with kernel 2.6.10-rc3-mm1 during compile, gcc complained about 'pgd_offset' and 'remap_page_range', similar errors were reportet (dmesg) during loading the fglrx-module. afer searching the weg heavily, i found patches for this problems would be nice to see them in the next ebuild. ##################################### # 8.08-kernel-2.6.11-pgd_offset.patch ##################################### --- firegl_public.c.back 2005-01-19 14:37:50.000000000 +0000 +++ firegl_public.c 2005-01-19 14:42:03.000000000 +0000 @@ -1575,7 +1575,7 @@ pte_t pte; pte_linear = VMALLOC_VMADDR(virtual_addr); // convert to pte linear address (x86 => nop) - pgd_p = pgd_offset(mm, pte_linear); + pgd_p = pml4_pgd_offset(pml4_offset(mm, pte_linear), pte_linear); pmd_p = pmd_offset(pgd_p, pte_linear); #ifndef FGL_ATOMIC_PTE #if LINUX_VERSION_CODE > 0x020500 @@ -2062,7 +2062,7 @@ */ pgd_p = pgd_offset_k(pte_linear); #else - pgd_p = pgd_offset(&init_mm, pte_linear); /* locate page directory entry */ + pgd_p = pml4_pgd_offset(pml4_offset(&init_mm, pte_linear), pte_linear); /* locate page directory entry */ #endif if (!pgd_present(*pgd_p)) { @@ -2441,7 +2441,7 @@ pte_t pte; pte_linear = VMALLOC_VMADDR(virtual_addr); // convert to pte linear address (x86 => nop) - pgd_p = pgd_offset(vma->vm_mm, pte_linear); + pgd_p = pml4_pgd_offset(pml4_offset(vma->vm_mm, pte_linear), pte_linear); pmd_p = pmd_offset(pgd_p, pte_linear); #ifndef FGL_ATOMIC_PTE #if LINUX_VERSION_CODE > 0x020500 ### EOF ### ########################################### # 8.08-kernel-2.6.11-remap_page_range.patch ########################################### diff -urN build_mod/agpgart_be.c build_mod2/agpgart_be.c --- build_mod/agpgart_be.c 2005-01-17 23:34:04.000000000 +0000 +++ build_mod2/agpgart_be.c 2005-01-17 23:36:12.000000000 +0000 @@ -116,7 +116,14 @@ #include <linux/pagemap.h> #include <linux/miscdevice.h> #include <linux/pm.h> +#if LINUX_VERSION_CODE > KERNEL_VERSION(2,6,9) +#define pci_find_class pci_get_class + + + + +#endif #if (LINUX_VERSION_CODE >= 0x020400) #define FGL_PM_PRESENT #else diff -urN build_mod/firegl_public.c build_mod2/firegl_public.c --- build_mod/firegl_public.c 2005-01-17 23:34:04.000000000 +0000 +++ build_mod2/firegl_public.c 2005-01-17 23:36:02.000000000 +0000 @@ -2590,13 +2590,13 @@ #endif /* __ia64__ */ vma->vm_flags |= VM_IO; /* not in core dump */ } - if (remap_page_range(FGL_VMA_API_PASS + if (remap_pfn_range(FGL_VMA_API_PASS vma->vm_start, - __ke_vm_offset(vma), + vma->vm_pgoff, vma->vm_end - vma->vm_start, vma->vm_page_prot)) { - __KE_DEBUG("remap_page_range failed\n"); + __KE_DEBUG("remap_pfn_range failed\n"); return -EAGAIN; } vma->vm_flags |= VM_SHM | VM_RESERVED; /* Don't swap */ @@ -2655,15 +2655,15 @@ #else // else { - if (__ke_vm_offset(vma) >= __pa(high_memory)) + if (vma->vm_pgoff >= __pa(high_memory)) vma->vm_flags |= VM_IO; /* not in core dump */ - if (remap_page_range(FGL_VMA_API_PASS + if (remap_pfn_range(FGL_VMA_API_PASS vma->vm_start, - __ke_vm_offset(vma), + vma->vm_pgoff, vma->vm_end - vma->vm_start, vma->vm_page_prot)) { - __KE_DEBUG("remap_page_range failed\n"); + __KE_DEBUG("remap_pfn_range failed\n"); return -EAGAIN; } #ifdef __x86_64__ @@ -2692,15 +2692,15 @@ // else #else { - if (__ke_vm_offset(vma) >= __pa(high_memory)) + if (vma->vm_pgoff >= __pa(high_memory)) vma->vm_flags |= VM_IO; /* not in core dump */ - if (remap_page_range(FGL_VMA_API_PASS + if (remap_pfn_range(FGL_VMA_API_PASS vma->vm_start, - __ke_vm_offset(vma), + vma->vm_pgoff, vma->vm_end - vma->vm_start, vma->vm_page_prot)) { - __KE_DEBUG("remap_page_range failed\n"); + __KE_DEBUG("remap_pfn_range failed\n"); return -EAGAIN; } #ifdef __x86_64__ ### EOF ### i tested this on: 2.6.10-rc3-mm1 / gcc version 3.3.5 (Gentoo Linux 3.3.5-r1, ssp-3.3.2-3, pie-8.7.7.1) Reproducible: Always Steps to Reproduce: 1.emerge ati-drivers 2. 3.
please include urls for where you found these patches.
The __ke_vm_offset(vma) -> vma->vm_pgoff is wrong the right way is add another define for __ke_vm_offset()
My bad, I didn't see this bug first. I added a patch today for the 4-level pagetable stuff in 2.6.11. It should be backward compatible so I didn't think too much about dropping it in since it works fine in my vanilla 2.6.10 config. Test it out and see if it works for you. If not, flame on. :)
How about adding an note in the readme where the patches are from ? as simple as adding www.rage3d.com. cheers Jon p.s.: it would also be nice to give some of the gentoo-only patches back to the community, for example post them to rage3d forums the url with the first appearance of the pagetable patch: http://rage3d.com/board/showthread.php?t=33798874&page=2&pp=30
The gentoo only patches that could be useful are resent upstream and are available on mirrors. Nowadays, most of them are just minor fixes that other distributions and, probably, rage3d forums already have.
Created attachment 53004 [details, diff] Patch for the ati-drivers-8.10.19 ebuild for <2.6.10 kernels. The patch from rage3d about remap_page_range vs remap_pfn_range breaks the module on 2.6.9 kernels, though. The 8.10.19 ebuild patches all 2.6 kernels with it, though. Attached is a patch for the ebuild that only applies the remap patch if the kernel is greater than 2.6.9, not for all 2.6 kernels. Using this, the module compiles and runs clean on my machine (gentoo-dev-sources-2.6.9-r14, AMD64).
I'll try to sort out the right ifdef magic to avoid such kludge, the next time please open another bug since the problems are quite unrelated
Fixed long time ago, I think
