lkml.org 
[lkml]   [2007]   [Apr]   [29]   [last100]   RSS Feed
Views: [wrap][no wrap]   [headers]  [forward] 
 
Messages in this thread
Patch in this message
/
From
SubjectRe: [linux-dvb] Re: [PATCH] Re: More than 2Gb problem (dvb related) ?
Date
Jon Burgess wrote:
> On Sat, 2007-04-28 at 18:17 +0100, Jon Burgess wrote:
> > On Fri, 2007-04-27 at 18:06 -0400, Lee Revell wrote:
> > > On 4/27/07, Jon Burgess <jburgess777@googlemail.com> wrote:
> > > > Interesting - I see similar symptoms after upgrading my PC:
> > > > * old PC was AMD Athlon 64 3000 w/ 2GB of RAM which had no issues
> > > > * new PC is a Intel Core 2 Duo w/ 4GB of RAM and fails in the way you
> > > > describe.
> > >
> > > Driver using an incorrect DMA mask?
> > >
> > > Lee
> >
> > It appears the problem is that the driver is using streamed PCI and
> > needs to explicitly sync the data otherwise it breaks when the SWIOTLB
> > is in use. A call to pci_unmap_sg() was missing too.
> >
> > This patch is against the current v4l-dvb HG tree.
> >
> > I've only tested the change with a budget-ci card.
> >
> > Jon
> >
> > Signed-off-by: Jon Burgess <jburgess777@googlemail.com>
>
> While the above patch works, it seems the underlying causes is that
> vmalloc_32() is providing memory above 4Gb on x86-64 which is not what
> the driver expects. This same issue came up a few weeks ago with regards
> to DRM on radeon http://lkml.org/lkml/2007/4/1/257
>
> Andi Kleen included a patch to ensure vmalloc_32() returns memory <4Gb
> in a patch which is currently in -mm
> http://kernel.org/pub/linux/kernel/people/akpm/patches/2.6/2.6.21-rc7/2.6.21-rc7-mm2/broken-out/x86_64-mm-vmalloc-32.patch
>
> With this patch applied the current driver appears to work OK.
>
> Attached is a smaller patch against v4l-dvb which just adds the missing
> pci_unmap_sg() call.

Imo your patch is correct and should be applied.
Any objections?

After digging through the code, kernel DMA docs and the saa7146
datasheet, I think that we should remove the scatter-gatter voodoo
from the budget and av7110 driver. ;-)

What about the attached patch?
- easy to understand and maintain
- saves 1 page of memory (page table of the SAA7146 MMU)
- saves PCI bandwidth (SAA7146 does not read page table anymore)

Comments?

CU
Oliver

--
--------------------------------------------------------
VDR Remote Plugin 0.3.9 available at
http://www.escape-edv.de/endriss/vdr/
-------------------------------------------------------
diff -r e9e1b86b7892 linux/drivers/media/dvb/ttpci/budget-core.c
--- a/linux/drivers/media/dvb/ttpci/budget-core.c Mon Apr 23 07:41:13 2007 +0200
+++ b/linux/drivers/media/dvb/ttpci/budget-core.c Mon Apr 30 01:29:42 2007 +0200
@@ -74,6 +74,7 @@ static int start_ts_capture(struct budge
static int start_ts_capture(struct budget *budget)
{
struct saa7146_dev *dev = budget->dev;
+ u32 dma_base = (u32) budget->dma;

dprintk(2, "budget: %p\n", budget);

@@ -137,16 +138,16 @@ static int start_ts_capture(struct budge
saa7146_write(dev, MC2, (MASK_08 | MASK_24));
mdelay(10);

