| Message ID | 1418007589-3688-1-git-send-email-addy.ke@rock-chips.com (mailing list archive) |
|---|---|
| State | New, archived |
| Headers | show |
On 2014/12/8 10:59, Addy Ke wrote: > high_ns calculated from the low division of CLKDIV register is the sum > of actual measured high_ns and rise_ns. The rise time which related to > external pull-up resistor can be up to the maximum rise time in I2C spec. > > In my test, if external pull-up resistor is 4.7K, rise_ns is about > 700ns. So the actual measured high_ns is about 3900ns, which is less > than 4000ns(the minimum high_ns in I2C spec). > > To fix this bug, min_low_ns should include fall time and min_high_ns > should include rise time too. > > This patch merged the patch that Doug submitted to chromium, which > can get the rise and fall times for signals from the device tree. > This allows us to more accurately calculate timings. see: > https://chromium-review.googlesource.com/#/c/232774/ > > Signed-off-by: Addy Ke <addy.ke@rock-chips.com> > --- > Changes in v2: > - merged the patch that Doug submitted to chromium > Changes in v3: > - merged the patch that Doug submitted to chromium to change bindins > see: https://chromium-review.googlesource.com/#/c/232775/ Sorry, see https://chromium-review.googlesource.com/#/c/232774/ not 232775 > > Documentation/devicetree/bindings/i2c/i2c-rk3x.txt | 11 +++++ > drivers/i2c/busses/i2c-rk3x.c | 57 +++++++++++++++------- > 2 files changed, 50 insertions(+), 18 deletions(-) > > diff --git a/Documentation/devicetree/bindings/i2c/i2c-rk3x.txt b/Documentation/devicetree/bindings/i2c/i2c-rk3x.txt > index dde6c22..925d6eb 100644 > --- a/Documentation/devicetree/bindings/i2c/i2c-rk3x.txt > +++ b/Documentation/devicetree/bindings/i2c/i2c-rk3x.txt > @@ -21,6 +21,14 @@ Required on RK3066, RK3188 : > Optional properties : > > - clock-frequency : SCL frequency to use (in Hz). If omitted, 100kHz is used. > + - i2c-scl-rising-time-ns : Number of nanoseconds the signal takes to rise > + (t(r) in i2c spec). If not specified this is assumed to be the max the > + spec allows(1000 ns for standard mode, 300 ns for fast mode) which > + might cause slightly slower communication. > + - i2c-scl-falling-time-ns : Number of nanoseconds the signal takes to fall > + (t(f) in the i2c0 spec). If not specified this is assumed to be the > + max the spec allows (300 ns) which might cause slightly slower > + communication. > > Example: > > @@ -39,4 +47,7 @@ i2c0: i2c@2002d000 { > > clock-names = "i2c"; > clocks = <&cru PCLK_I2C0>; > + > + i2c-scl-rising-time-ns = <800>; > + i2c-scl-falling-time-ns = <100>; > }; > diff --git a/drivers/i2c/busses/i2c-rk3x.c b/drivers/i2c/busses/i2c-rk3x.c > index 0ee5802..50c1678 100644 > --- a/drivers/i2c/busses/i2c-rk3x.c > +++ b/drivers/i2c/busses/i2c-rk3x.c > @@ -102,6 +102,8 @@ struct rk3x_i2c { > > /* Settings */ > unsigned int scl_frequency; > + unsigned int rise_ns; > + unsigned int fall_ns; > > /* Synchronization & notification */ > spinlock_t lock; > @@ -435,6 +437,8 @@ out: > * > * @clk_rate: I2C input clock rate > * @scl_rate: Desired SCL rate > + * @rise_ns: How many ns it takes for signals to rise. > + * @fall_ns: How many ns it takes for signals to fall. > * @div_low: Divider output for low > * @div_high: Divider output for high > * > @@ -443,11 +447,14 @@ out: > * too high, we silently use the highest possible rate. > */ > static int rk3x_i2c_calc_divs(unsigned long clk_rate, unsigned long scl_rate, > + unsigned long rise_ns, unsigned long fall_ns, > unsigned long *div_low, unsigned long *div_high) > { > + unsigned long spec_min_low_ns, spec_min_high_ns; > + unsigned long spec_max_data_hold_ns; > + unsigned long spec_data_hold_buffer_ns; > + > unsigned