Skip to contentSkip to conversation

soc: qcom: geni: Don't ignore clk_round_rate() errors in geni_se_clk_tbl_get()

PATCH Superseded Douglas Anderson 3 participants 5 messages Sep 6, 2018

Subsystems (2)

v2 of 3 Superseded Latest is v3 →

lore
Share & tools
lore
#0 lore raw

PATCH v2 1/2 soc: qcom: geni: Don't ignore clk_round_rate() errors in geni_se_clk_tbl_get()

The function clk_round_rate() is defined to return a "long", not an "unsigned long". That's because it might return a negative error code. Change the call in geni_se_clk_tbl_get() to check for errors.

While we're at it, get rid of a useless init of "freq".

NOTE: overall the idea that we should iterate over clk_round_rate() to try to reconstruct a table already present in the clock driver is questionable. Specifically:
- This method relies on "clk_round_rate()" rounding up.
- This method only works if the table is sorted and has no duplicates. ...this patch doesn't try to fix those problems, it just makes the error handling more correct.

Fixes eddac5af0654 ("soc: qcom: Add GENI based QUP Wrapper driver")

---

Changes in v2:
- Get rid of unneeded init of "freq" (Matthias).
- Add Matthias tag.

 drivers/soc/qcom/qcom-geni-se.c | 4 ++--
 1 file changed, 2 insertions(+), 2 deletions(-)
drivers/soc/qcom/qcom-geni-se.c +2 -2
index feed3db21c10..d2d97f1d7428 100644
@@ -513,7 +513,7 @@ EXPORT_SYMBOL(geni_se_resources_on);
  */
 int geni_se_clk_tbl_get(struct geni_se *se, unsigned long **tbl)
 {
-	unsigned long freq = 0;
+	long freq;
 	int i;
 
 	if (se->clk_perf_tbl) {
@@ -529,7 +529,7 @@ int geni_se_clk_tbl_get(struct geni_se *se, unsigned long **tbl)
 
 	for (i = 0; i < MAX_CLK_PERF_LEVEL; i++) {
 		freq = clk_round_rate(se->clk, freq + 1);
-		if (!freq || freq == se->clk_perf_tbl[i - 1])
+		if (freq <= 0 || freq == se->clk_perf_tbl[i - 1])
 			break;
 		se->clk_perf_tbl[i] = freq;
 	}
Signature · 1 line
2.19.0.rc1.350.ge57e33dbd1-goog
#1 lore raw

PATCH v2 2/2 soc: qcom: geni: geni_se_clk_freq_match() should always accept multiples

The geni_se_clk_freq_match() has some strange semantics. Specifically it is defined with two modes:
1. It can find a clock that's an exact multiple of the requested rate
2. It can find a non-exact match but it can't handle multiples then

...but callers should always be able to handle a clock that is a multiple of the requested clock so mode #2 doesn't really make sense. Let's change the semantics so that the non-exact match can also accept multiples and then change the code to handle that.

The only caller of this code is the unlanded SPI driver [1] which currently passes "exact = True", thus it should be safe to change the semantics in this way. ...and, in fact, the SPI driver should likely be modified to pass "exact = False" (with the new semantics) since that will allow it to work with SPI devices that request a clock rate that doesn't exactly match a rate we can make.

[1] https://lkml.kernel.org/r/[email protected]

Fixes eddac5af0654 ("soc: qcom: Add GENI based QUP Wrapper driver")

---

Changes in v2:
- Init best_delta to ULONG_MAX and avoid extra test (Matthias).
- s/If/It in commit message.
- Add Matthias tag.

