| Message ID | 1433328587-20797-1-git-send-email-jarkko.nikula@linux.intel.com (mailing list archive) |
|---|---|
| State | New, archived |
| Headers | show |
On Wed, Jun 03, 2015 at 01:49:45PM +0300, Jarkko Nikula wrote: > static bool is_lpss_ssp(const struct driver_data *drv_data) > { > - return drv_data->ssp_type == LPSS_SSP; > + return drv_data->ssp_type == LPSS_LPT_SSP || > + drv_data->ssp_type == LPSS_BYT_SSP; > } switch statement please, it helps scale new types. > - switch (drv_data->ssp_type) { > - case QUARK_X1000_SSP: > + if (is_quark_x1000_ssp(drv_data)) { > tx_thres = TX_THRESH_QUARK_X1000_DFLT; > tx_hi_thres = 0; > rx_thres = RX_THRESH_QUARK_X1000_DFLT; > - break; > - case LPSS_SSP: > + } else if (is_lpss_ssp(drv_data)) { Why are we moving away from a switch statement here? Half the point of using them for variant selection is that it makes it easier to adjust the set of cases later. > + const struct acpi_device_id *id; > + int devid, type = SSP_UNDEFINED; Don't mix initialisations with other variable declarations please. > if (!ACPI_HANDLE(&pdev->dev) || > acpi_bus_get_device(ACPI_HANDLE(&pdev->dev), &adev)) > return NULL; > > + id = acpi_match_device(pdev->dev.driver->acpi_match_table, &pdev->dev); > + if (id) > + type = (int)id->driver_data; It'd be a bit clearer if the error case was an else here, though - on first read I'd expected to return an error if we couldn't identify the device and the initialisation is far enough away to appear in a different hunk.
On 06/03/2015 03:11 PM, Mark Brown wrote: > On Wed, Jun 03, 2015 at 01:49:45PM +0300, Jarkko Nikula wrote: > >> static bool is_lpss_ssp(const struct driver_data *drv_data) >> { >> - return drv_data->ssp_type == LPSS_SSP; >> + return drv_data->ssp_type == LPSS_LPT_SSP || >> + drv_data->ssp_type == LPSS_BYT_SSP; >> } > > switch statement please, it helps scale new types. > >> - switch (drv_data->ssp_type) { >> - case QUARK_X1000_SSP: >> + if (is_quark_x1000_ssp(drv_data)) { >> tx_thres = TX_THRESH_QUARK_X1000_DFLT; >> tx_hi_thres = 0; >> rx_thres = RX_THRESH_QUARK_X1000_DFLT; >> - break; >> - case LPSS_SSP: >> + } else if (is_lpss_ssp(drv_data)) { > > Why are we moving away from a switch statement here? Half the point of > using them for variant selection is that it makes it easier to adjust > the set of cases later. > It looked clever to reuse is_lpss_ssp() here. Onto another hand I don't see we'll expand a lot LPSS types so I'll keep the swich-case. >> + const struct acpi_device_id *id; >> + int devid, type = SSP_UNDEFINED; > > Don't mix initialisations with other variable declarations please. > >> if (!ACPI_HANDLE(&pdev->dev) || >> acpi_bus_get_device(ACPI_HANDLE(&pdev->dev), &adev)) >> return NULL; >> >> + id = acpi_match_device(pdev->dev.driver->acpi_match_table, &pdev->dev); >> + if (id) >> + type = (int)id->driver_data; > > It'd be a bit clearer if the error case was an else here, though - on > first read I'd expected to return an error if we couldn't identify the > device and the initialisation is far enough away to appear in a > different hunk. > Above appears to deserve a comment at least. Code gets here when device is either probed using those ACPI IDs in pxa2xx_spi_acpi_match[] or if it's registered as an platform device without platform data but with ACPI companion being set. Idea here is to set type from pxa2xx_spi_acpi_match[] only for current ACPI probed platforms and let the acpi_match_device() return NULL for the latter platform device case. Upcoming platforms will integrate host controller (SPI, I2C and UART) and a dedicated integrated DMA engine into same device with single PCI or ACPI ID. Our plan is to register them as a MFD child devices: http://marc.info/?l=linux-kernel&m=143317013403300&w=2 ACPI companion gets set through the MFD device registration. For spi-pxa2xx.c it means probe enters into pxa2xx_spi_acpi_get_pdata() because we don't set any platform data and the MMIO, clock and IRQ are still set there. My idea to detect these newer platforms and set the type accordingly is to check does the device have named resource "lpss_priv" in a code snip below @@ -1299,6 +1319,13 @@ pxa2xx_spi_acpi_get_pdata(struct platform_device *pdev) if (IS_ERR(ssp->mmio_base)) return NULL; + if (platform_get_resource_byname(pdev, IORESOURCE_MEM, "lpss_priv")) { + type = LPSS_SPT_SSP; + pdata->tx_param = pdev->dev.parent; + pdata->rx_param = pdev->dev.parent; + pdata->dma_filter = pxa2xx_spi_idma_filter; + } + ssp->clk = devm_clk_get(&pdev->dev, NULL); ssp->irq = platform_get_irq(pdev, 0); ssp->type = type; That made me thinking can I set the type to LPSS_SPT_SSP and needed DMA filter directly when acpi_match_device() returns NULL. At quick look it seems no any other platform device case than our new MFD registration enters here but have to look at more carefully.
