gem5/src/arch/arm/table_walker.cc
Ali Saidi e6c31ceb2b ARM: Don't return the result of a table walk the same cycle it's completed.
The L1 cache may have been accessed to provide this data, which confuses
it, if it ends up being accesses twice in one cycle. Instead wait 1 tick
which will force the timing simple CPU to forward to its next clock cycle
when the translation completes.

Also prevent multiple outstanding table walks from occuring at once.
2010-11-08 13:58:24 -06:00

705 lines
22 KiB
C++

/*
* Copyright (c) 2010 ARM Limited
* All rights reserved
*
* The license below extends only to copyright in the software and shall
* not be construed as granting a license to any other intellectual
* property including but not limited to intellectual property relating
* to a hardware implementation of the functionality of the software
* licensed hereunder. You may use the software subject to the license
* terms below provided that you ensure that this notice is replicated
* unmodified and in its entirety in all distributions of the software,
* modified or unmodified, in source code or in binary form.
*
* Redistribution and use in source and binary forms, with or without
* modification, are permitted provided that the following conditions are
* met: redistributions of source code must retain the above copyright
* notice, this list of conditions and the following disclaimer;
* 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;
* neither the name of the copyright holders nor the names of its
* contributors may be used to endorse or promote products derived from
* this software without specific prior written permission.
*
* THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS 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 COPYRIGHT
* OWNER 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.
*
* Authors: Ali Saidi
*/
#include "arch/arm/faults.hh"
#include "arch/arm/table_walker.hh"
#include "arch/arm/tlb.hh"
#include "dev/io_device.hh"
#include "cpu/base.hh"
#include "cpu/thread_context.hh"
using namespace ArmISA;
TableWalker::TableWalker(const Params *p)
: MemObject(p), port(NULL), tlb(NULL), currState(NULL), pending(false),
doL1DescEvent(this), doL2DescEvent(this), doProcessEvent(this)
{
sctlr = 0;
}
TableWalker::~TableWalker()
{
;
}
unsigned int
drain(Event *de)
{
panic("Not implemented\n");
}
Port*
TableWalker::getPort(const std::string &if_name, int idx)
{
if (if_name == "port") {
if (port != NULL)
fatal("%s: port already connected to %s",
name(), port->getPeer()->name());
System *sys = params()->sys;
Tick minb = params()->min_backoff;
Tick maxb = params()->max_backoff;
port = new DmaPort(this, sys, minb, maxb);
return port;
}
return NULL;
}
Fault
TableWalker::walk(RequestPtr _req, ThreadContext *_tc, uint8_t _cid, TLB::Mode _mode,
TLB::Translation *_trans, bool _timing)
{
if (!currState) {
// For atomic mode, a new WalkerState instance should be only created
// once per TLB. For timing mode, a new instance is generated for every
// TLB miss.
DPRINTF(TLBVerbose, "creating new instance of WalkerState\n");
currState = new WalkerState();
currState->tableWalker = this;
}
else if (_timing) {
panic("currState should always be empty in timing mode!\n");
}
currState->tc = _tc;
currState->transState = _trans;
currState->req = _req;
currState->fault = NoFault;
currState->contextId = _cid;
currState->timing = _timing;
currState->mode = _mode;
/** @todo These should be cached or grabbed from cached copies in