 drivers/soc/qcom/qcom-geni-se.c | 37 ++++++++++++++++++++-------------
 1 file changed, 22 insertions(+), 15 deletions(-)
drivers/soc/qcom/qcom-geni-se.c +22 -15
index d2d97f1d7428..033bd56edcbb 100644
@@ -544,16 +544,17 @@ EXPORT_SYMBOL(geni_se_clk_tbl_get);
  * @se:		Pointer to the concerned serial engine.
  * @req_freq:	Requested clock frequency.
  * @index:	Index of the resultant frequency in the table.
- * @res_freq:	Resultant frequency which matches or is closer to the
- *		requested frequency.
+ * @res_freq:	Resultant frequency of the source clock.
  * @exact:	Flag to indicate exact multiple requirement of the requested
  *		frequency.
  *
- * This function is called by the protocol drivers to determine the matching
- * or exact multiple of the requested frequency, as provided by the serial
- * engine clock in order to meet the performance requirements. If there is
- * no matching or exact multiple of the requested frequency found, then it
- * selects the closest floor frequency, if exact flag is not set.
+ * This function is called by the protocol drivers to determine the best match
+ * of the requested frequency as provided by the serial engine clock in order
+ * to meet the performance requirements.
+ *
+ * If we return success:
+ * - if @exact is true  then @res_freq / <an_integer> == @req_freq
+ * - if @exact is false then @res_freq / <an_integer> <= @req_freq
  *
  * Return: 0 on success, standard Linux error codes on failure.
  */
@@ -564,6 +565,9 @@ int geni_se_clk_freq_match(struct geni_se *se, unsigned long req_freq,
 	unsigned long *tbl;
 	int num_clk_levels;
 	int i;
+	unsigned long best_delta;
+	unsigned long new_delta;
+	unsigned int divider;
 
 	num_clk_levels = geni_se_clk_tbl_get(se, &tbl);
 	if (num_clk_levels < 0)
@@ -572,18 +576,21 @@ int geni_se_clk_freq_match(struct geni_se *se, unsigned long req_freq,
 	if (num_clk_levels == 0)
 		return -EINVAL;
 
-	*res_freq = 0;
+	best_delta = ULONG_MAX;
 	for (i = 0; i < num_clk_levels; i++) {
-		if (!(tbl[i] % req_freq)) {
+		divider = DIV_ROUND_UP(tbl[i], req_freq);
+		new_delta = req_freq - tbl[i] / divider;
+		if (new_delta < best_delta) {
+			/* We have a new best! */
 			*index = i;
 			*res_freq = tbl[i];
-			return 0;
-		}
 
-		if (!(*res_freq) || ((tbl[i] > *res_freq) &&
-				     (tbl[i] < req_freq))) {
-			*index = i;
-			*res_freq = tbl[i];
+			/* If the new best is exact then we're done */
+			if (new_delta == 0)
+				return 0;
+
+			/* Record how close we got */
+			best_delta = new_delta;
 		}
 	}
 
Signature · 1 line
2.19.0.rc1.350.ge57e33dbd1-goog
#2 lore raw
In reply to Douglas Anderson
Thank you for the patch! Perhaps something to improve:

Hi Douglas,

Thank you for the patch! Perhaps something to improve:

[auto build test WARNING on agross/for-next]
[also build test WARNING on v4.19-rc2 next-20180906]
[if your patch is applied to the wrong git tree, please drop us a note to help improve the system]

url:    https://github.com/0day-ci/linux/commits/Douglas-Anderson/soc-qcom-geni-Don-t-ignore-clk_round_rate-errors-in-geni_se_clk_tbl_get/20180907-045155
base:   https://git.kernel.org/pub/scm/linux/kernel/git/agross/linux.git for-next
config: i386-allmodconfig (attached as .config)
compiler: gcc-7 (Debian 7.3.0-1) 7.3.0
reproduce:
        # save the attached .config to linux build tree
        make ARCH=i386 

Note: it may well be a FALSE warning. FWIW you are at least aware of it now.
http://gcc.gnu.org/wiki/Better_Uninitialized_Warnings

All warnings (new ones prefixed by >>):

drivers/soc/qcom/qcom-geni-se.c: In function 'geni_se_clk_tbl_get':

