On 21.05.2015 09:16, Krzysztof Kozlowski wrote: > On 21.05.2015 05:30, Dinh Nguyen wrote: >> Hi Krzysztof, >> >> On 05/05/2015 02:22 PM, Dinh Nguyen wrote: >>> On 05/05/2015 09:56 AM, Dinh Nguyen wrote: >>>> On 05/04/2015 10:55 PM, Krzysztof Kozlowski wrote: >>>>> 2015-05-05 4:52 GMT+09:00 Dinh Nguyen <dinguyen@xxxxxxxxxxxxxxxxxxxxx>: >>>>>> On 05/04/2015 09:06 AM, Dinh Nguyen wrote: >>>>>>> +CC Olof >>>>>>> >>>>>>> On 5/4/15 8:50 AM, Krzysztof Kozlowski wrote: >>>>>>>> 2015-05-04 22:28 GMT+09:00 Dinh Nguyen <dinguyen@xxxxxxxxxxxxxxxxxxxxx>: >>>>>>>>> Hi Krzystof, >>>>>>>>> >>>>>>>>> On 5/4/15 12:30 AM, Krzysztof Kozlowski wrote: >>>>>>>>>> 2015-05-04 13:28 GMT+09:00 <dinguyen@xxxxxxxxxxxxxxxxxxxxx>: >>>>>>>>>>> From: Dinh Nguyen <dinguyen@xxxxxxxxxxxxxxxxxxxxx> >>>>>>>>>>> >>>>>>>>>>> Turn on the clock to the PL330 DMA if there is a clock node provided. >>>>>>>>>> >>>>>>>>>> Why? There is no explanation in the patch for this important question - why? >>>>>>>>>> >>>>>>>>>> Amba bus already does this and provide a wrapper function. >>>>>>>>>> Additionally that would mess up with runtime PM and clock >>>>>>>>>> enable/disable. >>>>>>>>> >>>>>>>>> I don't see the clock for the DMA getting turned on at all, which is why >>>>>>>>> after the kernel has booted, the filesystem tries to open up a serial >>>>>>>>> port using DMA and the system hangs. The failure is seen here: >>>>>>>>> >>>>>>>>> http://arm-soc.lixom.net/bootlogs/next/next-20150504/socfpga-arm-multi_v7_defconfig.html >>>>>>>> >>>>>>>> Thanks! >>>>>>>> >>>>>>>> The amba bus and pl330 should enable the clock and then disable it >>>>>>>> after probing: >>>>>>>> static int amba_probe(struct device *dev) >>>>>>>> { >>>>>>>> ... >>>>>>>> ret = amba_get_enable_pclk(pcdev); >>>>>>>> ... >>>>>>>> >>>>>>>> I wonder why do you think it is not enabled at all? >>>>>>> >>>>>>> I've checked it down to the register level that the gate for this clock >>>>>>> does not get set. >>>>>>> >>>>>>>> >>>>>>>>> >>>>>>>>> This only happens with the multi_v7_defconfig, because the PL330 DMA is >>>>>>>>> getting built into the kernel, while the socfpga_defconfig does not >>>>>>>>> enable the PL330. >>>>>>>> >>>>>>>> It makes sense. If pl330 driver is not enabled then necessary clocks >>>>>>>> are turned on by bootloader. Probing pl330 effectively disables the >>>>>>>> clock (if DMA is not used). >>>>>>>> >>>>>>>>> The DTS for the socfpga platform looks like this: >>>>>>>>> >>>>>>>>> pdma: pdma@ffe01000 { >>>>>>>>> compatible = "arm,pl330", "arm,primecell"; >>>>>>>>> reg = <0xffe01000 0x1000>; >>>>>>>>> interrupts = <0 104 4>, >>>>>>>>> <0 105 4>, >>>>>>>>> ... >>>>>>>>> #dma-cells = <1>; >>>>>>>>> #dma-channels = <8>; >>>>>>>>> #dma-requests = <32>; >>>>>>>>> clocks = <&l4_main_clk>; >>>>>>>>> clock-names = "apb_pclk"; >>>>>>>>> }; >>>>>>>>> >>>>>>>>> Perhaps I have the wrong designation for clock-names and the amba bus is >>>>>>>>> not able to pick up the correct clock? >>>>>>>> >>>>>>>> I have two ideas: >>>>>>>> 1. Is this really the clock for the DMA? If DMA is not used then >>>>>>>> disabling it should be OK. >>>>>>> >>>>>>> Yes, this is the clock for the DMA. Yeah, leaving this clock off is >>>>>>> fine, until the DMA gets used. Up until v4.0, SoCFPGA was not using the >>>>>>> DMA at all, but in v4.0, there was a patch to assign the UARTs to it's >>>>>>> DMA channel. >>>>>>> >>>>>>> https://git.kernel.org/cgit/linux/kernel/git/torvalds/linux.git/commit/arch/arm/boot/dts/socfpga.dtsi?id=78c03c7af89721bd8a4428408a8cc7b53972e4b8 >>>>>>> >>>>>>>> 2. Disabling the clock may effectively