Hi Heiko,
Sorry for a late reply.
Hello Lukasz,
Am 02.10.2013 17:11, schrieb Lukasz Majewski:
Hi Leela,
The current pmic i2c code assumes the current i2c bus is the same as the pmic device's bus. There is nothing ensuring that to be true. Therefore, select the proper bus before performing a transaction.
Signed-off-by: Aaron Durbinadurbin@chromium.org Signed-off-by: Simon Glasssjg@chromium.org Signed-off-by: Leela Krishna Amudalal.krishna@samsung.com Reviewed-by: Doug Andersondianders@google.com
drivers/power/power_i2c.c | 62 +++++++++++++++++++++++++++++++++++++++++---- 1 file changed, 57 insertions(+), 5 deletions(-)
diff --git a/drivers/power/power_i2c.c b/drivers/power/power_i2c.c index 47c606f..c22e01f 100644 --- a/drivers/power/power_i2c.c +++ b/drivers/power/power_i2c.c @@ -16,9 +16,45 @@ #include<i2c.h> #include<compiler.h>
+static int pmic_select(struct pmic *p) +{
- int ret, old_bus;
- old_bus = i2c_get_bus_num();
- if (old_bus != p->bus) {
debug("%s: Select bus %d\n", __func__, p->bus);ret = i2c_set_bus_num(p->bus);if (ret) {debug("%s: Cannot select pmic %s, err%d\n",
__func__, p->name, ret);return -1;}- }
- return old_bus;
+}
+static int pmic_deselect(int old_bus) +{
- int ret;
- if (old_bus != i2c_get_bus_num()) {
ret = i2c_set_bus_num(old_bus);debug("%s: Select bus %d\n", __func__, old_bus);if (ret) {debug("%s: Cannot restore i2c bus, err%d\n",
__func__, ret);return -1;}- }
- return 0;
+}
int pmic_reg_write(struct pmic *p, u32 reg, u32 val) { unsigned char buf[4] = { 0 };
int ret, old_bus;
if (check_reg(p, reg)) return -1;
@@ -52,23 +88,33 @@ int pmic_reg_write(struct pmic *p, u32 reg, u32 val) return -1; }
- if (i2c_write(pmic_i2c_addr, reg, 1, buf,
pmic_i2c_tx_num))
- old_bus = pmic_select(p);
- if (old_bus< 0) return -1;
- return 0;
- ret = i2c_write(pmic_i2c_addr, reg, 1, buf,
pmic_i2c_tx_num);
I'm wondering if setting the bus before each, single i2c write (when we usually perform several writes to one device) will not be an overkill (we search the ll_entry_count linker list each time to find max i2c adapter names) ?
Yes, maybe we could optimze this in drivers/i2c/i2c_core.c. It should be enough to detect the max adapter once ... but it is not a "search"... ll_entry_count() calculates the number ...
Yes, you are right. I've overlooked it.
With -Os compiler flag this compiles to a few ASM instructions. Obviously it is NOT a performance killer :-) (I made unnecessary fuzzz... sorry).
Looking in i2c_set_bus_num(), I think it can be optimized ... lets speaking code:
diff --git a/drivers/i2c/i2c_core.c b/drivers/i2c/i2c_core.c index d1072e8..170423a 100644 --- a/drivers/i2c/i2c_core.c +++ b/drivers/i2c/i2c_core.c @@ -278,20 +278,22 @@ unsigned int i2c_get_bus_num(void) */ int i2c_set_bus_num(unsigned int bus) {
int max = ll_entry_count(struct i2c_adapter, i2c);
int max;if ((bus == I2C_BUS) && (I2C_ADAP->init_done > 0))return 0;
This looks nice.
if (I2C_ADAPTER(bus) >= max) {printf("Error, wrong i2c adapter %d max %dpossible\n",
I2C_ADAPTER(bus), max);return -2;}#ifndef CONFIG_SYS_I2C_DIRECT_BUS if (bus >= CONFIG_SYS_NUM_I2C_BUSES) return -1; #endif
if ((bus == I2C_BUS) && (I2C_ADAP->init_done > 0))return 0;
max = ll_entry_count(struct i2c_adapter, i2c);if (I2C_ADAPTER(bus) >= max) {printf("Error, wrong i2c adapter %d max %dpossible\n",
I2C_ADAPTER(bus), max);return -2;}#ifndef CONFIG_SYS_I2C_DIRECT_BUS i2c_mux_disconnet_all();
So, first check, if we are on the correct bus, and return immediately! What do you think?
I think that it is acceptable.
Beside of that, pmic_select() does the check, if we are on the correct bus too, and calls i2c_set_bus_num() only, if not ... so this is here no problem ...
Yes, I see.
but exactly I want to get rid of this code as it is in pmic_select() someday, when all i2c drivers converted to the new i2c framework.
My 2 cents. I understand that pmic_select() preserves old i2c bus number, when PMIC performs transmission. This is probably done to not break the legacy code (where one driver assumed, that it is alone).
If this is necessary, then I'm OK with this. However I personally think, that drivers shall call API functions from i2c core (like i2c_bus_num()) only with bus number to switch and do not store and preserve the i2c value. This is my personal comment.
i2c_set_bus_num() should go static then in drivers/i2c/i2c_core.c and i2c_read/write/... become a new "int bus" parameter ... but this will be a big api change ... but will prevent exactly such code all over the u-boot code ...
bye, Heiko