[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index]
Re: [PATCH v3 17/36] plugins: force slow path when plugins instrument me
From: |
Alex Bennée |
Subject: |
Re: [PATCH v3 17/36] plugins: force slow path when plugins instrument memory ops |
Date: |
Thu, 29 Jun 2023 12:59:50 +0100 |
User-agent: |
mu4e 1.11.7; emacs 29.0.92 |
Richard Henderson <richard.henderson@linaro.org> writes:
> On 6/28/23 11:06, Alex Bennée wrote:
>>> I thought we dropped this patch until we could do something with TLB
>>> accesses.
>> I did suggest something like:
>> --8<---------------cut here---------------start------------->8---
>> modified include/hw/core/cpu.h
>> @@ -80,10 +80,24 @@ DECLARE_CLASS_CHECKERS(CPUClass, CPU,
>> typedef struct ArchCPU CpuInstanceType; \
>> OBJECT_DECLARE_TYPE(ArchCPU, CpuClassType, CPU_MODULE_OBJ_NAME);
>> +/**
>> + * typedef MMUAccessType - describe the type of access for cputlb
>> + *
>> + * When handling the access to memory we need to know the type of
>> + * access we are doing. Loads and store rely on read and write page
>> + * permissions where as the instruction fetch relies on execute
>> + * permissions. Additional bits are used for TLB access so we can
>> + * suppress instrumentation of memory when the CPU is probing.
>> + */
>> typedef enum MMUAccessType {
>> MMU_DATA_LOAD = 0,
>> MMU_DATA_STORE = 1,
>> - MMU_INST_FETCH = 2
>> + MMU_INST_FETCH = 2,
>> + /* MMU Mask */
>> + MMU_VALID_MASK = (MMU_DATA_LOAD | MMU_DATA_STORE | MMU_INST_FETCH),
>> + /* Represents the CPU walking the page table */
>> + MMU_TLB_ACCESS = 0x4,
>> + MMU_TLB_LOAD = MMU_DATA_LOAD | MMU_TLB_ACCESS
>> } MMUAccessType;
>> typedef struct CPUWatchpoint CPUWatchpoint;
>> modified accel/tcg/cputlb.c
>> @@ -1503,11 +1503,12 @@ static void notdirty_write(CPUState *cpu, vaddr
>> mem_vaddr, unsigned size,
>> }
>> static int probe_access_internal(CPUArchState *env, target_ulong
>> addr,
>> - int fault_size, MMUAccessType access_type,
>> + int fault_size, MMUAccessType
>> full_access_type,
>> int mmu_idx, bool nonfault,
>> void **phost, CPUTLBEntryFull **pfull,
>> uintptr_t retaddr)
>> {
>> + MMUAccessType access_type = full_access_type & MMU_VALID_MASK;
>> uintptr_t index = tlb_index(env, mmu_idx, addr);
>> CPUTLBEntry *entry = tlb_entry(env, mmu_idx, addr);
>> target_ulong tlb_addr = tlb_read_idx(entry, access_type);
>> @@ -1546,7 +1547,9 @@ static int probe_access_internal(CPUArchState *env,
>> target_ulong addr,
>> /* Fold all "mmio-like" bits into TLB_MMIO. This is not RAM. */
>> if (unlikely(flags & ~(TLB_WATCHPOINT | TLB_NOTDIRTY))
>> ||
>> - (access_type != MMU_INST_FETCH &&
>> cpu_plugin_mem_cbs_enabled(env_cpu(env)))) {
>> + (access_type != MMU_INST_FETCH &&
>> + !(full_access_type & MMU_TLB_ACCESS) &&
>> + cpu_plugin_mem_cbs_enabled(env_cpu(env)))) {
>> *phost = NULL;
>> return TLB_MMIO;
>> }
>> --8<---------------cut here---------------end--------------->8---
>> and then we can apply MMU_TLB_LOAD as the type in the page walking
>> code.
>> I wanted to know if that was the sort of thing you where thinking off or
>> if that is too ugly.
>
> It's not implausible, but probably not ideal.
>
>> The other option is a specific probe_access_* function for TLB type
>> operations.
>
> Or that, yes.
>
> I'm confused that you'd simply re-include this patch as-is when it has
> known errors.
I was hoping for some testing from the people that reported the fault.
