Hyperion Entertainment Message Boards

xeno74 wrote:

Roland wrote:I noticed that Kernel 5.3 has created a problem which does not exist in 4.20 or earlier... If there is a Dawicontrol DC 2976 UW SCSI board (PCI) installed, the system does not boot unless the RAM is limited to 3500M. This happens now even when there is nothing connected to the controller!

Hi Roland,

Thanks a lot for reporting the issue with the Dawicontrol DC 2976 UW SCSI board (PCI).

Could you please test the RC2 of kernel 5.4 with this SCSI board? I would like to know if the problem still exists in the latest kernel.

Please post the results in the kernel 5.4 test thread.

Thanks in advance,

Christian

Roland wrote:The memory issue with Dawicontrol DC 2976 UW is present also in Kernel 5.4 rc2.

xeno74 wrote: Thanks for testing the RC2! OK, we know that this issue is still exists in the latest kernel. We have to figure out which code change is responsible for this issue. You told me that the stable kernel 4.20 doesn't have this issue. Could you please test the alpha1 of kernel 4.21/5.0?

Roland wrote: 4.21-alpha1 does not have that issue, I can use the full 8 Gb memory with it.

Roland wrote: I tested this with unitrd-4.19.3.

I tested also 5.1 (first stable version), and found out that if I use the uinitrd-4.19.3, I can use full RAM, but if unitrd-5.3, the issue is present.

4.20.17 seems to work ok both with unitrd-4.19.3 and unitrd-5.3.

xeno74 wrote:
Thanks for testing the alpha1. Please test the alpha2 of kernel 4.21/5.0. I would like to know when the code changed.

Roland wrote: I tested now the alphas... 4.21-alpha2 is last which works - the issue is present starting from alpha3.

diff -rupN a/arch/powerpc/sysdev/fsl_pci.c b/arch/powerpc/sysdev/fsl_pci.c
--- a/arch/powerpc/sysdev/fsl_pci.c	2018-12-28 08:20:11.651183458 +0100
+++ b/arch/powerpc/sysdev/fsl_pci.c	2018-12-28 08:17:54.971034055 +0100
@@ -127,9 +127,6 @@ static inline void setup_swiotlb_ops(str
 
 static int fsl_pci_dma_set_mask(struct device *dev, u64 dma_mask)
 {
-	if (!dev->dma_mask || !dma_supported(dev, dma_mask))
-		return -EIO;
-
 	/*
 	 * Fix up PCI devices that are able to DMA to the large inbound
 	 * mapping that allows addressing any RAM address from across PCI.

• Linux 5.4-rc3 Released Ahead Of Official Kernel Debut In November -- phoronix.com
• PowerPC fixes 5.4-3
• Torvalds' release announcement
• Linux Git log
• Phoronix articles, reviews and news stories covering Linux 5.4

xeno74 wrote: I tried to revert the changes in the file arch/powerpc/kernel/dma.c. The problem is, this file doesn’t exist in the latest kernels anymore. That means we can’t fix anything which doesn’t exist anymore. I think the issue was in these files but today it is somewhere in the other files.

Please create a bug report because the changes to the code are too long ago.

It’s very important to test the alphas and release candidates because the sooner we notice an error, the higher is the chance of a fast solution.