long min_low_ns, min_high_ns; > - unsigned long max_data_hold_ns; > - unsigned long data_hold_buffer_ns; > unsigned long max_low_ns, min_total_ns; > > unsigned long clk_rate_khz, scl_rate_khz; > @@ -470,28 +477,30 @@ static int rk3x_i2c_calc_divs(unsigned long clk_rate, unsigned long scl_rate, > > /* > * min_low_ns: The minimum number of ns we need to hold low > - * to meet i2c spec > + * to meet i2c spec, should include fall time. > * min_high_ns: The minimum number of ns we need to hold high > - * to meet i2c spec > + * to meet i2c spec, should include rise time. > * max_low_ns: The maximum number of ns we can hold low > - * to meet i2c spec > + * to meet i2c spec. > * > * Note: max_low_ns should be (max data hold time * 2 - buffer) > * This is because the i2c host on Rockchip holds the data line > * for half the low time. > */ > if (scl_rate <= 100000) { > - min_low_ns = 4700; > - min_high_ns = 4000; > - max_data_hold_ns = 3450; > - data_hold_buffer_ns = 50; > + spec_min_low_ns = 4700; > + spec_min_high_ns = 4000; > + spec_max_data_hold_ns = 3450; > + spec_data_hold_buffer_ns = 50; > } else { > - min_low_ns = 1300; > - min_high_ns = 600; > - max_data_hold_ns = 900; > - data_hold_buffer_ns = 50; > + spec_min_low_ns = 1300; > + spec_min_high_ns = 600; > + spec_max_data_hold_ns = 900; > + spec_data_hold_buffer_ns = 50; > } > - max_low_ns = max_data_hold_ns * 2 - data_hold_buffer_ns; > + min_low_ns = spec_min_low_ns + fall_ns; > + min_high_ns = spec_min_high_ns + rise_ns; > + max_low_ns = spec_max_data_hold_ns * 2 - spec_data_hold_buffer_ns; > min_total_ns = min_low_ns + min_high_ns; > > /* Adjust to avoid overflow */ > @@ -588,8 +597,8 @@ static void rk3x_i2c_adapt_div(struct rk3x_i2c *i2c, unsigned long clk_rate) > u64 t_low_ns, t_high_ns; > int ret; > > - ret = rk3x_i2c_calc_divs(clk_rate, i2c->scl_frequency, &div_low, > - &div_high); > + ret = rk3x_i2c_calc_divs(clk_rate, i2c->scl_frequency, i2c->rise_ns, > + i2c->fall_ns, &div_low, &div_high); > > WARN_ONCE(ret != 0, "Could not reach SCL freq %u", i2c->scl_frequency); > > @@ -633,9 +642,9 @@ static int rk3x_i2c_clk_notifier_cb(struct notifier_block *nb, unsigned long > switch (event) { > case PRE_RATE_CHANGE: > if (rk3x_i2c_calc_divs(ndata->new_rate, i2c->scl_frequency, > - &div_low, &div_high) != 0) { > + i2c->rise_ns, i2c->fall_ns, &div_low, > + &div_high) != 0) > return NOTIFY_STOP; > - } > > /* scale up */ > if (ndata->new_rate > ndata->old_rate) > @@ -859,6 +868,18 @@ static int rk3x_i2c_probe(struct platform_device *pdev) > i2c->scl_frequency = DEFAULT_SCL_RATE; > } > > + /* Read rise and fall ns; if not there use the default max from spec */ > + if (of_property_read_u32(pdev->dev.of_node, "i2c-scl-rising-time-ns", > + &i2c->rise_ns)) { > + if (i2c->scl_frequency <= 100000) > + i2c->rise_ns = 1000; > + else > + i2c->rise_ns = 300; > + } > + if (of_property_read_u32(pdev->dev.of_node, "i2c-scl-falling-time-ns", > + &i2c->fall_ns)) > + i2c->fall_ns = 300; > + > strlcpy(i2c->adap.name, "rk3x-i2c", sizeof(i2c->adap.name)); > i2c->adap.owner = THIS_MODULE; > i2c->adap.algo = &rk3x_i2c_algorithm; >
Hello Addy, On Mon, Dec 08, 2014 at 10:59:49AM +0800, Addy Ke wrote: > high_ns calculated from the low division of CLKDIV register is the sum > of actual measured high_ns and rise_ns. The rise time which related to I think "actual measured" is misleading here. (The driver doesn't dermine these parameters automatically, right?) Maybe something like the following is more correct and understandable?: The number of clock cycles to be written into the XYZ register that determines the i2c clk high phase includes the rise time. So to meet the timing requirements defined in the i2c specification