the TLB, all these miscreg reads are expensive */
currState->vaddr = currState->req->getVaddr();
currState->sctlr = currState->tc->readMiscReg(MISCREG_SCTLR);
sctlr = currState->sctlr;
currState->N = currState->tc->readMiscReg(MISCREG_TTBCR);
currState->isFetch = (currState->mode == TLB::Execute);
currState->isWrite = (currState->mode == TLB::Write);
if (!currState->timing)
return processWalk();
if (pending) {
pendingQueue.push_back(currState);
currState = NULL;
} else {
pending = true;
processWalk();
}
return NoFault;
}
void
TableWalker::processWalkWrapper()
{
assert(!currState);
assert(pendingQueue.size());
currState = pendingQueue.front();
pendingQueue.pop_front();
pending = true;
processWalk();
}
Fault
TableWalker::processWalk()
{
Addr ttbr = 0;
// If translation isn't enabled, we shouldn't be here
assert(currState->sctlr.m);
DPRINTF(TLB, "Begining table walk for address %#x, TTBCR: %#x, bits:%#x\n",
currState->vaddr, currState->N, mbits(currState->vaddr, 31,
32-currState->N));
if (currState->N == 0 || !mbits(currState->vaddr, 31, 32-currState->N)) {
DPRINTF(TLB, " - Selecting TTBR0\n");
ttbr = currState->tc->readMiscReg(MISCREG_TTBR0);
} else {
DPRINTF(TLB, " - Selecting TTBR1\n");
ttbr = currState->tc->readMiscReg(MISCREG_TTBR1);
currState->N = 0;
}
Addr l1desc_addr = mbits(ttbr, 31, 14-currState->N) |
(bits(currState->vaddr,31-currState->N,20) << 2);
DPRINTF(TLB, " - Descriptor at address %#x\n", l1desc_addr);
// Trickbox address check
Fault f;
f = tlb->walkTrickBoxCheck(l1desc_addr, currState->vaddr, sizeof(uint32_t),
currState->isFetch, currState->isWrite, 0, true);
if (f) {
if (currState->timing) {
currState->transState->finish(f, currState->req,
currState->tc, currState->mode);
pending = false;
nextWalk(currState->tc);
currState = NULL;
} else {
currState->tc = NULL;
currState->req = NULL;
}
return f;
}
if (currState->timing) {
port->dmaAction(MemCmd::ReadReq, l1desc_addr, sizeof(uint32_t),
&doL1DescEvent, (uint8_t*)&currState->l1Desc.data,
currState->tc->getCpuPtr()->ticks(1));
DPRINTF(TLBVerbose, "Adding to walker fifo: queue size before adding: %d\n",
stateQueueL1.size());
stateQueueL1.push_back(currState);
currState = NULL;
} else {
Request::Flags flag = 0;
if (currState->sctlr.c == 0){
flag = Request::UNCACHEABLE;
}
port->dmaAction(MemCmd::ReadReq, l1desc_addr, sizeof(uint32_t),
NULL, (uint8_t*)&currState->l1Desc.data,
currState->tc->getCpuPtr()->ticks(1), flag);
doL1Descriptor();
f = currState->fault;
}
return f;
}
void
TableWalker::memAttrs(ThreadContext *tc, TlbEntry &te, SCTLR sctlr,
uint8_t texcb, bool s)
{
// Note: tc and sctlr local variables are hiding tc and sctrl class
// variables
DPRINTF(TLBVerbose, "memAttrs texcb:%d s:%d\n", texcb, s);
te.shareable = false; // default value
te.nonCacheable = false;
bool outer_shareable = false;
if (sctlr.tre == 0 || ((sctlr.tre == 1) && (sctlr.m == 0))) {
switch(texcb) {
case 0: // Stongly-ordered
te.nonCacheable = true;
te.mtype = TlbEntry::StronglyOrdered;
te.shareable = true;
te.innerAttrs = 1;
te.outerAttrs = 0;
break;
case 1: // Shareable Device
te.nonCacheable = true;
te.mtype = TlbEntry::Device;
te.shareable = true;
te.innerAttrs = 3;
te.outerAttrs = 0;
break;
case 2: // Outer and Inner Write-Through, no Write-Allocate
te.mtype = TlbEntry::Normal;
te.shareable = s;
te.innerAttrs = 6;
te.outerAttrs = bits(texcb, 1, 0);
break;
case 3: // Outer and Inner Write-Back, no Write-Allocate
te.mtype = TlbEntry::Normal;
te.shareable = s;
te.innerAttrs = 7;
