Re: [PATCH v3 2/2] riscv: Fix text patching when IPI are used
Conor Dooley writes: > On Tue, Mar 05, 2024 at 08:33:30AM +0530, Anup Patel wrote: >> On Tue, Mar 5, 2024 at 1:54 AM Björn Töpel wrote: >> > >> > Conor Dooley writes: >> > >> > > On Thu, Feb 29, 2024 at 01:10:56PM +0100, Alexandre Ghiti wrote: >> > >> For now, we use stop_machine() to patch the text and when we use IPIs >> > >> for >> > >> remote icache flushes (which is emitted in patch_text_nosync()), the >> > >> system >> > >> hangs. >> > >> >> > >> So instead, make sure every CPU executes the stop_machine() patching >> > >> function and emit a local icache flush there. >> > >> >> > >> Co-developed-by: Björn Töpel >> > >> Signed-off-by: Björn Töpel >> > >> Signed-off-by: Alexandre Ghiti >> > >> Reviewed-by: Andrea Parri >> > > >> > > What commit does this fix? >> > >> > Hmm. The bug is exposed when the AIA IPI are introduced, and used >> > (instead of the firmware-based). >> > >> > I'm not sure this is something we'd like backported, but rather a >> > prerequisite to AIA. >> > >> > @Anup @Alex WDYT? >> > >> >> The current text patching never considered IPIs being injected >> directly in S-mode from hart to another so we are seeing this >> issue now with AIA IPIs. >> >> We certainly don't need to backport this fix since it's more >> of a preparatory fix for AIA IPIs. > > Whether or not this is backportable, if it fixes a bug, it should get > a Fixes: tag for the commit that it fixes. Fixes: isn't "the backport" > tag, cc: stable is. I guess the question is if this *is* a fix, or rather a change required for AIA (not a fix).
Re: [PATCH v3 2/2] riscv: Fix text patching when IPI are used
On Tue, Mar 05, 2024 at 08:33:30AM +0530, Anup Patel wrote: > On Tue, Mar 5, 2024 at 1:54 AM Björn Töpel wrote: > > > > Conor Dooley writes: > > > > > On Thu, Feb 29, 2024 at 01:10:56PM +0100, Alexandre Ghiti wrote: > > >> For now, we use stop_machine() to patch the text and when we use IPIs for > > >> remote icache flushes (which is emitted in patch_text_nosync()), the > > >> system > > >> hangs. > > >> > > >> So instead, make sure every CPU executes the stop_machine() patching > > >> function and emit a local icache flush there. > > >> > > >> Co-developed-by: Björn Töpel > > >> Signed-off-by: Björn Töpel > > >> Signed-off-by: Alexandre Ghiti > > >> Reviewed-by: Andrea Parri > > > > > > What commit does this fix? > > > > Hmm. The bug is exposed when the AIA IPI are introduced, and used > > (instead of the firmware-based). > > > > I'm not sure this is something we'd like backported, but rather a > > prerequisite to AIA. > > > > @Anup @Alex WDYT? > > > > The current text patching never considered IPIs being injected > directly in S-mode from hart to another so we are seeing this > issue now with AIA IPIs. > > We certainly don't need to backport this fix since it's more > of a preparatory fix for AIA IPIs. Whether or not this is backportable, if it fixes a bug, it should get a Fixes: tag for the commit that it fixes. Fixes: isn't "the backport" tag, cc: stable is. signature.asc Description: PGP signature
Re: [PATCH v3 2/2] riscv: Fix text patching when IPI are used
