qemu-e2k/include/exec/cpu-defs.h

211 lines
6.9 KiB
C
Raw Normal View History

/*
* common defines for all CPUs
*
* Copyright (c) 2003 Fabrice Bellard
*
* This library is free software; you can redistribute it and/or
* modify it under the terms of the GNU Lesser General Public
* License as published by the Free Software Foundation; either
* version 2 of the License, or (at your option) any later version.
*
* This library is distributed in the hope that it will be useful,
* but WITHOUT ANY WARRANTY; without even the implied warranty of
* MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
* Lesser General Public License for more details.
*
* You should have received a copy of the GNU Lesser General Public
* License along with this library; if not, see <http://www.gnu.org/licenses/>.
*/
#ifndef CPU_DEFS_H
#define CPU_DEFS_H
#ifndef NEED_CPU_H
#error cpu.h included from common code
#endif
#include "qemu/host-utils.h"
cputlb: serialize tlb updates with env->tlb_lock Currently we rely on atomic operations for cross-CPU invalidations. There are two cases that these atomics miss: cross-CPU invalidations can race with either (1) vCPU threads flushing their TLB, which happens via memset, or (2) vCPUs calling tlb_reset_dirty on their TLB, which updates .addr_write with a regular store. This results in undefined behaviour, since we're mixing regular and atomic ops on concurrent accesses. Fix it by using tlb_lock, a per-vCPU lock. All updaters of tlb_table and the corresponding victim cache now hold the lock. The readers that do not hold tlb_lock must use atomic reads when reading .addr_write, since this field can be updated by other threads; the conversion to atomic reads is done in the next patch. Note that an alternative fix would be to expand the use of atomic ops. However, in the case of TLB flushes this would have a huge performance impact, since (1) TLB flushes can happen very frequently and (2) we currently use a full memory barrier to flush each TLB entry, and a TLB has many entries. Instead, acquiring the lock is barely slower than a full memory barrier since it is uncontended, and with a single lock acquisition we can flush the entire TLB. Tested-by: Alex Bennée <alex.bennee@linaro.org> Reviewed-by: Alex Bennée <alex.bennee@linaro.org> Signed-off-by: Emilio G. Cota <cota@braap.org> Message-Id: <20181009174557.16125-6-cota@braap.org> Signed-off-by: Richard Henderson <richard.henderson@linaro.org>
2018-10-09 19:45:56 +02:00
#include "qemu/thread.h"
#include "qemu/queue.h"
#ifdef CONFIG_TCG
#include "tcg-target.h"
#endif
#ifndef CONFIG_USER_ONLY
#include "exec/hwaddr.h"
#endif
#include "exec/memattrs.h"
#ifndef TARGET_LONG_BITS
#error TARGET_LONG_BITS must be defined before including this header
#endif
#define TARGET_LONG_SIZE (TARGET_LONG_BITS / 8)
/* target_ulong is the type of a virtual address */
#if TARGET_LONG_SIZE == 4
typedef int32_t target_long;
typedef uint32_t target_ulong;
#define TARGET_FMT_lx "%08x"
#define TARGET_FMT_ld "%d"
#define TARGET_FMT_lu "%u"
#elif TARGET_LONG_SIZE == 8
typedef int64_t target_long;
typedef uint64_t target_ulong;
#define TARGET_FMT_lx "%016" PRIx64
#define TARGET_FMT_ld "%" PRId64
#define TARGET_FMT_lu "%" PRIu64
#else
#error TARGET_LONG_SIZE undefined
#endif
#if !defined(CONFIG_USER_ONLY) && defined(CONFIG_TCG)
implementing victim TLB for QEMU system emulated TLB QEMU system mode page table walks are expensive. Taken by running QEMU qemu-system-x86_64 system mode on Intel PIN , a TLB miss and walking a 4-level page tables in guest Linux OS takes ~450 X86 instructions on average. QEMU system mode TLB is implemented using a directly-mapped hashtable. This structure suffers from conflict misses. Increasing the associativity of the TLB may not be the solution to conflict misses as all the ways may have to be walked in serial. A victim TLB is a TLB used to hold translations evicted from the primary TLB upon replacement. The victim TLB lies between the main TLB and its refill path. Victim TLB is of greater associativity (fully associative in this patch). It takes longer to lookup the victim TLB, but its likely better than a full page table walk. The memory translation path is changed as follows : Before Victim TLB: 1. Inline TLB lookup 2. Exit code cache on TLB miss. 3. Check for unaligned, IO accesses 4. TLB refill. 5. Do the memory access. 