- saa7146_write(dev, BASE_ODD3, 0);
+ saa7146_write(dev, BASE_ODD3, dma_base);
if (budget->buffer_size > budget->buffer_height * budget->buffer_width) {
// using odd/even buffers
- saa7146_write(dev, BASE_EVEN3, budget->buffer_height * budget->buffer_width);
+ saa7146_write(dev, BASE_EVEN3, dma_base + budget->buffer_height * budget->buffer_width);
} else {
// using a single buffer
- saa7146_write(dev, BASE_EVEN3, 0);
- }
- saa7146_write(dev, PROT_ADDR3, budget->buffer_size);
- saa7146_write(dev, BASE_PAGE3, budget->pt.dma | ME1 | 0x90);
+ saa7146_write(dev, BASE_EVEN3, dma_base);
+ }
+ saa7146_write(dev, PROT_ADDR3, dma_base + budget->buffer_size);
+ saa7146_write(dev, BASE_PAGE3, 0x90);

saa7146_write(dev, PITCH3, budget->buffer_width);
saa7146_write(dev, NUM_LINE_BYTE3,
@@ -190,9 +191,9 @@ static void vpeirq(unsigned long data)
static void vpeirq(unsigned long data)
{
struct budget *budget = (struct budget *) data;
- u8 *mem = (u8 *) (budget->grabbing);
+ u8 *mem = budget->grabbing;
u32 olddma = budget->ttbp;
- u32 newdma = saa7146_read(budget->dev, PCI_VDP3);
+ u32 newdma = saa7146_read(budget->dev, PCI_VDP3) - (u32) budget->dma;
u32 count;

/* nearest lower position divisible by 188 */
@@ -510,8 +511,9 @@ int ttpci_budget_init(struct budget *bud

ttpci_eeprom_parse_mac(&budget->i2c_adap, budget->dvb_adapter.proposed_mac);

- if (NULL ==
- (budget->grabbing = saa7146_vmalloc_build_pgtable(dev->pci, budget->buffer_size, &budget->pt))) {
+ budget->grabbing = pci_alloc_consistent(dev->pci, budget->buffer_size,
+ &budget->dma);
+ if (NULL == budget->grabbing) {
ret = -ENOMEM;
goto err;
}
@@ -526,13 +528,15 @@ int ttpci_budget_init(struct budget *bud
if (bi->type != BUDGET_FS_ACTIVY)
saa7146_setgpio(dev, 2, SAA7146_GPIO_OUTHI);

- if (budget_register(budget) == 0) {
+ if (budget_register(budget) == 0)
return 0;
- }
+
err:
i2c_del_adapter(&budget->i2c_adap);

- vfree(budget->grabbing);
+ if (budget->grabbing)
+ pci_free_consistent(dev->pci, budget->buffer_size,
+ budget->grabbing, budget->dma);

dvb_unregister_adapter(&budget->dvb_adapter);

@@ -561,9 +565,8 @@ int ttpci_budget_deinit(struct budget *b

tasklet_kill(&budget->vpe_tasklet);

- saa7146_pgtable_free(dev->pci, &budget->pt);
-
- vfree(budget->grabbing);
+ pci_free_consistent(dev->pci, budget->buffer_size,
+ budget->grabbing, budget->dma);

return 0;
}
diff -r e9e1b86b7892 linux/drivers/media/dvb/ttpci/budget.h
--- a/linux/drivers/media/dvb/ttpci/budget.h Mon Apr 23 07:41:13 2007 +0200
+++ b/linux/drivers/media/dvb/ttpci/budget.h Mon Apr 30 01:32:53 2007 +0200
@@ -48,9 +48,6 @@ struct budget {
struct i2c_adapter i2c_adap;
struct budget_info *card;

- unsigned char *grabbing;
- struct saa7146_pgtable pt;
-
struct tasklet_struct fidb_tasklet;
struct tasklet_struct vpe_tasklet;

@@ -63,6 +60,8 @@ struct budget {
int ci_present;
int video_port;

+ void *grabbing;
+ dma_addr_t dma;
u32 buffer_width;
u32 buffer_height;
u32 buffer_size;
\
 
 \ /
  Last update: 2007-04-30 02:27    [W:0.114 / U:0.228 seconds]
©2003-2020 Jasper Spaans|hosted at Digital Ocean and TransIP|Read the blog|Advertise on this site