>> drivers/soc/qcom/qcom-geni-se.c:531:39: warning: 'freq' may be used uninitialized in this function [-Wmaybe-uninitialized]
      freq = clk_round_rate(se->clk, freq + 1);
                                     ~~~~~^~~

vim +/freq +531 drivers/soc/qcom/qcom-geni-se.c

eddac5af Karthikeyan Ramasubramanian 2018-03-30  500  
eddac5af Karthikeyan Ramasubramanian 2018-03-30  501  /**
eddac5af Karthikeyan Ramasubramanian 2018-03-30  502   * geni_se_clk_tbl_get() - Get the clock table to program DFS
eddac5af Karthikeyan Ramasubramanian 2018-03-30  503   * @se:		Pointer to the concerned serial engine.
eddac5af Karthikeyan Ramasubramanian 2018-03-30  504   * @tbl:	Table in which the output is returned.
eddac5af Karthikeyan Ramasubramanian 2018-03-30  505   *
eddac5af Karthikeyan Ramasubramanian 2018-03-30  506   * This function is called by the protocol drivers to determine the different
eddac5af Karthikeyan Ramasubramanian 2018-03-30  507   * clock frequencies supported by serial engine core clock. The protocol
eddac5af Karthikeyan Ramasubramanian 2018-03-30  508   * drivers use the output to determine the clock frequency index to be
eddac5af Karthikeyan Ramasubramanian 2018-03-30  509   * programmed into DFS.
eddac5af Karthikeyan Ramasubramanian 2018-03-30  510   *
eddac5af Karthikeyan Ramasubramanian 2018-03-30  511   * Return: number of valid performance levels in the table on success,
eddac5af Karthikeyan Ramasubramanian 2018-03-30  512   *	   standard Linux error codes on failure.
eddac5af Karthikeyan Ramasubramanian 2018-03-30  513   */
eddac5af Karthikeyan Ramasubramanian 2018-03-30  514  int geni_se_clk_tbl_get(struct geni_se *se, unsigned long **tbl)
eddac5af Karthikeyan Ramasubramanian 2018-03-30  515  {
9747c0a2 Douglas Anderson            2018-09-06  516  	long freq;
eddac5af Karthikeyan Ramasubramanian 2018-03-30  517  	int i;
eddac5af Karthikeyan Ramasubramanian 2018-03-30  518  
eddac5af Karthikeyan Ramasubramanian 2018-03-30  519  	if (se->clk_perf_tbl) {
eddac5af Karthikeyan Ramasubramanian 2018-03-30  520  		*tbl = se->clk_perf_tbl;
eddac5af Karthikeyan Ramasubramanian 2018-03-30  521  		return se->num_clk_levels;
eddac5af Karthikeyan Ramasubramanian 2018-03-30  522  	}
eddac5af Karthikeyan Ramasubramanian 2018-03-30  523  
eddac5af Karthikeyan Ramasubramanian 2018-03-30  524  	se->clk_perf_tbl = devm_kcalloc(se->dev, MAX_CLK_PERF_LEVEL,
eddac5af Karthikeyan Ramasubramanian 2018-03-30  525  					sizeof(*se->clk_perf_tbl),
eddac5af Karthikeyan Ramasubramanian 2018-03-30  526  					GFP_KERNEL);
eddac5af Karthikeyan Ramasubramanian 2018-03-30  527  	if (!se->clk_perf_tbl)
eddac5af Karthikeyan Ramasubramanian 2018-03-30  528  		return -ENOMEM;
eddac5af Karthikeyan Ramasubramanian 2018-03-30  529  
eddac5af Karthikeyan Ramasubramanian 2018-03-30  530  	for (i = 0; i < MAX_CLK_PERF_LEVEL; i++) {
eddac5af Karthikeyan Ramasubramanian 2018-03-30 @531  		freq = clk_round_rate(se->clk, freq + 1);
9747c0a2 Douglas Anderson            2018-09-06  532  		if (freq <= 0 || freq == se->clk_perf_tbl[i - 1])
eddac5af Karthikeyan Ramasubramanian 2018-03-30  533  			break;
eddac5af Karthikeyan Ramasubramanian 2018-03-30  534  		se->clk_perf_tbl[i] = freq;
eddac5af Karthikeyan Ramasubramanian 2018-03-30  535  	}
eddac5af Karthikeyan Ramasubramanian 2018-03-30  536  	se->num_clk_levels = i;
eddac5af Karthikeyan Ramasubramanian 2018-03-30  537  	*tbl = se->clk_perf_tbl;
eddac5af Karthikeyan Ramasubramanian 2018-03-30  538  	return se->num_clk_levels;
eddac5af Karthikeyan Ramasubramanian 2018-03-30  539  }
eddac5af Karthikeyan Ramasubramanian 2018-03-30  540  EXPORT_SYMBOL(geni_se_clk_tbl_get);

