d/freebsd-nq
1
0
Fork 0
Code Issues Pull Requests Packages Projects Releases Wiki Activity
a81c96922d
freebsd-nq/lib/libkvm/kvm_minidump_powerpc64_hpt.c
Leandro Lupori d3c34fc0f4 [PPC64] Initial libkvm minidump implementation 
This change adds PowerPC64 support for minidumps on libkvm.

Address translation, page walk, and data retrieval were tested and seem to be
working correctly.

Reviewed by:	jhibbits
Differential Revision:	https://reviews.freebsd.org/D21555
2019-12-09 13:59:36 +00:00

667 lines
16 KiB
C
Raw Blame History

/*-
* Copyright (c) 2006 Peter Wemm
* Copyright (c) 2019 Leandro Lupori
*
* Redistribution and use in source and binary forms, with or without
* modification, are permitted provided that the following conditions
* are met:
* 1. Redistributions of source code must retain the above copyright
* notice, this list of conditions and the following disclaimer.
* 2. Redistributions in binary form must reproduce the above copyright
* notice, this list of conditions and the following disclaimer in the
* documentation and/or other materials provided with the distribution.
*
* THIS SOFTWARE IS PROVIDED BY THE AUTHOR AND CONTRIBUTORS ``AS IS'' AND
* ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
* IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
* ARE DISCLAIMED. IN NO EVENT SHALL THE AUTHOR OR CONTRIBUTORS BE LIABLE
* FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
* DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
* OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
* HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
* LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
* OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
* SUCH DAMAGE.
*
* From: FreeBSD: src/lib/libkvm/kvm_minidump_riscv.c
*/
#include <sys/cdefs.h>
__FBSDID("$FreeBSD$");
#include <sys/param.h>
#include <vm/vm.h>
#include <kvm.h>
#include <limits.h>
#include <stdint.h>
#include <stdlib.h>
#include <string.h>
#include <unistd.h>
#include "../../sys/powerpc/include/minidump.h"
#include "kvm_private.h"
#include "kvm_powerpc64.h"
/*
* PowerPC64 HPT machine dependent routines for kvm and minidumps.
*
* Address Translation parameters:
*
* b = 12 (SLB base page size: 4 KB)
* b = 24 (SLB base page size: 16 MB)
* p = 12 (page size: 4 KB)
* p = 24 (page size: 16 MB)
* s = 28 (segment size: 256 MB)
*/
/* Large (huge) page params */
#define LP_PAGE_SHIFT 24
#define LP_PAGE_SIZE (1ULL << LP_PAGE_SHIFT)
#define LP_PAGE_MASK 0x00ffffffULL
/* SLB */
#define SEGMENT_LENGTH 0x10000000ULL
#define round_seg(x) roundup2((uint64_t)(x), SEGMENT_LENGTH)
/* Virtual real-mode VSID in LPARs */
#define VSID_VRMA 0x1ffffffULL
#define SLBV_L 0x0000000000000100ULL /* Large page selector */
#define SLBV_CLASS 0x0000000000000080ULL /* Class selector */
#define SLBV_LP_MASK 0x0000000000000030ULL
#define SLBV_VSID_MASK 0x3ffffffffffff000ULL /* Virtual SegID mask */
#define SLBV_VSID_SHIFT 12
#define SLBE_B_MASK 0x0000000006000000ULL
#define SLBE_B_256MB 0x0000000000000000ULL
#define SLBE_VALID 0x0000000008000000ULL /* SLB entry valid */
#define SLBE_INDEX_MASK 0x0000000000000fffULL /* SLB index mask */
#define SLBE_ESID_MASK 0xfffffffff0000000ULL /* Effective SegID mask */
#define SLBE_ESID_SHIFT 28
/* PTE */
#define LPTEH_VSID_SHIFT 12
#define LPTEH_AVPN_MASK 0xffffffffffffff80ULL
#define LPTEH_B_MASK 0xc000000000000000ULL
#define LPTEH_B_256MB 0x0000000000000000ULL
#define LPTEH_BIG 0x0000000000000004ULL /* 4KB/16MB page */
#define LPTEH_HID 0x0000000000000002ULL
#define LPTEH_VALID 0x0000000000000001ULL
#define LPTEL_RPGN 0xfffffffffffff000ULL
#define LPTEL_LP_MASK 0x00000000000ff000ULL
#define LPTEL_NOEXEC 0x0000000000000004ULL
/* Supervisor (U: RW, S: RW) */
#define LPTEL_BW 0x0000000000000002ULL
/* Both Read Only (U: RO, S: RO) */
#define LPTEL_BR 0x0000000000000003ULL
#define LPTEL_RW LPTEL_BW
#define LPTEL_RO LPTEL_BR
/*
* PTE AVA field manipulation macros.
