| Message ID | 20230124092744.27370-1-tiwai@suse.de (mailing list archive) |
|---|---|
| Headers | show |
| Series | ALSA: memalloc: Fix for Xen PV and non-IOMMU systems | expand |
On Tue, 24 Jan 2023 10:27:42 +0100, Takashi Iwai wrote: > > Hi, > > this is a patch series to address the recent regression on Xen PV (and > possibly non-IOMMU) systems about the SG-buffer memory allocation. > We switched to use dma_alloc_noncontiguous() as hoped it handling > everything right, but it turned out that this doesn't work always. > So this is one step back, use the explicit SG-buffer with > dma_alloc_coherent() calls, but in a bit more optimized ways, and also > applying only for those systems. It seems that the second patch causes a problem; at least I see casual Oopses on my system after using the patch. Let's scratch. I'll resubmit the fix. Marek, could you try that later and report back if it still works and doesn't break things again? thanks, Takashi
On Wed, Jan 25, 2023 at 04:29:19PM +0100, Takashi Iwai wrote: > On Tue, 24 Jan 2023 10:27:42 +0100, > Takashi Iwai wrote: > > > > Hi, > > > > this is a patch series to address the recent regression on Xen PV (and > > possibly non-IOMMU) systems about the SG-buffer memory allocation. > > We switched to use dma_alloc_noncontiguous() as hoped it handling > > everything right, but it turned out that this doesn't work always. > > So this is one step back, use the explicit SG-buffer with > > dma_alloc_coherent() calls, but in a bit more optimized ways, and also > > applying only for those systems. > > It seems that the second patch causes a problem; at least I see casual > Oopses on my system after using the patch. Let's scratch. > > I'll resubmit the fix. Marek, could you try that later and report > back if it still works and doesn't break things again? Sure, just cc me.
Hi, this is a patch series to address the recent regression on Xen PV (and possibly non-IOMMU) systems about the SG-buffer memory allocation. We switched to use dma_alloc_noncontiguous() as hoped it handling everything right, but it turned out that this doesn't work always. So this is one step back, use the explicit SG-buffer with dma_alloc_coherent() calls, but in a bit more optimized ways, and also applying only for those systems. Takashi === Takashi Iwai (2): ALSA: memalloc: Explicit SG-allocations for Xen PV and non-IOMMU systems ALSA: memalloc: Use coherent DMA allocation for fallback again sound/core/memalloc.c | 68 +++++++++++++++++++++++++++++++++---------- 1 file changed, 53 insertions(+), 15 deletions(-)