Re: [PATCH v3 09/10] mfd: bd96801: Add ERRB IRQ

From: Matti Vaittinen
Date: Fri Jun 14 2024 - 02:56:26 EST


On 6/13/24 19:32, Lee Jones wrote:
On Tue, 04 Jun 2024, Matti Vaittinen wrote:

The ROHM BD96801 "scalable PMIC" provides two physical IRQs. The ERRB
handling can in many cases be omitted because it is used to inform fatal
IRQs, which usually kill the power from the SOC.

There may however be use-cases where the SOC has a 'back-up' emergency
power source which allows some very short time of operation to try to
gracefully shut down sensitive hardware. Furthermore, it is possible the
processor controlling the PMIC is not powered by the PMIC. In such cases
handling the ERRB IRQs may be beneficial.

Add support for ERRB IRQs.

Signed-off-by: Matti Vaittinen <mazziesaccount@xxxxxxxxx>
---
Revision history:
v2 =>:
- No changes
v1 => v2:
- New patch
---
drivers/mfd/rohm-bd96801.c | 291 ++++++++++++++++++++++++++++++++-----
1 file changed, 253 insertions(+), 38 deletions(-)

diff --git a/drivers/mfd/rohm-bd96801.c b/drivers/mfd/rohm-bd96801.c
index 1c2a9591be7b..b7f073318873 100644
--- a/drivers/mfd/rohm-bd96801.c
+++ b/drivers/mfd/rohm-bd96801.c
@@ -5,13 +5,9 @@
* ROHM BD96801 PMIC driver
*
* This version of the "BD86801 scalable PMIC"'s driver supports only very
- * basic set of the PMIC features. Most notably, there is no support for
- * the ERRB interrupt and the configurations which should be done when the
- * PMIC is in STBY mode.
- *
- * Supporting the ERRB interrupt would require dropping the regmap-IRQ
- * usage or working around (or accepting a presense of) a naming conflict
- * in debugFS IRQs.

Why bother adding all that blurb in the first place?

Because, I assume there are users who would like to have the ERRB in use. The main purpose of this comment is that any such users could
a) see this version does not support ERRB.
b) can find the original RFC with ERRB supportn and a workaround.
c) know why this version does not work with ERRB and thus fix this

It seems this ERRB support may be missing from upstream for a while, hence I think having this note is worthy until (if) this ERRB patch lands in upstream.


+ * basic set of the PMIC features.
+ * Most notably, there is no support for the configurations which should
+ * be done when the PMIC is in STBY mode.
*
* Being able to reliably do the configurations like changing the
* regulator safety limits (like limits for the over/under -voltages, over
@@ -23,16 +19,14 @@
* be the need to configure these safety limits. Hence it's not simple to
* come up with a generic solution.
*
- * Users who require the ERRB handling and STBY state configurations can
- * have a look at the original RFC:
+ * Users who require the STBY state configurations can have a look at the
+ * original RFC:
* https://lore.kernel.org/all/cover.1712920132.git.mazziesaccount@xxxxxxxxx/
- * which implements a workaround to debugFS naming conflict and some of
- * the safety limit configurations - but leaves the state change handling
- * and synchronization to be implemented.
+ * which implements some of the safety limit configurations - but leaves the
+ * state change handling and synchronization to be implemented.
*
* It would be great to hear (and receive a patch!) if you implement the
- * STBY configuration support or a proper fix to the debugFS naming
- * conflict in your downstream driver ;)
+ * STBY configuration support or a proper fix in your downstream driver ;)
*/

...

static int bd96801_i2c_probe(struct i2c_client *i2c)
{
- struct regmap_irq_chip_data *intb_irq_data;
+ int i, ret, intb_irq, errb_irq, num_regu_irqs, num_intb, num_errb = 0;
+ int wdg_irq_no;
+ struct regmap_irq_chip_data *intb_irq_data, *errb_irq_data;
+ struct irq_domain *intb_domain, *errb_domain;
+ struct resource wdg_irq;
const struct fwnode_handle *fwnode;
- struct irq_domain *intb_domain;
+ struct resource *regulator_res;
struct regmap *regmap;
- int ret, intb_irq;
fwnode = dev_fwnode(&i2c->dev);
if (!fwnode)
@@ -212,10 +364,28 @@ static int bd96801_i2c_probe(struct i2c_client *i2c)
if (intb_irq < 0)
return dev_err_probe(&i2c->dev, intb_irq, "INTB IRQ not configured\n");
+ num_intb = ARRAY_SIZE(regulator_intb_irqs);
+
+ /* ERRB may be omitted if processor is powered by the PMIC */
+ errb_irq = fwnode_irq_get_byname(fwnode, "errb");
+ if (errb_irq < 0)
+ errb_irq = 0;
+
+ if (errb_irq)
+ num_errb = ARRAY_SIZE(regulator_errb_irqs);
+
+ num_regu_irqs = num_intb + num_errb;
+
+ regulator_res = kcalloc(num_regu_irqs, sizeof(*regulator_res), GFP_KERNEL);