6. Return to code cache. After Victim TLB: 1. Inline TLB lookup 2. Exit code cache on TLB miss. 3. Check for unaligned, IO accesses 4. Victim TLB lookup. 5. If victim TLB misses, TLB refill 6. Do the memory access. 7. Return to code cache The advantage is that victim TLB can offer more associativity to a directly mapped TLB and thus potentially fewer page table walks while still keeping the time taken to flush within reasonable limits. However, placing a victim TLB before the refill path increase TLB refill path as the victim TLB is consulted before the TLB refill. The performance results demonstrate that the pros outweigh the cons. some performance results taken on SPECINT2006 train datasets and kernel boot and qemu configure script on an Intel(R) Xeon(R) CPU E5620 @ 2.40GHz Linux machine are shown in the Google Doc link below. https://docs.google.com/spreadsheets/d/1eiItzekZwNQOal_h-5iJmC4tMDi051m9qidi5_nwvH4/edit?usp=sharing In summary, victim TLB improves the performance of qemu-system-x86_64 by 11% on average on SPECINT2006, kernelboot and qemu configscript and with highest improvement of in 26% in 456.hmmer. And victim TLB does not result in any performance degradation in any of the measured benchmarks. Furthermore, the implemented victim TLB is architecture independent and is expected to benefit other architectures in QEMU as well. Although there are measurement fluctuations, the performance improvement is very significant and by no means in the range of noises. Signed-off-by: Xin Tong <trent.tong@gmail.com> Message-id: 1407202523-23553-1-git-send-email-trent.tong@gmail.com Reviewed-by: Peter Maydell <peter.maydell@linaro.org> Signed-off-by: Peter Maydell <peter.maydell@linaro.org>
2014-08-05 03:35:23 +02:00
/* use a fully associative victim tlb of 8 entries */
#define CPU_VTLB_SIZE 8
#if HOST_LONG_BITS == 32 && TARGET_LONG_BITS == 32
#define CPU_TLB_ENTRY_BITS 4
#else
#define CPU_TLB_ENTRY_BITS 5
#endif
#define CPU_TLB_DYN_MIN_BITS 6
#define CPU_TLB_DYN_DEFAULT_BITS 8
# if HOST_LONG_BITS == 32
/* Make sure we do not require a double-word shift for the TLB load */
# define CPU_TLB_DYN_MAX_BITS (32 - TARGET_PAGE_BITS)
# else /* HOST_LONG_BITS == 64 */
/*
* Assuming TARGET_PAGE_BITS==12, with 2**22 entries we can cover 2**(22+12) ==
* 2**34 == 16G of address space. This is roughly what one would expect a
* TLB to cover in a modern (as of 2018) x86_64 CPU. For instance, Intel
* Skylake's Level-2 STLB has 16 1G entries.
* Also, make sure we do not size the TLB past the guest's address space.
*/
# define CPU_TLB_DYN_MAX_BITS \
MIN(22, TARGET_VIRT_ADDR_SPACE_BITS - TARGET_PAGE_BITS)
# endif
typedef struct CPUTLBEntry {
/* bit TARGET_LONG_BITS to TARGET_PAGE_BITS : virtual address
bit TARGET_PAGE_BITS-1..4 : Nonzero for accesses that should not
go directly to ram.
bit 3 : indicates that the entry is invalid
bit 2..0 : zero
*/
union {
struct {
target_ulong addr_read;
target_ulong addr_write;
target_ulong addr_code;
/* Addend to virtual address to get host address. IO accesses
use the corresponding iotlb value. */
uintptr_t addend;
};
/* padding to get a power of two size */
uint8_t dummy[1 << CPU_TLB_ENTRY_BITS];
};
} CPUTLBEntry;
QEMU_BUILD_BUG_ON(sizeof(CPUTLBEntry) != (1 << CPU_TLB_ENTRY_BITS));
/* The IOTLB is not accessed directly inline by generated TCG code,
* so the CPUIOTLBEntry layout is not as critical as that of the
* CPUTLBEntry. (This is also why we don't want to combine the two
* structs into one.)
*/
typedef struct CPUIOTLBEntry {
/*
* @addr contains:
* - in the lower TARGET_PAGE_BITS, a physical section number
* - with the lower TARGET_PAGE_BITS masked off, an offset which
* must be added to the virtual address to obtain:
* + the ram_addr_t of the target RAM (if the physical section
* number is PHYS_SECTION_NOTDIRTY or PHYS_SECTION_ROM)
* + the offset within the target MemoryRegion (otherwise)
*/
hwaddr addr;
MemTxAttrs attrs;
} CPUIOTLBEntry;
/**
* struct CPUTLBWindow
* @begin_ns: host time (in ns) at the beginning of the time window
* @max_entries: maximum number of entries observed in the window
*
* See also: tlb_mmu_resize_locked()
*/
typedef struct CPUTLBWindow {
int64_t begin_ns;
size_t max_entries;
} CPUTLBWindow;
typedef struct CPUTLBDesc {
/*
* Describe a region covering all of the large pages allocated
* into the tlb. When any page within this region is flushed,
* we must flush the entire tlb. The region is matched if
* (addr & large_page_mask) == large_page_addr.