*
* AVA[0:54] = PTEH[2:56]
* AVA[VSID] = AVA[0:49] = PTEH[2:51]
* AVA[PAGE] = AVA[50:54] = PTEH[52:56]
*/
#define PTEH_AVA_VSID_MASK 0x3ffffffffffff000UL
#define PTEH_AVA_VSID_SHIFT 12
#define PTEH_AVA_VSID(p) \
(((p) & PTEH_AVA_VSID_MASK) >> PTEH_AVA_VSID_SHIFT)
#define PTEH_AVA_PAGE_MASK 0x0000000000000f80UL
#define PTEH_AVA_PAGE_SHIFT 7
#define PTEH_AVA_PAGE(p) \
(((p) & PTEH_AVA_PAGE_MASK) >> PTEH_AVA_PAGE_SHIFT)
/* Masks to obtain the Physical Address from PTE low 64-bit word. */
#define PTEL_PA_MASK 0x0ffffffffffff000UL
#define PTEL_LP_PA_MASK 0x0fffffffff000000UL
#define PTE_HASH_MASK 0x0000007fffffffffUL
/*
* Number of AVA/VA page bits to shift right, in order to leave only the
* ones that should be considered.
*
* q = MIN(54, 77-b) (PowerISA v2.07B, 5.7.7.3)
* n = q + 1 - 50 (VSID size in bits)
* s(ava) = 5 - n
* s(va) = (28 - b) - n
*
* q: bit number of lower limit of VA/AVA bits to compare
* n: number of AVA/VA page bits to compare
* s: shift amount
* 28 - b: VA page size in bits
*/
#define AVA_PAGE_SHIFT(b) (5 - (MIN(54, 77-(b)) + 1 - 50))
#define VA_PAGE_SHIFT(b) (28 - (b) - (MIN(54, 77-(b)) + 1 - 50))
/* Kernel ESID -> VSID mapping */
#define KERNEL_VSID_BIT 0x0000001000000000UL /* Bit set in all kernel VSIDs */
#define KERNEL_VSID(esid) ((((((uint64_t)esid << 8) | ((uint64_t)esid >> 28)) \
* 0x13bbUL) & (KERNEL_VSID_BIT - 1)) | \
KERNEL_VSID_BIT)
/* Types */
typedef uint64_t ppc64_physaddr_t;
typedef struct {
uint64_t slbv;
uint64_t slbe;
} ppc64_slb_entry_t;
typedef struct {
uint64_t pte_hi;
uint64_t pte_lo;
} ppc64_pt_entry_t;
struct hpt_data {
ppc64_slb_entry_t *slbs;
uint32_t slbsize;
};
static void
slb_fill(ppc64_slb_entry_t *slb, uint64_t ea, uint64_t i)
{
uint64_t esid;
esid = ea >> SLBE_ESID_SHIFT;
slb->slbv = KERNEL_VSID(esid) << SLBV_VSID_SHIFT;
slb->slbe = (esid << SLBE_ESID_SHIFT) | SLBE_VALID | i;
}
static int
slb_init(kvm_t *kd)
{
struct minidumphdr *hdr;
struct hpt_data *data;
ppc64_slb_entry_t *slb;
uint32_t slbsize;
uint64_t ea, i, maxmem;
hdr = &kd->vmst->hdr;
data = PPC64_MMU_DATA(kd);
/* Alloc SLBs */
maxmem = hdr->bitmapsize * 8 * PPC64_PAGE_SIZE;
slbsize = round_seg(hdr->kernend + 1 - hdr->kernbase + maxmem) /
SEGMENT_LENGTH * sizeof(ppc64_slb_entry_t);
data->slbs = _kvm_malloc(kd, slbsize);
if (data->slbs == NULL) {
_kvm_err(kd, kd->program, "cannot allocate slbs");
return (-1);
}
data->slbsize = slbsize;
dprintf("%s: maxmem=0x%jx, segs=%jd, slbsize=0x%jx\n",
__func__, (uintmax_t)maxmem,
(uintmax_t)slbsize / sizeof(ppc64_slb_entry_t), (uintmax_t)slbsize);
/*
* Generate needed SLB entries.
