1843e1988Sjohnlev /* 2843e1988Sjohnlev * CDDL HEADER START 3843e1988Sjohnlev * 4843e1988Sjohnlev * The contents of this file are subject to the terms of the 5843e1988Sjohnlev * Common Development and Distribution License (the "License"). 6843e1988Sjohnlev * You may not use this file except in compliance with the License. 7843e1988Sjohnlev * 8843e1988Sjohnlev * You can obtain a copy of the license at usr/src/OPENSOLARIS.LICENSE 9843e1988Sjohnlev * or http://www.opensolaris.org/os/licensing. 10843e1988Sjohnlev * See the License for the specific language governing permissions 11843e1988Sjohnlev * and limitations under the License. 12843e1988Sjohnlev * 13843e1988Sjohnlev * When distributing Covered Code, include this CDDL HEADER in each 14843e1988Sjohnlev * file and include the License file at usr/src/OPENSOLARIS.LICENSE. 15843e1988Sjohnlev * If applicable, add the following below this CDDL HEADER, with the 16843e1988Sjohnlev * fields enclosed by brackets "[]" replaced with your own identifying 17843e1988Sjohnlev * information: Portions Copyright [yyyy] [name of copyright owner] 18843e1988Sjohnlev * 19843e1988Sjohnlev * CDDL HEADER END 20843e1988Sjohnlev */ 21843e1988Sjohnlev /* 22*0d928757SGary Mills * Copyright (c) 2012 Gary Mills 23*0d928757SGary Mills * 2441afdfa7SKrishnendu Sadhukhan - Sun Microsystems * Copyright (c) 2010, Oracle and/or its affiliates. All rights reserved. 25843e1988Sjohnlev */ 26843e1988Sjohnlev 27843e1988Sjohnlev #include <sys/types.h> 28843e1988Sjohnlev #include <sys/clock.h> 29843e1988Sjohnlev #include <sys/psm.h> 30843e1988Sjohnlev #include <sys/archsystm.h> 31843e1988Sjohnlev #include <sys/machsystm.h> 32843e1988Sjohnlev #include <sys/compress.h> 33843e1988Sjohnlev #include <sys/modctl.h> 34843e1988Sjohnlev #include <sys/trap.h> 35843e1988Sjohnlev #include <sys/panic.h> 36843e1988Sjohnlev #include <sys/regset.h> 37843e1988Sjohnlev #include <sys/frame.h> 38843e1988Sjohnlev #include <sys/kobj.h> 39843e1988Sjohnlev #include <sys/apic.h> 4041afdfa7SKrishnendu Sadhukhan - Sun Microsystems #include <sys/apic_timer.h> 41843e1988Sjohnlev #include <sys/dumphdr.h> 42843e1988Sjohnlev #include <sys/mem.h> 43843e1988Sjohnlev #include <sys/x86_archext.h> 44843e1988Sjohnlev #include <sys/xpv_panic.h> 45843e1988Sjohnlev #include <sys/boot_console.h> 46843e1988Sjohnlev #include <sys/bootsvcs.h> 47843e1988Sjohnlev #include <sys/consdev.h> 48843e1988Sjohnlev #include <vm/hat_pte.h> 49843e1988Sjohnlev #include <vm/hat_i86.h> 50843e1988Sjohnlev 51843e1988Sjohnlev /* XXX: need to add a PAE version too, if we ever support both PAE and non */ 52843e1988Sjohnlev #if defined(__i386) 53843e1988Sjohnlev #define XPV_FILENAME "/boot/xen-syms" 54843e1988Sjohnlev #else 55843e1988Sjohnlev #define XPV_FILENAME "/boot/amd64/xen-syms" 56843e1988Sjohnlev #endif 57843e1988Sjohnlev #define XPV_MODNAME "xpv" 58843e1988Sjohnlev 59843e1988Sjohnlev int xpv_panicking = 0; 60843e1988Sjohnlev 61843e1988Sjohnlev struct module *xpv_module; 62843e1988Sjohnlev struct modctl *xpv_modctl; 63843e1988Sjohnlev 64843e1988Sjohnlev #define ALIGN(x, a) ((a) == 0 ? (uintptr_t)(x) : \ 65843e1988Sjohnlev (((uintptr_t)(x) + (uintptr_t)(a) - 1l) & ~((uintptr_t)(a) - 1l))) 66843e1988Sjohnlev 67843e1988Sjohnlev /* Pointer to the xpv_panic_info structure handed to us by Xen. */ 68843e1988Sjohnlev static struct panic_info *xpv_panic_info = NULL; 69843e1988Sjohnlev 70843e1988Sjohnlev /* Timer support */ 71843e1988Sjohnlev #define NSEC_SHIFT 5 72843e1988Sjohnlev #define T_XPV_TIMER 0xd1 73843e1988Sjohnlev #define XPV_TIMER_INTERVAL 1000 /* 1000 microseconds */ 74843e1988Sjohnlev static uint32_t *xpv_apicadr = NULL; 75843e1988Sjohnlev static uint_t nsec_scale; 76843e1988Sjohnlev 77843e1988Sjohnlev /* IDT support */ 78843e1988Sjohnlev #pragma align 16(xpv_panic_idt) 79843e1988Sjohnlev static gate_desc_t xpv_panic_idt[NIDT]; /* interrupt descriptor table */ 80843e1988Sjohnlev 81843e1988Sjohnlev /* Xen pagetables mapped into our HAT's ptable windows */ 82843e1988Sjohnlev static pfn_t ptable_pfn[MAX_NUM_LEVEL]; 83843e1988Sjohnlev 84843e1988Sjohnlev /* Number of MMU_PAGESIZE pages we're adding to the Solaris dump */ 85843e1988Sjohnlev static int xpv_dump_pages; 86843e1988Sjohnlev 87843e1988Sjohnlev /* 88a576ab5bSrab * There are up to two large swathes of RAM that we don't want to include 89a576ab5bSrab * in the dump: those that comprise the Xen version of segkpm. On 32-bit 90a576ab5bSrab * systems there is no such region of memory. On 64-bit systems, there 91a576ab5bSrab * should be just a single contiguous region that corresponds to all of 92a576ab5bSrab * physical memory. The tricky bit is that Xen's heap sometimes lives in 93a576ab5bSrab * the middle of their segkpm, and is mapped using only kpm-like addresses. 94a576ab5bSrab * In that case, we need to skip the swathes before and after Xen's heap. 95a576ab5bSrab */ 96a576ab5bSrab uintptr_t kpm1_low = 0; 97a576ab5bSrab uintptr_t kpm1_high = 0; 98a576ab5bSrab uintptr_t kpm2_low = 0; 99a576ab5bSrab uintptr_t kpm2_high = 0; 100a576ab5bSrab 101a576ab5bSrab /* 102843e1988Sjohnlev * Some commonly used values that we don't want to recompute over and over. 103843e1988Sjohnlev */ 104843e1988Sjohnlev static int xpv_panic_nptes[MAX_NUM_LEVEL]; 105843e1988Sjohnlev static ulong_t xpv_panic_cr3; 106843e1988Sjohnlev static uintptr_t xpv_end; 107843e1988Sjohnlev 108843e1988Sjohnlev static void xpv_panic_console_print(const char *fmt, ...); 109843e1988Sjohnlev static void (*xpv_panic_printf)(const char *, ...) = xpv_panic_console_print; 110843e1988Sjohnlev 111843e1988Sjohnlev #define CONSOLE_BUF_SIZE 256 112843e1988Sjohnlev static char console_buffer[CONSOLE_BUF_SIZE]; 113843e1988Sjohnlev static boolean_t use_polledio; 114843e1988Sjohnlev 115e4b86885SCheng Sean Ye /* 116e4b86885SCheng Sean Ye * Pointers to machine check panic info (if any). 