te.outerAttrs = bits(texcb, 1, 0);
break;
case 4: // Outer and Inner Non-cacheable
te.nonCacheable = true;
te.mtype = TlbEntry::Normal;
te.shareable = s;
te.innerAttrs = 0;
te.outerAttrs = bits(texcb, 1, 0);
break;
case 5: // Reserved
panic("Reserved texcb value!\n");
break;
case 6: // Implementation Defined
panic("Implementation-defined texcb value!\n");
break;
case 7: // Outer and Inner Write-Back, Write-Allocate
te.mtype = TlbEntry::Normal;
te.shareable = s;
te.innerAttrs = 5;
te.outerAttrs = 1;
break;
case 8: // Non-shareable Device
te.nonCacheable = true;
te.mtype = TlbEntry::Device;
te.shareable = false;
te.innerAttrs = 3;
te.outerAttrs = 0;
break;
case 9 ... 15: // Reserved
panic("Reserved texcb value!\n");
break;
case 16 ... 31: // Cacheable Memory
te.mtype = TlbEntry::Normal;
te.shareable = s;
if (bits(texcb, 1,0) == 0 || bits(texcb, 3,2) == 0)
te.nonCacheable = true;
te.innerAttrs = bits(texcb, 1, 0);
te.outerAttrs = bits(texcb, 3, 2);
break;
default:
panic("More than 32 states for 5 bits?\n");
}
} else {
assert(tc);
PRRR prrr = tc->readMiscReg(MISCREG_PRRR);
NMRR nmrr = tc->readMiscReg(MISCREG_NMRR);
DPRINTF(TLBVerbose, "memAttrs PRRR:%08x NMRR:%08x\n", prrr, nmrr);
uint8_t curr_tr = 0, curr_ir = 0, curr_or = 0;
switch(bits(texcb, 2,0)) {
case 0:
curr_tr = prrr.tr0;
curr_ir = nmrr.ir0;
curr_or = nmrr.or0;
outer_shareable = (prrr.nos0 == 0);
break;
case 1:
curr_tr = prrr.tr1;
curr_ir = nmrr.ir1;
curr_or = nmrr.or1;
outer_shareable = (prrr.nos1 == 0);
break;
case 2:
curr_tr = prrr.tr2;
curr_ir = nmrr.ir2;
curr_or = nmrr.or2;
outer_shareable = (prrr.nos2 == 0);
break;
case 3:
curr_tr = prrr.tr3;
curr_ir = nmrr.ir3;
curr_or = nmrr.or3;
outer_shareable = (prrr.nos3 == 0);
break;
case 4:
curr_tr = prrr.tr4;
curr_ir = nmrr.ir4;
curr_or = nmrr.or4;
outer_shareable = (prrr.nos4 == 0);
break;
case 5:
curr_tr = prrr.tr5;
curr_ir = nmrr.ir5;
curr_or = nmrr.or5;
outer_shareable = (prrr.nos5 == 0);
break;
case 6:
panic("Imp defined type\n");
case 7:
curr_tr = prrr.tr7;
curr_ir = nmrr.ir7;
curr_or = nmrr.or7;
outer_shareable = (prrr.nos7 == 0);
break;
}
switch(curr_tr) {
case 0:
DPRINTF(TLBVerbose, "StronglyOrdered\n");
te.mtype = TlbEntry::StronglyOrdered;
te.nonCacheable = true;
te.innerAttrs = 1;
te.outerAttrs = 0;
te.shareable = true;
break;
case 1:
DPRINTF(TLBVerbose, "Device ds1:%d ds0:%d s:%d\n",
prrr.ds1, prrr.ds0, s);
te.mtype = TlbEntry::Device;
te.nonCacheable = true;
te.innerAttrs = 3;
te.outerAttrs = 0;
if (prrr.ds1 && s)
te.shareable = true;
if (prrr.ds0 && !s)
te.shareable = true;
break;
case 2:
DPRINTF(TLBVerbose, "Normal ns1:%d ns0:%d s:%d\n",
prrr.ns1, prrr.ns0, s);
te.mtype = TlbEntry::Normal;
if (prrr.ns1 && s)
te.shareable = true;
if (prrr.ns0 && !s)
te.shareable = true;
break;
case 3:
panic("Reserved type");
}
if (te.mtype == TlbEntry::Normal){
switch(curr_ir) {
case 0:
te.nonCacheable = true;
te.innerAttrs = 0;
break;
case 1:
te.innerAttrs = 5;
break;
case 2:
te.innerAttrs = 6;
break;
case 3:
te.innerAttrs = 7;
break;
}
switch(curr_or) {
case 0:
te.nonCacheable = true;
te.outerAttrs = 0;
break;
case 1:
te.outerAttrs = 1;
break;
case 2:
te.outerAttrs = 2;
break;
case 3:
te.outerAttrs = 3;
break;
}
}
}
DPRINTF(TLBVerbose, "memAttrs: shareable: %d, innerAttrs: %d, \
outerAttrs: %d\n",
te.shareable, te.innerAttrs, te.outerAttrs);
/** Formatting for Physical Address Register (PAR)
* Only including lower bits (TLB info here)
* PAR:
* PA [31:12]
* Reserved [11]
* TLB info [10:1]
* NOS [10] (Not Outer Sharable)
