Hi, I'm trying to figure out how to patch srmmu.c the easiest way so that 32k size pages can be used for Sparc-Leon. The Leon-Sparc processor supports that page size. Studying the code I see that 16 hardware PTE-tables are joined together to form a 1024 entry software PTE-table exactly 4k big. The source of pgtsrmmu.h states: /* 40 * To support pagetables in highmem, Linux introduces APIs which 41 * return struct page* and generally manipulate page tables when 42 * they are not mapped into kernel space. Our hardware page tables 43 * are smaller than pages. We lump hardware tabes into big, page sized 44 * software tables. 45 * 46 * PMD_SHIFT determines the size of the area a second-level page table entry 47 * can map, and our pmd_t is 16 times larger than normal. The values which 48 * were once defined here are now generic for 4c and srmmu, so they're 49 * found in pgtable.h. 50 */ The problem I face now is that when using 32k pages the hardware PTE-table has only 8 entries (when using SRMMU-like index layout with bit [[31-24][23 - 18][17-15]]). So I could at max form a 8*64=512 entry software PTE-table that would only be 2k. However the page size would be 32k anyway. Could someone explain why there is a need to join hardware PTE-Tables together to form chunks of 4k (PAGE_SIZE, but in my case 32k). Is this only a performance consideration or would I run into trouble if using only 2k chunks of 512 enties? If I only use pages tables in nocache mem at 0xfc000000 (and I think this is the case for the Sparc-Leon port using linux-2.6.21.1), can the pgtsrmmu.h comment then be ignored? If I understand it right that page tables are always allocated through __srmmu_get_nocache() and I can allocate chunks smaller than PAGE_SIZE. Does someone see a easy path to implement 32k pages? -- Greetings Konrad -- To unsubscribe from this list: send the line "unsubscribe sparclinux" in the body of a message to majordomo@xxxxxxxxxxxxxxx More majordomo info at http://vger.kernel.org/majordomo-info.html