Why not devm_* and omit the kfree()?

I used kcalloc() because this memory is only temporarily needed. It is not needed after devm_mfd_add_devices() returns.

Sure the devm_* would simplify the error paths... Thanks!


+ if (!regulator_res)
+ return -ENOMEM;
+
regmap = devm_regmap_init_i2c(i2c, &bd96801_regmap_config);
- if (IS_ERR(regmap))
- return dev_err_probe(&i2c->dev, PTR_ERR(regmap),
+ if (IS_ERR(regmap)) {
+ ret = dev_err_probe(&i2c->dev, PTR_ERR(regmap),
"Regmap initialization failed\n");
+ goto free_out;
+ }
ret = regmap_write(regmap, BD96801_LOCK_REG, BD96801_UNLOCK);
if (ret)
@@ -224,18 +394,63 @@ static int bd96801_i2c_probe(struct i2c_client *i2c)
ret = devm_regmap_add_irq_chip(&i2c->dev, regmap, intb_irq,
IRQF_ONESHOT, 0, &bd96801_irq_chip_intb,
&intb_irq_data);
- if (ret)
- return dev_err_probe(&i2c->dev, ret, "Failed to add INTB IRQ chip\n");
+ if (ret) {
+ dev_err_probe(&i2c->dev, ret, "Failed to add INTB irq_chip\n");
+ goto free_out;
+ }
intb_domain = regmap_irq_get_domain(intb_irq_data);
- ret = devm_mfd_add_devices(&i2c->dev, PLATFORM_DEVID_AUTO,
- bd96801_mfd_cells,
- ARRAY_SIZE(bd96801_mfd_cells), NULL, 0,
- intb_domain);
-
+ /*
+ * MFD core code is built to handle only one IRQ domain. BD96801
+ * has two domains so we do IRQ mapping here and provide the
+ * already mapped IRQ numbers to sub-devices.
+ */
+ for (i = 0; i < num_intb; i++) {
+ struct resource *res = &regulator_res[i];
+
+ *res = regulator_intb_irqs[i];
+ res->start = res->end = irq_create_mapping(intb_domain,
+ res->start);
+ }
+
+ wdg_irq_no = irq_create_mapping(intb_domain, BD96801_WDT_ERR_STAT);
+ wdg_irq = DEFINE_RES_IRQ_NAMED(wdg_irq_no, "bd96801-wdg");
+ bd96801_mfd_cells[WDG_CELL].resources = &wdg_irq;
+ bd96801_mfd_cells[WDG_CELL].num_resources = 1;
+
+ if (num_errb) {

if (!num_errb)
goto skip_errb;

Ok, can do.


+ ret = devm_regmap_add_irq_chip(&i2c->dev, regmap, errb_irq,
+ IRQF_ONESHOT, 0,
+ &bd96801_irq_chip_errb,
+ &errb_irq_data);
+ if (ret) {
+ dev_err_probe(&i2c->dev, ret,
+ "Failed to add ERRB (%d) irq_chip\n",
+ errb_irq);
+ goto free_out;
+ }
+ errb_domain = regmap_irq_get_domain(errb_irq_data);
+
+ for (i = 0; i < num_errb; i++) {
+ struct resource *res = &regulator_res[num_intb + i];
+
+ *res = regulator_errb_irqs[i];
+ res->start = res->end = irq_create_mapping(errb_domain,
+ res->start);
+ }
+ }

skip_errb:

...

Thanks for comments Lee. Reworking this will have to wait for the irqdomain name suffix, which I will continue after Hervé has done his part of the irqdomain changes. I will omit this patch from the next re-spin of the series.

Yours,
-- Matti

--
Matti Vaittinen
Linux kernel developer at ROHM Semiconductors
Oulu Finland

~~ When things go utterly wrong vim users can always type :help! ~~