* NS [9] (Non-Secure)
* -- [8] (Implementation Defined)
* SH [7] (Sharable)
* Inner[6:4](Inner memory attributes)
* Outer[3:2](Outer memory attributes)
* SS [1] (SuperSection)
* F [0] (Fault, Fault Status in [6:1] if faulted)
*/
te.attributes = (
((outer_shareable ? 0:1) << 10) |
// TODO: NS Bit
((te.shareable ? 1:0) << 7) |
(te.innerAttrs << 4) |
(te.outerAttrs << 2)
// TODO: Supersection bit
// TODO: Fault bit
);
}
void
TableWalker::doL1Descriptor()
{
DPRINTF(TLB, "L1 descriptor for %#x is %#x\n",
currState->vaddr, currState->l1Desc.data);
TlbEntry te;
switch (currState->l1Desc.type()) {
case L1Descriptor::Ignore:
case L1Descriptor::Reserved:
if (!currState->delayed) {
currState->tc = NULL;
currState->req = NULL;
}
DPRINTF(TLB, "L1 Descriptor Reserved/Ignore, causing fault\n");
if (currState->isFetch)
currState->fault =
new PrefetchAbort(currState->vaddr, ArmFault::Translation0);
else
currState->fault =
new DataAbort(currState->vaddr, 0, currState->isWrite,
ArmFault::Translation0);
return;
case L1Descriptor::Section:
if (currState->sctlr.afe && bits(currState->l1Desc.ap(), 0) == 0) {
/** @todo: check sctlr.ha (bit[17]) if Hardware Access Flag is
* enabled if set, do l1.Desc.setAp0() instead of generating
* AccessFlag0
*/
currState->fault = new DataAbort(currState->vaddr,
currState->l1Desc.domain(), currState->isWrite,
ArmFault::AccessFlag0);
}
if (currState->l1Desc.supersection()) {
panic("Haven't implemented supersections\n");
}
te.N = 20;
te.pfn = currState->l1Desc.pfn();
te.size = (1<<te.N) - 1;
te.global = !currState->l1Desc.global();
te.valid = true;
te.vpn = currState->vaddr >> te.N;
te.sNp = true;
te.xn = currState->l1Desc.xn();
te.ap = currState->l1Desc.ap();
te.domain = currState->l1Desc.domain();
te.asid = currState->contextId;
memAttrs(currState->tc, te, currState->sctlr,
currState->l1Desc.texcb(), currState->l1Desc.shareable());
DPRINTF(TLB, "Inserting Section Descriptor into TLB\n");
DPRINTF(TLB, " - N:%d pfn:%#x size: %#x global:%d valid: %d\n",
te.N, te.pfn, te.size, te.global, te.valid);
DPRINTF(TLB, " - vpn:%#x sNp: %d xn:%d ap:%d domain: %d asid:%d nc:%d\n",
te.vpn, te.sNp, te.xn, te.ap, te.domain, te.asid,
te.nonCacheable);
DPRINTF(TLB, " - domain from l1 desc: %d data: %#x bits:%d\n",
currState->l1Desc.domain(), currState->l1Desc.data,
(currState->l1Desc.data >> 5) & 0xF );
if (!currState->timing) {
currState->tc = NULL;
currState->req = NULL;
}
tlb->insert(currState->vaddr, te);
return;
case L1Descriptor::PageTable:
Addr l2desc_addr;
l2desc_addr = currState->l1Desc.l2Addr() |
(bits(currState->vaddr, 19,12) << 2);
DPRINTF(TLB, "L1 descriptor points to page table at: %#x\n",
l2desc_addr);
// Trickbox address check
currState->fault = tlb->walkTrickBoxCheck(l2desc_addr, currState->vaddr,
sizeof(uint32_t), currState->isFetch, currState->isWrite,
currState->l1Desc.domain(), false);
if (currState->fault) {
if (!currState->timing) {
currState->tc = NULL;
currState->req = NULL;
}
return;
}
if (currState->timing) {
currState->delayed = true;
port->dmaAction(MemCmd::ReadReq, l2desc_addr, sizeof(uint32_t),
&doL2DescEvent, (uint8_t*)&currState->l2Desc.data,
currState->tc->getCpuPtr()->ticks(1));
} else {
port->dmaAction(MemCmd::ReadReq, l2desc_addr, sizeof(uint32_t),
NULL, (uint8_t*)&currState->l2Desc.data,
currState->tc->getCpuPtr()->ticks(1));
doL2Descriptor();
}
return;
default:
panic("A new type in a 2 bit field?\n");
}
}
void
TableWalker::doL2Descriptor()
{
DPRINTF(TLB, "L2 descriptor for %#x is %#x\n",