Anyway this is what I have now:
--8<---------------cut here---------------start------------->8---
plugins: force slow path when plugins instrument memory ops
The lack of SVE memory instrumentation has been an omission in plugin
handling since it was introduced. Fortunately we can utilise the
probe_* functions to force all all memory access to follow the slow
path. We do this by checking the access type and presence of plugin
memory callbacks and if set return the TLB_MMIO flag.
We have to jump through a few hoops in user mode to re-use the flag
but it was the desired effect:
./qemu-system-aarch64 -display none -serial mon:stdio \
-M virt -cpu max -semihosting-config enable=on \
-kernel ./tests/tcg/aarch64-softmmu/memory-sve \
-plugin ./contrib/plugins/libexeclog.so,ifilter=st1w,afilter=0x40001808 -d
plugin
gives (disas doesn't currently understand st1w):
0, 0x40001808, 0xe54342a0, ".byte 0xa0, 0x42, 0x43, 0xe5", store, 0x40213010,
RAM, store, 0x40213014, RAM, store, 0x40213018, RAM
And for user-mode:
./qemu-aarch64 \
-plugin contrib/plugins/libexeclog.so,afilter=0x4007c0 \
-d plugin \
./tests/tcg/aarch64-linux-user/sha512-sve
gives:
1..10
ok 1 - do_test(&tests[i])
0, 0x4007c0, 0xa4004b80, ".byte 0x80, 0x4b, 0x00, 0xa4", load, 0x5500800370,
load, 0x5500800371, load, 0x5500800372, load, 0x5500800373, load, 0x5500800374,
load, 0x5500800375, load, 0x5500800376, load, 0x5500800377, load, 0x5500800378,
load, 0x5500800379, load, 0x550080037a, load, 0x550080037b, load, 0x550080037c,
load, 0x550080037d, load, 0x550080037e, load, 0x550080037f, load, 0x5500800380,
load, 0x5500800381, load, 0x5500800382, load, 0x5500800383, load, 0x5500800384,
load, 0x5500800385, load, 0x5500800386, lo
ad, 0x5500800387, load, 0x5500800388, load, 0x5500800389, load, 0x550080038a,
load, 0x550080038b, load, 0x550080038c, load, 0x550080038d, load, 0x550080038e,
load, 0x550080038f, load, 0x5500800390, load, 0x5500800391, load, 0x5500800392,
load, 0x5500800393, load, 0x5500800394, load, 0x5500800395, load, 0x5500800396,
load, 0x5500800397, load, 0x5500800398, load, 0x5500800399, load, 0x550080039a,
load, 0x550080039b, load, 0x550080039c, load, 0x550080039d, load, 0x550080039e,
load, 0x550080039f, load, 0x55008003a0, load, 0x55008003a1, load, 0x55008003a2,
load, 0x55008003a3, load, 0x55008003a4, load, 0x55008003a5, load, 0x55008003a6,
load, 0x55008003a7, load, 0x55008003a8, load, 0x55008003a9, load, 0x55008003aa,
load, 0x55008003ab, load, 0x55008003ac, load, 0x55008003ad, load, 0x55008003ae,
load, 0x55008003af
(4007c0 is the ld1b in the sha512-sve)
Signed-off-by: Alex Bennée <alex.bennee@linaro.org>
Reviewed-by: Richard Henderson <richard.henderson@linaro.org>
Cc: Robert Henry <robhenry@microsoft.com>
Cc: Aaron Lindsay <aaron@os.amperecomputing.com>
---
v2
- allow TLB_MMIO to appear in user-mode probe_access
v3
- checkpatch cleanups
v4
- add new probe helper for mmu's, add check_mem_cbs to probe_internal
8 files changed, 94 insertions(+), 21 deletions(-)
include/exec/cpu-all.h | 2 +-
include/exec/exec-all.h | 23 ++++++++++++++++++++++
include/hw/core/cpu.h | 17 +++++++++++++++++
accel/tcg/cputlb.c | 40 ++++++++++++++++++++++++++++++++-------
accel/tcg/user-exec.c | 8 ++++++--
target/arm/ptw.c | 13 ++++++-------
target/arm/tcg/sve_helper.c | 4 ----
tests/tcg/aarch64/Makefile.target | 8 ++++++++
modified include/exec/cpu-all.h
@@ -301,7 +301,7 @@ CPUArchState *cpu_copy(CPUArchState *env);
* be signaled by probe_access_flags().
