X-Git-Url: http://pilppa.org/gitweb/?a=blobdiff_plain;f=drivers%2Flguest%2Fsegments.c;h=9e189cbec7dda8640ffea4eda82e35f499c42236;hb=47f76ca3a34cd6571a2de39da2926123ca39a4c1;hp=1b2cfe89dcd568546bfd93d4e780794759914d94;hpb=efffbeee5bc4168059683714b300d307f5193d69;p=linux-2.6-omap-h63xx.git diff --git a/drivers/lguest/segments.c b/drivers/lguest/segments.c index 1b2cfe89dcd..9e189cbec7d 100644 --- a/drivers/lguest/segments.c +++ b/drivers/lguest/segments.c @@ -1,16 +1,50 @@ +/*P:600 The x86 architecture has segments, which involve a table of descriptors + * which can be used to do funky things with virtual address interpretation. + * We originally used to use segments so the Guest couldn't alter the + * Guest<->Host Switcher, and then we had to trim Guest segments, and restore + * for userspace per-thread segments, but trim again for on userspace->kernel + * transitions... This nightmarish creation was contained within this file, + * where we knew not to tread without heavy armament and a change of underwear. + * + * In these modern times, the segment handling code consists of simple sanity + * checks, and the worst you'll experience reading this code is butterfly-rash + * from frolicking through its parklike serenity. :*/ #include "lg.h" -static int desc_ok(const struct desc_struct *gdt) -{ - /* MBZ=0, P=1, DT=1 */ - return ((gdt->b & 0x00209000) == 0x00009000); -} - -static int segment_present(const struct desc_struct *gdt) -{ - return gdt->b & 0x8000; -} - +/*H:600 + * Segments & The Global Descriptor Table + * + * (That title sounds like a bad Nerdcore group. Not to suggest that there are + * any good Nerdcore groups, but in high school a friend of mine had a band + * called Joe Fish and the Chips, so there are definitely worse band names). + * + * To refresh: the GDT is a table of 8-byte values describing segments. Once + * set up, these segments can be loaded into one of the 6 "segment registers". + * + * GDT entries are passed around as "struct desc_struct"s, which like IDT + * entries are split into two 32-bit members, "a" and "b". One day, someone + * will clean that up, and be declared a Hero. (No pressure, I'm just saying). + * + * Anyway, the GDT entry contains a base (the start address of the segment), a + * limit (the size of the segment - 1), and some flags. Sounds simple, and it + * would be, except those zany Intel engineers decided that it was too boring + * to put the base at one end, the limit at the other, and the flags in + * between. They decided to shotgun the bits at random throughout the 8 bytes, + * like so: + * + * 0 16 40 48 52 56 63 + * [ limit part 1 ][ base part 1 ][ flags ][li][fl][base ] + * mit ags part 2 + * part 2 + * + * As a result, this file contains a certain amount of magic numeracy. Let's + * begin. + */ + +/* There are several entries we don't let the Guest set. The TSS entry is the + * "Task State Segment" which controls all kinds of delicate things. The + * LGUEST_CS and LGUEST_DS entries are reserved for the Switcher, and the + * the Guest can't be trusted to deal with double faults. */ static int ignored_gdt(unsigned int num) { return (num == GDT_ENTRY_TSS @@ -19,107 +53,135 @@ static int ignored_gdt(unsigned int num) || num == GDT_ENTRY_DOUBLEFAULT_TSS); } -/* We don't allow removal of CS, DS or SS; it doesn't make sense. */ -static void check_segment_use(struct