]> git.kernelconcepts.de Git - karo-tx-linux.git/commitdiff
powerpc/fsl-booke64: Set vmemmap_psize to 4K
authorScott Wood <scottwood@freescale.com>
Tue, 6 May 2014 00:41:10 +0000 (19:41 -0500)
committerScott Wood <scottwood@freescale.com>
Thu, 22 May 2014 23:08:12 +0000 (18:08 -0500)
The only way Freescale booke chips support mappings larger than 4K
is via TLB1.  The only way we support (direct) TLB1 entries is via
hugetlb, which is not what map_kernel_page() does when given a large
page size.

Without this, a kernel with CONFIG_SPARSEMEM_VMEMMAP enabled crashes on
boot with messages such as:

PID hash table entries: 4096 (order: 3, 32768 bytes)
Sorting __ex_table...
BUG: Bad page state in process swapper  pfn:00a2f
page:8000040000023a48 count:0 mapcount:0 mapping:0000040000ffce48 index:0x40000ffbe50
page flags: 0x40000ffda40(active|arch_1|private|private_2|head|tail|swapcache|mappedtodisk|reclaim|swapbacked|unevictable|mlocked)
page dumped because: PAGE_FLAGS_CHECK_AT_FREE flag(s) set
bad because of flags:
page flags: 0x311840(active|private|private_2|swapcache|unevictable|mlocked)
Modules linked in:
CPU: 0 PID: 0 Comm: swapper Not tainted 3.15.0-rc1-00003-g7fa250c #299
Call Trace:
[c00000000098ba20] [c000000000008b3c] .show_stack+0x7c/0x1cc (unreliable)
[c00000000098baf0] [c00000000060aa50] .dump_stack+0x88/0xb4
[c00000000098bb70] [c0000000000c0468] .bad_page+0x144/0x1a0
[c00000000098bc10] [c0000000000c0628] .free_pages_prepare+0x164/0x17c
[c00000000098bcc0] [c0000000000c24cc] .free_hot_cold_page+0x48/0x214
[c00000000098bd60] [c00000000086c318] .free_all_bootmem+0x1fc/0x354
[c00000000098be70] [c00000000085da84] .mem_init+0xac/0xdc
[c00000000098bef0] [c0000000008547b0] .start_kernel+0x21c/0x4d4
[c00000000098bf90] [c000000000000448] .start_here_common+0x20/0x58

Signed-off-by: Scott Wood <scottwood@freescale.com>
arch/powerpc/mm/tlb_nohash.c

index ae3d5b799b9004f3089176c90a22b4dd8e76c84a..92cb18d52ea8d147ed1201d56eec2d776e28e978 100644 (file)
@@ -596,8 +596,13 @@ static void __early_init_mmu(int boot_cpu)
        /* XXX This should be decided at runtime based on supported
         * page sizes in the TLB, but for now let's assume 16M is
         * always there and a good fit (which it probably is)
+        *
+        * Freescale booke only supports 4K pages in TLB0, so use that.
         */
-       mmu_vmemmap_psize = MMU_PAGE_16M;
+       if (mmu_has_feature(MMU_FTR_TYPE_FSL_E))
+               mmu_vmemmap_psize = MMU_PAGE_4K;
+       else
+               mmu_vmemmap_psize = MMU_PAGE_16M;
 
        /* XXX This code only checks for TLB 0 capabilities and doesn't
         *     check what page size combos are supported by the HW. It