From: Jaehoon Chung <jh80.chung@samsung.com>
To: Heiko Stuebner <heiko@sntech.de>,
ulf.hansson@linaro.org, mturquette@baylibre.com,
sboyd@codeaurora.org
Cc: linux-mmc@vger.kernel.org, linux-clk@vger.kernel.org,
linux-arm-kernel@lists.infradead.org,
linux-rockchip@lists.infradead.org, linux-kernel@vger.kernel.org,
Alexandru M Stan <amstan@chromium.org>, CPGS <cpgs@samsung.com>
Subject: Re: [PATCH 6/8] mmc: dw_mmc: Generic MMC tuning with the clock phase framework
Date: Tue, 15 Sep 2015 17:25:38 +0900 [thread overview]
Message-ID: <55F7D602.70003@samsung.com> (raw)
In-Reply-To: <1441045446-30858-7-git-send-email-heiko@sntech.de>
Dear, Heiko.
On 09/01/2015 03:24 AM, Heiko Stuebner wrote:
> From: Alexandru M Stan <amstan@chromium.org>
>
> This algorithm will try 1 degree increments, since there's no way to tell
> what resolution the underlying phase code uses. As an added bonus, doing
> many tunings yields better results since some tests are run more than once
> (ex: if the underlying driver uses 45 degree increments, the tuning code
> will try the same angle more than once).
>
> It will then construct a list of good phase ranges (even ranges that cross
> 360/0), will pick the biggest range then it will set the sample_clk to the
> middle of that range.
>
> We do not touch ciu_drive (and by extension define default-drive-phase).
> Drive phase is mostly used to define minimum hold times, while one could
> write some code to determine what phase meets the minimum hold time (ex 10
> degrees) this will not work with the current clock phase framework (which
> floors angles, so we'll get 0 deg, and there's no way to know what
> resolution the floors happen at). We assume that the default drive angles
> set by the hardware are good enough.
>
> If a device has device specific code (like exynos) then that will still
> take precedence, otherwise this new code will execute. If the device wants
> to tune, but has no sample_clk defined we'll return EIO with an error
> message.
Which point is "_generic_"? I don't find the code that control the register relevant to CLK_DRV/SMPL PHASE.
It seems that posted the similar patches at u-boot mailing list..
Best Regards,
Jaehoon Chung
>
> Signed-off-by: Alexandru M Stan <amstan@chromium.org>
>
> Convert to mmc_send_tuning()
> Fold in from the ChromeOS-tree:
> - mmc: dw_mmc: Change tuning to only 16 phases
> - mmc: dw_mmc: Test more phases
> Signed-off-by: Heiko Stuebner <heiko@sntech.de>
> ---
> drivers/mmc/host/dw_mmc.c | 142 +++++++++++++++++++++++++++++++++++++++++++++
> include/linux/mmc/dw_mmc.h | 3 +
> 2 files changed, 145 insertions(+)
>
> diff --git a/drivers/mmc/host/dw_mmc.c b/drivers/mmc/host/dw_mmc.c
> index b1b7e7f..13bcde0 100644
> --- a/drivers/mmc/host/dw_mmc.c
> +++ b/drivers/mmc/host/dw_mmc.c
> @@ -1194,6 +1194,12 @@ static void dw_mci_set_ios(struct mmc_host *mmc, struct mmc_ios *ios)
> if (drv_data && drv_data->set_ios)
> drv_data->set_ios(slot->host, ios);
>
> + /* Make sure we use phases which we can enumerate with */
> + if (!IS_ERR(slot->host->sample_clk)) {
> + clk_set_phase(slot->host->sample_clk,
> + slot->host->default_sample_phase);
> + }
> +
> switch (ios->power_mode) {
> case MMC_POWER_UP:
> if (!IS_ERR(mmc->supply.vmmc)) {
> @@ -1414,6 +1420,127 @@ static void dw_mci_enable_sdio_irq(struct mmc_host *mmc, int enb)
> spin_unlock_irqrestore(&host->irq_lock, irqflags);
> }
>
