| Message ID | 1344489221-16687-1-git-send-email-andy.gross@ti.com (mailing list archive) |
|---|---|
| State | New, archived |
| Headers | show |
On Thu, Aug 9, 2012 at 12:13 AM, Andy Gross <andy.gross@ti.com> wrote: > Removed the unnecessary copy of the memory page addresses when > programming the DMM/PAT and all support code for the lut copy. > The original intent was to have this code in place for > suspend/resume functionality w.r.t. DEVICE_OFF. > > Performance analysis showed that the extra copy from uncached memory > led to a fairly hefty penalty when programming large 1D or 2D > buffers. This can be implemented in a more efficient manner when we > actually have to support DEVICE_OFF suspend/resume operations. This patch itself is ok, but I'd like to wait a bit and merge this together w/ a 2nd patch that handles saving the PAT state in the suspend path. BR, -R > Signed-off-by: Andy Gross <andy.gross@ti.com> > --- > drivers/staging/omapdrm/omap_dmm_priv.h | 6 ------ > drivers/staging/omapdrm/omap_dmm_tiler.c | 25 +------------------------ > 2 files changed, 1 insertions(+), 30 deletions(-) > > diff --git a/drivers/staging/omapdrm/omap_dmm_priv.h b/drivers/staging/omapdrm/omap_dmm_priv.h > index 08b22e9..09ebc50 100644 > --- a/drivers/staging/omapdrm/omap_dmm_priv.h > +++ b/drivers/staging/omapdrm/omap_dmm_priv.h > @@ -141,9 +141,6 @@ struct refill_engine { > /* only one trans per engine for now */ > struct dmm_txn txn; > > - /* offset to lut associated with container */ > - u32 *lut_offset; > - > wait_queue_head_t wait_for_refill; > > struct list_head idle_node; > @@ -176,9 +173,6 @@ struct dmm { > /* array of LUT - TCM containers */ > struct tcm **tcm; > > - /* LUT table storage */ > - u32 *lut; > - > /* allocation list and lock */ > struct list_head alloc_head; > }; > diff --git a/drivers/staging/omapdrm/omap_dmm_tiler.c b/drivers/staging/omapdrm/omap_dmm_tiler.c > index ec7a5c8..80d3f8a 100644 > --- a/drivers/staging/omapdrm/omap_dmm_tiler.c > +++ b/drivers/staging/omapdrm/omap_dmm_tiler.c > @@ -24,7 +24,6 @@ > #include <linux/interrupt.h> > #include <linux/dma-mapping.h> > #include <linux/slab.h> > -#include <linux/vmalloc.h> > #include <linux/delay.h> > #include <linux/mm.h> > #include <linux/time.h> > @@ -184,9 +183,6 @@ static int dmm_txn_append(struct dmm_txn *txn, struct pat_area *area, > int columns = (1 + area->x1 - area->x0); > int rows = (1 + area->y1 - area->y0); > int i = columns*rows; > - u32 *lut = omap_dmm->lut + (engine->tcm->lut_id * omap_dmm->lut_width * > - omap_dmm->lut_height) + > - (area->y0 * omap_dmm->lut_width) + area->x0; > > pat = alloc_dma(txn, sizeof(struct pat), &pat_pa); > > @@ -209,10 +205,6 @@ static int dmm_txn_append(struct dmm_txn *txn, struct pat_area *area, > page_to_phys(pages[n]) : engine->dmm->dummy_pa; > } > > - /* fill in lut with new addresses */ > - for (i = 0; i < rows; i++, lut += omap_dmm->lut_width) > - memcpy(lut, &data[i*columns], columns * sizeof(u32)); > - > txn->last_pat = pat; > > return 0; > @@ -504,8 +496,6 @@ static int omap_dmm_remove(struct platform_device *dev) > if (omap_dmm->dummy_page) > __free_page(omap_dmm->dummy_page); > > - vfree(omap_dmm->lut); > - > if (omap_dmm->irq > 0) > free_irq(omap_dmm->irq, omap_dmm); > > @@ -521,7 +511,7 @@ static int omap_dmm_probe(struct platform_device *dev) > { > int ret = -EFAULT, i; > struct tcm_area area = {0}; > - u32 hwinfo, pat_geom, lut_table_size; > + u32 hwinfo, pat_geom; > struct resource *mem; > > omap_dmm = kzalloc(sizeof(*omap_dmm), GFP_KERNEL); > @@ -593,16 +583,6 @@ static int omap_dmm_probe(struct platform_device *dev) > */ > writel(0x7e7e7e7e, omap_dmm->base + DMM_PAT_IRQENABLE_SET); > > - lut_table_size = omap_dmm->lut_width * omap_dmm->lut_height * > - omap_dmm->num_lut; > - > - omap_dmm->lut = vmalloc(lut_table_size * sizeof(*omap_dmm->lut)); > - if (!omap_dmm->lut) { > - dev_err(&dev->dev, "could not allocate lut table\n"); > - ret = -ENOMEM; > - goto fail; > - } > - > omap_dmm->dummy_page = alloc_page(GFP_KERNEL | __GFP_DMA32); > if (!omap_dmm->dummy_page) { > dev_err(&dev->dev, "could not allocate dummy page\n"); > @@ -685,9 +665,6 @@ static int omap_dmm_probe(struct platform_device *dev) > .p1.y = omap_dmm->container_height - 1, > }; > > - for (i = 0; i < lut_table_size; i++) > - omap_dmm->lut[i] = omap_dmm->dummy_pa; > - > /* initialize all LUTs to dummy page entries */ > for (i = 0; i < omap_dmm->num_lut; i++) { > area.tcm = omap_dmm->tcm[i]; > -- > 1.7.5.4 > > -- > To unsubscribe from this list: send the line "unsubscribe linux-omap" in > the body of a message to majordomo@vger.kernel.org > More majordomo info at http://vger.kernel.org/majordomo-info.html -- To unsubscribe from this list: send the line "unsubscribe linux-omap" in the body of a message to majordomo@vger.kernel.org More majordomo info at http://vger.kernel.org/majordomo-info.html
