On 20 November 2014 18:54, Mark Salter <msalter@xxxxxxxxxx> wrote: > On Thu, 2014-11-20 at 18:38 +0100, Ard Biesheuvel wrote: >> On 20 November 2014 18:28, Mark Salter <msalter@xxxxxxxxxx> wrote: >> > On Tue, 2014-11-18 at 13:57 +0100, Ard Biesheuvel wrote: >> >> diff --git a/drivers/firmware/efi/virtmap.c >> >> b/drivers/firmware/efi/virtmap.c >> >> index 98735fb43581..4b6a5c31629f 100644 >> >> --- a/drivers/firmware/efi/virtmap.c >> >> +++ b/drivers/firmware/efi/virtmap.c >> >> @@ -8,6 +8,7 @@ >> >> */ >> >> >> >> #include <linux/efi.h> >> >> +#include <linux/memblock.h> >> >> #include <linux/mm_types.h> >> >> #include <linux/rbtree.h> >> >> #include <linux/rwsem.h> >> >> @@ -97,8 +98,15 @@ void __init efi_virtmap_init(void) >> >> u64 paddr, npages, size; >> >> pgprot_t prot; >> >> >> >> - if (!efi_mem_is_usable_region(md)) >> >> + paddr = md->phys_addr; >> >> + npages = md->num_pages; >> >> + memrange_efi_to_native(&paddr, &npages); >> >> + size = npages << PAGE_SHIFT; >> >> + >> >> + if (!efi_mem_is_usable_region(md)) { >> >> efi_register_mem_resource(md); >> >> + memblock_remove(paddr, size); >> >> + } >> > >> > What exactly is the memblock_remove trying to accomplish? With it, I >> > get: >> > >> >> The idea was pfn_valid() will then return false for those pfns, >> allowing us to distinguish them from memory that the kernel may be >> using, primarily for /dev/mem filtering. But apparently it is causing >> problems for you, so I will have to figure out if there is a better >> way of addressing this. >> > Okay. Well I think that works for pfn_valid, but it is confusing the > mem_cgroup code. I think because you end up with multiple memory regions > after creating the holes. Earlier memory management code saw one memory > region. And because this comes after paging_init(), all of the memory > ends up in the kernel linear mapping which is something we were trying > to avoid. > I will drop the memblock_remove() then. I guess I could make the test in devmem_is_allowed() do if (!memblock_is_memory() || memblock_is_reserved()) instead of 'if (!pfn_valid())' so that reserved regions become accessible with having to remove them. Are you happy with the other change in this patch, i.e., using memblock_add() directly so that we don't have to deal with the rounding? -- Ard. >> >> > Unable to handle kernel NULL pointer dereference at virtual address 00000008 >> > pgd = fffffe03d5890000 >> > [00000008] *pgd=00000043d5ae0003, *pud=00000043d5ae0003, *pmd=00000043d5ae0003, *pte=0000000000000000 >> > Internal error: Oops: 96000007 [#1] SMP >> > Modules linked in: >> > CPU: 0 PID: 107 Comm: modprobe Not tainted 3.18.0-rc5+ #8 >> > Hardware name: APM X-Gene Mustang board (DT) >> > task: fffffe03d5770000 ti: fffffe03dd300000 task.ti: fffffe03dd300000 >> > PC is at mem_cgroup_begin_page_stat+0x40/0x110 >> > LR is at mem_cgroup_begin_page_stat+0x40/0x110 >> > pc : [<fffffe00001ec34c>] lr : [<fffffe00001ec34c>] pstate: 60000145 >> > sp : fffffe03dd303a80 >> > x29: fffffe03dd303a80 x28: 000000000000ffe0 >> > x27: fffffdfef0f56bc0 x26: 0000000000000000 >> > x25: 000003ff89b80000 x24: 0000000000000000 >> > x23: fffffe03dd303ae8 x22: fffffe03dd303ae7 >> > x21: 0000000000000000 x20: fffffe03d58a0420 >> > x19: fffffdfef0004500 x18: 0000000000000000 >> > x17: 000003ff89ba40dc x16: 000003ff89bbff90 >> > x15: 000003ff89ba3b44 x14: 31313030343d6e66 >> > x13: 7020303035343030 x12: 0000000000000006 >> > x11: 0000000000000000 x10: 0000000000000186 >> > x9 : 0000000000000187 x8 : 3030306665666466 >> > x7 : fffffe0001041c18 x6 : 00000000000000ff >> > x5 : fffffe00004aaf20 x4 : fffffe00004b209c >> > x3 : 0000000000000000 x2 : 0000000000000000 >> > x1 : fffffe0000f6a5d0 x0 : 0000000000000000 >> > >> > Process modprobe (pid: 107, stack limit = 0xfffffe03dd300058) >> > Stack: (0xfffffe03dd303a80 to 0xfffffe03dd304000) >> > 3a80: dd303ac0 fffffe03 001c3e20 fffffe00 f0004500 fffffdfe d58a0420 fffffe03 >> > 3aa0: f0004500 fffffdfe d58f4dc0 fffffe03 00000000 00000000 001b62d4 fffffe00 >> > 3ac0: dd303af0 fffffe03 001b9984 fffffe00 01140f53 00200040 d58a0420 fffffe03 >> > 3ae0: d5880000 fffffe03 d58a0420 fffffe03 dd303b40 fffffe03 001b9c4c fffffe00 >> > 3b00: d58a0420 fffffe03 89b80000 000003ff f0f56bf0 fffffdfe 00000000 00000000 >> > ... >> > 3fa0: 89b907b7 000003ff 89bbf000 000003ff 89bc16e8 000003ff eb753490 000003ff >> > 3fc0: 89b91f9c 000003ff eb753490 000003ff 89b91fac 000003ff 20000000 00000000 >> > 3fe0: 00000000 00000000 ffffffff ffffffff aaaaaaaa aaaaaaaa aaaaaaaa aaaaaaaa >> > Call trace: >> > [<fffffe00001ec34c>] mem_cgroup_begin_page_stat+0x40/0x110 >> > [<fffffe00001c3e1c>] page_add_file_rmap+0x24/0x80 >> > [<fffffe00001b9980>] do_set_pte+0x108/0x19c >> > [<fffffe00001b9c48>] do_read_fault.isra.93+0x234/0x28c >> > [<fffffe00001ba57c>] handle_mm_fault+0x488/0xd64 >> > [<fffffe00000a0ea0>] do_page_fault+0x224/0x358 >> > [<fffffe0000090128>] do_mem_abort+0x4c/0xb0 >> > Exception stack(0xfffffe03dd303e30 to 0xfffffe03dd303f50) >> > 3e20: 00000000 00000000 89bbf000 000003ff >> > 3e40: ffffffff ffffffff 89b91fac 000003ff 01043280 fffffe00 00000002 00000000 >> > 3e60: dd303e70 fffffe03 008ce3a0 fffffe00 dd303e90 fffffe03 001c0400 fffffe00 >> > 3e80: d5880000 fffffe03 000cbc60 fffffe00 eb753610 000003ff 00092fdc fffffe00 >> > 3ea0: 00000000 00000000 89bbf000 000003ff ffffffff ffffffff 89ba4654 000003ff >> > 3ec0: 80000000 00000000 d5521480 fffffe03 89bc16e8 000003ff 89bc1a40 000003ff >> > 3ee0: 00000004 00000000 89b80000 000003ff fefefeff fefefefe 00000000 00000000 >> > 3f00: 0000000f 00000000 89bc1028 000003ff 89ba7000 000003ff 0000079c 00000000 >> > 3f20: 00400000 00000000 ffff0000 ffffffff 004038a0 00000000 00010000 00000000 >> > 3f40: 89ba3c1c 000003ff 89ba3b44 000003ff >> > Code: b9405835 35000275 aa1303e0 9400065c (f9400413) >> > >> > > > -- To unsubscribe from this list: send the line "unsubscribe linux-efi" in the body of a message to majordomo@xxxxxxxxxxxxxxx More majordomo info at http://vger.kernel.org/majordomo-info.html