| Message ID | 1453401227-27135-1-git-send-email-srinivas.kandagatla@linaro.org (mailing list archive) |
|---|---|
| State | Not Applicable, archived |
| Delegated to: | Andy Gross |
| Headers | show |
On Thu, Jan 21, 2016 at 06:33:47PM +0000, Srinivas Kandagatla wrote: > This driver reuses pdev->id for spi bus numbers resulting in random > or very large bus numbering when used with device trees. pdev->id > is not the correct choice when using device trees. So add code to What makes you say this, why is pdev->id not "correct"? It is worrying if anything cares what number we pick. > get bus numbers via device tree aliases and if it fails then generate > a unique bus number. The other question is even if this is a good idea why is it something that should be open coded in individual drivers, if we want to change the policy we should be consistent between drivers.
On 21/01/16 18:38, Mark Brown wrote: > On Thu, Jan 21, 2016 at 06:33:47PM +0000, Srinivas Kandagatla wrote: > >> This driver reuses pdev->id for spi bus numbers resulting in random >> or very large bus numbering when used with device trees. pdev->id >> is not the correct choice when using device trees. So add code to > > What makes you say this, why is pdev->id not "correct"? It is worrying > if anything cares what number we pick. Issue is that using pdev->id for bus number, as pdev->id does not get populated in device tree cases. The end users who are reading the schematics would not be able to map the actual bus numbers on the schematics with the bus numbers allocated using pdev->id. It add more confusion. Without this patch the bus number allocated to this driver is 32766. This does not really reflect the actual bus numbers on the boards schematics. > >> get bus numbers via device tree aliases and if it fails then generate >> a unique bus number. > > The other question is even if this is a good idea why is it something > that should be open coded in individual drivers, if we want to change > the policy we should be consistent between drivers. Device tree aliases seems used very much in many drivers. The unique bus number scheme was actually inspired by the driver/tty/serial/msm_serial.c --srini > -- To unsubscribe from this list: send the line "unsubscribe linux-arm-msm" in the body of a message to majordomo@vger.kernel.org More majordomo info at http://vger.kernel.org/majordomo-info.html
On Thu, Jan 21, 2016 at 06:47:34PM +0000, Srinivas Kandagatla wrote: > On 21/01/16 18:38, Mark Brown wrote: > >On Thu, Jan 21, 2016 at 06:33:47PM +0000, Srinivas Kandagatla wrote: > >>This driver reuses pdev->id for spi bus numbers resulting in random > >>or very large bus numbering when used with device trees. pdev->id > >>is not the correct choice when using device trees. So add code to > >What makes you say this, why is pdev->id not "correct"? It is worrying > >if anything cares what number we pick. > Issue is that using pdev->id for bus number, as pdev->id does not get > populated in device tree cases. That's a statement of what currently happens... > The end users who are reading the schematics would not be able to map the > actual bus numbers on the schematics with the bus numbers allocated using > pdev->id. It add more confusion. > Without this patch the bus number allocated to this driver is 32766. > This does not really reflect the actual bus numbers on the boards > schematics. Is this really causing anyone any confusion? Normally people are looking at the devices on the SPI bus rather than the bus itself... In any case if this *is* causing confusion should we not be doing something at the bus core level that allows us to assign a descriptive name since this doesn't seem in the least bit SPI specific but could apply to any bus? There's also the problem that if someone has decided to label the bus with a descriptive name in their schematic (eg, "SPI_FLASH") then being able to assign a number doesn't do much to help, we'd need to be able to provide strings. A brief survey of schematics I have to hand suggests that this is a thing people do. > >>get bus numbers via device tree aliases and if it fails then generate > >>a unique bus number. > >The other question is even if this is a good idea why is it something > >that should be open coded in individual drivers, if we want to change > >the policy we should be consistent between drivers. > Device tree aliases seems used very much in many drivers. > The unique bus number scheme was actually inspired by the > driver/tty/serial/msm_serial.c That doesn't help explain why it is a good idea to open code this in individual drivers. I was asking why it's a good idea to do this in a single driver rather than at a higher level.