*
* When translating addresses from EA to VA to PA, the needed SLB
* entry could be generated on the fly, but this is not the case
* for the walk_pages method, that needs to search the SLB entry
* by VSID, in order to find out the EA from a PTE.
*/
/* VM area */
for (ea = hdr->kernbase, i = 0, slb = data->slbs;
ea < hdr->kernend; ea += SEGMENT_LENGTH, i++, slb++)
slb_fill(slb, ea, i);
/* DMAP area */
for (ea = hdr->dmapbase;
ea < MIN(hdr->dmapend, hdr->dmapbase + maxmem);
ea += SEGMENT_LENGTH, i++, slb++) {
slb_fill(slb, ea, i);
if (hdr->hw_direct_map)
slb->slbv |= SLBV_L;
}
return (0);
}
static void
ppc64mmu_hpt_cleanup(kvm_t *kd)
{
struct hpt_data *data;
if (kd->vmst == NULL)
return;
data = PPC64_MMU_DATA(kd);
free(data->slbs);
free(data);
PPC64_MMU_DATA(kd) = NULL;
}
static int
ppc64mmu_hpt_init(kvm_t *kd)
{
struct hpt_data *data;
struct minidumphdr *hdr;
hdr = &kd->vmst->hdr;
/* Alloc MMU data */
data = _kvm_malloc(kd, sizeof(*data));
if (data == NULL) {
_kvm_err(kd, kd->program, "cannot allocate MMU data");
return (-1);
}
data->slbs = NULL;
PPC64_MMU_DATA(kd) = data;
if (slb_init(kd) == -1)
goto failed;
return (0);
failed:
ppc64mmu_hpt_cleanup(kd);
return (-1);
}
static ppc64_slb_entry_t *
slb_search(kvm_t *kd, kvaddr_t ea)
{
struct hpt_data *data;
ppc64_slb_entry_t *slb;
int i, n;
data = PPC64_MMU_DATA(kd);
slb = data->slbs;
n = data->slbsize / sizeof(ppc64_slb_entry_t);
/* SLB search */
for (i = 0; i < n; i++, slb++) {
if ((slb->slbe & SLBE_VALID) == 0)
continue;
/* Compare 36-bit ESID of EA with segment one (64-s) */
if ((slb->slbe & SLBE_ESID_MASK) != (ea & SLBE_ESID_MASK))
continue;
/* Match found */
dprintf("SEG#%02d: slbv=0x%016jx, slbe=0x%016jx\n",
i, (uintmax_t)slb->slbv, (uintmax_t)slb->slbe);
break;
}
/* SLB not found */
if (i == n) {
_kvm_err(kd, kd->program, "%s: segment not found for EA 0x%jx",
__func__, (uintmax_t)ea);
return (NULL);
}
return (slb);
}
static ppc64_pt_entry_t
pte_get(kvm_t *kd, u_long ptex)
{
ppc64_pt_entry_t pte, *p;
p = _kvm_pmap_get(kd, ptex, sizeof(pte));