On Wed, Jun 03, 2015 at 05:07:48PM +0300, Jarkko Nikula wrote: > On 06/03/2015 03:11 PM, Mark Brown wrote: > My idea to detect these newer platforms and set the type accordingly is to > check does the device have named resource "lpss_priv" in a code snip below > @@ -1299,6 +1319,13 @@ pxa2xx_spi_acpi_get_pdata(struct platform_device > *pdev) > if (IS_ERR(ssp->mmio_base)) > return NULL; > > + if (platform_get_resource_byname(pdev, IORESOURCE_MEM, "lpss_priv")) { > + type = LPSS_SPT_SSP; > + pdata->tx_param = pdev->dev.parent; > + pdata->rx_param = pdev->dev.parent; > + pdata->dma_filter = pxa2xx_spi_idma_filter; > + } > + Why not just pass some platform data? The above seems like it's abusing the API a bit and might upset generic code. > ssp->clk = devm_clk_get(&pdev->dev, NULL); > ssp->irq = platform_get_irq(pdev, 0); > ssp->type = type; > That made me thinking can I set the type to LPSS_SPT_SSP and needed DMA > filter directly when acpi_match_device() returns NULL. At quick look it > seems no any other platform device case than our new MFD registration enters > here but have to look at more carefully. That might work too, checking for DT as well would help defensiveness though you'd have to watch out for board file registraitons too since I'm not sure all the PXA platforms were DTified. Or just register the device with a different ID and use that?
On 06/03/2015 05:39 PM, Mark Brown wrote: > On Wed, Jun 03, 2015 at 05:07:48PM +0300, Jarkko Nikula wrote: >> On 06/03/2015 03:11 PM, Mark Brown wrote: > >> My idea to detect these newer platforms and set the type accordingly is to >> check does the device have named resource "lpss_priv" in a code snip below > >> @@ -1299,6 +1319,13 @@ pxa2xx_spi_acpi_get_pdata(struct platform_device >> *pdev) >> if (IS_ERR(ssp->mmio_base)) >> return NULL; >> >> + if (platform_get_resource_byname(pdev, IORESOURCE_MEM, "lpss_priv")) { >> + type = LPSS_SPT_SSP; >> + pdata->tx_param = pdev->dev.parent; >> + pdata->rx_param = pdev->dev.parent; >> + pdata->dma_filter = pxa2xx_spi_idma_filter; >> + } >> + > > Why not just pass some platform data? The above seems like it's abusing > the API a bit and might upset generic code. > I wanted to avoid that in order to keep MFD patch set and SPI changes independent from each other from both build and runtime. It would also need some code arrangements between pxa2xx_spi_probe() and pxa2xx_spi_acpi_get_pdata() but that's a minor reason. Of course relying on "lpss_priv" is not fully future proof either but it looks platform after Sunrisepoint won't need any extra type but just looking at differences from device registers. >> ssp->clk = devm_clk_get(&pdev->dev, NULL); >> ssp->irq = platform_get_irq(pdev, 0); >> ssp->type = type; > >> That made me thinking can I set the type to LPSS_SPT_SSP and needed DMA >> filter directly when acpi_match_device() returns NULL. At quick look it >> seems no any other platform device case than our new MFD registration enters >> here but have to look at more carefully. > > That might work too, checking for