117e4b86885SCheng Sean Ye */ 118e4b86885SCheng Sean Ye xpv_mca_panic_data_t *xpv_mca_panic_data = NULL; 119e4b86885SCheng Sean Ye 120843e1988Sjohnlev static void 121843e1988Sjohnlev xpv_panic_putc(int m) 122843e1988Sjohnlev { 123843e1988Sjohnlev struct cons_polledio *c = cons_polledio; 124843e1988Sjohnlev 125843e1988Sjohnlev /* This really shouldn't happen */ 126*0d928757SGary Mills if (boot_console_type(NULL) == CONS_HYPERVISOR) 127843e1988Sjohnlev return; 128843e1988Sjohnlev 129843e1988Sjohnlev if (use_polledio == B_TRUE) 130843e1988Sjohnlev c->cons_polledio_putchar(c->cons_polledio_argument, m); 131843e1988Sjohnlev else 132843e1988Sjohnlev bcons_putchar(m); 133843e1988Sjohnlev } 134843e1988Sjohnlev 135843e1988Sjohnlev static void 136843e1988Sjohnlev xpv_panic_puts(char *msg) 137843e1988Sjohnlev { 138843e1988Sjohnlev char *m; 139843e1988Sjohnlev 140843e1988Sjohnlev dump_timeleft = dump_timeout; 141843e1988Sjohnlev for (m = msg; *m; m++) 142843e1988Sjohnlev xpv_panic_putc((int)*m); 143843e1988Sjohnlev } 144843e1988Sjohnlev 145843e1988Sjohnlev static void 146843e1988Sjohnlev xpv_panic_console_print(const char *fmt, ...) 147843e1988Sjohnlev { 148843e1988Sjohnlev va_list ap; 149843e1988Sjohnlev 150843e1988Sjohnlev va_start(ap, fmt); 151843e1988Sjohnlev (void) vsnprintf(console_buffer, sizeof (console_buffer), fmt, ap); 152843e1988Sjohnlev va_end(ap); 153843e1988Sjohnlev 154843e1988Sjohnlev xpv_panic_puts(console_buffer); 155843e1988Sjohnlev } 156843e1988Sjohnlev 157843e1988Sjohnlev static void 158843e1988Sjohnlev xpv_panic_map(int level, pfn_t pfn) 159843e1988Sjohnlev { 160843e1988Sjohnlev x86pte_t pte, *pteptr; 161843e1988Sjohnlev 162843e1988Sjohnlev /* 163843e1988Sjohnlev * The provided pfn represents a level 'level' page table. Map it 164843e1988Sjohnlev * into the 'level' slot in the list of page table windows. 165843e1988Sjohnlev */ 166843e1988Sjohnlev pteptr = (x86pte_t *)PWIN_PTE_VA(level); 167843e1988Sjohnlev pte = pfn_to_pa(pfn) | PT_VALID; 168843e1988Sjohnlev 169843e1988Sjohnlev XPV_ALLOW_PAGETABLE_UPDATES(); 170843e1988Sjohnlev if (mmu.pae_hat) 171843e1988Sjohnlev *pteptr = pte; 172843e1988Sjohnlev else 173843e1988Sjohnlev *(x86pte32_t *)pteptr = pte; 174843e1988Sjohnlev XPV_DISALLOW_PAGETABLE_UPDATES(); 175843e1988Sjohnlev 176843e1988Sjohnlev mmu_tlbflush_entry(PWIN_VA(level)); 177843e1988Sjohnlev } 178843e1988Sjohnlev 179843e1988Sjohnlev /* 180843e1988Sjohnlev * Walk the page tables to find the pfn mapped by the given va. 181843e1988Sjohnlev */ 182843e1988Sjohnlev static pfn_t 183843e1988Sjohnlev xpv_va_walk(uintptr_t *vaddr) 184843e1988Sjohnlev { 185843e1988Sjohnlev int l, idx; 186843e1988Sjohnlev pfn_t pfn; 187843e1988Sjohnlev x86pte_t pte; 188843e1988Sjohnlev x86pte_t *ptep; 189843e1988Sjohnlev uintptr_t va = *vaddr; 190843e1988Sjohnlev uintptr_t scan_va; 191843e1988Sjohnlev caddr_t ptable_window; 192843e1988Sjohnlev static pfn_t toplevel_pfn; 193843e1988Sjohnlev static uintptr_t lastva; 194843e1988Sjohnlev 195843e1988Sjohnlev /* 196843e1988Sjohnlev * If we do anything other than a simple scan through memory, don't 197843e1988Sjohnlev * trust the mapped page tables. 198843e1988Sjohnlev */ 199843e1988Sjohnlev if (va != lastva + MMU_PAGESIZE) 200843e1988Sjohnlev for (l = mmu.max_level; l >= 0; l--) 201843e1988Sjohnlev ptable_pfn[l] = PFN_INVALID; 202843e1988Sjohnlev 203843e1988Sjohnlev toplevel_pfn = mmu_btop(xpv_panic_cr3); 204843e1988Sjohnlev 205843e1988Sjohnlev while (va < xpv_end && va >= *vaddr) { 206843e1988Sjohnlev /* Find the lowest table with any entry for va */ 207843e1988Sjohnlev pfn = toplevel_pfn; 208843e1988Sjohnlev for (l = mmu.max_level; l >= 0; l--) { 209843e1988Sjohnlev if (ptable_pfn[l] != pfn) { 210843e1988Sjohnlev xpv_panic_map(l, pfn); 211843e1988Sjohnlev ptable_pfn[l] = pfn; 212843e1988Sjohnlev } 213843e1988Sjohnlev 214843e1988Sjohnlev /* 215843e1988Sjohnlev * Search this pagetable for any mapping to an 216843e1988Sjohnlev * address >= va. 217843e1988Sjohnlev */ 218843e1988Sjohnlev ptable_window = PWIN_VA(l); 219843e1988Sjohnlev if (l == mmu.max_level && mmu.pae_hat) 220843e1988Sjohnlev ptable_window += 221843e1988Sjohnlev (xpv_panic_cr3 & MMU_PAGEOFFSET); 222843e1988Sjohnlev 223843e1988Sjohnlev idx = (va >> LEVEL_SHIFT(l)) & (xpv_panic_nptes[l] - 1); 224843e1988Sjohnlev scan_va = va; 225843e1988Sjohnlev while (idx < xpv_panic_nptes[l] && scan_va < xpv_end && 226843e1988Sjohnlev scan_va >= *vaddr) { 227843e1988Sjohnlev ptep = (x86pte_t *)(ptable_window + 228843e1988Sjohnlev (idx << mmu.pte_size_shift)); 229843e1988Sjohnlev pte = GET_PTE(ptep); 230843e1988Sjohnlev if (pte & PTE_VALID) 231843e1988Sjohnlev break; 232843e1988Sjohnlev idx++; 233843e1988Sjohnlev scan_va += mmu.level_size[l]; 234843e1988Sjohnlev } 235843e1988Sjohnlev 236843e1988Sjohnlev /* 237843e1988Sjohnlev * If there are no valid mappings in this table, we 238843e1988Sjohnlev * can skip to the end of the VA range it covers. 239843e1988Sjohnlev */ 240843e1988Sjohnlev if (idx == xpv_panic_nptes[l]) { 241843e1988Sjohnlev va = NEXT_ENTRY_VA(va, l + 1); 242843e1988Sjohnlev break; 243843e1988Sjohnlev } 244843e1988Sjohnlev 245a576ab5bSrab va = scan_va; 246a576ab5bSrab /* 247a576ab5bSrab * See if we've hit the end of the range. 248a576ab5bSrab */ 249a576ab5bSrab if (va >= xpv_end || va < *vaddr) 250a576ab5bSrab break; 251a576ab5bSrab 252843e1988Sjohnlev /* 253843e1988Sjohnlev * If this mapping is for a pagetable, we drop down 254843e1988Sjohnlev * to the next level in the hierarchy and look for 255843e1988Sjohnlev * a mapping in it. 256843e1988Sjohnlev */ 257843e1988Sjohnlev pfn = PTE2MFN(pte, l); 258843e1988Sjohnlev if (!PTE_ISPAGE(pte, l)) 259843e1988Sjohnlev continue; 260843e1988Sjohnlev 261843e1988Sjohnlev /* 262843e1988Sjohnlev * The APIC page is magic. Nothing to see here; 263843e1988Sjohnlev * move along. 