pte.pte_hi = be64toh(p->pte_hi);
pte.pte_lo = be64toh(p->pte_lo);
return (pte);
}
static int
pte_search(kvm_t *kd, ppc64_slb_entry_t *slb, uint64_t hid, kvaddr_t ea,
ppc64_pt_entry_t *p)
{
uint64_t hash, hmask;
uint64_t pteg, ptex;
uint64_t va_vsid, va_page;
int b;
int ava_pg_shift, va_pg_shift;
ppc64_pt_entry_t pte;
/*
* Get VA:
*
* va(78) = va_vsid(50) || va_page(s-b) || offset(b)
*
* va_vsid: 50-bit VSID (78-s)
* va_page: (s-b)-bit VA page
*/
b = slb->slbv & SLBV_L? LP_PAGE_SHIFT : PPC64_PAGE_SHIFT;
va_vsid = (slb->slbv & SLBV_VSID_MASK) >> SLBV_VSID_SHIFT;
va_page = (ea & ~SLBE_ESID_MASK) >> b;
dprintf("%s: hid=0x%jx, ea=0x%016jx, b=%d, va_vsid=0x%010jx, "
"va_page=0x%04jx\n",
__func__, (uintmax_t)hid, (uintmax_t)ea, b,
(uintmax_t)va_vsid, (uintmax_t)va_page);
/*
* Get hash:
*
* Primary hash: va_vsid(11:49) ^ va_page(s-b)
* Secondary hash: ~primary_hash
*/
hash = (va_vsid & PTE_HASH_MASK) ^ va_page;
if (hid)
hash = ~hash & PTE_HASH_MASK;
/*
* Get PTEG:
*
* pteg = (hash(0:38) & hmask) << 3
*
* hmask (hash mask): mask generated from HTABSIZE || 11*0b1
* hmask = number_of_ptegs - 1
*/
hmask = kd->vmst->hdr.pmapsize / (8 * sizeof(ppc64_pt_entry_t)) - 1;
pteg = (hash & hmask) << 3;
ava_pg_shift = AVA_PAGE_SHIFT(b);
va_pg_shift = VA_PAGE_SHIFT(b);
dprintf("%s: hash=0x%010jx, hmask=0x%010jx, (hash & hmask)=0x%010jx, "
"pteg=0x%011jx, ava_pg_shift=%d, va_pg_shift=%d\n",
__func__, (uintmax_t)hash, (uintmax_t)hmask,
(uintmax_t)(hash & hmask), (uintmax_t)pteg,
ava_pg_shift, va_pg_shift);
/* Search PTEG */
for (ptex = pteg; ptex < pteg + 8; ptex++) {
pte = pte_get(kd, ptex);
/* Check H, V and B */
if ((pte.pte_hi & LPTEH_HID) != hid ||
(pte.pte_hi & LPTEH_VALID) == 0 ||
(pte.pte_hi & LPTEH_B_MASK) != LPTEH_B_256MB)
continue;
/* Compare AVA with VA */
if (PTEH_AVA_VSID(pte.pte_hi) != va_vsid ||
(PTEH_AVA_PAGE(pte.pte_hi) >> ava_pg_shift) !=
(va_page >> va_pg_shift))
continue;
/*
* Check if PTE[L] matches SLBV[L].
*
* Note: this check ignores PTE[LP], as does the kernel.