*/
#define TLB_INVALID_MASK (1 << (TARGET_PAGE_BITS_MIN - 1))
-#define TLB_MMIO 0
+#define TLB_MMIO (1 << (TARGET_PAGE_BITS_MIN - 2))
#define TLB_WATCHPOINT 0
#else
modified include/exec/exec-all.h
@@ -464,6 +464,29 @@ int probe_access_full(CPUArchState *env, vaddr addr, int
size,
MMUAccessType access_type, int mmu_idx,
bool nonfault, void **phost,
CPUTLBEntryFull **pfull, uintptr_t retaddr);
+
+/**
+ * probe_access_mmu() - Like probe_access_full except cannot fault and
+ * doesn't trigger instrumentation.
+ *
+ * @env: CPUArchState
+ * @vaddr: virtual address to probe
+ * @size: size of the probe
+ * @access_type: read, write or execute permission
+ * @mmu_idx: softmmu index
+ * @phost: ptr to return value host address or NULL
+ * @pfull: ptr to return value CPUTLBEntryFull structure or NULL
+ *
+ * The CPUTLBEntryFull structure returned via @pfull is transient
+ * and must be consumed or copied immediately, before any further
+ * access or changes to TLB @mmu_idx.
+ *
+ * Returns: TLB flags as per probe_access_flags()
+ */
+int probe_access_full_mmu(CPUArchState *env, vaddr addr, int size,
+ MMUAccessType access_type, int mmu_idx,
+ void **phost, CPUTLBEntryFull **pfull);
+
#endif
/* Hide the qatomic_read to make code a little easier on the eyes */
modified include/hw/core/cpu.h
@@ -976,6 +976,23 @@ void cpu_watchpoint_remove_by_ref(CPUState *cpu,
CPUWatchpoint *watchpoint);
void cpu_watchpoint_remove_all(CPUState *cpu, int mask);
#endif
+/**
+ * cpu_plugin_mem_cbs_enabled() - are plugin memory callbacks enabled?
+ * @cs: CPUState pointer
+ *
+ * The memory callbacks are installed if a plugin has instrumented an
+ * instruction for memory. This can be useful to know if you want to
+ * force a slow path for a series of memory accesses.
+ */
+static inline bool cpu_plugin_mem_cbs_enabled(const CPUState *cpu)
+{
+#ifdef CONFIG_PLUGIN
+ return !!cpu->plugin_mem_cbs;
+#else
+ return false;
+#endif
+}
+
/**
* cpu_get_address_space:
* @cpu: CPU to get address space from
modified accel/tcg/cputlb.c
@@ -1513,13 +1513,14 @@ static int probe_access_internal(CPUArchState *env,
vaddr addr,
int fault_size, MMUAccessType access_type,
int mmu_idx, bool nonfault,
void **phost, CPUTLBEntryFull **pfull,
- uintptr_t retaddr)
+ uintptr_t retaddr, bool check_mem_cbs)
{
uintptr_t index = tlb_index(env, mmu_idx, addr);
CPUTLBEntry *entry = tlb_entry(env, mmu_idx, addr);
uint64_t tlb_addr = tlb_read_idx(entry, access_type);
vaddr page_addr = addr & TARGET_PAGE_MASK;
int flags = TLB_FLAGS_MASK & ~TLB_FORCE_SLOW;
+ bool force_mmio = check_mem_cbs &&
cpu_plugin_mem_cbs_enabled(env_cpu(env));
CPUTLBEntryFull *full;
if (!tlb_hit_page(tlb_addr, page_addr)) {
@@ -1553,7 +1554,9 @@ static int probe_access_internal(CPUArchState *env, vaddr
addr,
flags |= full->slow_flags[access_type];
/* Fold all "mmio-like" bits into TLB_MMIO. This is not RAM. */
- if (unlikely(flags & ~(TLB_WATCHPOINT | TLB_NOTDIRTY))) {
+ if (unlikely(flags & ~(TLB_WATCHPOINT | TLB_NOTDIRTY))
+ ||
+ (access_type != MMU_INST_FETCH && force_mmio)) {
*phost = NULL;
return TLB_MMIO;
}
@@ -1569,7 +1572,7 @@ int probe_access_full(CPUArchState *env, vaddr addr, int
size,
uintptr_t retaddr)
{