> +#define NUM_PHASES 360
> +#define TUNING_ITERATION_TO_PHASE(i) (DIV_ROUND_UP((i) * 360, NUM_PHASES))
> +
> +static int dw_mci_execute_generic_tuning(struct dw_mci_slot *slot)
> +{
> + struct dw_mci *host = slot->host;
> + struct mmc_host *mmc = slot->mmc;
> + int ret = 0;
> + int i;
> + bool v, prev_v = 0, first_v;
> + struct range_t {
> + int start;
> + int end; /* inclusive */
> + };
> + struct range_t *ranges;
> + unsigned int range_count = 0;
> + int longest_range_len = -1;
> + int longest_range = -1;
> + int middle_phase;
> +
> + if (IS_ERR(host->sample_clk)) {
> + dev_err(host->dev, "Tuning clock (sample_clk) not defined.\n");
> + return -EIO;
> + }
> +
> + ranges = kmalloc_array(NUM_PHASES / 2 + 1, sizeof(*ranges), GFP_KERNEL);
> + if (!ranges)
> + return -ENOMEM;
> +
> + /* Try each phase and extract good ranges */
> + for (i = 0; i < NUM_PHASES; ) {
> + clk_set_phase(host->sample_clk, TUNING_ITERATION_TO_PHASE(i));
> +
> + v = !mmc_send_tuning(mmc);
> +
> + if (i == 0)
> + first_v = v;
> +
> + if ((!prev_v) && v) {
> + range_count++;
> + ranges[range_count-1].start = i;
> + }
> + if (v) {
> + ranges[range_count-1].end = i;
> + i++;
> + } else if (i == NUM_PHASES - 1) {
> + /* No extra skipping rules if we're at the end */
> + i++;
> + } else {
> + /*
> + * No need to check too close to an invalid
> + * one since testing bad phases is slow. Skip
> + * 20 degrees.
> + */
> + i += DIV_ROUND_UP(20 * NUM_PHASES, 360);
> +
> + /* Always test the last one */
> + if (i >= NUM_PHASES)
> + i = NUM_PHASES - 1;
> + }
> +
> + prev_v = v;
> + }
> +
> + if (range_count == 0) {
> + dev_warn(host->dev, "All phases bad!");
> + ret = -EIO;
> + goto free;
> + }
> +
> + /* wrap around case, merge the end points */
> + if ((range_count > 1) && first_v && v) {
> + ranges[0].start = ranges[range_count-1].start;
> + range_count--;
> + }
> +
> + if (ranges[0].start == 0 && ranges[0].end == NUM_PHASES - 1) {
> + clk_set_phase(host->sample_clk, host->default_sample_phase);
> + dev_info(host->dev, "All phases work, using default phase %d.",
> + host->default_sample_phase);
> + goto free;
> + }
> +
> + /* Find the longest range */
> + for (i = 0; i < range_count; i++) {
> + int len = (ranges[i].end - ranges[i].start + 1);
> +
> + if (len < 0)
> + len += NUM_PHASES;
> +
> + if (longest_range_len < len) {
> + longest_range_len = len;
> + longest_range = i;
> + }
> +
> + dev_dbg(host->dev, "Good phase range %d-%d (%d len)\n",
> + TUNING_ITERATION_TO_PHASE(ranges[i].start),
> + TUNING_ITERATION_TO_PHASE(ranges[i].end),
> + len
> + );
> + }
> +
> + dev_dbg(host->dev, "Best phase range %d-%d (%d len)\n",
> + TUNING_ITERATION_TO_PHASE(ranges[longest_range].start),
> + TUNING_ITERATION_TO_PHASE(ranges[longest_range].end),
> + longest_range_len
> + );
> +
> + middle_phase = ranges[longest_range].start + longest_range_len / 2;
> + middle_phase %= NUM_PHASES;
> + dev_info(host->dev, "Successfully tuned phase to %d\n",
> + TUNING_ITERATION_TO_PHASE(middle_phase));
> +
> + clk_set_phase(host->sample_clk,
> + TUNING_ITERATION_TO_PHASE(middle_phase));
> +
> +free:
> + kfree(ranges);
> + return ret;
> +}
> +
> static int dw_mci_execute_tuning(struct mmc_host *mmc, u32 opcode)
> {
> struct dw_mci_slot *slot = mmc_priv(mmc);
> @@ -1423,6 +1550,8 @@ static int dw_mci_execute_tuning(struct mmc_host *mmc, u32 opcode)
>
> if (drv_data && drv_data->execute_tuning)