DT as well would help defensiveness > though you'd have to watch out for board file registraitons too since > I'm not sure all the PXA platforms were DTified. Or just register the > device with a different ID and use that? > Looks like all users are arch/arm/mach-pxa/ board files (no ACPI handle), PCI enumarated that set platform data in pxa2xx_spi_pci_probe() and ACPI matched devices in pxa2xx_spi_acpi_match[] (no platform data). Messing with device ID doesn't look clear either as we need to match the clock name and it's more clear to keep the host controller and DMA instance IDs the same. I realized I really need to handle NULL from acpi_match_device() in this patch. Assuming our MFD set is in the tree before adding the SPI Sunrisepoint support. Then code gets through the ACPI_HANDLE() test in pxa2xx_spi_acpi_get_pdata() (no platform data but ACPI companion is set by the MFD core) and goes setting the enable_dma flag for instance. This is not fatal but pxa2xx_spi_dma_setup() picks up then the any first available DMA channel because DMA filter function is not in place. It obviously won't work since SPI is hard-wired to its integrated DMA so transfers will be halted. I'll send version 2 in a minute.
diff --git a/drivers/spi/spi-pxa2xx-pci.c b/drivers/spi/spi-pxa2xx-pci.c index fa7399e84bbb..3cfd4357489a 100644 --- a/drivers/spi/spi-pxa2xx-pci.c +++ b/drivers/spi/spi-pxa2xx-pci.c @@ -62,7 +62,7 @@ static struct pxa_spi_info spi_info_configs[] = { .max_clk_rate = 3686400, }, [PORT_BYT] = { - .type = LPSS_SSP, + .type = LPSS_BYT_SSP, .port_id = 0, .num_chipselect = 1, .max_clk_rate = 50000000, @@ -70,7 +70,7 @@ static struct pxa_spi_info spi_info_configs[] = { .rx_param = &byt_rx_param, }, [PORT_BSW0] = { - .type = LPSS_SSP, + .type = LPSS_BYT_SSP, .port_id = 0, .num_chipselect = 1, .max_clk_rate = 50000000, @@ -78,7 +78,7 @@ static struct pxa_spi_info spi_info_configs[] = { .rx_param = &bsw0_rx_param, }, [PORT_BSW1] = { - .type = LPSS_SSP, + .type = LPSS_BYT_SSP, .port_id = 1, .num_chipselect = 1, .max_clk_rate = 50000000, @@ -86,7 +86,7 @@ static struct pxa_spi_info spi_info_configs[] = { .rx_param = &bsw1_rx_param, }, [PORT_BSW2] = { - .type = LPSS_SSP, + .type = LPSS_BYT_SSP, .port_id = 2, .num_chipselect = 1, .max_clk_rate = 50000000, diff --git a/drivers/spi/spi-pxa2xx.c b/drivers/spi/spi-pxa2xx.c index e3223ac75a7c..13ec36b1c9aa 100644 --- a/drivers/spi/spi-pxa2xx.c +++ b/drivers/spi/spi-pxa2xx.c @@ -74,7 +74,8 @@ MODULE_ALIAS("platform:pxa2xx-spi"); static bool is_lpss_ssp(const struct driver_data *drv_data) { - return drv_data->ssp_type == LPSS_SSP; + return drv_data->ssp_type == LPSS_LPT_SSP || + drv_data->ssp_type == LPSS_BYT_SSP; } static bool is_quark_x1000_ssp(const struct driver_data *drv_data) @@ -1079,22 +1080,18 @@ static int setup(struct spi_device *spi) unsigned int clk_div; uint tx_thres, tx_hi_thres, rx_thres; - switch (drv_data->ssp_type) { - case QUARK_X1000_SSP: + if (is_quark_x1000_ssp(drv_data)) { tx_thres = TX_THRESH_QUARK_X1000_DFLT; tx_hi_thres = 0; rx_thres = RX_THRESH_QUARK_X1000_DFLT; - break; - case LPSS_SSP: + } else if (is_lpss_ssp(drv_data)) { tx_thres = LPSS_TX_LOTHRESH_DFLT; tx_hi_thres = LPSS_TX_HITHRESH_DFLT; rx_thres = LPSS_RX_THRESH_DFLT; - break; - default: + } else { tx_thres = TX_THRESH_DFLT; tx_hi_thres = 0; rx_thres = RX_THRESH_DFLT; - break; } /* Only alloc on first setup */ @@ -1242,6 +1239,18 @@ static void cleanup(struct spi_device *spi) } #ifdef CONFIG_ACPI + +static struct acpi_device_id pxa2xx_spi_acpi_match[] = { + { "INT33C0", LPSS_LPT_SSP }, + { "INT33C1", LPSS_LPT_SSP }, + { "INT3430", LPSS_LPT_SSP }, + { "INT3431", LPSS_LPT_SSP }, + { "80860F0E", LPSS_BYT_SSP }, + { "8086228E", LPSS_BYT_SSP }, + { }, +}; +MODULE_DEVICE_TABLE(acpi, pxa2xx_spi_acpi_match); + static struct pxa2xx_spi_master * pxa2xx_spi_acpi_get_pdata(struct platform_device *pdev) { @@ -1249,12 +1258,17 @@ pxa2xx_spi_acpi_get_pdata(struct platform_device *pdev) struct acpi_device *adev; struct ssp_device *ssp; struct resource *res; - int devid; + const struct acpi_device_id *id; + int devid, type = SSP_UNDEFINED; if (!ACPI_HANDLE(&pdev->dev) || acpi_bus_get_device(ACPI_HANDLE(&pdev->dev), &adev)) return NULL; + id = acpi_match_device(pdev->dev.driver->acpi_match_table, &pdev->dev); + if (id) + type = (int)id->driver_data; + pdata = devm_kzalloc(&pdev->dev, sizeof(*pdata), GFP_KERNEL); if (!pdata) return NULL; @@ -1272,7 +1286,7 @@ pxa2xx_spi_acpi_get_pdata(struct platform_device *pdev) ssp->clk = devm_clk_get(&pdev->dev, NULL); ssp->irq = platform_get_irq(pdev, 0); - ssp->type = LPSS_SSP; + ssp->type = type; ssp->pdev = pdev; ssp->port_id = -1; @@ -1285,16 +1299,6 @@ pxa2xx_spi_acpi_get_pdata(struct platform_device *pdev) return pdata; } -static struct acpi_device_id pxa2xx_spi_acpi_match[] = { - { "INT33C0", 0 }, - { "INT33C1", 0 }, - { "INT3430", 0 }, - { "INT3431", 0 }, - { "80860F0E", 0 }, - { "8086228E", 0 }, - { }, -}; -MODULE_DEVICE_TABLE(acpi, pxa2xx_spi_acpi_match); #else static inline struct pxa2xx_spi_master * pxa2xx_spi_acpi_get_pdata(struct platform_device *pdev) diff --git a/include/linux/pxa2xx_ssp.h b/include/linux/pxa2xx_ssp.h index dab545bb66b3..95a4b3bd7a5c 100644 --- a/include/linux/pxa2xx_ssp.h +++ b/include/linux/pxa2xx_ssp.h @@ -194,8 +194,9 @@ enum pxa_ssp_type { PXA168_SSP, PXA910_SSP, CE4100_SSP, - LPSS_SSP, QUARK_X1000_SSP, + LPSS_LPT_SSP, + LPSS_BYT_SSP, }; struct ssp_device {
Intel LPSS SPI properties differ between between platforms. Now private registers offset 0x400 or 0x800 is autodetected but there is need to support also other offset and handle a few other differences. Prepare for that by splitting the LPSS_SSP type into compatible hardware types and set it now based on PCI or ACPI ID. That type will be used to set properties that differ between current and upcoming platforms. Signed-off-by: Jarkko Nikula <jarkko.nikula@linux.intel.com> --- drivers/spi/spi-pxa2xx-pci.c | 8 ++++---- drivers/spi/spi-pxa2xx.c | 44 ++++++++++++++++++++++++-------------------- include/linux/pxa2xx_ssp.h | 3 ++- 3 files changed, 30 insertions(+), 25 deletions(-)