which defines the minimal time SCL has to be high, the rise time has to taken into account. The same applies to the low phase with falling time. > external pull-up resistor can be up to the maximum rise time in I2C spec. > > In my test, if external pull-up resistor is 4.7K, rise_ns is about It would be nice to point out the actual board you used (if it's a publically available machine). > 700ns. So the actual measured high_ns is about 3900ns, which is less > than 4000ns(the minimum high_ns in I2C spec). s/s(/s (/; s/spec/specification for Standard-mode/ There are different capitalisation of i2c in the patch and the commit log. I don't know what Wolfram prefers here, but using the same spelling everywhere would be nice. > To fix this bug, min_low_ns should include fall time and min_high_ns > should include rise time too. > > This patch merged the patch that Doug submitted to chromium, which Isn't Chromium a web browser? Does it really contain i2c stuff? I bet it's about Chromium OS. > can get the rise and fall times for signals from the device tree. > This allows us to more accurately calculate timings. see: > https://chromium-review.googlesource.com/#/c/232774/ > > Signed-off-by: Addy Ke <addy.ke@rock-chips.com> > --- > Changes in v2: > - merged the patch that Doug submitted to chromium > Changes in v3: > - merged the patch that Doug submitted to chromium to change bindins > see: https://chromium-review.googlesource.com/#/c/232775/ > > Documentation/devicetree/bindings/i2c/i2c-rk3x.txt | 11 +++++ > drivers/i2c/busses/i2c-rk3x.c | 57 +++++++++++++++------- > 2 files changed, 50 insertions(+), 18 deletions(-) > > diff --git a/Documentation/devicetree/bindings/i2c/i2c-rk3x.txt b/Documentation/devicetree/bindings/i2c/i2c-rk3x.txt > index dde6c22..925d6eb 100644 > --- a/Documentation/devicetree/bindings/i2c/i2c-rk3x.txt > +++ b/Documentation/devicetree/bindings/i2c/i2c-rk3x.txt > @@ -21,6 +21,14 @@ Required on RK3066, RK3188 : > Optional properties : > > - clock-frequency : SCL frequency to use (in Hz). If omitted, 100kHz is used. > + - i2c-scl-rising-time-ns : Number of nanoseconds the signal takes to rise > + (t(r) in i2c spec). If not specified this is assumed to be the max the s/spec/specification/; s/max/maximum/ > + spec allows(1000 ns for standard mode, 300 ns for fast mode) which s/s(/s (/; s/standard mode/Standard-mode/; s/fast mode/Fast-mode/; > + might cause slightly slower communication. > + - i2c-scl-falling-time-ns : Number of nanoseconds the signal takes to fall > + (t(f) in the i2c0 spec). If not specified this is assumed to be the s/i2c0/i2c/ > + max the spec allows (300 ns) which might cause slightly slower > + communication. > > Example: > > @@ -39,4 +47,7 @@ i2c0: i2c@2002d000 { > > clock-names = "i2c"; > clocks = <&cru PCLK_I2C0>; > + > + i2c-scl-rising-time-ns = <800>; > + i2c-scl-falling-time-ns = <100>; > }; > diff --git a/drivers/i2c/busses/i2c-rk3x.c b/drivers/i2c/busses/i2c-rk3x.c > index 0ee5802..50c1678 100644 > --- a/drivers/i2c/busses/i2c-rk3x.c > +++ b/drivers/i2c/busses/i2c-rk3x.c > @@ -470,28 +477,30 @@ static int rk3x_i2c_calc_divs(unsigned long clk_rate, unsigned long scl_rate, > > /* > * min_low_ns: The minimum number of ns we need to hold low > - * to meet i2c spec > + * to meet i2c spec, should include fall time. > * min_high_ns: The minimum number of ns we need to hold high > - * to meet i2c spec > + * to meet i2c spec, should include rise time. > * max_low_ns: The maximum number of ns we can hold low > - * to meet i2c spec > + * to meet i2c spec. > * > * Note: max_low_ns should be (max data hold time * 2 - buffer) > * This is because the i2c host on Rockchip holds the data line > * for half the low time. > */ > if (scl_rate <= 100000) { /* Standard-mode */ > - min_low_ns = 4700; > - min_high_ns = 4000; > - max_data_hold_ns = 3450; > - data_hold_buffer_ns = 50; > + spec_min_low_ns = 4700; > + spec_min_high_ns = 4000; > + spec_max_data_hold_ns = 3450; The specification calls this "data valid time" (t_(VD;DAT)). > + spec_data_hold_buffer_ns = 50; I didn't find this parameter in the spec, also it doesn't differ for Standard-mode vs. Fast-mode. What is this? > } else { /* Fast-mode */ > - min_low_ns = 1300; > - min_high_ns = 600; > - max_data_hold_ns = 900; > - data_hold_buffer_ns = 50; > + spec_min_low_ns = 1300; > + spec_min_high_ns = 600; > + spec_max_data_hold_ns = 900; > + spec_data_hold_buffer_ns = 50; > } > - max_low_ns = max_data_hold_ns * 2 - data_hold_buffer_ns; > + min_low_ns = spec_min_low_ns + fall_ns; > + min_high_ns = spec_min_high_ns + rise_ns; > + max_low_ns = spec_max_data_hold_ns * 2 - spec_data_hold_buffer_ns; > min_total_ns = min_low_ns + min_high_ns; > > /* Adjust to avoid overflow */ > @@ -588,8 +597,8 @@ static void rk3x_i2c_adapt_div(struct rk3x_i2c *i2c, unsigned long clk_rate) > u64 t_low_ns, t_high_ns; > int ret; > > - ret = rk3x_i2c_calc_divs(clk_rate, i2c->scl_frequency, &div_low, > - &div_high); > + ret = rk3x_i2c_calc_divs(clk_rate, i2c->scl_frequency, i2c->rise_ns, > + i2c->fall_ns, &div_low, &div_high); > > WARN_ONCE(ret != 0, "Could not reach SCL freq %u", i2c->scl_frequency); > > @@ -633,9 +642,9 @@ static int rk3x_i2c_clk_notifier_cb(struct notifier_block *nb, unsigned long > switch (event) { > case PRE_RATE_CHANGE: > if (rk3x_i2c_calc_divs(ndata->new_rate, i2c->scl_frequency, > - &div_low, &div_high) != 0) { > + i2c->rise_ns, i2c->fall_ns, &div_low, > + &div_high) != 0) > return NOTIFY_STOP; > - } > > /* scale up */ > if (ndata->new_rate > ndata->old_rate) > @@ -859,6 +868,18 @@ static int rk3x_i2c_probe(struct platform_device *pdev) > i2c->scl_frequency = DEFAULT_SCL_RATE; > } > > + /* Read rise and fall ns; if not there use the default max from spec */ > + if (of_property_read_u32(pdev->dev.of_node, "i2c-scl-rising-time-ns", > + &i2c->rise_ns)) { > + if (i2c->scl_frequency <= 100000) > + i2c->rise_ns = 1000; > + else > + i2c->rise_ns = 300; > + } > + if (of_property_read_u32(pdev->dev.of_node, "i2c-scl-falling-time-ns", > + &i2c->fall_ns)) > + i2c->fall_ns = 300; > + I don't know if other drivers do the same (I assume they should). If so, moving this logic into the core would be nice. I guess this can still be done later. The i2c specification also has a minimum for t(r) and t(f). I don't know why these are limited, but I think it would be good to check for these limits, too. > strlcpy(i2c->adap.name, "rk3x-i2c", sizeof(i2c->adap.name)); > i2c->adap.owner = THIS_MODULE; > i2c->adap.algo = &rk3x_i2c_algorithm; Best regards Uwe
Hi, On Mon, Dec 8, 2014 at 12:52 AM, Uwe Kleine-König <u.kleine-koenig@pengutronix.de> wrote: > Hello Addy, > > On Mon, Dec 08, 2014 at 10:59:49AM +0800, Addy Ke wrote: >> high_ns calculated from the low division of CLKDIV register is the sum >> of actual measured high_ns and rise_ns. The rise time which related to > I think "actual measured" is misleading here. (The driver doesn't > dermine these parameters automatically, right?) > > Maybe something like the following is more correct and understandable?: > > The number of clock cycles to be written into the XYZ register > that determines the i2c clk high phase includes the rise time. > So to meet the timing requirements defined in the i2c > specification which defines the minimal time SCL has to be high, > the rise time has to taken into account. The same applies to the > low phase with falling time. I agree that the prose of the commit