264843e1988Sjohnlev */ 265843e1988Sjohnlev if (((uintptr_t)xpv_apicadr & MMU_PAGEMASK) == 266843e1988Sjohnlev (va & MMU_PAGEMASK)) { 267843e1988Sjohnlev va += MMU_PAGESIZE; 268843e1988Sjohnlev break; 269843e1988Sjohnlev } 270843e1988Sjohnlev 271a576ab5bSrab /* 272a576ab5bSrab * See if the address is within one of the two 273a576ab5bSrab * kpm-like regions we want to skip. 274a576ab5bSrab */ 275a576ab5bSrab if (va >= kpm1_low && va < kpm1_high) { 276a576ab5bSrab va = kpm1_high; 277a576ab5bSrab break; 278a576ab5bSrab } 279a576ab5bSrab if (va >= kpm2_low && va < kpm2_high) { 280a576ab5bSrab va = kpm2_high; 281843e1988Sjohnlev break; 282843e1988Sjohnlev } 283843e1988Sjohnlev 284843e1988Sjohnlev /* 285843e1988Sjohnlev * The Xen panic code only handles small pages. If 286843e1988Sjohnlev * this mapping is for a large page, we need to 287843e1988Sjohnlev * identify the consituent page that covers the 288843e1988Sjohnlev * specific VA we were looking for. 289843e1988Sjohnlev */ 290843e1988Sjohnlev if (l > 0) { 291843e1988Sjohnlev if (l > 1) 292843e1988Sjohnlev panic("Xen panic can't cope with " 293843e1988Sjohnlev "giant pages."); 294843e1988Sjohnlev idx = (va >> LEVEL_SHIFT(0)) & 295843e1988Sjohnlev (xpv_panic_nptes[0] - 1); 296843e1988Sjohnlev pfn += idx; 297843e1988Sjohnlev } 298843e1988Sjohnlev 299843e1988Sjohnlev *vaddr = va; 300843e1988Sjohnlev lastva = va; 301843e1988Sjohnlev return (pfn | PFN_IS_FOREIGN_MFN); 302843e1988Sjohnlev } 303843e1988Sjohnlev } 304843e1988Sjohnlev return (PFN_INVALID); 305843e1988Sjohnlev } 306843e1988Sjohnlev 307843e1988Sjohnlev /* 308843e1988Sjohnlev * Walk through the Xen VA space, finding pages that are mapped in. 309843e1988Sjohnlev * 310843e1988Sjohnlev * These pages all have MFNs rather than PFNs, meaning they may be outside 311843e1988Sjohnlev * the physical address space the kernel knows about, or they may collide 312843e1988Sjohnlev * with PFNs the kernel is using. 313843e1988Sjohnlev * 314843e1988Sjohnlev * The obvious trick of just adding the PFN_IS_FOREIGN_MFN bit to the MFNs 315843e1988Sjohnlev * to avoid collisions doesn't work. The pages need to be written to disk 316843e1988Sjohnlev * in PFN-order or savecore gets confused. We can't allocate memory to 317843e1988Sjohnlev * contruct a sorted pfn->VA reverse mapping, so we have to write the pages 318843e1988Sjohnlev * to disk in VA order. 319843e1988Sjohnlev * 320843e1988Sjohnlev * To square this circle, we simply make up PFNs for each of Xen's pages. 321843e1988Sjohnlev * We assign each mapped page a fake PFN in ascending order. These fake 322843e1988Sjohnlev * PFNs each have the FOREIGN bit set, ensuring that they fall outside the 323843e1988Sjohnlev * range of Solaris PFNs written by the kernel. 324843e1988Sjohnlev */ 325843e1988Sjohnlev int 326843e1988Sjohnlev dump_xpv_addr() 327843e1988Sjohnlev { 328843e1988Sjohnlev uintptr_t va; 329843e1988Sjohnlev mem_vtop_t mem_vtop; 330843e1988Sjohnlev 331843e1988Sjohnlev xpv_dump_pages = 0; 332843e1988Sjohnlev va = xen_virt_start; 333843e1988Sjohnlev 334843e1988Sjohnlev while (xpv_va_walk(&va) != PFN_INVALID) { 335843e1988Sjohnlev mem_vtop.m_as = &kas; 336843e1988Sjohnlev mem_vtop.m_va = (void *)va; 337843e1988Sjohnlev mem_vtop.m_pfn = (pfn_t)xpv_dump_pages | PFN_IS_FOREIGN_MFN; 338843e1988Sjohnlev 339843e1988Sjohnlev dumpvp_write(&mem_vtop, sizeof (mem_vtop_t)); 340843e1988Sjohnlev xpv_dump_pages++; 341843e1988Sjohnlev 342843e1988Sjohnlev va += MMU_PAGESIZE; 343843e1988Sjohnlev } 344843e1988Sjohnlev 345843e1988Sjohnlev /* 346843e1988Sjohnlev * Add the shared_info page. This page actually ends up in the 347843e1988Sjohnlev * dump twice: once for the Xen va and once for the Solaris va. 348843e1988Sjohnlev * This isn't ideal, but we don't know the address Xen is using for 349843e1988Sjohnlev * the page, so we can't share it. 350843e1988Sjohnlev */ 351843e1988Sjohnlev mem_vtop.m_as = &kas; 352843e1988Sjohnlev mem_vtop.m_va = HYPERVISOR_shared_info; 353843e1988Sjohnlev mem_vtop.m_pfn = (pfn_t)xpv_dump_pages | PFN_IS_FOREIGN_MFN; 354843e1988Sjohnlev dumpvp_write(&mem_vtop, sizeof (mem_vtop_t)); 355843e1988Sjohnlev xpv_dump_pages++; 356843e1988Sjohnlev 357843e1988Sjohnlev return (xpv_dump_pages); 358843e1988Sjohnlev } 359843e1988Sjohnlev 360843e1988Sjohnlev void 361843e1988Sjohnlev dump_xpv_pfn() 362843e1988Sjohnlev { 363843e1988Sjohnlev pfn_t pfn; 364843e1988Sjohnlev int cnt; 365843e1988Sjohnlev 366843e1988Sjohnlev for (cnt = 0; cnt < xpv_dump_pages; cnt++) { 367843e1988Sjohnlev pfn = (pfn_t)cnt | PFN_IS_FOREIGN_MFN; 368843e1988Sjohnlev dumpvp_write(&pfn, sizeof (pfn)); 369843e1988Sjohnlev } 370843e1988Sjohnlev } 371843e1988Sjohnlev 372843e1988Sjohnlev int 373843e1988Sjohnlev dump_xpv_data(void *dump_cbuf) 374843e1988Sjohnlev { 375843e1988Sjohnlev uintptr_t va; 376843e1988Sjohnlev uint32_t csize; 377843e1988Sjohnlev int cnt = 0; 378843e1988Sjohnlev 379843e1988Sjohnlev /* 380843e1988Sjohnlev * XXX: we should probably run this data through a UE check. The 381843e1988Sjohnlev * catch is that the UE code relies on on_trap() and getpfnum() 382843e1988Sjohnlev * working. 