currState->vaddr, currState->l2Desc.data);
TlbEntry te;
if (currState->l2Desc.invalid()) {
DPRINTF(TLB, "L2 descriptor invalid, causing fault\n");
if (!currState->delayed) {
currState->tc = NULL;
currState->req = NULL;
}
if (currState->isFetch)
currState->fault =
new PrefetchAbort(currState->vaddr, ArmFault::Translation1);
else
currState->fault =
new DataAbort(currState->vaddr, currState->l1Desc.domain(),
currState->isWrite, ArmFault::Translation1);
return;
}
if (currState->sctlr.afe && bits(currState->l2Desc.ap(), 0) == 0) {
/** @todo: check sctlr.ha (bit[17]) if Hardware Access Flag is enabled
* if set, do l2.Desc.setAp0() instead of generating AccessFlag0
*/
currState->fault =
new DataAbort(currState->vaddr, 0, currState->isWrite,
ArmFault::AccessFlag1);
}
if (currState->l2Desc.large()) {
te.N = 16;
te.pfn = currState->l2Desc.pfn();
} else {
te.N = 12;
te.pfn = currState->l2Desc.pfn();
}
te.valid = true;
te.size = (1 << te.N) - 1;
te.asid = currState->contextId;
te.sNp = false;
te.vpn = currState->vaddr >> te.N;
te.global = currState->l2Desc.global();
te.xn = currState->l2Desc.xn();
te.ap = currState->l2Desc.ap();
te.domain = currState->l1Desc.domain();
memAttrs(currState->tc, te, currState->sctlr, currState->l2Desc.texcb(),
currState->l2Desc.shareable());
if (!currState->delayed) {
currState->tc = NULL;
currState->req = NULL;
}
tlb->insert(currState->vaddr, te);
}
void
TableWalker::doL1DescriptorWrapper()
{
currState = stateQueueL1.front();
currState->delayed = false;
DPRINTF(TLBVerbose, "L1 Desc object host addr: %p\n",&currState->l1Desc.data);
DPRINTF(TLBVerbose, "L1 Desc object data: %08x\n",currState->l1Desc.data);
DPRINTF(TLBVerbose, "calling doL1Descriptor for vaddr:%#x\n", currState->vaddr);
doL1Descriptor();
stateQueueL1.pop_front();
// Check if fault was generated
if (currState->fault != NoFault) {
currState->transState->finish(currState->fault, currState->req,
currState->tc, currState->mode);
pending = false;
nextWalk(currState->tc);
currState->req = NULL;
currState->tc = NULL;
currState->delayed = false;
}
else if (!currState->delayed) {
// delay is not set so there is no L2 to do
DPRINTF(TLBVerbose, "calling translateTiming again\n");
currState->fault = tlb->translateTiming(currState->req, currState->tc,
currState->transState, currState->mode);
pending = false;
nextWalk(currState->tc);
currState->req = NULL;
currState->tc = NULL;
currState->delayed = false;
delete currState;
} else {
// need to do L2 descriptor
stateQueueL2.push_back(currState);
}
currState = NULL;
}
void
TableWalker::doL2DescriptorWrapper()
{
currState = stateQueueL2.front();
assert(currState->delayed);
DPRINTF(TLBVerbose, "calling doL2Descriptor for vaddr:%#x\n",
currState->vaddr);
doL2Descriptor();
// Check if fault was generated
if (currState->fault != NoFault) {
currState->transState->finish(currState->fault, currState->req,
currState->tc, currState->mode);
}
else {
DPRINTF(TLBVerbose, "calling translateTiming again\n");
currState->fault = tlb->translateTiming(currState->req, currState->tc,
currState->transState, currState->mode);
}
stateQueueL2.pop_front();
pending = false;
nextWalk(currState->tc);
currState->req = NULL;
currState->tc = NULL;
currState->delayed = false;
delete currState;
currState = NULL;
}
void
TableWalker::nextWalk(ThreadContext *tc)
{
if (pendingQueue.size())
schedule(doProcessEvent, tc->getCpuPtr()->nextCycle(curTick+1));
}
ArmISA::TableWalker *
ArmTableWalkerParams::create()
{
return new ArmISA::TableWalker(this);
}