message could be cleaned up. I don't think English is Addy's primary language. Your proposals sound fine to me, though I would probably replace XYZ by "CLKDIV". You should capitalize I2C here, too. >> external pull-up resistor can be up to the maximum rise time in I2C spec. >> >> In my test, if external pull-up resistor is 4.7K, rise_ns is about > It would be nice to point out the actual board you used (if it's a > publically available machine). He probably tested on rk3288-pinky, which is not an upstream board yet. >> 700ns. So the actual measured high_ns is about 3900ns, which is less >> than 4000ns(the minimum high_ns in I2C spec). > s/s(/s (/; > s/spec/specification for Standard-mode/ > > There are different capitalisation of i2c in the patch and the commit log. I > don't know what Wolfram prefers here, but using the same spelling > everywhere would be nice. Can you please point out? IIRC you should always capitalize I2C in prose (descriptions, comments, documentation, etc). However when used in places which specific capitalization standards it should be lowercase. That means file names, directory names, device tree property names, etc should be lower case. I think Addy got it right in most cases and places where it's wrong are probably my fault. I think I missed a few in my patches. :( >> To fix this bug, min_low_ns should include fall time and min_high_ns >> should include rise time too. >> >> This patch merged the patch that Doug submitted to chromium, which > Isn't Chromium a web browser? Does it really contain i2c stuff? > I bet it's about Chromium OS. Chromium OS is part of the Chromium project. >> + (t(r) in i2c spec). If not specified this is assumed to be the max the > s/spec/specification/; s/max/maximum/ This should have been I2C, not i2c (capital). >> + might cause slightly slower communication. >> + - i2c-scl-falling-time-ns : Number of nanoseconds the signal takes to fall >> + (t(f) in the i2c0 spec). If not specified this is assumed to be the > s/i2c0/i2c/ Same here: I2C. >> * min_low_ns: The minimum number of ns we need to hold low >> - * to meet i2c spec >> + * to meet i2c spec, should include fall time. >> * min_high_ns: The minimum number of ns we need to hold high >> - * to meet i2c spec >> + * to meet i2c spec, should include rise time. >> * max_low_ns: The maximum number of ns we can hold low >> - * to meet i2c spec >> + * to meet i2c spec. ...and those should be I2C. >> * >> * Note: max_low_ns should be (max data hold time * 2 - buffer) >> * This is because the i2c host on Rockchip holds the data line >> * for half the low time. >> */ >> if (scl_rate <= 100000) { > /* Standard-mode */ >> - min_low_ns = 4700; >> - min_high_ns = 4000; >> - max_data_hold_ns = 3450; >> - data_hold_buffer_ns = 50; >> + spec_min_low_ns = 4700; >> + spec_min_high_ns = 4000; >> + spec_max_data_hold_ns = 3450; > The specification calls this "data valid time" (t_(VD;DAT)). Could change this if need be, but the old code had it as "data hold". Maybe we could argue about it in a future patch? >> + spec_data_hold_buffer_ns = 50; > I didn't find this parameter in the spec, also it doesn't differ for > Standard-mode vs. Fast-mode. What is this? I believe I put it in an earlier bit of sample code when Addy was worried about things being too close to the margins. I'm not sure what specifically Addy was worried about. Maybe this should actually be the falling time for data? Again, this is not something new so perhaps we could debate in a future patch? >> + /* Read rise and fall ns; if not there use the default max from spec */ >> + if (of_property_read_u32(pdev->dev.of_node, "i2c-scl-rising-time-ns", >> + &i2c->rise_ns)) { >> + if (i2c->scl_frequency <= 