383843e1988Sjohnlev */ 384843e1988Sjohnlev va = xen_virt_start; 385843e1988Sjohnlev 386843e1988Sjohnlev while (xpv_va_walk(&va) != PFN_INVALID) { 387843e1988Sjohnlev csize = (uint32_t)compress((void *)va, dump_cbuf, PAGESIZE); 388843e1988Sjohnlev dumpvp_write(&csize, sizeof (uint32_t)); 389843e1988Sjohnlev dumpvp_write(dump_cbuf, csize); 390843e1988Sjohnlev if (dump_ioerr) { 391843e1988Sjohnlev dumphdr->dump_flags &= ~DF_COMPLETE; 392843e1988Sjohnlev return (cnt); 393843e1988Sjohnlev } 394843e1988Sjohnlev cnt++; 395843e1988Sjohnlev va += MMU_PAGESIZE; 396843e1988Sjohnlev } 397843e1988Sjohnlev 398843e1988Sjohnlev /* 399843e1988Sjohnlev * Finally, dump the shared_info page 400843e1988Sjohnlev */ 401843e1988Sjohnlev csize = (uint32_t)compress((void *)HYPERVISOR_shared_info, dump_cbuf, 402843e1988Sjohnlev PAGESIZE); 403843e1988Sjohnlev dumpvp_write(&csize, sizeof (uint32_t)); 404843e1988Sjohnlev dumpvp_write(dump_cbuf, csize); 405843e1988Sjohnlev if (dump_ioerr) 406843e1988Sjohnlev dumphdr->dump_flags &= ~DF_COMPLETE; 407843e1988Sjohnlev cnt++; 408843e1988Sjohnlev 409843e1988Sjohnlev return (cnt); 410843e1988Sjohnlev } 411843e1988Sjohnlev 412843e1988Sjohnlev static void * 413843e1988Sjohnlev showstack(void *fpreg, int xpv_only) 414843e1988Sjohnlev { 415843e1988Sjohnlev struct frame *fpp; 416843e1988Sjohnlev ulong_t off; 417843e1988Sjohnlev char *sym; 418843e1988Sjohnlev uintptr_t pc, fp, lastfp; 419843e1988Sjohnlev uintptr_t minaddr = min(KERNELBASE, xen_virt_start); 420843e1988Sjohnlev 421843e1988Sjohnlev fp = (uintptr_t)fpreg; 422843e1988Sjohnlev if (fp < minaddr) { 423843e1988Sjohnlev xpv_panic_printf("Bad frame ptr: 0x%p\n", fpreg); 424843e1988Sjohnlev return (fpreg); 425843e1988Sjohnlev } 426843e1988Sjohnlev 427843e1988Sjohnlev do { 428843e1988Sjohnlev fpp = (struct frame *)fp; 429843e1988Sjohnlev pc = fpp->fr_savpc; 430843e1988Sjohnlev 431843e1988Sjohnlev if ((xpv_only != 0) && 432843e1988Sjohnlev (fp > xpv_end || fp < xen_virt_start)) 433843e1988Sjohnlev break; 434843e1988Sjohnlev if ((sym = kobj_getsymname(pc, &off)) != NULL) 435843e1988Sjohnlev xpv_panic_printf("%08lx %s:%s+%lx\n", fp, 436843e1988Sjohnlev mod_containing_pc((caddr_t)pc), sym, off); 437843e1988Sjohnlev else if ((pc >= xen_virt_start) && (pc <= xpv_end)) 438843e1988Sjohnlev xpv_panic_printf("%08lx 0x%lx (in Xen)\n", fp, pc); 439843e1988Sjohnlev else 440843e1988Sjohnlev xpv_panic_printf("%08lx %lx\n", fp, pc); 441843e1988Sjohnlev 442843e1988Sjohnlev lastfp = fp; 443843e1988Sjohnlev fp = fpp->fr_savfp; 444843e1988Sjohnlev 445843e1988Sjohnlev /* 446843e1988Sjohnlev * Xen marks an exception frame by inverting the frame 447843e1988Sjohnlev * pointer. 448843e1988Sjohnlev */ 449843e1988Sjohnlev if (fp < lastfp) { 450843e1988Sjohnlev if ((~fp > minaddr) && ((~fp) ^ lastfp) < 0xfff) 451843e1988Sjohnlev fp = ~fp; 452843e1988Sjohnlev } 453843e1988Sjohnlev } while (fp > lastfp); 454843e1988Sjohnlev return ((void *)fp); 455843e1988Sjohnlev } 456843e1988Sjohnlev 457843e1988Sjohnlev void * 458843e1988Sjohnlev xpv_traceback(void *fpreg) 459843e1988Sjohnlev { 460843e1988Sjohnlev return (showstack(fpreg, 1)); 461843e1988Sjohnlev } 462843e1988Sjohnlev 463843e1988Sjohnlev #if defined(__amd64) 464843e1988Sjohnlev static void 465843e1988Sjohnlev xpv_panic_hypercall(ulong_t call) 466843e1988Sjohnlev { 467843e1988Sjohnlev panic("Illegally issued hypercall %d during panic!\n", (int)call); 468843e1988Sjohnlev } 469843e1988Sjohnlev #endif 470843e1988Sjohnlev 471843e1988Sjohnlev void 472843e1988Sjohnlev xpv_die(struct regs *rp) 473843e1988Sjohnlev { 474843e1988Sjohnlev struct panic_trap_info ti; 475843e1988Sjohnlev struct cregs creg; 476843e1988Sjohnlev 477843e1988Sjohnlev ti.trap_regs = rp; 478843e1988Sjohnlev ti.trap_type = rp->r_trapno; 479843e1988Sjohnlev 480843e1988Sjohnlev curthread->t_panic_trap = &ti; 481843e1988Sjohnlev if (ti.trap_type == T_PGFLT) { 482843e1988Sjohnlev getcregs(&creg); 483843e1988Sjohnlev ti.trap_addr = (caddr_t)creg.cr_cr2; 484843e1988Sjohnlev panic("Fatal pagefault at 0x%lx. fault addr=0x%p rp=0x%p", 485903a11ebSrh87107 rp->r_pc, (void *)ti.trap_addr, (void *)rp); 486843e1988Sjohnlev } else { 487843e1988Sjohnlev ti.trap_addr = (caddr_t)rp->r_pc; 488843e1988Sjohnlev panic("Fatal trap %ld at 0x%lx. rp=0x%p", rp->r_trapno, 489903a11ebSrh87107 rp->r_pc, (void *)rp); 490843e1988Sjohnlev } 491843e1988Sjohnlev } 492843e1988Sjohnlev 493843e1988Sjohnlev /* 494843e1988Sjohnlev * Build IDT to handle a Xen panic 495843e1988Sjohnlev */ 496843e1988Sjohnlev static void 497843e1988Sjohnlev switch_to_xpv_panic_idt() 498843e1988Sjohnlev { 499843e1988Sjohnlev int i; 500843e1988Sjohnlev desctbr_t idtr; 501843e1988Sjohnlev gate_desc_t *idt = xpv_panic_idt; 502843e1988Sjohnlev selector_t cs = get_cs_register(); 503843e1988Sjohnlev 504843e1988Sjohnlev for (i = 0; i < 32; i++) 5059844da31SSeth Goldberg set_gatesegd(&idt[i], &xpv_invaltrap, cs, SDT_SYSIGT, TRP_XPL, 5069844da31SSeth Goldberg 0); 507843e1988Sjohnlev 5089844da31SSeth Goldberg set_gatesegd(&idt[T_ZERODIV], &xpv_div0trap, cs, SDT_SYSIGT, TRP_XPL, 5099844da31SSeth Goldberg 0); 5109844da31SSeth Goldberg set_gatesegd(&idt[T_SGLSTP], &xpv_dbgtrap, cs, SDT_SYSIGT, TRP_XPL, 0); 5119844da31SSeth Goldberg set_gatesegd(&idt[T_NMIFLT], &xpv_nmiint, cs, SDT_SYSIGT, TRP_XPL, 0); 512843e1988Sjohnlev set_gatesegd(&idt[T_BOUNDFLT], &xpv_boundstrap, cs, SDT_SYSIGT, 5139844da31SSeth Goldberg TRP_XPL, 0); 5149844da31SSeth Goldberg set_gatesegd(&idt[T_ILLINST], &xpv_invoptrap, cs, SDT_SYSIGT, TRP_XPL, 5159844da31SSeth Goldberg 0); 5169844da31SSeth Goldberg set_gatesegd(&idt[T_NOEXTFLT], &xpv_ndptrap, cs, SDT_SYSIGT, TRP_XPL, 5179844da31SSeth Goldberg 0); 5189844da31SSeth Goldberg set_gatesegd(&idt[T_TSSFLT], &xpv_invtsstrap, cs, SDT_SYSIGT, TRP_XPL, 5199844da31SSeth Goldberg 0); 5209844da31SSeth Goldberg set_gatesegd(&idt[T_SEGFLT], &xpv_segnptrap, cs, SDT_SYSIGT, TRP_XPL, 5219844da31SSeth Goldberg 0); 5229844da31SSeth Goldberg set_gatesegd(&idt[T_STKFLT], &xpv_stktrap, cs, SDT_SYSIGT, TRP_XPL, 0); 5239844da31SSeth Goldberg set_gatesegd(&idt[T_GPFLT], &xpv_gptrap, cs, SDT_SYSIGT, TRP_XPL, 0); 5249844da31SSeth Goldberg set_gatesegd(&idt[T_PGFLT], &xpv_pftrap, cs, SDT_SYSIGT, TRP_XPL, 0); 5259844da31SSeth Goldberg set_gatesegd(&idt[T_EXTERRFLT], &xpv_ndperr, cs, SDT_SYSIGT, TRP_XPL, 5269844da31SSeth Goldberg 0); 5279844da31SSeth Goldberg set_gatesegd(&idt[T_ALIGNMENT], &xpv_achktrap, cs, SDT_SYSIGT, TRP_XPL, 5289844da31SSeth Goldberg 0); 5299844da31SSeth Goldberg set_gatesegd(&idt[T_MCE], &xpv_mcetrap, cs, SDT_SYSIGT, TRP_XPL, 0); 5309844da31SSeth Goldberg set_gatesegd(&idt[T_SIMDFPE], &xpv_xmtrap, cs, SDT_SYSIGT, TRP_XPL, 0); 531843e1988Sjohnlev 532843e1988Sjohnlev /* 533843e1988Sjohnlev * We have