int flags = probe_access_internal(env, addr, size, access_type, mmu_idx,
- nonfault, phost, pfull, retaddr);
+ nonfault, phost, pfull, retaddr, true);
/* Handle clean RAM pages. */
if (unlikely(flags & TLB_NOTDIRTY)) {
@@ -1580,6 +1583,29 @@ int probe_access_full(CPUArchState *env, vaddr addr, int
size,
return flags;
}
+int probe_access_full_mmu(CPUArchState *env, vaddr addr, int size,
+ MMUAccessType access_type, int mmu_idx,
+ void **phost, CPUTLBEntryFull **pfull)
+{
+ void *discard_phost;
+ CPUTLBEntryFull *discard_tlb;
+
+ /* privately handle users that don't need full results */
+ phost = phost ? phost : &discard_phost;
+ pfull = pfull ? pfull : &discard_tlb;
+
+ int flags = probe_access_internal(env, addr, size, access_type, mmu_idx,
+ true, phost, pfull, 0, false);
+
+ /* Handle clean RAM pages. */
+ if (unlikely(flags & TLB_NOTDIRTY)) {
+ notdirty_write(env_cpu(env), addr, 1, *pfull, 0);
+ flags &= ~TLB_NOTDIRTY;
+ }
+
+ return flags;
+}
+
int probe_access_flags(CPUArchState *env, vaddr addr, int size,
MMUAccessType access_type, int mmu_idx,
bool nonfault, void **phost, uintptr_t retaddr)
@@ -1590,7 +1616,7 @@ int probe_access_flags(CPUArchState *env, vaddr addr, int
size,
g_assert(-(addr | TARGET_PAGE_MASK) >= size);
flags = probe_access_internal(env, addr, size, access_type, mmu_idx,
- nonfault, phost, &full, retaddr);
+ nonfault, phost, &full, retaddr, true);
/* Handle clean RAM pages. */
if (unlikely(flags & TLB_NOTDIRTY)) {
@@ -1611,7 +1637,7 @@ void *probe_access(CPUArchState *env, vaddr addr, int
size,
g_assert(-(addr | TARGET_PAGE_MASK) >= size);
flags = probe_access_internal(env, addr, size, access_type, mmu_idx,
- false, &host, &full, retaddr);
+ false, &host, &full, retaddr, true);
/* Per the interface, size == 0 merely faults the access. */
if (size == 0) {
@@ -1644,7 +1670,7 @@ void *tlb_vaddr_to_host(CPUArchState *env, abi_ptr addr,
int flags;
flags = probe_access_internal(env, addr, 0, access_type,
- mmu_idx, true, &host, &full, 0);
+ mmu_idx, true, &host, &full, 0, false);
/* No combination of flags are expected by the caller. */
return flags ? NULL : host;
@@ -1667,7 +1693,7 @@ tb_page_addr_t get_page_addr_code_hostp(CPUArchState
*env, vaddr addr,
void *p;
(void)probe_access_internal(env, addr, 1, MMU_INST_FETCH,
- cpu_mmu_index(env, true), false, &p, &full, 0);
+ cpu_mmu_index(env, true), false, &p, &full, 0,
false);
if (p == NULL) {
return -1;
}
modified accel/tcg/user-exec.c
@@ -745,6 +745,10 @@ static int probe_access_internal(CPUArchState *env, vaddr
addr,
if (guest_addr_valid_untagged(addr)) {
int page_flags = page_get_flags(addr);
if (page_flags & acc_flag) {
+ if ((acc_flag == PAGE_READ || acc_flag == PAGE_WRITE)
+ && cpu_plugin_mem_cbs_enabled(env_cpu(env))) {
+ return TLB_MMIO;
+ }
return 0; /* success */
}
maperr = !(page_flags & PAGE_VALID);
@@ -767,7 +771,7 @@ int probe_access_flags(CPUArchState *env, vaddr addr, int
size,
g_assert(-(addr | TARGET_PAGE_MASK) >= size);
flags = probe_access_internal(env, addr, size, access_type, nonfault, ra);
- *phost = flags ? NULL : g2h(env_cpu(env), addr);
+ *phost = (flags & TLB_INVALID_MASK) ? NULL : g2h(env_cpu(env), addr);
return flags;
}
@@ -778,7 +782,7 @@ void *probe_access(CPUArchState *env, vaddr addr, int size,