100000) >> + i2c->rise_ns = 1000; >> + else >> + i2c->rise_ns = 300; >> + } >> + if (of_property_read_u32(pdev->dev.of_node, "i2c-scl-falling-time-ns", >> + &i2c->fall_ns)) >> + i2c->fall_ns = 300; >> + > I don't know if other drivers do the same (I assume they should). If so, > moving this logic into the core would be nice. I guess this can still be > done later. > > The i2c specification also has a minimum for t(r) and t(f). I don't know > why these are limited, but I think it would be good to check for these > limits, too. One of these two properties is in another driver (the Designware one). However, different hardware seems to take in different parameters for I2C. It seems like for now the best thing to do is to keep drivers consistent. If we notice a pattern then stuff should move to the core. -Doug
> > There are different capitalisation of i2c in the patch and the commit log. I > > don't know what Wolfram prefers here, but using the same spelling > > everywhere would be nice. > > Can you please point out? IIRC you should always capitalize I2C in > prose (descriptions, comments, documentation, etc). However when used > in places which specific capitalization standards it should be > lowercase. That means file names, directory names, device tree > property names, etc should be lower case. This is what I prefer, too. However, I am not too strict about it. I mean, both are readable. > >> + /* Read rise and fall ns; if not there use the default max from spec */ > >> + if (of_property_read_u32(pdev->dev.of_node, "i2c-scl-rising-time-ns", > >> + &i2c->rise_ns)) { > >> + if (i2c->scl_frequency <= 100000) > >> + i2c->rise_ns = 1000; > >> + else > >> + i2c->rise_ns = 300; > >> + } > >> + if (of_property_read_u32(pdev->dev.of_node, "i2c-scl-falling-time-ns", > >> + &i2c->fall_ns)) > >> + i2c->fall_ns = 300; > >> + > > I don't know if other drivers do the same (I assume they should). If so, > > moving this logic into the core would be nice. I guess this can still be > > done later. Let's leave that for later. If we do it, we should start with bus speed setting first, probably.
Hi, On Mon, Dec 8, 2014 at 9:34 AM, Wolfram Sang <wsa@the-dreams.de> wrote: > >> > There are different capitalisation of i2c in the patch and the commit log. I >> > don't know what Wolfram prefers here, but using the same spelling >> > everywhere would be nice. >> >> Can you please point out? IIRC you should always capitalize I2C in >> prose (descriptions, comments, documentation, etc). However when used >> in places which specific capitalization standards it should be >> lowercase. That means file names, directory names, device tree >> property names, etc should be lower case. > > This is what I prefer, too. However, I am not too strict about it. > I mean, both are readable. Glad my understanding is OK. :) > >> >> + /* Read rise and fall ns; if not there use the default max from spec */ >> >> + if (of_property_read_u32(pdev->dev.of_node, "i2c-scl-rising-time-ns", >> >> + &i2c->rise_ns)) { >> >> + if (i2c->scl_frequency <= 100000) >> >> + i2c->rise_ns = 1000; >> >> + else >> >> + i2c->rise_ns = 300; >> >> + } >> >> + if (of_property_read_u32(pdev->dev.of_node, "i2c-scl-falling-time-ns", >> >> + &i2c->fall_ns)) >> >> + i2c->fall_ns = 300; >> >> + >> > I don't know if other drivers do the same (I assume they should). If so, >> > moving this logic into the core would be nice. I guess this can still be >> > done later. > > Let's leave that for later. If we do it, we should start with bus speed > setting first, probably. OK, great. So just to summarize for Addy, I think you're being asked to spin one more time. 1. Update the patch description as per Uwe. Fix other typos pointed out by him. 2. Since you're spinning anyway, adjust "i2c" capitalization to "I2C". 3. If you agree with Uwe, rename spec_max_data_hold_ns to spec_data_valid_ns. If not, we should debate in a followup patch. 4. If you think spec_data_hold_buffer_ns could be better represented in another way (should this be i2c-sda-falling-time-ns"?), please do so. If not, we should debate in a followup patch. -Doug