> err = drv_data->execute_tuning(slot);
> + else
> + err = dw_mci_execute_generic_tuning(slot);
> return err;
> }
>
> @@ -2741,6 +2870,11 @@ static struct dw_mci_board *dw_mci_parse_dt(struct dw_mci *host)
> if (!of_property_read_u32(np, "clock-frequency", &clock_frequency))
> pdata->bus_hz = clock_frequency;
>
> + if (of_property_read_u32(np, "default-sample-phase",
> + &host->default_sample_phase)) {
> + host->default_sample_phase = 0;
> + }
> +
> if (drv_data && drv_data->parse_dt) {
> ret = drv_data->parse_dt(host);
> if (ret)
> @@ -2843,6 +2977,14 @@ int dw_mci_probe(struct dw_mci *host)
> host->bus_hz = clk_get_rate(host->ciu_clk);
> }
>
> + host->drv_clk = devm_clk_get(host->dev, "ciu_drv");
> + if (IS_ERR(host->drv_clk))
> + dev_dbg(host->dev, "ciu_drv not available\n");
> +
> + host->sample_clk = devm_clk_get(host->dev, "ciu_sample");
> + if (IS_ERR(host->sample_clk))
> + dev_dbg(host->dev, "ciu_sample not available\n");
> +
> if (!host->bus_hz) {
> dev_err(host->dev,
> "Platform data must supply bus speed\n");
> diff --git a/include/linux/mmc/dw_mmc.h b/include/linux/mmc/dw_mmc.h
> index 134c574..40187ba 100644
> --- a/include/linux/mmc/dw_mmc.h
> +++ b/include/linux/mmc/dw_mmc.h
> @@ -176,7 +176,10 @@ struct dw_mci {
> void *priv;
> struct clk *biu_clk;
> struct clk *ciu_clk;
> + struct clk *drv_clk;
> + struct clk *sample_clk;
> struct dw_mci_slot *slot[MAX_MCI_SLOTS];
> + int default_sample_phase;
>
> /* FIFO push and pull */
> int fifo_depth;
>
WARNING: multiple messages have this Message-ID (diff)
From: jh80.chung@samsung.com (Jaehoon Chung)
To: linux-arm-kernel@lists.infradead.org
Subject: [PATCH 6/8] mmc: dw_mmc: Generic MMC tuning with the clock phase framework
Date: Tue, 15 Sep 2015 17:25:38 +0900 [thread overview]
Message-ID: <55F7D602.70003@samsung.com> (raw)
In-Reply-To: <1441045446-30858-7-git-send-email-heiko@sntech.de>
Dear, Heiko.
On 09/01/2015 03:24 AM, Heiko Stuebner wrote:
> From: Alexandru M Stan <amstan@chromium.org>
>
> This algorithm will try 1 degree increments, since there's no way to tell
> what resolution the underlying phase code uses. As an added bonus, doing
> many tunings yields better results since some tests are run more than once
> (ex: if the underlying driver uses 45 degree increments, the tuning code
> will try the same angle more than once).
>
> It will then construct a list of good phase ranges (even ranges that cross
> 360/0), will pick the biggest range then it will set the sample_clk to the
> middle of that range.
>
> We do not touch ciu_drive (and by extension define default-drive-phase).
> Drive phase is mostly used to define minimum hold times, while one could
> write some code to determine what phase meets the minimum hold time (ex 10
> degrees) this will not work with the current clock phase framework (which
> floors angles, so we'll get 0 deg, and there's no way to know what
> resolution the floors happen at). We assume that the default drive angles
> set by the hardware are good enough.
>
> If a device has device specific code (like exynos) then that will still
> take precedence, otherwise this new code will execute. If the device wants
> to tune, but has no sample_clk defined we'll return EIO with an error
> message.
Which point is "_generic_"? I don't find the code that control the register relevant to CLK_DRV/SMPL PHASE.
It seems that posted the similar patches at u-boot mailing list..