*/
if (b == PPC64_PAGE_SHIFT) {
if (pte.pte_hi & LPTEH_BIG)
continue;
} else if ((pte.pte_hi & LPTEH_BIG) == 0)
continue;
/* Match found */
dprintf("%s: PTE found: ptex=0x%jx, pteh=0x%016jx, "
"ptel=0x%016jx\n",
__func__, (uintmax_t)ptex, (uintmax_t)pte.pte_hi,
(uintmax_t)pte.pte_lo);
break;
}
/* Not found? */
if (ptex == pteg + 8) {
/* Try secondary hash */
if (hid == 0)
return (pte_search(kd, slb, LPTEH_HID, ea, p));
else {
_kvm_err(kd, kd->program,
"%s: pte not found", __func__);
return (-1);
}
}
/* PTE found */
*p = pte;
return (0);
}
static int
pte_lookup(kvm_t *kd, kvaddr_t ea, ppc64_pt_entry_t *pte)
{
ppc64_slb_entry_t *slb;
/* First, find SLB */
if ((slb = slb_search(kd, ea)) == NULL)
return (-1);
/* Next, find PTE */
return (pte_search(kd, slb, 0, ea, pte));
}
static int
ppc64mmu_hpt_kvatop(kvm_t *kd, kvaddr_t va, off_t *pa)
{
struct minidumphdr *hdr;
struct vmstate *vm;
ppc64_pt_entry_t pte;
ppc64_physaddr_t pgoff, pgpa;
off_t ptoff;
int err;
vm = kd->vmst;
hdr = &vm->hdr;
pgoff = va & PPC64_PAGE_MASK;
dprintf("%s: va=0x%016jx\n", __func__, (uintmax_t)va);
/*
* A common use case of libkvm is to first find a symbol address
* from the kernel image and then use kvatop to translate it and
* to be able to fetch its corresponding data.
*
* The problem is that, in PowerPC64 case, the addresses of relocated
* data won't match those in the kernel image. This is handled here by
* adding the relocation offset to those addresses.
*/
if (va < hdr->dmapbase)
va += hdr->startkernel - PPC64_KERNBASE;
/* Handle DMAP */
if (va >= hdr->dmapbase && va <= hdr->dmapend) {
pgpa = (va & ~hdr->dmapbase) & ~PPC64_PAGE_MASK;
ptoff = _kvm_pt_find(kd, pgpa, PPC64_PAGE_SIZE);
if (ptoff == -1) {
_kvm_err(kd, kd->program, "%s: "
"direct map address 0x%jx not in minidump",
__func__, (uintmax_t)va);
goto invalid;
}
*pa = ptoff + pgoff;
return (PPC64_PAGE_SIZE - pgoff);
/* Translate VA to PA */
} else if (va >= hdr->kernbase) {
if ((err = pte_lookup(kd, va, &pte)) == -1) {
_kvm_err(kd, kd->program,
"%s: pte not valid", __func__);
goto invalid;
}
if (pte.pte_hi & LPTEH_BIG)
pgpa = (pte.pte_lo & PTEL_LP_PA_MASK) |
(va & ~PPC64_PAGE_MASK & LP_PAGE_MASK);
else
pgpa = pte.pte_lo & PTEL_PA_MASK;
dprintf("%s: pgpa=0x%016jx\n", __func__, (uintmax_t)pgpa);
ptoff = _kvm_pt_find(kd, pgpa, PPC64_PAGE_SIZE);
if (ptoff == -1) {
_kvm_err(kd, kd->program, "%s: "
"physical address 0x%jx not in minidump",
__func__, (uintmax_t)pgpa);
goto invalid;
}
*pa = ptoff + pgoff;
return (PPC64_PAGE_SIZE - pgoff);
} else {
_kvm_err(kd, kd->program,
"%s: virtual address 0x%jx not minidumped",
__func__, (uintmax_t)va);
goto invalid;
}
invalid:
_kvm_err(kd, 0, "invalid address (0x%jx)", (uintmax_t)va);
return (0);
}
static vm_prot_t
entry_to_prot(ppc64_pt_entry_t *pte)
{
vm_prot_t prot = VM_PROT_READ;
if (pte->pte_lo & LPTEL_RW)
prot |= VM_PROT_WRITE;
if ((pte->pte_lo & LPTEL_NOEXEC) != 0)
prot |= VM_PROT_EXECUTE;
return (prot);
}
static ppc64_slb_entry_t *
slb_vsid_search(kvm_t *kd, uint64_t vsid)
{
struct hpt_data *data;
ppc64_slb_entry_t *slb;
int i, n;
data = PPC64_MMU_DATA(kd);
slb = data->slbs;
n = data->slbsize / sizeof(ppc64_slb_entry_t);
vsid <<= SLBV_VSID_SHIFT;
/* SLB search */
for (i = 0; i < n; i++, slb++) {
/* Check if valid and compare VSID */
if ((slb->slbe & SLBE_VALID) &&
(slb->slbv & SLBV_VSID_MASK) == vsid)
break;
}
/* SLB not found */
if (i == n) {
_kvm_err(kd, kd->program,
"%s: segment not found for VSID 0x%jx",
__func__, (uintmax_t)vsid >> SLBV_VSID_SHIFT);
return (NULL);
}
return (slb);
}
static u_long
get_ea(kvm_t *kd, ppc64_pt_entry_t *pte, u_long ptex)
{
ppc64_slb_entry_t *slb;
uint64_t ea, hash, vsid;
int b, shift;
/* Find SLB */
vsid = PTEH_AVA_VSID(pte->pte_hi);
if ((slb = slb_vsid_search(kd, vsid)) == NULL)
return (~0UL);
/* Get ESID part of EA */
ea = slb->slbe & SLBE_ESID_MASK;
b = slb->slbv & SLBV_L? LP_PAGE_SHIFT : PPC64_PAGE_SHIFT;
/*
* If there are less than 64K PTEGs (16-bit), the upper bits of
* EA page must be obtained from PTEH's AVA.
