> That isn't a limit, just a reporting clause - the real check is this
> line a little above:
>
> if (unlikely(dev && !dma_capable(dev, dma_addr, size))) {
>
> which is purely based on the dma mask. So for some reason we must
> be in 32-bit only mode for the dma-mask, and not actually
Christoph,
here is all of the newly patched dmesg output. I also added
'aic7xxx.a9c7xxx=verbose' for extra information.
Matthew
[0.00] Linux version 4.18.12.pentium4-xeon-christoph+
(root@pentium4) (gcc version 5.4.0 (Gentoo 5.4.0-r4 p1.8, pie-0.6.5))
#525 SMP PREEMPT Sat Oct 13
> it seems like we do for some reason never actually enable swiotlb
> for 32-bit x86. Before my commit the block bounce buffering papered
> over that for networking, Please try this patch:
>
> diff --git a/arch/x86/kernel/pci-swiotlb.c b/arch/x86/kernel/pci-swiotlb.c
> index
Hannes,
I'm getting the following error in a custom configured 4.18 32-bit
x86 kernel supporting PAE, with 16GiB physical memory. It loops
infinitely on the error.
aic7xxx :00:03.0: dma_direct_map_sg: overflow
0x0003ff80+65536 of device mask
I have tried enabling the
Christoph,
I was able to bisect this to your patch "scsi: reduce use of block
bounce buffers". I am getting the error on a 32-bit Dell PowerEdge
6650. It has the aic7xxx integrated onto the motherboard.
Again, here is the error:
aic7xxx :00:03.0: dma_direct_map_sg: overflow
5 matches
Mail list logo