eddac5af Karthikeyan Ramasubramanian 2018-03-30 541

:::::: The code at line 531 was first introduced by commit
:::::: eddac5af06546d2e7a0730e3dc02dde3dc91098a soc: qcom: Add GENI based QUP Wrapper driver

:::::: TO: Karthikeyan Ramasubramanian <[email protected]> :::::: CC: Andy Gross <[email protected]>

---
0-DAY kernel test infrastructure Open Source Technology Center
https://lists.01.org/pipermail/kbuild-all Intel Corporation

#3 lore raw

Hi

Quoted 19 lines · kbuild test robot

Hi Douglas,

Thank you for the patch! Perhaps something to improve:

[auto build test WARNING on agross/for-next]
[also build test WARNING on v4.19-rc2 next-20180906]
[if your patch is applied to the wrong git tree, please drop us a note to help improve the system]

url:    https://github.com/0day-ci/linux/commits/Douglas-Anderson/soc-qcom-geni-Don-t-ignore-clk_round_rate-errors-in-geni_se_clk_tbl_get/20180907-045155
base:   https://git.kernel.org/pub/scm/linux/kernel/git/agross/linux.git for-next
config: i386-allmodconfig (attached as .config)
compiler: gcc-7 (Debian 7.3.0-1) 7.3.0
reproduce:
        # save the attached .config to linux build tree
        make ARCH=i386

Note: it may well be a FALSE warning. FWIW you are at least aware of it now.
http://gcc.gnu.org/wiki/Better_Uninitialized_Warnings

Quoted 6 lines

All warnings (new ones prefixed by >>):

drivers/soc/qcom/qcom-geni-se.c: In function 'geni_se_clk_tbl_get':

Quoted 1 line
Quoted 1 line

drivers/soc/qcom/qcom-geni-se.c:531:39: warning: 'freq' may be used uninitialized in this function [-Wmaybe-uninitialized]

      freq = clk_round_rate(se->clk, freq + 1);
                                     ~~~~~^~~

Oh crud. Yeah. ...so v1 was right after all. :(

-Doug

#4 lore raw
Quoted 22 lines · Doug Anderson

Hi

Quoted 19 lines · kbuild test robot

Hi Douglas,

Thank you for the patch! Perhaps something to improve:

[auto build test WARNING on agross/for-next]
[also build test WARNING on v4.19-rc2 next-20180906]
[if your patch is applied to the wrong git tree, please drop us a note to help improve the system]

url:    https://github.com/0day-ci/linux/commits/Douglas-Anderson/soc-qcom-geni-Don-t-ignore-clk_round_rate-errors-in-geni_se_clk_tbl_get/20180907-045155
base:   https://git.kernel.org/pub/scm/linux/kernel/git/agross/linux.git for-next
config: i386-allmodconfig (attached as .config)
compiler: gcc-7 (Debian 7.3.0-1) 7.3.0
reproduce:
        # save the attached .config to linux build tree
        make ARCH=i386

Note: it may well be a FALSE warning. FWIW you are at least aware of it now.
http://gcc.gnu.org/wiki/Better_Uninitialized_Warnings

Quoted 8 lines
Quoted 6 lines

All warnings (new ones prefixed by >>):

drivers/soc/qcom/qcom-geni-se.c: In function 'geni_se_clk_tbl_get':

Quoted 1 line
Quoted 1 line

drivers/soc/qcom/qcom-geni-se.c:531:39: warning: 'freq' may be used uninitialized in this function [-Wmaybe-uninitialized]

      freq = clk_round_rate(se->clk, freq + 1);
                                     ~~~~~^~~

Oh crud. Yeah. ...so v1 was right after all. :(

Sorry about the misguided suggestion :/