g_assert(-(addr | TARGET_PAGE_MASK) >= size);
flags = probe_access_internal(env, addr, size, access_type, false, ra);
- g_assert(flags == 0);
+ g_assert((flags & ~TLB_MMIO) == 0);
return size ? g2h(env_cpu(env), addr) : NULL;
}
modified target/arm/ptw.c
@@ -489,9 +489,9 @@ static bool S1_ptw_translate(CPUARMState *env, S1Translate
*ptw,
int flags;
env->tlb_fi = fi;
- flags = probe_access_full(env, addr, 0, MMU_DATA_LOAD,
- arm_to_core_mmu_idx(s2_mmu_idx),
- true, &ptw->out_host, &full, 0);
+ flags = probe_access_full_mmu(env, addr, 0, MMU_DATA_LOAD,
+ arm_to_core_mmu_idx(s2_mmu_idx),
+ &ptw->out_host, &full);
env->tlb_fi = NULL;
if (unlikely(flags & TLB_INVALID_MASK)) {
@@ -644,12 +644,11 @@ static uint64_t arm_casq_ptw(CPUARMState *env, uint64_t
old_val,
*/
if (unlikely(!ptw->out_rw)) {
int flags;
- void *discard;
env->tlb_fi = fi;
- flags = probe_access_flags(env, ptw->out_virt, 0, MMU_DATA_STORE,
- arm_to_core_mmu_idx(ptw->in_ptw_idx),
- true, &discard, 0);
+ flags = probe_access_full_mmu(env, ptw->out_virt, 0,
+ MMU_DATA_STORE,
arm_to_core_mmu_idx(ptw->in_ptw_idx),
+ NULL, NULL);
env->tlb_fi = NULL;
if (unlikely(flags & TLB_INVALID_MASK)) {
modified target/arm/tcg/sve_helper.c
@@ -5688,9 +5688,6 @@ void sve_ldN_r(CPUARMState *env, uint64_t *vg, const
target_ulong addr,
flags = info.page[0].flags | info.page[1].flags;
if (unlikely(flags != 0)) {
-#ifdef CONFIG_USER_ONLY
- g_assert_not_reached();
-#else
/*
* At least one page includes MMIO.
* Any bus operation can fail with cpu_transaction_failed,
@@ -5727,7 +5724,6 @@ void sve_ldN_r(CPUARMState *env, uint64_t *vg, const
target_ulong addr,
memcpy(&env->vfp.zregs[(rd + i) & 31], &scratch[i], reg_max);
}
return;
-#endif
}
/* The entire operation is in RAM, on valid pages. */
--8<---------------cut here---------------end--------------->8---
--
Alex Bennée
Virtualisation Tech Lead @ Linaro
- [PATCH v3 09/36] tests/docker: add test-fuzz, (continued)
- [PATCH v3 09/36] tests/docker: add test-fuzz, Alex Bennée, 2023/06/27
- [PATCH v3 13/36] tests/lcitool: add an explicit gcc-native package, Alex Bennée, 2023/06/27
- [PATCH v3 34/36] gdbstub: Add support for info proc mappings, Alex Bennée, 2023/06/27
- [PATCH v3 30/36] linux-user: Add "safe" parameter to do_guest_openat(), Alex Bennée, 2023/06/27
- [PATCH v3 10/36] Makefile: add lcitool-refresh to UNCHECKED_GOALS, Alex Bennée, 2023/06/27
- [PATCH v3 07/36] scripts/oss-fuzz: add a suppression for keymap, Alex Bennée, 2023/06/27
- [PATCH v3 17/36] plugins: force slow path when plugins instrument memory ops, Alex Bennée, 2023/06/27
[PATCH v3 22/36] include/hw/qdev-core: fixup kerneldoc annotations, Alex Bennée, 2023/06/27
[PATCH v3 19/36] plugins: update lockstep to use g_memdup2, Alex Bennée, 2023/06/27
[PATCH v3 24/36] docs/devel: split qom-api reference into new file, Alex Bennée, 2023/06/27
[PATCH v3 20/36] docs/devel: add some front matter to the devel index, Alex Bennée, 2023/06/27
[PATCH v3 33/36] gdbstub: Report the actual qemu-user pid, Alex Bennée, 2023/06/27
[PATCH v3 23/36] docs/devel/qom.rst: Correct code style, Alex Bennée, 2023/06/27
[PATCH v3 36/36] tests/tcg: Add a test for info proc mappings, Alex Bennée, 2023/06/27