Hello, On Mon, Dec 08, 2014 at 10:53:44AM -0800, Doug Anderson wrote: > So just to summarize for Addy, I think you're being asked to spin one more time. > > 1. Update the patch description as per Uwe. Fix other typos pointed out by him. > > 2. Since you're spinning anyway, adjust "i2c" capitalization to "I2C". > > 3. If you agree with Uwe, rename spec_max_data_hold_ns to > spec_data_valid_ns. If not, we should debate in a followup patch. > > 4. If you think spec_data_hold_buffer_ns could be better represented > in another way (should this be i2c-sda-falling-time-ns"?), please do > so. If not, we should debate in a followup patch. Sounds all reasonable. I'd not address 3 and 4 in this patch, but do this in separate one. (That's what I intended from the beginning, just failed to point that out explicitly.) Uwe
diff --git a/Documentation/devicetree/bindings/i2c/i2c-rk3x.txt b/Documentation/devicetree/bindings/i2c/i2c-rk3x.txt index dde6c22..925d6eb 100644 --- a/Documentation/devicetree/bindings/i2c/i2c-rk3x.txt +++ b/Documentation/devicetree/bindings/i2c/i2c-rk3x.txt @@ -21,6 +21,14 @@ Required on RK3066, RK3188 : Optional properties : - clock-frequency : SCL frequency to use (in Hz). If omitted, 100kHz is used. + - i2c-scl-rising-time-ns : Number of nanoseconds the signal takes to rise + (t(r) in i2c spec). If not specified this is assumed to be the max the + spec allows(1000 ns for standard mode, 300 ns for fast mode) which + might cause slightly slower communication. + - i2c-scl-falling-time-ns : Number of nanoseconds the signal takes to fall + (t(f) in the i2c0 spec). If not specified this is assumed to be the + max the spec allows (300 ns) which might cause slightly slower + communication. Example: @@ -39,4 +47,7 @@ i2c0: i2c@2002d000 { clock-names = "i2c"; clocks = <&cru PCLK_I2C0>; + + i2c-scl-rising-time-ns = <800>; + i2c-scl-falling-time-ns = <100>; }; diff --git a/drivers/i2c/busses/i2c-rk3x.c b/drivers/i2c/busses/i2c-rk3x.c index 0ee5802..50c1678 100644 --- a/drivers/i2c/busses/i2c-rk3x.c +++ b/drivers/i2c/busses/i2c-rk3x.c @@ -102,6 +102,8 @@ struct rk3x_i2c { /* Settings */ unsigned int scl_frequency; + unsigned int rise_ns; + unsigned int fall_ns; /* Synchronization & notification */ spinlock_t lock; @@ -435,6 +437,8 @@ out: * * @clk_rate: I2C input clock rate * @scl_rate: Desired SCL rate + * @rise_ns: How many ns it takes for signals to rise. + * @fall_ns: How many ns it takes for signals to fall. * @div_low: Divider output for low * @div_high: Divider output for high * @@ -443,11 +447,14 @@ out: * too high, we silently use the highest possible rate. */ static int rk3x_i2c_calc_divs(unsigned long clk_rate, unsigned long scl_rate, + unsigned long rise_ns, unsigned long fall_ns, unsigned long *div_low, unsigned long *div_high) { + unsigned long spec_min_low_ns, spec_min_high_ns; + unsigned long spec_max_data_hold_ns; + unsigned long spec_data_hold_buffer_ns; + unsigned long min_low_ns, min_high_ns; - unsigned long max_data_hold_ns; - unsigned long data_hold_buffer_ns; unsigned long max_low_ns, min_total_ns; unsigned long clk_rate_khz, scl_rate_khz; @@ -470,28 +477,30 @@ static int rk3x_i2c_calc_divs(unsigned long clk_rate, unsigned long scl_rate, /* * min_low_ns: The minimum number of ns we need to hold low - * to meet i2c spec + * to meet i2c