lguest *lg, unsigned int desc) -{ - if (lg->regs->gs / 8 == desc) - lg->regs->gs = 0; - if (lg->regs->fs / 8 == desc) - lg->regs->fs = 0; - if (lg->regs->es / 8 == desc) - lg->regs->es = 0; - if (lg->regs->ds / 8 == desc - || lg->regs->cs / 8 == desc - || lg->regs->ss / 8 == desc) - kill_guest(lg, "Removed live GDT entry %u", desc); -} - +/*H:630 Once the Guest gave us new GDT entries, we fix them up a little. We + * don't care if they're invalid: the worst that can happen is a General + * Protection Fault in the Switcher when it restores a Guest segment register + * which tries to use that entry. Then we kill the Guest for causing such a + * mess: the message will be "unhandled trap 256". */ static void fixup_gdt_table(struct lguest *lg, unsigned start, unsigned end) { unsigned int i; for (i = start; i < end; i++) { - /* We never copy these ones to real gdt */ + /* We never copy these ones to real GDT, so we don't care what + * they say */ if (ignored_gdt(i)) continue; - /* We could fault in switch_to_guest if they are using - * a removed segment. */ - if (!segment_present(&lg->gdt[i])) { - check_segment_use(lg, i); - continue; - } - - if (!desc_ok(&lg->gdt[i])) - kill_guest(lg, "Bad GDT descriptor %i", i); - - /* DPL 0 presumably means "for use by guest". */ - if ((lg->gdt[i].b & 0x00006000) == 0) - lg->gdt[i].b |= (GUEST_PL << 13); - - /* Set accessed bit, since gdt isn't writable. */ - lg->gdt[i].b |= 0x00000100; + /* Segment descriptors contain a privilege level: the Guest is + * sometimes careless and leaves this as 0, even though it's + * running at privilege level 1. If so, we fix it here. */ + if ((lg->arch.gdt[i].b & 0x00006000) == 0) + lg->arch.gdt[i].b |= (GUEST_PL << 13); + + /* Each descriptor has an "accessed" bit. If we don't set it + * now, the CPU will try to set it when the Guest first loads + * that entry into a segment register. But the GDT isn't + * writable by the Guest, so bad things can happen. */ + lg->arch.gdt[i].b |= 0x00000100; } } +/*H:610 Like the IDT, we never simply use the GDT the Guest gives us. We keep + * a GDT for each CPU, and copy across the Guest's entries each time we want to + * run the Guest on that CPU. + * + * This routine is called at boot or modprobe time for each CPU to set up the + * constant GDT entries: the ones which are the same no matter what Guest we're + * running. */ void setup_default_gdt_entries(struct lguest_ro_state *state) { struct desc_struct *gdt = state->guest_gdt; unsigned long tss = (unsigned long)&state->guest_tss; - /* Hypervisor segments. */ + /* The Switcher segments are full 0-4G segments, privilege level 0 */ gdt[GDT_ENTRY_LGUEST_CS] = FULL_EXEC_SEGMENT; gdt[GDT_ENTRY_LGUEST_DS] = FULL_SEGMENT; - /* This is the one which we *cannot* copy from guest, since tss - is depended on this lguest_ro_state, ie. this cpu. */ + /* The TSS segment refers to the TSS entry for this particular CPU. + * Forgive the magic flags: the 0x8900 means the entry is Present, it's + * privilege level 0 Available 386 TSS system segment, and the 0x67 + * means Saturn is eclipsed by Mercury in the twelfth house. */ gdt[GDT_ENTRY_TSS].a = 0x00000067 | (tss << 16); gdt[GDT_ENTRY_TSS].b = 0x00008900 | (tss & 0xFF000000) | ((tss >> 16) & 0x000000FF); } +/* This routine sets up the initial Guest GDT for booting. All entries start + * as 0 (unusable). */ void