On Tue, Mar 5, 2024 at 1:54 AM Björn Töpel wrote: > > Conor Dooley writes: > > > On Thu, Feb 29, 2024 at 01:10:56PM +0100, Alexandre Ghiti wrote: > >> For now, we use stop_machine() to patch the text and when we use IPIs for > >> remote icache flushes (which is emitted in patch_text_nosync()), the system > >> hangs. > >> > >> So instead, make sure every CPU executes the stop_machine() patching > >> function and emit a local icache flush there. > >> > >> Co-developed-by: Björn Töpel > >> Signed-off-by: Björn Töpel > >> Signed-off-by: Alexandre Ghiti > >> Reviewed-by: Andrea Parri > > > > What commit does this fix? > > Hmm. The bug is exposed when the AIA IPI are introduced, and used > (instead of the firmware-based). > > I'm not sure this is something we'd like backported, but rather a > prerequisite to AIA. > > @Anup @Alex WDYT? > The current text patching never considered IPIs being injected directly in S-mode from hart to another so we are seeing this issue now with AIA IPIs. We certainly don't need to backport this fix since it's more of a preparatory fix for AIA IPIs. Regards, Anup
Re: [PATCH v3 2/2] riscv: Fix text patching when IPI are used
Conor Dooley writes: > On Thu, Feb 29, 2024 at 01:10:56PM +0100, Alexandre Ghiti wrote: >> For now, we use stop_machine() to patch the text and when we use IPIs for >> remote icache flushes (which is emitted in patch_text_nosync()), the system >> hangs. >> >> So instead, make sure every CPU executes the stop_machine() patching >> function and emit a local icache flush there. >> >> Co-developed-by: Björn Töpel >> Signed-off-by: Björn Töpel >> Signed-off-by: Alexandre Ghiti >> Reviewed-by: Andrea Parri > > What commit does this fix? Hmm. The bug is exposed when the AIA IPI are introduced, and used (instead of the firmware-based). I'm not sure this is something we'd like backported, but rather a prerequisite to AIA. @Anup @Alex WDYT?
Re: [PATCH v3 2/2] riscv: Fix text patching when IPI are used
On Thu, Feb 29, 2024 at 01:10:56PM +0100, Alexandre Ghiti wrote: > For now, we use stop_machine() to patch the text and when we use IPIs for > remote icache flushes (which is emitted in patch_text_nosync()), the system > hangs. > > So instead, make sure every CPU executes the stop_machine() patching > function and emit a local icache flush there. > > Co-developed-by: Björn Töpel > Signed-off-by: Björn Töpel > Signed-off-by: Alexandre Ghiti > Reviewed-by: Andrea Parri What commit does this fix? signature.asc Description: PGP signature
[PATCH v3 2/2] riscv: Fix text patching when IPI are used
For now, we use stop_machine() to patch the text and when we use IPIs for
remote icache flushes (which is emitted in patch_text_nosync()), the system
hangs.
So instead, make sure every CPU executes the stop_machine() patching
function and emit a local icache flush there.
Co-developed-by: Björn Töpel
Signed-off-by: Björn Töpel
Signed-off-by: Alexandre Ghiti
Reviewed-by: Andrea Parri
---
arch/riscv/include/asm/patch.h | 1 +
arch/riscv/kernel/ftrace.c | 44 ++
arch/riscv/kernel/patch.c | 16 +
3 files changed, 53 insertions(+), 8 deletions(-)
diff --git a/arch/riscv/include/asm/patch.h b/arch/riscv/include/asm/patch.h
index e88b52d39eac..9f5d6e14c405 100644
--- a/arch/riscv/include/asm/patch.h
+++ b/arch/riscv/include/asm/patch.h
@@ -6,6 +6,7 @@
#ifndef _ASM_RISCV_PATCH_H
#define _ASM_RISCV_PATCH_H
+int patch_insn_write(void *addr, const void *insn, size_t len);
int patch_text_nosync(void *addr, const void *insns, size_t len);