no double fault handler. Any single fault represents a 534843e1988Sjohnlev * catastrophic failure for us, so there is no attempt to handle 535843e1988Sjohnlev * them cleanly: we just print a message and reboot. If we 536843e1988Sjohnlev * encounter a second fault while doing that, there is nothing 537843e1988Sjohnlev * else we can do. 538843e1988Sjohnlev */ 539843e1988Sjohnlev 540843e1988Sjohnlev /* 541843e1988Sjohnlev * Be prepared to absorb any stray device interrupts received 542843e1988Sjohnlev * while writing the core to disk. 543843e1988Sjohnlev */ 544843e1988Sjohnlev for (i = 33; i < NIDT; i++) 545843e1988Sjohnlev set_gatesegd(&idt[i], &xpv_surprise_intr, cs, SDT_SYSIGT, 5469844da31SSeth Goldberg TRP_XPL, 0); 547843e1988Sjohnlev 548843e1988Sjohnlev /* The one interrupt we expect to get is from the APIC timer. */ 549843e1988Sjohnlev set_gatesegd(&idt[T_XPV_TIMER], &xpv_timer_trap, cs, SDT_SYSIGT, 5509844da31SSeth Goldberg TRP_XPL, 0); 551843e1988Sjohnlev 552843e1988Sjohnlev idtr.dtr_base = (uintptr_t)xpv_panic_idt; 553843e1988Sjohnlev idtr.dtr_limit = sizeof (xpv_panic_idt) - 1; 554843e1988Sjohnlev wr_idtr(&idtr); 555843e1988Sjohnlev 556843e1988Sjohnlev #if defined(__amd64) 557843e1988Sjohnlev /* Catch any hypercalls. */ 558843e1988Sjohnlev wrmsr(MSR_AMD_LSTAR, (uintptr_t)xpv_panic_hypercall); 559843e1988Sjohnlev wrmsr(MSR_AMD_CSTAR, (uintptr_t)xpv_panic_hypercall); 560843e1988Sjohnlev #endif 561843e1988Sjohnlev } 562843e1988Sjohnlev 563843e1988Sjohnlev static void 564843e1988Sjohnlev xpv_apic_clkinit() 565843e1988Sjohnlev { 566843e1988Sjohnlev uint_t apic_ticks = 0; 567843e1988Sjohnlev 568843e1988Sjohnlev /* 569843e1988Sjohnlev * Measure how many APIC ticks there are within a fixed time 570843e1988Sjohnlev * period. We're going to be fairly coarse here. This timer is 571843e1988Sjohnlev * just being used to detect a stalled panic, so as long as we have 572843e1988Sjohnlev * the right order of magnitude, everything should be fine. 573843e1988Sjohnlev */ 574843e1988Sjohnlev xpv_apicadr[APIC_SPUR_INT_REG] = AV_UNIT_ENABLE | APIC_SPUR_INTR; 575843e1988Sjohnlev xpv_apicadr[APIC_LOCAL_TIMER] = AV_MASK; 576843e1988Sjohnlev xpv_apicadr[APIC_INT_VECT0] = AV_MASK; /* local intr reg 0 */ 577843e1988Sjohnlev 578843e1988Sjohnlev xpv_apicadr[APIC_DIVIDE_REG] = 0; 579843e1988Sjohnlev xpv_apicadr[APIC_INIT_COUNT] = APIC_MAXVAL; 580843e1988Sjohnlev drv_usecwait(XPV_TIMER_INTERVAL); 581843e1988Sjohnlev apic_ticks = APIC_MAXVAL - xpv_apicadr[APIC_CURR_COUNT]; 582843e1988Sjohnlev 583843e1988Sjohnlev /* 584843e1988Sjohnlev * apic_ticks now represents roughly how many apic ticks comprise 585843e1988Sjohnlev * one timeout interval. Program the timer to send us an interrupt 586843e1988Sjohnlev * every time that interval expires. 587843e1988Sjohnlev */ 58841afdfa7SKrishnendu Sadhukhan - Sun Microsystems xpv_apicadr[APIC_LOCAL_TIMER] = T_XPV_TIMER | AV_PERIODIC; 589843e1988Sjohnlev xpv_apicadr[APIC_INIT_COUNT] = apic_ticks; 590843e1988Sjohnlev xpv_apicadr[APIC_EOI_REG] = 0; 591843e1988Sjohnlev } 592843e1988Sjohnlev 593843e1988Sjohnlev void 594843e1988Sjohnlev xpv_timer_tick(void) 595843e1988Sjohnlev { 596843e1988Sjohnlev static int ticks = 0; 597843e1988Sjohnlev 598843e1988Sjohnlev if (ticks++ >= MICROSEC / XPV_TIMER_INTERVAL) { 599843e1988Sjohnlev ticks = 0; 600843e1988Sjohnlev if (dump_timeleft && (--dump_timeleft == 0)) 601843e1988Sjohnlev panic("Xen panic timeout\n"); 602843e1988Sjohnlev } 603843e1988Sjohnlev xpv_apicadr[APIC_EOI_REG] = 0; 604843e1988Sjohnlev } 605843e1988Sjohnlev 606843e1988Sjohnlev void 607843e1988Sjohnlev xpv_interrupt(void) 608843e1988Sjohnlev { 609843e1988Sjohnlev #ifdef DEBUG 610843e1988Sjohnlev static int cnt = 0; 611843e1988Sjohnlev 612843e1988Sjohnlev if (cnt++ < 10) 613843e1988Sjohnlev xpv_panic_printf("Unexpected interrupt received.\n"); 614843e1988Sjohnlev if ((cnt < 1000) && ((cnt % 100) == 0)) 615843e1988Sjohnlev xpv_panic_printf("%d unexpected interrupts received.\n", cnt); 616843e1988Sjohnlev #endif 617843e1988Sjohnlev 618843e1988Sjohnlev xpv_apicadr[APIC_EOI_REG] = 0; 619843e1988Sjohnlev } 620843e1988Sjohnlev 621843e1988Sjohnlev /* 622843e1988Sjohnlev * Managing time in panic context is trivial. We only have a single CPU, 623843e1988Sjohnlev * we never get rescheduled, we never get suspended. We just need to 624843e1988Sjohnlev * convert clock ticks into nanoseconds. 625843e1988Sjohnlev */ 626843e1988Sjohnlev static hrtime_t 627843e1988Sjohnlev xpv_panic_gethrtime(void) 628843e1988Sjohnlev { 629843e1988Sjohnlev hrtime_t tsc, hrt; 630843e1988Sjohnlev unsigned int *l = (unsigned int *)&(tsc); 631843e1988Sjohnlev 632843e1988Sjohnlev tsc = __rdtsc_insn(); 633843e1988Sjohnlev hrt = (mul32(l[1], nsec_scale) << NSEC_SHIFT) + 634843e1988Sjohnlev (mul32(l[0], nsec_scale) >> (32 - NSEC_SHIFT)); 635843e1988Sjohnlev 636843e1988Sjohnlev return (hrt); 637843e1988Sjohnlev } 638843e1988Sjohnlev 639843e1988Sjohnlev static void 640843e1988Sjohnlev xpv_panic_time_init() 641843e1988Sjohnlev { 642843e1988Sjohnlev nsec_scale = 643843e1988Sjohnlev CPU->cpu_m.mcpu_vcpu_info->time.tsc_to_system_mul >> NSEC_SHIFT; 644843e1988Sjohnlev 645843e1988Sjohnlev gethrtimef = xpv_panic_gethrtime; 646843e1988Sjohnlev } 647843e1988Sjohnlev 648843e1988Sjohnlev static void 649843e1988Sjohnlev xpv_panicsys(struct regs *rp, char *fmt, ...) 