Best Regards,
Jaehoon Chung
>
> Signed-off-by: Alexandru M Stan <amstan@chromium.org>
>
> Convert to mmc_send_tuning()
> Fold in from the ChromeOS-tree:
> - mmc: dw_mmc: Change tuning to only 16 phases
> - mmc: dw_mmc: Test more phases
> Signed-off-by: Heiko Stuebner <heiko@sntech.de>
> ---
> drivers/mmc/host/dw_mmc.c | 142 +++++++++++++++++++++++++++++++++++++++++++++
> include/linux/mmc/dw_mmc.h | 3 +
> 2 files changed, 145 insertions(+)
>
> diff --git a/drivers/mmc/host/dw_mmc.c b/drivers/mmc/host/dw_mmc.c
> index b1b7e7f..13bcde0 100644
> --- a/drivers/mmc/host/dw_mmc.c
> +++ b/drivers/mmc/host/dw_mmc.c
> @@ -1194,6 +1194,12 @@ static void dw_mci_set_ios(struct mmc_host *mmc, struct mmc_ios *ios)
> if (drv_data && drv_data->set_ios)
> drv_data->set_ios(slot->host, ios);
>
> + /* Make sure we use phases which we can enumerate with */
> + if (!IS_ERR(slot->host->sample_clk)) {
> + clk_set_phase(slot->host->sample_clk,
> + slot->host->default_sample_phase);
> + }
> +
> switch (ios->power_mode) {
> case MMC_POWER_UP:
> if (!IS_ERR(mmc->supply.vmmc)) {
> @@ -1414,6 +1420,127 @@ static void dw_mci_enable_sdio_irq(struct mmc_host *mmc, int enb)
> spin_unlock_irqrestore(&host->irq_lock, irqflags);
> }
>
> +#define NUM_PHASES 360
> +#define TUNING_ITERATION_TO_PHASE(i) (DIV_ROUND_UP((i) * 360, NUM_PHASES))
> +
> +static int dw_mci_execute_generic_tuning(struct dw_mci_slot *slot)
> +{
> + struct dw_mci *host = slot->host;
> + struct mmc_host *mmc = slot->mmc;
> + int ret = 0;
> + int i;
> + bool v, prev_v = 0, first_v;
> + struct range_t {
> + int start;
> + int end; /* inclusive */
> + };
> + struct range_t *ranges;
> + unsigned int range_count = 0;
> + int longest_range_len = -1;
> + int longest_range = -1;
> + int middle_phase;
> +
> + if (IS_ERR(host->sample_clk)) {
> + dev_err(host->dev, "Tuning clock (sample_clk) not defined.\n");
> + return -EIO;
> + }
> +
> + ranges = kmalloc_array(NUM_PHASES / 2 + 1, sizeof(*ranges), GFP_KERNEL);
> + if (!ranges)
> + return -ENOMEM;
> +
> + /* Try each phase and extract good ranges */
> + for (i = 0; i < NUM_PHASES; ) {
> + clk_set_phase(host->sample_clk, TUNING_ITERATION_TO_PHASE(i));
> +
> + v = !mmc_send_tuning(mmc);
> +
> + if (i == 0)
> + first_v = v;
> +
> + if ((!prev_v) && v) {
> + range_count++;
> + ranges[range_count-1].start = i;
> + }
> + if (v) {
> + ranges[range_count-1].end = i;
> + i++;
> + } else if (i == NUM_PHASES - 1) {
> + /* No extra skipping rules if we're at the end */
> + i++;
> + } else {
> + /*
> + * No need to check too close to an invalid
> + * one since testing bad phases is slow. Skip
> + * 20 degrees.
