i2p/haiku-beta6
1
0
Fork 0
Code Issues Pull Requests Actions Packages Projects Releases Wiki Activity

updated kerneldriver again for DMA cmdbuffer in mainmem: added MTRR-WC mapping now. Actually copying some stuff from Thomas :-), interesting read. I'll try to go without further cloning of that DMA area for now. (learning..)

Browse Source 
git-svn-id: file:///srv/svn/repos/haiku/haiku/trunk@12837 a95241bf-73f2-0310-859d-f6bbb57e9c96
This commit is contained in:
Rudolf Cornelissen
2005-05-26 11:41:33 +00:00
parent 395332ac33
commit 37dd6ec29e
1 changed files with 54 additions and 24 deletions
Download Patch File Download Diff File Expand all files Collapse all files
src/add-ons/kernel/drivers/graphics/nvidia/driver.c
+54 -24
View File
@@ -843,6 +843,7 @@ static status_t open_hook (const char* name, uint32 flags, void** cookie) {
char shared_name[B_OS_NAME_LENGTH]; char shared_name[B_OS_NAME_LENGTH];
physical_entry map[1]; physical_entry map[1];
size_t net_buf_size; size_t net_buf_size;
void *unaligned_dma_buffer;
/* find the device name in the list of devices */ /* find the device name in the list of devices */
/* we're never passed a name we didn't publish */ /* we're never passed a name we didn't publish */
@@ -881,28 +882,46 @@ static status_t open_hook (const char* name, uint32 flags, void** cookie) {
/* we want to setup a 1Mb buffer (size must be multiple of B_PAGE_SIZE) */ /* we want to setup a 1Mb buffer (size must be multiple of B_PAGE_SIZE) */
net_buf_size = ((1 * 1024 * 1024) + (B_PAGE_SIZE-1)) & ~(B_PAGE_SIZE-1); net_buf_size = ((1 * 1024 * 1024) + (B_PAGE_SIZE-1)) & ~(B_PAGE_SIZE-1);
/* create the area that will hold the DMA command buffer */ /* create the area that will hold the DMA command buffer */
si->dma_buffer_area = si->unaligned_dma_area =
create_area("NV accelerant DMA cmd buffer", create_area("NV DMA cmd buffer",
(void **)&(si->dma_buffer), (void **)&unaligned_dma_buffer,
B_ANY_KERNEL_ADDRESS, B_ANY_KERNEL_ADDRESS,
2 * net_buf_size, /* take twice the net size so we can have MTRR-WC even on old systems */ 2 * net_buf_size, /* take twice the net size so we can have MTRR-WC even on old systems */
B_FULL_LOCK | B_CONTIGUOUS, /* both properties needed: GPU always needs access */ B_FULL_LOCK | B_CONTIGUOUS, /* both properties needed: GPU always needs access */
B_USER_CLONEABLE_AREA); B_USER_CLONEABLE_AREA | B_READ_AREA | B_WRITE_AREA);
/* on error, abort */ /* on error, abort */
if (si->dma_buffer_area < 0) if (si->unaligned_dma_area < 0)
{ {
/* free the already created shared_info area, and return the error */ /* free the already created shared_info area, and return the error */
result = si->dma_buffer_area; result = si->unaligned_dma_area;
goto free_shared; goto free_shared;
} }
/* we also need the physical adress our DMA buffer resides in, as this needs to be /* we (also) need the physical adress our DMA buffer is at, as this needs to be
* fed into the GPU's engine later on. So get the first page of our assigned buffer */ * fed into the GPU's engine later on. Get an aligned adress so we can use MTRR-WC
get_memory_map(si->dma_buffer, B_PAGE_SIZE, map, 1); * even on older CPU's. */
si->dma_buffer_pci = (void*)((uint32)(map[0].address)); get_memory_map(unaligned_dma_buffer, B_PAGE_SIZE, map, 1);
/* set our net buffer at an aligned adress for MTRR-WC mapping even by older CPU's */ si->dma_buffer_pci = (void*)
/* (using MTRR-WC mapping will speed-up placing commands..) */ ((((uint32)(map[0].address)) + net_buf_size - 1) & ~(net_buf_size - 1));
si->engine.dma.cmdbuffer = (uint32*)
((((uint32)(si->dma_buffer_pci)) + net_buf_size - 1) & ~(net_buf_size - 1)); /* map the net DMA command buffer into vmem, using Write Combining */
si->dma_area = map_physical_memory(
"NV aligned DMA cmd buffer", si->dma_buffer_pci, net_buf_size,
B_ANY_KERNEL_BLOCK_ADDRESS | B_MTR_WC,
B_READ_AREA | B_WRITE_AREA, &(si->dma_buffer));
/* if failed with write combining try again without */
if (si->dma_area < 0) {
si->dma_area = map_physical_memory(
"NV aligned DMA cmd buffer", si->dma_buffer_pci, net_buf_size,
B_ANY_KERNEL_BLOCK_ADDRESS,
B_READ_AREA | B_WRITE_AREA, &(si->dma_buffer));
}
/* if there was an error, delete our other areas and pass on error*/
if (si->dma_area < 0)
{
/* free the already created areas, and return the error */
result = si->dma_area;
goto free_shared_and_uadma;
}
/* save the vendor and device IDs */ /* save the vendor and device IDs */
si->vendor_id = di->pcii.vendor_id; si->vendor_id = di->pcii.vendor_id;
@@ -938,7 +957,7 @@ static status_t open_hook (const char* name, uint32 flags, void** cookie) {
/* map the device */ /* map the device */
result = map_device(di); result = map_device(di);
if (result < 0) goto free_shared_and_dma; if (result < 0) goto free_shared_and_alldma;
result = B_OK; result = B_OK;
/* create a semaphore for vertical blank management */ /* create a semaphore for vertical blank management */
@@ -995,11 +1014,17 @@ delete_the_sem:
unmap: unmap:
unmap_device(di); unmap_device(di);
free_shared_and_dma: free_shared_and_alldma:
/* clean up our DMA area */ /* clean up our aligned DMA area */
delete_area(si->dma_buffer_area); delete_area(si->dma_area);
si->dma_buffer_area = -1; si->dma_area = -1;
si->dma_buffer = si->dma_buffer_pci = si->engine.dma.cmdbuffer = NULL; si->dma_buffer = NULL;
free_shared_and_uadma:
/* clean up our unaligned DMA area */
delete_area(si->unaligned_dma_area);
si->unaligned_dma_area = -1;
si->dma_buffer_pci = NULL;
free_shared: free_shared:
/* clean up our shared area */ /* clean up our shared area */
@@ -1074,10 +1099,15 @@ free_hook (void* dev) {
/* free regs and framebuffer areas */ /* free regs and framebuffer areas */
unmap_device(di); unmap_device(di);
/* clean up our DMA area */ /* clean up our aligned DMA area */
delete_area(si->dma_buffer_area); delete_area(si->dma_area);
si->dma_buffer_area = -1; si->dma_area = -1;
si->dma_buffer = si->dma_buffer_pci = si->engine.dma.cmdbuffer = NULL; si->dma_buffer = NULL;
/* clean up our unaligned DMA area */
delete_area(si->unaligned_dma_area);
si->unaligned_dma_area = -1;
si->dma_buffer_pci = NULL;
/* clean up our shared area */ /* clean up our shared area */
delete_area(di->shared_area); delete_area(di->shared_area);