diff --git a/drivers/staging/omapdrm/omap_dmm_priv.h b/drivers/staging/omapdrm/omap_dmm_priv.h index 08b22e9..09ebc50 100644 --- a/drivers/staging/omapdrm/omap_dmm_priv.h +++ b/drivers/staging/omapdrm/omap_dmm_priv.h @@ -141,9 +141,6 @@ struct refill_engine { /* only one trans per engine for now */ struct dmm_txn txn; - /* offset to lut associated with container */ - u32 *lut_offset; - wait_queue_head_t wait_for_refill; struct list_head idle_node; @@ -176,9 +173,6 @@ struct dmm { /* array of LUT - TCM containers */ struct tcm **tcm; - /* LUT table storage */ - u32 *lut; - /* allocation list and lock */ struct list_head alloc_head; }; diff --git a/drivers/staging/omapdrm/omap_dmm_tiler.c b/drivers/staging/omapdrm/omap_dmm_tiler.c index ec7a5c8..80d3f8a 100644 --- a/drivers/staging/omapdrm/omap_dmm_tiler.c +++ b/drivers/staging/omapdrm/omap_dmm_tiler.c @@ -24,7 +24,6 @@ #include <linux/interrupt.h> #include <linux/dma-mapping.h> #include <linux/slab.h> -#include <linux/vmalloc.h> #include <linux/delay.h> #include <linux/mm.h> #include <linux/time.h> @@ -184,9 +183,6 @@ static int dmm_txn_append(struct dmm_txn *txn, struct pat_area *area, int columns = (1 + area->x1 - area->x0); int rows = (1 + area->y1 - area->y0); int i = columns*rows; - u32 *lut = omap_dmm->lut + (engine->tcm->lut_id * omap_dmm->lut_width * - omap_dmm->lut_height) + - (area->y0 * omap_dmm->lut_width) + area->x0; pat = alloc_dma(txn, sizeof(struct pat), &pat_pa); @@ -209,10 +205,6 @@ static int dmm_txn_append(struct dmm_txn *txn, struct pat_area *area, page_to_phys(pages[n]) : engine->dmm->dummy_pa; } - /* fill in lut with new addresses */ - for (i = 0; i < rows; i++, lut += omap_dmm->lut_width) - memcpy(lut, &data[i*columns], columns * sizeof(u32)); - txn->last_pat = pat; return 0; @@ -504,8 +496,6 @@ static int omap_dmm_remove(struct platform_device *dev) if (omap_dmm->dummy_page) __free_page(omap_dmm->dummy_page); - vfree(omap_dmm->lut); - if (omap_dmm->irq > 0) free_irq(omap_dmm->irq, omap_dmm); @@ -521,7 +511,7 @@ static int omap_dmm_probe(struct platform_device *dev) { int ret = -EFAULT, i; struct tcm_area area = {0}; - u32 hwinfo, pat_geom, lut_table_size; + u32 hwinfo, pat_geom; struct resource *mem; omap_dmm = kzalloc(sizeof(*omap_dmm), GFP_KERNEL); @@ -593,16 +583,6 @@ static int omap_dmm_probe(struct platform_device *dev) */ writel(0x7e7e7e7e, omap_dmm->base + DMM_PAT_IRQENABLE_SET); - lut_table_size = omap_dmm->lut_width * omap_dmm->lut_height * - omap_dmm->num_lut; - - omap_dmm->lut = vmalloc(lut_table_size * sizeof(*omap_dmm->lut)); - if (!omap_dmm->lut) { - dev_err(&dev->dev, "could not allocate lut table\n"); - ret = -ENOMEM; - goto fail; - } - omap_dmm->dummy_page = alloc_page(GFP_KERNEL | __GFP_DMA32); if (!omap_dmm->dummy_page) { dev_err(&dev->dev, "could not allocate dummy page\n"); @@ -685,9 +665,6 @@ static int omap_dmm_probe(struct platform_device *dev) .p1.y = omap_dmm->container_height - 1, }; - for (i = 0; i < lut_table_size; i++) - omap_dmm->lut[i] = omap_dmm->dummy_pa; - /* initialize all LUTs to dummy page entries */ for (i = 0; i < omap_dmm->num_lut; i++) { area.tcm = omap_dmm->tcm[i];
Removed the unnecessary copy of the memory page addresses when programming the DMM/PAT and all support code for the lut copy. The original intent was to have this code in place for suspend/resume functionality w.r.t. DEVICE_OFF. Performance analysis showed that the extra copy from uncached memory led to a fairly hefty penalty when programming large 1D or 2D buffers. This can be implemented in a more efficient manner when we actually have to support DEVICE_OFF suspend/resume operations. Signed-off-by: Andy Gross <andy.gross@ti.com> --- drivers/staging/omapdrm/omap_dmm_priv.h | 6 ------ drivers/staging/omapdrm/omap_dmm_tiler.c | 25 +------------------------ 2 files changed, 1 insertions(+), 30 deletions(-)