Hi Ziji, On Sat, 30 Sep 2017 10:47:39 +0800 Ziji Hu wrote: > Hi Gregory, > > 2017-09-28 23:05 GMT+08:00 Gregory CLEMENT <gregory.clement@xxxxxxxxxxxxxxxxxx>: > > On Armada 7K/8K we need to explicitly enable the bus clock. The bus clock > > is optional because not all the SoCs need them but at least for Armada > > 7K/8K it is actually mandatory. > > > > The binding documentation is updating accordingly. > > > > Without this patch the kernel hand during boot if the mvpp2.2 network > > driver was not present in the kernel. Indeed the clock needed by the > > xenon controller was set by the network driver. > > > > Fixes: 3a3748dba881 ("mmc: sdhci-xenon: Add Marvell Xenon SDHC core > > functionality)" > > CC: Stable <stable@xxxxxxxxxxxxxxx> > > Signed-off-by: Gregory CLEMENT <gregory.clement@xxxxxxxxxxxxxxxxxx> > > --- > > Hi Ulf and Adrian, > > > > This patch should be merged in 4.14-rc, as it fixes real issues. > > > > This patch maye looks like just as a nice clean-up but it is not. On > > the earlier version of the series adding the support for the xenon > > controller there was already this axi bus clock. But at this moment > > the documentation for the clock on the Armada 7K/8K was missing and > > there was no driver for it, so we just use the clock set by the > > bootloader and never change them that's why it worked without any > > visible problem. > > > > Then as explained in the commit log the network driver enabled the > > clock for us, and it happened that it was setup before the xenon > > driver. It is only thanks the new information we received on the > > clocks of the Sov and with more exhaustive testes that we found this > > issue and the fix for it. > > I know very little about clk in Linux. > But I just happened to review this very issue before I left Marvell. > I had some concerns about this patch at that time. > Please check the following. > > 1. If that network device is used again, so that clock will be setup *twice* > since there are two duplicated clock setup in that network driver > and Xenon driver. Right? The first "setup" will be executed, the second one just increase the enable and prepare count. > 2. If that network device is used again and Xenon later exists and > shutdown the clock, > will that network device be corrupted? No. The enable and prepare count is reduced to 1, nothing else could happen. The clock won't be shutdown until there's no users. > 3. In my very own opinion, such a shared AXI bus clock might be > controlled in u-boot or > a "SoC-level" init routine. IMHO, it's the driver's responsibility to manage its own clocks. Thanks -- To unsubscribe from this list: send the line "unsubscribe linux-mmc" in the body of a message to majordomo@xxxxxxxxxxxxxxx More majordomo info at http://vger.kernel.org/majordomo-info.html