650843e1988Sjohnlev { 651843e1988Sjohnlev extern void panicsys(const char *, va_list, struct regs *, int); 652843e1988Sjohnlev va_list alist; 653843e1988Sjohnlev 654843e1988Sjohnlev va_start(alist, fmt); 655843e1988Sjohnlev panicsys(fmt, alist, rp, 1); 656843e1988Sjohnlev va_end(alist); 657843e1988Sjohnlev } 658843e1988Sjohnlev 659843e1988Sjohnlev void 660843e1988Sjohnlev xpv_do_panic(void *arg) 661843e1988Sjohnlev { 662843e1988Sjohnlev struct panic_info *pip = (struct panic_info *)arg; 663843e1988Sjohnlev int l; 664843e1988Sjohnlev struct cregs creg; 665843e1988Sjohnlev #if defined(__amd64) 666843e1988Sjohnlev extern uintptr_t postbootkernelbase; 667843e1988Sjohnlev #endif 668843e1988Sjohnlev 669843e1988Sjohnlev if (xpv_panicking++ > 0) 670843e1988Sjohnlev panic("multiple calls to xpv_do_panic()"); 671843e1988Sjohnlev 672843e1988Sjohnlev /* 673843e1988Sjohnlev * Indicate to the underlying panic framework that a panic has been 674843e1988Sjohnlev * initiated. This is ordinarily done as part of vpanic(). Since 675843e1988Sjohnlev * we already have all the register state saved by the hypervisor, 676843e1988Sjohnlev * we skip that and jump straight into the panic processing code. 677e4b86885SCheng Sean Ye * 678e4b86885SCheng Sean Ye * XXX If another thread grabs and wins the panic_quiesce trigger 679e4b86885SCheng Sean Ye * then we'll have two threads in panicsys believing they are in 680e4b86885SCheng Sean Ye * charge of the panic attempt! 681843e1988Sjohnlev */ 682843e1988Sjohnlev (void) panic_trigger(&panic_quiesce); 683843e1988Sjohnlev 684843e1988Sjohnlev #if defined(__amd64) 685843e1988Sjohnlev /* 686843e1988Sjohnlev * bzero() and bcopy() get unhappy when asked to operate on 687843e1988Sjohnlev * addresses outside of the kernel. At this point Xen is really a 688843e1988Sjohnlev * part of the kernel, so we update the routines' notion of where 689843e1988Sjohnlev * the kernel starts. 690843e1988Sjohnlev */ 691843e1988Sjohnlev postbootkernelbase = xen_virt_start; 692843e1988Sjohnlev #endif 693843e1988Sjohnlev 694843e1988Sjohnlev #if defined(HYPERVISOR_VIRT_END) 695843e1988Sjohnlev xpv_end = HYPERVISOR_VIRT_END; 696843e1988Sjohnlev #else 697843e1988Sjohnlev xpv_end = (uintptr_t)UINTPTR_MAX - sizeof (uintptr_t); 698843e1988Sjohnlev #endif 699843e1988Sjohnlev 700843e1988Sjohnlev /* 701843e1988Sjohnlev * If we were redirecting console output to the hypervisor, we have 702843e1988Sjohnlev * to stop. 703843e1988Sjohnlev */ 704843e1988Sjohnlev use_polledio = B_FALSE; 705*0d928757SGary Mills if (boot_console_type(NULL) == CONS_HYPERVISOR) { 706843e1988Sjohnlev bcons_device_change(CONS_HYPERVISOR); 707843e1988Sjohnlev } else if (cons_polledio != NULL && 708843e1988Sjohnlev cons_polledio->cons_polledio_putchar != NULL) { 709843e1988Sjohnlev if (cons_polledio->cons_polledio_enter != NULL) 710843e1988Sjohnlev cons_polledio->cons_polledio_enter( 711843e1988Sjohnlev cons_polledio->cons_polledio_argument); 712843e1988Sjohnlev use_polledio = 1; 713843e1988Sjohnlev } 714843e1988Sjohnlev 715843e1988Sjohnlev /* Make sure we handle all console output from here on. */ 716843e1988Sjohnlev sysp->bsvc_putchar = xpv_panic_putc; 717843e1988Sjohnlev 718843e1988Sjohnlev /* 719843e1988Sjohnlev * If we find an unsupported panic_info structure, there's not much 720843e1988Sjohnlev * we can do other than complain, plow on, and hope for the best. 721843e1988Sjohnlev */ 722843e1988Sjohnlev if (pip->pi_version != PANIC_INFO_VERSION) 723843e1988Sjohnlev xpv_panic_printf("Warning: Xen is using an unsupported " 724843e1988Sjohnlev "version of the panic_info structure.\n"); 725843e1988Sjohnlev 726843e1988Sjohnlev xpv_panic_info = pip; 727843e1988Sjohnlev 728a576ab5bSrab #if defined(__amd64) 729a576ab5bSrab kpm1_low = (uintptr_t)xpv_panic_info->pi_ram_start; 730a576ab5bSrab if (xpv_panic_info->pi_xen_start == NULL) { 731a576ab5bSrab kpm1_high = (uintptr_t)xpv_panic_info->pi_ram_end; 732a576ab5bSrab } else { 733a576ab5bSrab kpm1_high = (uintptr_t)xpv_panic_info->pi_xen_start; 734a576ab5bSrab kpm2_low = (uintptr_t)xpv_panic_info->pi_xen_end; 735a576ab5bSrab kpm2_high = (uintptr_t)xpv_panic_info->pi_ram_end; 736a576ab5bSrab } 737a576ab5bSrab #endif 738a576ab5bSrab 739843e1988Sjohnlev /* 740843e1988Sjohnlev * Make sure we are running on the Solaris %gs. The Xen panic code 741843e1988Sjohnlev * should already have set up the GDT properly. 742843e1988Sjohnlev */ 743843e1988Sjohnlev xpv_panic_resetgs(); 744843e1988Sjohnlev #if defined(__amd64) 745843e1988Sjohnlev wrmsr(MSR_AMD_GSBASE, (uint64_t)&cpus[0]); 746843e1988Sjohnlev #endif 747843e1988Sjohnlev 748843e1988Sjohnlev xpv_panic_time_init(); 749843e1988Sjohnlev 750843e1988Sjohnlev /* 751843e1988Sjohnlev * Switch to our own IDT, avoiding any accidental returns to Xen 752843e1988Sjohnlev * world. 753843e1988Sjohnlev */ 754843e1988Sjohnlev switch_to_xpv_panic_idt(); 755843e1988Sjohnlev 756843e1988Sjohnlev /* 757843e1988Sjohnlev * Initialize the APIC timer, which is used to detect a hung dump 758843e1988Sjohnlev * attempt. 759843e1988Sjohnlev */ 760843e1988Sjohnlev xpv_apicadr = pip->pi_apic; 761843e1988Sjohnlev xpv_apic_clkinit(); 762843e1988Sjohnlev 763843e1988Sjohnlev /* 764843e1988Sjohnlev * Set up a few values that we'll need repeatedly. 765843e1988Sjohnlev */ 766843e1988Sjohnlev getcregs(&creg); 767843e1988Sjohnlev xpv_panic_cr3 = creg.cr_cr3; 768843e1988Sjohnlev for (l = mmu.max_level; l >= 0; l--) 769843e1988Sjohnlev xpv_panic_nptes[l] = mmu.ptes_per_table; 770843e1988Sjohnlev #ifdef __i386 771843e1988Sjohnlev if (mmu.pae_hat) 772843e1988Sjohnlev xpv_panic_nptes[mmu.max_level] = 4; 773843e1988Sjohnlev #endif 774843e1988Sjohnlev 775843e1988Sjohnlev /* Add the fake Xen module to the module list */ 776843e1988Sjohnlev if (xpv_module != NULL) { 777843e1988Sjohnlev extern int last_module_id; 778843e1988Sjohnlev 779843e1988Sjohnlev xpv_modctl->mod_id = last_module_id++; 780843e1988Sjohnlev xpv_modctl->mod_next = &modules; 781843e1988Sjohnlev xpv_modctl->mod_prev = modules.mod_prev; 782843e1988Sjohnlev modules.mod_prev->mod_next = xpv_modctl; 783843e1988Sjohnlev modules.mod_prev = xpv_modctl; 784843e1988Sjohnlev } 785e4b86885SCheng Sean Ye 786e4b86885SCheng Sean Ye if (pip->pi_mca.mpd_magic == MCA_PANICDATA_MAGIC) 787e4b86885SCheng Sean Ye xpv_mca_panic_data = &pip->pi_mca; 788e4b86885SCheng Sean Ye 789843e1988Sjohnlev xpv_panic_printf = printf; 790843e1988Sjohnlev xpv_panicsys((struct regs *)pip->pi_regs, pip->pi_panicstr); 791843e1988Sjohnlev xpv_panic_printf("Failed to reboot following panic.