Francisco Jerez
2009-Oct-22 00:36 UTC
[Nouveau] [PATCH] nv04-nv40/exa: Reorder the commands in PrepareCopy to match the blob.
This fixes a somewhat indeterministic corruption problem on nv17 when there is stuff going on the other fifos (e.g. gallium but I've also reproduced it with an app just SIFM-ing memory around): in some cases it made the blits the X server had scheduled use the wrong pitch. Signed-off-by: Francisco Jerez <currojerez at riseup.net> --- src/nv04_exa.c | 22 ++++++++++++---------- 1 files changed, 12 insertions(+), 10 deletions(-) diff --git a/src/nv04_exa.c b/src/nv04_exa.c index 132c9d5..7384cbc 100644 --- a/src/nv04_exa.c +++ b/src/nv04_exa.c @@ -197,16 +197,9 @@ NV04EXAPrepareCopy(PixmapPtr pSrcPixmap, PixmapPtr pDstPixmap, int dx, int dy, return FALSE; planemask |= ~0 << pDstPixmap->drawable.bitsPerPixel; - if (planemask != ~0 || alu != GXcopy) { - if (pDstPixmap->drawable.bitsPerPixel == 32) - return FALSE; - BEGIN_RING(chan, blit, NV04_IMAGE_BLIT_OPERATION, 1); - OUT_RING (chan, 1); /* ROP_AND */ - NV04EXASetROP(pScrn, alu, planemask); - } else { - BEGIN_RING(chan, blit, NV04_IMAGE_BLIT_OPERATION, 1); - OUT_RING (chan, 3); /* SRCCOPY */ - } + if ((planemask != ~0 || alu != GXcopy) + && pDstPixmap->drawable.bitsPerPixel == 32) + return FALSE; if (!NVAccelGetCtxSurf2DFormatFromPixmap(pDstPixmap, &fmt)) return FALSE; @@ -218,6 +211,15 @@ NV04EXAPrepareCopy(PixmapPtr pSrcPixmap, PixmapPtr pDstPixmap, int dx, int dy, OUT_RELOCl(chan, src_bo, src_delta, NOUVEAU_BO_VRAM | NOUVEAU_BO_RD); OUT_RELOCl(chan, dst_bo, dst_delta, NOUVEAU_BO_VRAM | NOUVEAU_BO_WR); + if (planemask != ~0 || alu != GXcopy) { + BEGIN_RING(chan, blit, NV04_IMAGE_BLIT_OPERATION, 1); + OUT_RING (chan, 1); /* ROP_AND */ + NV04EXASetROP(pScrn, alu, planemask); + } else { + BEGIN_RING(chan, blit, NV04_IMAGE_BLIT_OPERATION, 1); + OUT_RING (chan, 3); /* SRCCOPY */ + } + pNv->pspix = pSrcPixmap; pNv->pdpix = pDstPixmap; pNv->alu = alu; -- 1.6.4.4
Francisco Jerez
2009-Nov-06 15:14 UTC
[Nouveau] [PATCHv2] nv04-nv40/exa: Match the blob behavior more closely on PrepareCopy.
Fix a somewhat indeterministic corruption problem on nv17 when there is stuff going on the other fifos (e.g. gallium but I've also reproduced it with an app just SIFM-ing memory around): in some cases it made the blits the X server had scheduled fail corrupting the nearby screen areas. Signed-off-by: Francisco Jerez <currojerez at riseup.net> --- src/nv04_exa.c | 6 ++++++ 1 files changed, 6 insertions(+), 0 deletions(-) diff --git a/src/nv04_exa.c b/src/nv04_exa.c index 71a4ac2..f4966a4 100644 --- a/src/nv04_exa.c +++ b/src/nv04_exa.c @@ -209,10 +209,16 @@ NV04EXAPrepareCopy(PixmapPtr pSrcPixmap, PixmapPtr pDstPixmap, int dx, int dy, MARK_UNDO(chan); return FALSE; } + + BEGIN_RING(chan, blit, NV04_IMAGE_BLIT_SURFACE, 1); + OUT_RING (chan, pNv->NvContextSurfaces->handle); BEGIN_RING(chan, blit, NV04_IMAGE_BLIT_OPERATION, 1); OUT_RING (chan, 1); /* ROP_AND */ + NV04EXASetROP(pScrn, alu, planemask); } else { + BEGIN_RING(chan, blit, NV04_IMAGE_BLIT_SURFACE, 1); + OUT_RING (chan, pNv->NvContextSurfaces->handle); BEGIN_RING(chan, blit, NV04_IMAGE_BLIT_OPERATION, 1); OUT_RING (chan, 3); /* SRCCOPY */ } -- 1.6.4.4
Maybe Matching Threads
- [PATCH 1/2] Check before trying a solid fill
- [PATCH] nv/exa: fix 15/16 bits solid fill
- [Bug 48954] New: nv25 PGRAPH error and X freeze
- [Bug 30603] New: X crash in NVAccelDownloadM2MF with KDE desktop effects enabled.
- [PATCH] G72 doesn't need ExaCopy workaround