*/
if (kd->vmst->hdr.pmapsize / (8 * sizeof(ppc64_pt_entry_t)) <
0x10000U) {
/*
* Add 0 to 5 EA bits, right after VSID.
* b == 12: 5 bits
* b == 24: 4 bits
*/
shift = AVA_PAGE_SHIFT(b);
ea |= (PTEH_AVA_PAGE(pte->pte_hi) >> shift) <<
(SLBE_ESID_SHIFT - 5 + shift);
}
/* Get VA page from hash and add to EA. */
hash = (ptex & ~7) >> 3;
if (pte->pte_hi & LPTEH_HID)
hash = ~hash & PTE_HASH_MASK;
ea |= ((hash ^ (vsid & PTE_HASH_MASK)) << b) & ~SLBE_ESID_MASK;
return (ea);
}
static int
ppc64mmu_hpt_walk_pages(kvm_t *kd, kvm_walk_pages_cb_t *cb, void *arg)
{
struct vmstate *vm;
int ret;
unsigned int pagesz;
u_long dva, pa, va;
u_long ptex, nptes;
uint64_t vsid;
ret = 0;
vm = kd->vmst;
nptes = vm->hdr.pmapsize / sizeof(ppc64_pt_entry_t);
/* Walk through PTEs */
for (ptex = 0; ptex < nptes; ptex++) {
ppc64_pt_entry_t pte = pte_get(kd, ptex);
if ((pte.pte_hi & LPTEH_VALID) == 0)
continue;
/* Skip non-kernel related pages, as well as VRMA ones */
vsid = PTEH_AVA_VSID(pte.pte_hi);
if ((vsid & KERNEL_VSID_BIT) == 0 ||
(vsid >> PPC64_PAGE_SHIFT) == VSID_VRMA)
continue;
/* Retrieve page's VA (EA on PPC64 terminology) */
if ((va = get_ea(kd, &pte, ptex)) == ~0UL)
goto out;
/* Get PA and page size */
if (pte.pte_hi & LPTEH_BIG) {
pa = pte.pte_lo & PTEL_LP_PA_MASK;
pagesz = LP_PAGE_SIZE;
} else {
pa = pte.pte_lo & PTEL_PA_MASK;
pagesz = PPC64_PAGE_SIZE;
}
/* Get DMAP address */
dva = vm->hdr.dmapbase + pa;
if (!_kvm_visit_cb(kd, cb, arg, pa, va, dva,
entry_to_prot(&pte), pagesz, 0))
goto out;
}
ret = 1;
out:
return (ret);
}
static struct ppc64_mmu_ops ops = {
.init = ppc64mmu_hpt_init,
.cleanup = ppc64mmu_hpt_cleanup,
.kvatop = ppc64mmu_hpt_kvatop,
.walk_pages = ppc64mmu_hpt_walk_pages,
};
struct ppc64_mmu_ops *ppc64_mmu_ops_hpt = &ops;
Reference in New Issue View Git Blame Copy Permalink