disable its parent or >>>>>>>> grandparent if there are not more users. Maybe some other driver needs >>>>>>>> these parents to be enabled? This was the issue for at least one >>>>>>>> similar error (on Exynos boards). >>>>>>>> >>>>>>> >>>>>>> I'll check up on these issues. When I was debugging this issue, the >>>>>>> l4_main_clk is only used by the DMA, so it was not getting turned on by >>>>>>> an other drivers. >>>>>>> >>>>>> >>>>>> Ah, it looks like perhaps there's a problem with the serial driver and >>>>>> suspend/resume? If disable CONFIG_PM, then the DMA seems to be working >>>>>> fine with the debug uart. It appears the DMA is getting suspended and >>>>>> doesn't get resumed. >>>>>> >>>>> >>>>> You mean runtime PM suspend and resume or system sleep? During boot >>>>> only the first one should happen. >>>> >>>> It's runtime PM suspend/resume. >>>> >>>>> >>>>> Could you test the DMA with dmatest? Disable the DMA in UART and >>>>> compile with CONFIG_DMATEST. Syntax for testing is here: >>>>> Documentation/dmaengine/dmatest.txt >>>>> >>>> >>>> # echo Y > /sys/module/dmatest/parameters/run >>>> [ 93.143775] dmatest: Started 1 threads using dma0chan0 >>>> [ 93.149227] pm_generic_runtime_resume >>>> [ 93.153334] dmatest: Started 1 threads using dma0chan1 >>>> [ 93.159380] dmatest: Started 1 threads using dma0chan2 >>>> [ 93.165041] dmatest: Started 1 threads using dma0chan3 >>>> [ 93.170280] dmatest: Started 1 threads using dma0chan4 >>>> [ 93.175996] dmatest: Started 1 threads using dma0chan5 >>>> [ 93.181642] dmatest: Started 1 threads using dma0chan6 >>>> [ 93.188754] dmatest: dma0chan1-copy0: summary 10 tests, 0 failures >>>> 282 iops 2008 KB/s (0) >>>> [ 93.197091] dmatest: Started 1 threads using dma0chan7 >>>> [ 93.199353] dmatest: dma0chan3-copy0: summary 10 tests, 0 failures >>>> 297 iops 2260 KB/s (0) >>>> [ 93.205407] dmatest: dma0chan0-copy0: summary 10 tests, 0 failures >>>> 177 iops 1364 KB/s (0) >>>> [ 93.215599] dmatest: dma0chan2-copy0: summary 10 tests, 0 failures >>>> 196 iops 1450 KB/s (0) >>>> [ 93.219994] dmatest: dma0chan4-copy0: summary 10 tests, 0 failures >>>> 225 iops 1554 KB/s (0) >>>> [ 93.224322] dmatest: dma0chan5-copy0: summary 10 tests, 0 failures >>>> 231 iops 1948 KB/s (0) >>>> [ 93.230065] dmatest: dma0chan6-copy0: summary 10 tests, 0 failures >>>> 231 iops 1759 KB/s (0) >>>> [ 93.231251] dmatest: dma0chan7-copy0: summary 10 tests, 0 failures >>>> 298 iops 2331 KB/s (0) >>>> [ 93.243523] pm_generic_runtime_suspend >>>> root@socfpga_cyclone5:~# >>>> >>> >>> If I run dmatest the 2nd time it fails. It does not look like >>> amba_pm_runtime_resume() is getting called to turn on the clocks on the >>> subsequent tries. >>> >> >> >> I managed to track this down the call dmaengine_terminate_all(), which >> then calls into pl330_terminate_all(). So in pl330_terminate_all(), it >> call _stop, which hits a infinite loop, UNTIL. But since the >> amba_pm_runtime_resume() has not been called yet, the clock is turned >> off. Thus, we're stuck in an infinite loop. >> >> I'm not sure what would be right approach to fix this? > > Good catch. I confirmed that device is not runtime resumed. I wonder why > it works in my case (pl330 on Exynos4412)... > > Anyway I have an idea to fix it. I'll send a patch. I sent a patch. Could you test on your board and confirm that this fixes the issue? Best regards, Krzysztof -- To unsubscribe from this list: send the line "unsubscribe dmaengine" in the body of a message to majordomo@xxxxxxxxxxxxxxx More majordomo info at http://vger.kernel.org/majordomo-info.html