> + */
> + i += DIV_ROUND_UP(20 * NUM_PHASES, 360);
> +
> + /* Always test the last one */
> + if (i >= NUM_PHASES)
> + i = NUM_PHASES - 1;
> + }
> +
> + prev_v = v;
> + }
> +
> + if (range_count == 0) {
> + dev_warn(host->dev, "All phases bad!");
> + ret = -EIO;
> + goto free;
> + }
> +
> + /* wrap around case, merge the end points */
> + if ((range_count > 1) && first_v && v) {
> + ranges[0].start = ranges[range_count-1].start;
> + range_count--;
> + }
> +
> + if (ranges[0].start == 0 && ranges[0].end == NUM_PHASES - 1) {
> + clk_set_phase(host->sample_clk, host->default_sample_phase);
> + dev_info(host->dev, "All phases work, using default phase %d.",
> + host->default_sample_phase);
> + goto free;
> + }
> +
> + /* Find the longest range */
> + for (i = 0; i < range_count; i++) {
> + int len = (ranges[i].end - ranges[i].start + 1);
> +
> + if (len < 0)
> + len += NUM_PHASES;
> +
> + if (longest_range_len < len) {
> + longest_range_len = len;
> + longest_range = i;
> + }
> +
> + dev_dbg(host->dev, "Good phase range %d-%d (%d len)\n",
> + TUNING_ITERATION_TO_PHASE(ranges[i].start),
> + TUNING_ITERATION_TO_PHASE(ranges[i].end),
> + len
> + );
> + }
> +
> + dev_dbg(host->dev, "Best phase range %d-%d (%d len)\n",
> + TUNING_ITERATION_TO_PHASE(ranges[longest_range].start),
> + TUNING_ITERATION_TO_PHASE(ranges[longest_range].end),
> + longest_range_len
> + );
> +
> + middle_phase = ranges[longest_range].start + longest_range_len / 2;
> + middle_phase %= NUM_PHASES;
> + dev_info(host->dev, "Successfully tuned phase to %d\n",
> + TUNING_ITERATION_TO_PHASE(middle_phase));
> +
> + clk_set_phase(host->sample_clk,
> + TUNING_ITERATION_TO_PHASE(middle_phase));
> +
> +free:
> + kfree(ranges);
> + return ret;
> +}
> +
> static int dw_mci_execute_tuning(struct mmc_host *mmc, u32 opcode)
> {
> struct dw_mci_slot *slot = mmc_priv(mmc);
> @@ -1423,6 +1550,8 @@ static int dw_mci_execute_tuning(struct mmc_host *mmc, u32 opcode)
>
> if (drv_data && drv_data->execute_tuning)
> err = drv_data->execute_tuning(slot);
> + else
> + err = dw_mci_execute_generic_tuning(slot);
> return err;
> }
>
> @@ -2741,6 +2870,11 @@ static struct dw_mci_board *dw_mci_parse_dt(struct dw_mci *host)
> if (!of_property_read_u32(np, "clock-frequency", &clock_frequency))
> pdata->bus_hz = clock_frequency;
>
> + if (of_property_read_u32(np, "default-sample-phase",
> + &host->default_sample_phase)) {
> + host->default_sample_phase = 0;
> + }
> +
> if (drv_data && drv_data->parse_dt) {
> ret = drv_data->parse_dt(host);
> if (ret)
> @@ -2843,6 +2977,14 @@ int dw_mci_probe(struct dw_mci *host)
> host->bus_hz = clk_get_rate(host->ciu_clk);
> }
>
> + host->drv_clk = devm_clk_get(host->dev, "ciu_drv");
> + if (IS_ERR(host->drv_clk))
> + dev_dbg(host->dev, "ciu_drv not available\n");
> +
> + host->sample_clk = devm_clk_get(host->dev, "ciu_sample");
> + if (IS_ERR(host->sample_clk))
> + dev_dbg(host->dev, "ciu_sample not available\n");
> +
> if (!host->bus_hz) {
> dev_err(host->dev,
> "Platform data must supply bus speed\n");
> diff --git a/include/linux/mmc/dw_mmc.h b/include/linux/mmc/dw_mmc.h
> index 134c574..40187ba 100644
> --- a/include/linux/mmc/dw_mmc.h
> +++ b/include/linux/mmc/dw_mmc.h
> @@ -176,7 +176,10 @@ struct dw_mci {
