENOMEM;
> +
> +	xilinx_wwdt_wdd = &xdev->xilinx_wwdt_wdd;
> +	xilinx_wwdt_wdd->info = &xilinx_wwdt_ident;
> +	xilinx_wwdt_wdd->ops = &xilinx_wwdt_ops;
> +	xilinx_wwdt_wdd->parent = dev;
> +
> +	xdev->base = devm_platform_ioremap_resource(pdev, 0);
> +	if (IS_ERR(xdev->base))
> +		return PTR_ERR(xdev->base);
> +
> +	ret = of_property_read_u32(dev->of_node, "pretimeout-sec",
> +				   &pre_timeout);
> +	if (ret)
> +		dev_dbg(dev,
> +			"Parameter \"pretimeout-sec\" not found\n");
> +
> +	xdev->clk = devm_clk_get(dev, NULL);
> +	if (IS_ERR(xdev->clk))
> +		return PTR_ERR(xdev->clk);
> +
> +	ret = clk_get_rate(xdev->clk);
> +	if (!ret)
> +		return -EINVAL;
> +
> +	ret = clk_prepare_enable(xdev->clk);
> +	if (ret) {
> +		dev_err(dev, "unable to enable clock\n");
> +		return ret;
> +	}
> +	ret = devm_add_action_or_reset(dev, xwwdt_clk_disable_unprepare,
> +				       xdev->clk);
> +	if (ret)
> +		goto err_clk_disable;
> +

Seems pointless. What happens when you unload the driver ?

> +	xilinx_wwdt_wdd->pretimeout = pre_timeout;
> +	xilinx_wwdt_wdd->timeout = XWWDT_DEFAULT_TIMEOUT;
> +	xilinx_wwdt_wdd->min_timeout = XWWDT_MIN_TIMEOUT;
> +	xilinx_wwdt_wdd->max_timeout = XWWDT_MAX_TIMEOUT;
> +
> +	xdev->irq = platform_get_irq_byname(pdev, "wdt");
> +	if (xdev->irq < 0) {
> +		ret = xdev->irq;
> +		goto err_clk_disable;
> +	}
> +
> +	if (!devm_request_irq(dev, xdev->irq, xilinx_wwdt_isr,
> +			      0, dev_name(dev), xdev)) {
> +		xilinx_wwdt_wdd->info = &xilinx_wwdt_pretimeout_ident;
> +	}
> +
> +	ret = watchdog_init_timeout(xilinx_wwdt_wdd,
> +				    wwdt_timeout, &pdev->dev);
> +	if (ret)
> +		dev_info(&pdev->dev, "Configured default timeout value\n");
> +
> +	spin_lock_init(&xdev->spinlock);
> +	watchdog_set_drvdata(xilinx_wwdt_wdd, xdev);
> +
> +	ret = devm_watchdog_register_device(dev, xilinx_wwdt_wdd);
> +	if (ret)
> +		goto err_clk_disable;
> +
At this point the watchdog is fully instantiated, and userspace may have enabled it.

> +	clk_disable_unprepare(xdev->clk);
> +
And now its clock is getting stopped. Actually, it is getting unprepared.
Does this even work ?

> +	dev_info(dev, "Xilinx Window Watchdog Timer with timeout %ds\n",
> +		 xilinx_wwdt_wdd->timeout);
> +
> +	return 0;
> +
> +err_clk_disable:
> +	clk_disable_unprepare(xdev->clk);

devm_add_action_or_reset() already handles this.

> +	return ret;
> +}
> +
> +/* Mat for of_platform binding */
> +static const struct of_device_id xwwdt_of_match[] = {
> +	{ .compatible = "xlnx,versal-wwdt-1.0", },
> +	{},
> +};
> +MODULE_DEVICE_TABLE(of, xwwdt_of_match);
> +
> +static struct platform_driver xwwdt_driver = {
> +	.probe = xwwdt_probe,
> +	.driver = {
> +		.name = "Xilinx Window Watchdog",
> +		.of_match_table = xwwdt_of_match,
> +	},
> +};
> +
> +module_platform_driver(xwwdt_driver);
> +
> +MODULE_AUTHOR("Neeli Srinivas <sneeli@xilinx.com>");
> +MODULE_DESCRIPTION("Xilinx Window Watchdog driver");
> +MODULE_LICENSE("GPL");
> 


  reply	other threads:[~2021-05-27 14:39 UTC|newest]

Thread overview: 9+ messages / expand[flat|nested]  mbox.gz  Atom feed  top
2021-05-27  9:57 [PATCH 0/3] watchdog: Add window watchdog driver Srinivas Neeli
2021-05-27  9:57 ` [PATCH 1/3] watchdog: bindings: xlnx,versal-wwdt: Add binding documentation for xilinx window watchdog device Srinivas Neeli
2021-05-27  9:57 ` [PATCH 2/3] watchdog: xilinx_wwdt: Add Versal Window watchdog mode Srinivas Neeli
2021-05-27 14:39   ` Guenter Roeck [this message]
     [not found]     ` <DM6PR02MB538635505ABE30C42100E059AF369@DM6PR02MB5386.namprd02.prod.outlook.com>
     [not found]       ` <DM6PR02MB5386C805459902F9C28391AAAF139@DM6PR02MB5386.namprd02.prod.outlook.com>
2021-07-14 11:31         ` Guenter Roeck
2021-09-29  7:01           ` Srinivas Neeli
2021-10-18  9:23             ` Srinivas Neeli
2021-10-20  5:09               ` Guenter Roeck
2021-05-27  9:57 ` [PATCH 3/3] MAINTAINERS: Add fragment for xilinx watchdog drivers Srinivas Neeli

Reply instructions:

You may reply publicly to this message via plain-text email
using any one of the following methods:

* Save the following mbox file, import it into your mail client,
  and reply-to-all from there: mbox

  Avoid top-posting and favor interleaved quoting:
  https://en.wikipedia.org/wiki/Posting_style#Interleaved_style

* Reply using the --to, --cc, and --in-reply-to
  switches of git-send-email(1):

  git send-email \
    --in-reply-to=fb83065e-7870-c830-7dde-efb3af8cc25c@roeck-us.net \
    --to=linux@roeck-us.net \
    --cc=git@xilinx.com \
    --cc=linux-kernel@vger.kernel.org \
    --cc=linux-watchdog@vger.kernel.org \
    --cc=michal.simek@xilinx.com \
    --cc=sgoud@xilinx.com \
    --cc=shubhraj@xilinx.com \
    --cc=srinivas.neeli@xilinx.com \
    --cc=wim@linux-watchdog.org \
    /path/to/YOUR_REPLY

  https://kernel.org/pub/software/scm/git/docs/git-send-email.html

* If your mail client supports setting the In-Reply-To header
  via mailto: links, try the mailto: link
Be sure your reply has a Subject: header at the top and a blank line before the message body.
This is a public inbox, see mirroring instructions
for how to clone and mirror all data and code used for this inbox;
as well as URLs for NNTP newsgroup(s).