spec, should include fall time. * min_high_ns: The minimum number of ns we need to hold high - * to meet i2c spec + * to meet i2c spec, should include rise time. * max_low_ns: The maximum number of ns we can hold low - * to meet i2c spec + * to meet i2c spec. * * Note: max_low_ns should be (max data hold time * 2 - buffer) * This is because the i2c host on Rockchip holds the data line * for half the low time. */ if (scl_rate <= 100000) { - min_low_ns = 4700; - min_high_ns = 4000; - max_data_hold_ns = 3450; - data_hold_buffer_ns = 50; + spec_min_low_ns = 4700; + spec_min_high_ns = 4000; + spec_max_data_hold_ns = 3450; + spec_data_hold_buffer_ns = 50; } else { - min_low_ns = 1300; - min_high_ns = 600; - max_data_hold_ns = 900; - data_hold_buffer_ns = 50; + spec_min_low_ns = 1300; + spec_min_high_ns = 600; + spec_max_data_hold_ns = 900; + spec_data_hold_buffer_ns = 50; } - max_low_ns = max_data_hold_ns * 2 - data_hold_buffer_ns; + min_low_ns = spec_min_low_ns + fall_ns; + min_high_ns = spec_min_high_ns + rise_ns; + max_low_ns = spec_max_data_hold_ns * 2 - spec_data_hold_buffer_ns; min_total_ns = min_low_ns + min_high_ns; /* Adjust to avoid overflow */ @@ -588,8 +597,8 @@ static void rk3x_i2c_adapt_div(struct rk3x_i2c *i2c, unsigned long clk_rate) u64 t_low_ns, t_high_ns; int ret; - ret = rk3x_i2c_calc_divs(clk_rate, i2c->scl_frequency, &div_low, - &div_high); + ret = rk3x_i2c_calc_divs(clk_rate, i2c->scl_frequency, i2c->rise_ns, + i2c->fall_ns, &div_low, &div_high); WARN_ONCE(ret != 0, "Could not reach SCL freq %u", i2c->scl_frequency); @@ -633,9 +642,9 @@ static int rk3x_i2c_clk_notifier_cb(struct notifier_block *nb, unsigned long switch (event) { case PRE_RATE_CHANGE: if (rk3x_i2c_calc_divs(ndata->new_rate, i2c->scl_frequency, - &div_low, &div_high) != 0) { + i2c->rise_ns, i2c->fall_ns, &div_low, + &div_high) != 0) return NOTIFY_STOP; - } /* scale up */ if (ndata->new_rate > ndata->old_rate) @@ -859,6 +868,18 @@ static int rk3x_i2c_probe(struct platform_device *pdev) i2c->scl_frequency = DEFAULT_SCL_RATE; } + /* Read rise and fall ns; if not there use the default max from spec */ + if (of_property_read_u32(pdev->dev.of_node, "i2c-scl-rising-time-ns", + &i2c->rise_ns)) { + if (i2c->scl_frequency <= 100000) + i2c->rise_ns = 1000; + else + i2c->rise_ns = 300; + } + if (of_property_read_u32(pdev->dev.of_node, "i2c-scl-falling-time-ns", + &i2c->fall_ns)) + i2c->fall_ns = 300; + strlcpy(i2c->adap.name, "rk3x-i2c", sizeof(i2c->adap.name)); i2c->adap.owner = THIS_MODULE; i2c->adap.algo = &rk3x_i2c_algorithm;
high_ns calculated from the low division of CLKDIV register is the sum of actual measured high_ns and rise_ns. The rise time which related to external pull-up resistor can be up to the maximum rise time in I2C spec. In my test, if external pull-up resistor is 4.7K, rise_ns is about 700ns. So the actual measured high_ns is about 3900ns, which is less than 4000ns(the minimum high_ns in I2C spec). To fix this bug, min_low_ns should include fall time and min_high_ns should include rise time too. This patch merged the patch that Doug submitted to chromium, which can get the rise and fall times for signals from the device tree. This allows us to more accurately calculate timings. see: https://chromium-review.googlesource.com/#/c/232774/ Signed-off-by: Addy Ke <addy.ke@rock-chips.com> --- Changes in v2: - merged the patch that Doug submitted to chromium Changes in v3: - merged the patch that Doug submitted to chromium to change bindins see: https://chromium-review.googlesource.com/#/c/232775/ Documentation/devicetree/bindings/i2c/i2c-rk3x.txt | 11 +++++ drivers/i2c/busses/i2c-rk3x.c | 57 +++++++++++++++------- 2 files changed, 50 insertions(+), 18 deletions(-)