\n"); 792843e1988Sjohnlev for (;;) 793843e1988Sjohnlev ; 794843e1988Sjohnlev } 795843e1988Sjohnlev 796843e1988Sjohnlev /* 797843e1988Sjohnlev * Set up the necessary data structures to pretend that the Xen hypervisor 798843e1988Sjohnlev * is a loadable module, allowing mdb to find the Xen symbols in a crash 799843e1988Sjohnlev * dump. Since these symbols all map to VA space Solaris doesn't normally 800843e1988Sjohnlev * have access to, we don't link these structures into the kernel's lists 801843e1988Sjohnlev * until/unless we hit a Xen panic. 802843e1988Sjohnlev * 803843e1988Sjohnlev * The observant reader will note a striking amount of overlap between this 804843e1988Sjohnlev * code and that found in krtld. While it would be handy if we could just 805843e1988Sjohnlev * ask krtld to do this work for us, it's not that simple. Among the 806843e1988Sjohnlev * complications: we're not actually loading the text here (grub did it at 807843e1988Sjohnlev * boot), the .text section is writable, there are no relocations to do, 808843e1988Sjohnlev * none of the module text/data is in readable memory, etc. Training krtld 809843e1988Sjohnlev * to deal with this weird module is as complicated, and more risky, than 810843e1988Sjohnlev * reimplementing the necessary subset of it here. 811843e1988Sjohnlev */ 812843e1988Sjohnlev static void 813843e1988Sjohnlev init_xen_module() 814843e1988Sjohnlev { 815843e1988Sjohnlev struct _buf *file = NULL; 816843e1988Sjohnlev struct module *mp; 817843e1988Sjohnlev struct modctl *mcp; 818843e1988Sjohnlev int i, shn; 819843e1988Sjohnlev Shdr *shp, *ctf_shp; 820843e1988Sjohnlev char *names = NULL; 821843e1988Sjohnlev size_t n, namesize, text_align, data_align; 822843e1988Sjohnlev #if defined(__amd64) 823843e1988Sjohnlev const char machine = EM_AMD64; 824843e1988Sjohnlev #else 825843e1988Sjohnlev const char machine = EM_386; 826843e1988Sjohnlev #endif 827843e1988Sjohnlev 828843e1988Sjohnlev /* Allocate and init the module structure */ 829843e1988Sjohnlev mp = kmem_zalloc(sizeof (*mp), KM_SLEEP); 830843e1988Sjohnlev mp->filename = kobj_zalloc(strlen(XPV_FILENAME) + 1, KM_SLEEP); 831843e1988Sjohnlev (void) strcpy(mp->filename, XPV_FILENAME); 832843e1988Sjohnlev 833843e1988Sjohnlev /* Allocate and init the modctl structure */ 834843e1988Sjohnlev mcp = kmem_zalloc(sizeof (*mcp), KM_SLEEP); 835843e1988Sjohnlev mcp->mod_modname = kobj_zalloc(strlen(XPV_MODNAME) + 1, KM_SLEEP); 836843e1988Sjohnlev (void) strcpy(mcp->mod_modname, XPV_MODNAME); 837843e1988Sjohnlev mcp->mod_filename = kobj_zalloc(strlen(XPV_FILENAME) + 1, KM_SLEEP); 838843e1988Sjohnlev (void) strcpy(mcp->mod_filename, XPV_FILENAME); 839843e1988Sjohnlev mcp->mod_inprogress_thread = (kthread_id_t)-1; 840843e1988Sjohnlev mcp->mod_ref = 1; 841843e1988Sjohnlev mcp->mod_loaded = 1; 842843e1988Sjohnlev mcp->mod_loadcnt = 1; 843843e1988Sjohnlev mcp->mod_mp = mp; 844843e1988Sjohnlev 845843e1988Sjohnlev /* 846843e1988Sjohnlev * Try to open a Xen image that hasn't had its symbol and CTF 847843e1988Sjohnlev * information stripped off. 848843e1988Sjohnlev */ 849843e1988Sjohnlev file = kobj_open_file(XPV_FILENAME); 850843e1988Sjohnlev if (file == (struct _buf *)-1) { 851843e1988Sjohnlev file = NULL; 852843e1988Sjohnlev goto err; 853843e1988Sjohnlev } 854843e1988Sjohnlev 855843e1988Sjohnlev /* 856843e1988Sjohnlev * Read the header and ensure that this is an ELF file for the 857843e1988Sjohnlev * proper ISA. If it's not, somebody has done something very 858843e1988Sjohnlev * stupid. Why bother? See Mencken. 859843e1988Sjohnlev */ 860843e1988Sjohnlev if (kobj_read_file(file, (char *)&mp->hdr, sizeof (mp->hdr), 0) < 0) 861843e1988Sjohnlev goto err; 862843e1988Sjohnlev for (i = 0; i < SELFMAG; i++) 863843e1988Sjohnlev if (mp->hdr.e_ident[i] != ELFMAG[i]) 864843e1988Sjohnlev goto err; 865843e1988Sjohnlev if ((mp->hdr.e_ident[EI_DATA] != ELFDATA2LSB) || 866843e1988Sjohnlev (mp->hdr.e_machine != machine)) 867843e1988Sjohnlev goto err; 868843e1988Sjohnlev 869843e1988Sjohnlev /* Read in the section headers */ 870843e1988Sjohnlev n = mp->hdr.e_shentsize * mp->hdr.e_shnum; 871843e1988Sjohnlev mp->shdrs = kmem_zalloc(n, KM_SLEEP); 872843e1988Sjohnlev if (kobj_read_file(file, mp->shdrs, n, mp->hdr.e_shoff) < 0) 873843e1988Sjohnlev goto err; 874843e1988Sjohnlev 875843e1988Sjohnlev /* Read the section names */ 876843e1988Sjohnlev shp = (Shdr *)(mp->shdrs + mp->hdr.e_shstrndx * mp->hdr.e_shentsize); 877843e1988Sjohnlev namesize = shp->sh_size; 878843e1988Sjohnlev names = kmem_zalloc(shp->sh_size, KM_SLEEP); 879843e1988Sjohnlev if (kobj_read_file(file, names, shp->sh_size, shp->sh_offset) < 0) 880843e1988Sjohnlev goto err; 881843e1988Sjohnlev 882843e1988Sjohnlev /* 883843e1988Sjohnlev * Fill in the text and data size fields. 884843e1988Sjohnlev */ 885843e1988Sjohnlev ctf_shp = NULL; 886843e1988Sjohnlev text_align = data_align = 0; 887843e1988Sjohnlev for (shn = 1; shn < mp->hdr.e_shnum; shn++) { 888843e1988Sjohnlev shp = (Shdr *)(mp->shdrs + shn * mp->hdr.e_shentsize); 889843e1988Sjohnlev 890843e1988Sjohnlev /* Sanity check the offset of the section name */ 891843e1988Sjohnlev if (shp->sh_name >= namesize) 892843e1988Sjohnlev continue; 893843e1988Sjohnlev 894843e1988Sjohnlev /* If we find the symtab section, remember it for later. */ 895843e1988Sjohnlev if (shp->sh_type == SHT_SYMTAB) { 896843e1988Sjohnlev mp->symtbl_section = shn; 897843e1988Sjohnlev mp->symhdr = shp; 898843e1988Sjohnlev continue; 899843e1988Sjohnlev } 900843e1988Sjohnlev 901843e1988Sjohnlev /* If we find the CTF section, remember it for later. */ 902843e1988Sjohnlev if ((shp->sh_size != 0) && 903843e1988Sjohnlev (strcmp(names + shp->sh_name, ".SUNW_ctf") == 0)) { 904843e1988Sjohnlev ctf_shp = shp; 905843e1988Sjohnlev continue; 906843e1988Sjohnlev } 907843e1988Sjohnlev 908843e1988Sjohnlev if (!