On 21/01/16 19:03, Mark Brown wrote: > On Thu, Jan 21, 2016 at 06:47:34PM +0000, Srinivas Kandagatla wrote: >> On 21/01/16 18:38, Mark Brown wrote: >>> On Thu, Jan 21, 2016 at 06:33:47PM +0000, Srinivas Kandagatla wrote: > >>>> This driver reuses pdev->id for spi bus numbers resulting in random >>>> or very large bus numbering when used with device trees. pdev->id >>>> is not the correct choice when using device trees. So add code to > >>> What makes you say this, why is pdev->id not "correct"? It is worrying >>> if anything cares what number we pick. > >> Issue is that using pdev->id for bus number, as pdev->id does not get >> populated in device tree cases. > > That's a statement of what currently happens... > >> The end users who are reading the schematics would not be able to map the >> actual bus numbers on the schematics with the bus numbers allocated using >> pdev->id. It add more confusion. > >> Without this patch the bus number allocated to this driver is 32766. >> This does not really reflect the actual bus numbers on the boards >> schematics. > > Is this really causing anyone any confusion? Normally people are > looking at the devices on the SPI bus rather than the bus itself... In > any case if this *is* causing confusion should we not be doing something > at the bus core level that allows us to assign a descriptive name since > this doesn't seem in the least bit SPI specific but could apply to any > bus? > > There's also the problem that if someone has decided to label the bus > with a descriptive name in their schematic (eg, "SPI_FLASH") then being > able to assign a number doesn't do much to help, we'd need to be able to > provide strings. A brief survey of schematics I have to hand suggests > that this is a thing people do. > >>>> get bus numbers via device tree aliases and if it fails then generate >>>> a unique bus number. > >>> The other question is even if this is a good idea why is it something >>> that should be open coded in individual drivers, if we want to change >>> the policy we should be consistent between drivers. > >> Device tree aliases seems used very much in many drivers. >> The unique bus number scheme was actually inspired by the >> driver/tty/serial/msm_serial.c > > That doesn't help explain why it is a good idea to open code this in > individual drivers. I was asking why it's a good idea to do this in a > single driver rather than at a higher level. Oops!!, I should have looked at spi.c which already does exactly same thing. I think the logic did not get triggered because (int)-1 overflowed into s16 busnum. --srini > -- To unsubscribe from this list: send the line "unsubscribe linux-arm-msm" in the body of a message to majordomo@vger.kernel.org More majordomo info at http://vger.kernel.org/majordomo-info.html
On Thu, Jan 21, 2016 at 07:26:19PM +0000, Srinivas Kandagatla wrote: > On 21/01/16 19:03, Mark Brown wrote: > >That doesn't help explain why it is a good idea to open code this in > >individual drivers. I was asking why it's a good idea to do this in a > >single driver rather than at a higher level. > Oops!!, I should have looked at spi.c which already does exactly same thing. > I think the logic did not get triggered because (int)-1 overflowed into s16 > busnum. Ah, that sounds like a bug... I'd actually forgotten that we did that, it should work though since the code is there and it (hopefully) worked at some point.
diff --git a/drivers/spi/spi-qup.c b/drivers/spi/spi-qup.c index 810a7fa..2bc67a9 100644 --- a/drivers/spi/spi-qup.c +++ b/drivers/spi/spi-qup.c @@ -127,6 +127,8 @@ #define SPI_DELAY_THRESHOLD 1 #define SPI_DELAY_RETRY 10 +static atomic_t spi_qup_next_id = ATOMIC_INIT(0); + struct spi_qup { void __iomem *base; struct device *dev; @@ -759,7 +761,7 @@ static int spi_qup_probe(struct platform_device *pdev) struct device *dev; void __iomem *base; u32 max_freq, iomode, num_cs; - int ret, irq, size; + int ret, irq, size, bus_num; dev = &pdev->dev; res = platform_get_resource(pdev, IORESOURCE_MEM, 0); @@ -816,7 +818,15 @@ static int spi_qup_probe(struct platform_device *pdev) else master->num_chipselect = num_cs; - master->bus_num = pdev->id; + if (dev->of_node) + bus_num = of_alias_get_id(dev->of_node, "spi"); + else + bus_num = pdev->id; + + if (bus_num < 0) + bus_num = atomic_inc_return(&spi_qup_next_id) - 1; + + master->bus_num = bus_num; master->mode_bits = SPI_CPOL | SPI_CPHA | SPI_CS_HIGH | SPI_LOOP; master->bits_per_word_mask = SPI_BPW_RANGE_MASK(4, 32); master->max_speed_hz = max_freq;
This driver reuses pdev->id for spi bus numbers resulting in random or very large bus numbering when used with device trees. pdev->id is not the correct choice when using device trees. So add code to get bus numbers via device tree aliases and if it fails then generate a unique bus number. Without this patch the driver get a random and useless bus number. Signed-off-by: Srinivas Kandagatla <srinivas.kandagatla@linaro.org> --- drivers/spi/spi-qup.c | 14 ++++++++++++-- 1 file changed, 12 insertions(+), 2 deletions(-)