> void *priv;
> struct clk *biu_clk;
> struct clk *ciu_clk;
> + struct clk *drv_clk;
> + struct clk *sample_clk;
> struct dw_mci_slot *slot[MAX_MCI_SLOTS];
> + int default_sample_phase;
>
> /* FIFO push and pull */
> int fifo_depth;
>
next prev parent reply other threads:[~2015-09-15 8:25 UTC|newest]
Thread overview: 47+ messages / expand[flat|nested] mbox.gz Atom feed top
2015-08-31 18:23 [PATCH 0/8] mmc: dw_mmc: allow tuning using the clk-phase api Heiko Stuebner
2015-08-31 18:23 ` Heiko Stuebner
2015-08-31 18:23 ` [PATCH 1/8] clk: rockchip: Allow more precision for some mmc clock phases Heiko Stuebner
2015-08-31 18:23 ` Heiko Stuebner
2015-08-31 18:24 ` [PATCH 2/8] clk: rockchip: Make calculations use rounding Heiko Stuebner
2015-08-31 18:24 ` Heiko Stuebner
2015-08-31 18:24 ` [PATCH 3/8] mmc: core: Add mmc_regulator_set_vqmmc() Heiko Stuebner
2015-08-31 18:24 ` Heiko Stuebner
2015-08-31 18:24 ` Heiko Stuebner
2015-09-02 11:38 ` Ulf Hansson
2015-09-02 11:38 ` Ulf Hansson
2015-09-02 16:20 ` Doug Anderson
2015-09-02 16:20 ` Doug Anderson
2015-09-10 12:40 ` Ulf Hansson
2015-09-10 12:40 ` Ulf Hansson
2015-08-31 18:24 ` [PATCH 4/8] mmc: dw_mmc: Use mmc_regulator_set_vqmmc in start_signal_voltage_switch Heiko Stuebner
2015-08-31 18:24 ` Heiko Stuebner
2015-08-31 18:24 ` Heiko Stuebner
2015-08-31 18:24 ` [PATCH 5/8] mmc: dw_mmc: dt-binding: Add tuning related things Heiko Stuebner
2015-08-31 18:24 ` Heiko Stuebner
2015-08-31 18:24 ` Heiko Stuebner
2015-09-02 5:01 ` Jaehoon Chung
2015-09-02 5:01 ` Jaehoon Chung
2015-09-02 5:01 ` Jaehoon Chung
2015-09-02 7:41 ` Heiko Stuebner
2015-09-02 7:41 ` Heiko Stuebner
2015-09-02 7:41 ` Heiko Stuebner
2015-08-31 18:24 ` [PATCH 6/8] mmc: dw_mmc: Generic MMC tuning with the clock phase framework Heiko Stuebner
2015-08-31 18:24 ` Heiko Stuebner
2015-09-15 8:25 ` Jaehoon Chung [this message]
2015-09-15 8:25 ` Jaehoon Chung
2015-09-15 22:09 ` Heiko Stübner
2015-09-15 22:09 ` Heiko Stübner
2015-09-15 22:09 ` Heiko Stübner
2015-09-16 2:30 ` Jaehoon Chung
2015-09-16 2:30 ` Jaehoon Chung
2015-09-16 14:52 ` Heiko Stübner
2015-09-16 14:52 ` Heiko Stübner
2015-09-16 14:52 ` Heiko Stübner
2015-09-17 2:03 ` Jaehoon Chung
2015-09-17 2:03 ` Jaehoon Chung
2015-09-17 2:03 ` Jaehoon Chung
2015-08-31 18:24 ` [PATCH 7/8] ARM: dts: rockchip: Add drive/sample clocks for rk3288 dw_mmc devices Heiko Stuebner
2015-08-31 18:24 ` Heiko Stuebner
2015-08-31 18:24 ` [PATCH 8/8] ARM: dts: rockchip: add tuning related settings to veyron devices Heiko Stuebner
2015-08-31 18:24 ` Heiko Stuebner
2015-08-31 18:24 ` Heiko Stuebner
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=55F7D602.70003@samsung.com \
--to=jh80.chung@samsung.com \
--cc=amstan@chromium.org \
--cc=cpgs@samsung.com \
--cc=heiko@sntech.de \
--cc=linux-arm-kernel@lists.infradead.org \
--cc=linux-clk@vger.kernel.org \
--cc=linux-kernel@vger.kernel.org \
--cc=linux-mmc@vger.kernel.org \
--cc=linux-rockchip@lists.infradead.org \
--cc=mturquette@baylibre.com \
--cc=sboyd@codeaurora.org \
--cc=ulf.hansson@linaro.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 an external index of several public inboxes,
see mirroring instructions on how to clone and mirror
all data and code used by this external index.