setup_guest_gdt(struct lguest *lg) { - lg->gdt[GDT_ENTRY_KERNEL_CS] = FULL_EXEC_SEGMENT; - lg->gdt[GDT_ENTRY_KERNEL_DS] = FULL_SEGMENT; - lg->gdt[GDT_ENTRY_KERNEL_CS].b |= (GUEST_PL << 13); - lg->gdt[GDT_ENTRY_KERNEL_DS].b |= (GUEST_PL << 13); + /* Start with full 0-4G segments... */ + lg->arch.gdt[GDT_ENTRY_KERNEL_CS] = FULL_EXEC_SEGMENT; + lg->arch.gdt[GDT_ENTRY_KERNEL_DS] = FULL_SEGMENT; + /* ...except the Guest is allowed to use them, so set the privilege + * level appropriately in the flags. */ + lg->arch.gdt[GDT_ENTRY_KERNEL_CS].b |= (GUEST_PL << 13); + lg->arch.gdt[GDT_ENTRY_KERNEL_DS].b |= (GUEST_PL << 13); } -/* This is a fast version for the common case where only the three TLS entries - * have changed. */ +/*H:650 An optimization of copy_gdt(), for just the three "thead-local storage" + * entries. */ void copy_gdt_tls(const struct lguest *lg, struct desc_struct *gdt) { unsigned int i; for (i = GDT_ENTRY_TLS_MIN; i <= GDT_ENTRY_TLS_MAX; i++) - gdt[i] = lg->gdt[i]; + gdt[i] = lg->arch.gdt[i]; } +/*H:640 When the Guest is run on a different CPU, or the GDT entries have + * changed, copy_gdt() is called to copy the Guest's GDT entries across to this + * CPU's GDT. */ void copy_gdt(const struct lguest *lg, struct desc_struct *gdt) { unsigned int i; + /* The default entries from setup_default_gdt_entries() are not + * replaced. See ignored_gdt() above. */ for (i = 0; i < GDT_ENTRIES; i++) if (!ignored_gdt(i)) - gdt[i] = lg->gdt[i]; + gdt[i] = lg->arch.gdt[i]; } +/*H:620 This is where the Guest asks us to load a new GDT (LHCALL_LOAD_GDT). + * We copy it from the Guest and tweak the entries. */ void load_guest_gdt(struct lguest *lg, unsigned long table, u32 num) { - if (num > ARRAY_SIZE(lg->gdt)) + /* We assume the Guest has the same number of GDT entries as the + * Host, otherwise we'd have to dynamically allocate the Guest GDT. */ + if (num > ARRAY_SIZE(lg->arch.gdt)) kill_guest(lg, "too many gdt entries %i", num); - lgread(lg, lg->gdt, table, num * sizeof(lg->gdt[0])); - fixup_gdt_table(lg, 0, ARRAY_SIZE(lg->gdt)); + /* We read the whole thing in, then fix it up. */ + __lgread(lg, lg->arch.gdt, table, num * sizeof(lg->arch.gdt[0])); + fixup_gdt_table(lg, 0, ARRAY_SIZE(lg->arch.gdt)); + /* Mark that the GDT changed so the core knows it has to copy it again, + * even if the Guest is run on the same CPU. */ lg->changed |= CHANGED_GDT; } +/* This is the fast-track version for just changing the three TLS entries. + * Remember that this happens on every context switch, so it's worth + * optimizing. But wouldn't it be neater to have a single hypercall to cover + * both cases? */ void guest_load_tls(struct lguest *lg, unsigned long gtls) { - struct desc_struct *tls = &lg->gdt[GDT_ENTRY_TLS_MIN]; + struct desc_struct *tls = &lg->arch.gdt[GDT_ENTRY_TLS_MIN]; - lgread(lg, tls, gtls, sizeof(*tls)*GDT_ENTRY_TLS_ENTRIES); + __lgread(lg, tls, gtls, sizeof(*tls)*GDT_ENTRY_TLS_ENTRIES); fixup_gdt_table(lg, GDT_ENTRY_TLS_MIN, GDT_ENTRY_TLS_MAX+1); + /* Note that just the TLS entries have changed. */ lg->changed |= CHANGED_GDT_TLS; } +/*:*/ + +/*H:660 + * With this, we have finished the Host. + * + * Five of the seven parts of our task are complete. You have made it through + * the Bit of Despair (I think that's somewhere in the page table code, + * myself). + * + * Next, we examine "make Switcher". It's short, but intense. + */