int patch_text_set_nosync(void *addr, u8 c, size_t len);
int patch_text(void *addr, u32 *insns, int ninsns);
diff --git a/arch/riscv/kernel/ftrace.c b/arch/riscv/kernel/ftrace.c
index f5aa24d9e1c1..4f4987a6d83d 100644
--- a/arch/riscv/kernel/ftrace.c
+++ b/arch/riscv/kernel/ftrace.c
@@ -8,6 +8,7 @@
#include
#include
#include
+#include
#include
#include
@@ -75,8 +76,7 @@ static int __ftrace_modify_call(unsigned long hook_pos,
unsigned long target,
make_call_t0(hook_pos, target, call);
/* Replace the auipc-jalr pair at once. Return -EPERM on write error. */
- if (patch_text_nosync
- ((void *)hook_pos, enable ? call : nops, MCOUNT_INSN_SIZE))
+ if (patch_insn_write((void *)hook_pos, enable ? call : nops,
MCOUNT_INSN_SIZE))
return -EPERM;
return 0;
@@ -88,7 +88,7 @@ int ftrace_make_call(struct dyn_ftrace *rec, unsigned long
addr)
make_call_t0(rec->ip, addr, call);
- if (patch_text_nosync((void *)rec->ip, call, MCOUNT_INSN_SIZE))
+ if (patch_insn_write((void *)rec->ip, call, MCOUNT_INSN_SIZE))
return -EPERM;
return 0;
@@ -99,7 +99,7 @@ int ftrace_make_nop(struct module *mod, struct dyn_ftrace
*rec,
{
unsigned int nops[2] = {NOP4, NOP4};
- if (patch_text_nosync((void *)rec->ip, nops, MCOUNT_INSN_SIZE))
+ if (patch_insn_write((void *)rec->ip, nops, MCOUNT_INSN_SIZE))
return -EPERM;
return 0;
@@ -134,6 +134,42 @@ int ftrace_update_ftrace_func(ftrace_func_t func)
return ret;
}
+
+struct ftrace_modify_param {
+ int command;
+ atomic_t cpu_count;
+};
+
+static int __ftrace_modify_code(void *data)
+{
+ struct ftrace_modify_param *param = data;
+
+ if (atomic_inc_return(>cpu_count) == num_online_cpus()) {
+ ftrace_modify_all_code(param->command);
+ /*
+* Make sure the patching store is effective *before* we
+* increment the counter which releases all waiting CPUs
+* by using the release variant of atomic increment. The
+* release pairs with the call to local_flush_icache_all()
+* on the waiting CPU.
+*/
+ atomic_inc_return_release(>cpu_count);
+ } else {
+ while (atomic_read(>cpu_count) <= num_online_cpus())
+ cpu_relax();
+ }
+
+ local_flush_icache_all();
+
+ return 0;
+}
+
+void arch_ftrace_update_code(int command)
+{
+ struct ftrace_modify_param param = { command, ATOMIC_INIT(0) };
+
+ stop_machine(__ftrace_modify_code, , cpu_online_mask);
+}
#endif
#ifdef CONFIG_DYNAMIC_FTRACE_WITH_REGS
diff --git a/arch/riscv/kernel/patch.c b/arch/riscv/kernel/patch.c
index 0b5c16dfe3f4..9a1bce1adf5a 100644
--- a/arch/riscv/kernel/patch.c
+++ b/arch/riscv/kernel/patch.c
@@ -188,7 +188,7 @@ int patch_text_set_nosync(void *addr, u8 c, size_t len)
}
NOKPROBE_SYMBOL(patch_text_set_nosync);
-static int patch_insn_write(void *addr, const void *insn, size_t len)
+int patch_insn_write(void *addr, const void *insn, size_t len)
{
size_t patched = 0;
size_t size;
@@ -232,15 +232,23 @@ static int patch_text_cb(void *data)
if (atomic_inc_return(>cpu_count) == num_online_cpus()) {
for (i = 0; ret == 0 && i < patch->ninsns; i++) {
len = GET_INSN_LENGTH(patch->insns[i]);
- ret = patch_text_nosync(patch->addr + i * len,
- >insns[i], len);
+ ret = patch_insn_write(patch->addr + i * len,
>insns[i], len);
}
- atomic_inc(>cpu_count);
+ /*
+* Make sure the patching store is effective *before* we
+* increment the counter which releases all waiting CPUs
+* by using the release variant of