*/
target_ulong large_page_addr;
target_ulong large_page_mask;
/* The next index to use in the tlb victim table. */
size_t vindex;
CPUTLBWindow window;
size_t n_used_entries;
} CPUTLBDesc;
/*
* Data elements that are shared between all MMU modes.
*/
typedef struct CPUTLBCommon {
/* Serialize updates to tlb_table and tlb_v_table, and others as noted. */
QemuSpin lock;
/*
* Within dirty, for each bit N, modifications have been made to
* mmu_idx N since the last time that mmu_idx was flushed.
* Protected by tlb_c.lock.
*/
uint16_t dirty;
/*
* Statistics. These are not lock protected, but are read and
* written atomically. This allows the monitor to print a snapshot
* of the stats without interfering with the cpu.
*/
size_t full_flush_count;
size_t part_flush_count;
size_t elide_flush_count;
} CPUTLBCommon;
# define CPU_TLB \
/* tlb_mask[i] contains (n_entries - 1) << CPU_TLB_ENTRY_BITS */ \
uintptr_t tlb_mask[NB_MMU_MODES]; \
CPUTLBEntry *tlb_table[NB_MMU_MODES];
# define CPU_IOTLB \
CPUIOTLBEntry *iotlb[NB_MMU_MODES];
/*
* The meaning of each of the MMU modes is defined in the target code.
* Note that NB_MMU_MODES is not yet defined; we can only reference it
* within preprocessor defines that will be expanded later.
*/
#define CPU_COMMON_TLB \
CPUTLBCommon tlb_c; \
CPUTLBDesc tlb_d[NB_MMU_MODES]; \
CPU_TLB \
implementing victim TLB for QEMU system emulated TLB QEMU system mode page table walks are expensive. Taken by running QEMU qemu-system-x86_64 system mode on Intel PIN , a TLB miss and walking a 4-level page tables in guest Linux OS takes ~450 X86 instructions on average. QEMU system mode TLB is implemented using a directly-mapped hashtable. This structure suffers from conflict misses. Increasing the associativity of the TLB may not be the solution to conflict misses as all the ways may have to be walked in serial. A victim TLB is a TLB used to hold translations evicted from the primary TLB upon replacement. The victim TLB lies between the main TLB and its refill path. Victim TLB is of greater associativity (fully associative in this patch). It takes longer to lookup the victim TLB, but its likely better than a full page table walk. The memory translation path is changed as follows : Before Victim TLB: 1. Inline TLB lookup 2. Exit code cache on TLB miss. 3. Check for unaligned, IO accesses 4. TLB refill. 5. Do the memory access. 6. Return to code cache. After Victim TLB: 1. Inline TLB lookup 2. Exit code cache on TLB miss. 3. Check for unaligned, IO accesses 4. Victim TLB lookup. 5. If victim TLB misses, TLB refill 6. Do the memory access. 7. Return to code cache The advantage is that victim TLB can offer more associativity to a directly mapped TLB and thus potentially fewer page table walks while still keeping the time taken to flush within reasonable limits. However, placing a victim TLB before the refill path increase TLB refill path as the victim TLB is consulted before the TLB refill. The performance results demonstrate that the pros outweigh the cons. some performance results taken on SPECINT2006 train datasets and kernel boot and qemu configure script on an Intel(R) Xeon(R) CPU E5620 @ 2.40GHz Linux machine are shown in the Google Doc link below. https://docs.google.com/spreadsheets/d/1eiItzekZwNQOal_h-5iJmC4tMDi051m9qidi5_nwvH4/edit?usp=sharing In summary, victim TLB improves the performance of qemu-system-x86_64 by 11% on average on SPECINT2006, kernelboot and qemu configscript and with highest improvement of in 26% in 456.hmmer. And victim TLB does not result in any performance degradation in any of the measured benchmarks. Furthermore, the implemented victim TLB is architecture independent and is expected to benefit other architectures in QEMU as well. Although there are measurement fluctuations, the performance improvement is very significant and by no means in the range of noises. Signed-off-by: Xin Tong <trent.tong@gmail.com> Message-id: 1407202523-23553-1-git-send-email-trent.tong@gmail.com Reviewed-by: Peter Maydell <peter.maydell@linaro.org> Signed-off-by: Peter Maydell <peter.maydell@linaro.org>
2014-08-05 03:35:23 +02:00
CPUTLBEntry tlb_v_table[NB_MMU_MODES][CPU_VTLB_SIZE]; \
CPU_IOTLB \
CPUIOTLBEntry iotlb_v[NB_MMU_MODES][CPU_VTLB_SIZE];
#else
#define CPU_COMMON_TLB
#endif
#define CPU_COMMON \
/* soft mmu support */ \
CPU_COMMON_TLB \
#endif