(shp->sh_flags & SHF_ALLOC)) 909843e1988Sjohnlev continue; 910843e1988Sjohnlev 911843e1988Sjohnlev /* 912843e1988Sjohnlev * Xen marks its text section as writable, so we need to 913843e1988Sjohnlev * look for the name - not just the flag. 914843e1988Sjohnlev */ 915843e1988Sjohnlev if ((strcmp(&names[shp->sh_name], ".text") != NULL) && 916843e1988Sjohnlev (shp->sh_flags & SHF_WRITE) != 0) { 917843e1988Sjohnlev if (shp->sh_addralign > data_align) 918843e1988Sjohnlev data_align = shp->sh_addralign; 919843e1988Sjohnlev mp->data_size = ALIGN(mp->data_size, data_align); 920843e1988Sjohnlev mp->data_size += ALIGN(shp->sh_size, 8); 921843e1988Sjohnlev if (mp->data == NULL || mp->data > (char *)shp->sh_addr) 922843e1988Sjohnlev mp->data = (char *)shp->sh_addr; 923843e1988Sjohnlev } else { 924843e1988Sjohnlev if (shp->sh_addralign > text_align) 925843e1988Sjohnlev text_align = shp->sh_addralign; 926843e1988Sjohnlev mp->text_size = ALIGN(mp->text_size, text_align); 927843e1988Sjohnlev mp->text_size += ALIGN(shp->sh_size, 8); 928843e1988Sjohnlev if (mp->text == NULL || mp->text > (char *)shp->sh_addr) 929843e1988Sjohnlev mp->text = (char *)shp->sh_addr; 930843e1988Sjohnlev } 931843e1988Sjohnlev } 932843e1988Sjohnlev kmem_free(names, namesize); 933843e1988Sjohnlev names = NULL; 9348cdfbd11Snn35248 shp = NULL; 935843e1988Sjohnlev mcp->mod_text = mp->text; 936843e1988Sjohnlev mcp->mod_text_size = mp->text_size; 937843e1988Sjohnlev 938843e1988Sjohnlev /* 939843e1988Sjohnlev * If we have symbol table and string table sections, read them in 940843e1988Sjohnlev * now. If we don't, we just plow on. We'll still get a valid 941843e1988Sjohnlev * core dump, but finding anything useful will be just a bit 942843e1988Sjohnlev * harder. 943843e1988Sjohnlev * 944843e1988Sjohnlev * Note: we don't bother with a hash table. We'll never do a 945843e1988Sjohnlev * symbol lookup unless we crash, and then mdb creates its own. We 946843e1988Sjohnlev * also don't try to perform any relocations. Xen should be loaded 947843e1988Sjohnlev * exactly where the ELF file indicates, and the symbol information 948843e1988Sjohnlev * in the file should be complete and correct already. Static 949843e1988Sjohnlev * linking ain't all bad. 950843e1988Sjohnlev */ 951843e1988Sjohnlev if ((mp->symhdr != NULL) && (mp->symhdr->sh_link < mp->hdr.e_shnum)) { 952843e1988Sjohnlev mp->strhdr = (Shdr *) 953843e1988Sjohnlev (mp->shdrs + mp->symhdr->sh_link * mp->hdr.e_shentsize); 954843e1988Sjohnlev mp->nsyms = mp->symhdr->sh_size / mp->symhdr->sh_entsize; 955843e1988Sjohnlev 956843e1988Sjohnlev /* Allocate space for the symbol table and strings. */ 957843e1988Sjohnlev mp->symsize = mp->symhdr->sh_size + 958843e1988Sjohnlev mp->nsyms * sizeof (symid_t) + mp->strhdr->sh_size; 959843e1988Sjohnlev mp->symspace = kmem_zalloc(mp->symsize, KM_SLEEP); 960843e1988Sjohnlev mp->symtbl = mp->symspace; 961843e1988Sjohnlev mp->strings = (char *)(mp->symtbl + mp->symhdr->sh_size); 962843e1988Sjohnlev 963843e1988Sjohnlev if ((kobj_read_file(file, mp->symtbl, 964843e1988Sjohnlev mp->symhdr->sh_size, mp->symhdr->sh_offset) < 0) || 965843e1988Sjohnlev (kobj_read_file(file, mp->strings, 966843e1988Sjohnlev mp->strhdr->sh_size, mp->strhdr->sh_offset) < 0)) 967843e1988Sjohnlev goto err; 968843e1988Sjohnlev } 969843e1988Sjohnlev 970843e1988Sjohnlev /* 971843e1988Sjohnlev * Read in the CTF section 972843e1988Sjohnlev */ 973843e1988Sjohnlev if ((ctf_shp != NULL) && ((moddebug & MODDEBUG_NOCTF) == 0)) { 9748cdfbd11Snn35248 mp->ctfdata = kmem_zalloc(ctf_shp->sh_size, KM_SLEEP); 975843e1988Sjohnlev mp->ctfsize = ctf_shp->sh_size; 976843e1988Sjohnlev if (kobj_read_file(file, mp->ctfdata, mp->ctfsize, 977843e1988Sjohnlev ctf_shp->sh_offset) < 0) 978843e1988Sjohnlev goto err; 979843e1988Sjohnlev } 980843e1988Sjohnlev 981843e1988Sjohnlev kobj_close_file(file); 982843e1988Sjohnlev 983843e1988Sjohnlev xpv_module = mp; 984843e1988Sjohnlev xpv_modctl = mcp; 985843e1988Sjohnlev return; 986843e1988Sjohnlev 987843e1988Sjohnlev err: 988843e1988Sjohnlev cmn_err(CE_WARN, "Failed to initialize xpv module."); 989843e1988Sjohnlev if (file != NULL) 990843e1988Sjohnlev kobj_close_file(file); 991843e1988Sjohnlev 992843e1988Sjohnlev kmem_free(mp->filename, strlen(XPV_FILENAME) + 1); 993843e1988Sjohnlev if (mp->shdrs != NULL) 994843e1988Sjohnlev kmem_free(mp->shdrs, mp->hdr.e_shentsize * mp->hdr.e_shnum); 995843e1988Sjohnlev if (mp->symspace != NULL) 996843e1988Sjohnlev kmem_free(mp->symspace, mp->symsize); 997843e1988Sjohnlev if (mp->ctfdata != NULL) 998843e1988Sjohnlev kmem_free(mp->ctfdata, mp->ctfsize); 999843e1988Sjohnlev kmem_free(mp, sizeof (*mp)); 1000843e1988Sjohnlev kmem_free(mcp->mod_filename, strlen(XPV_FILENAME) + 1); 1001843e1988Sjohnlev kmem_free(mcp->mod_modname, strlen(XPV_MODNAME) + 1); 1002843e1988Sjohnlev kmem_free(mcp, sizeof (*mcp)); 1003843e1988Sjohnlev if (names != NULL) 1004843e1988Sjohnlev kmem_free(names, namesize); 1005843e1988Sjohnlev } 1006843e1988Sjohnlev 1007843e1988Sjohnlev void 1008843e1988Sjohnlev xpv_panic_init() 1009843e1988Sjohnlev { 1010843e1988Sjohnlev xen_platform_op_t op; 1011843e1988Sjohnlev int i; 1012843e1988Sjohnlev 1013843e1988Sjohnlev ASSERT(DOMAIN_IS_INITDOMAIN(xen_info)); 1014843e1988Sjohnlev 1015843e1988Sjohnlev for (i = 0; i < mmu.num_level; i++) 1016843e1988Sjohnlev ptable_pfn[i] = PFN_INVALID; 1017843e1988Sjohnlev 1018843e1988Sjohnlev /* Let Xen know where to jump if/when it panics. */ 1019843e1988Sjohnlev op.cmd = XENPF_panic_init; 1020843e1988Sjohnlev op.interface_version = XENPF_INTERFACE_VERSION; 1021843e1988Sjohnlev op.u.panic_init.panic_addr = (unsigned long)xpv_panic_hdlr; 1022843e1988Sjohnlev 1023843e1988Sjohnlev (void) HYPERVISOR_platform_op(&op); 1024843e1988Sjohnlev 1025843e1988Sjohnlev init_xen_module(); 1026843e1988Sjohnlev } 1027