[PATCH 5.12 641/700] powerpc/powernv: Fix machine check reporting of async store errors

From: Greg Kroah-Hartman
Date: Mon Jul 12 2021 - 04:05:10 EST


From: Nicholas Piggin <npiggin@xxxxxxxxx>

[ Upstream commit 3729e0ec59a20825bd4c8c70996b2df63915e1dd ]

POWER9 and POWER10 asynchronous machine checks due to stores have their
cause reported in SRR1 but SRR1[42] is set, which in other cases
indicates DSISR cause.

Check for these cases and clear SRR1[42], so the cause matching uses
the i-side (SRR1) table.

Fixes: 7b9f71f974a1 ("powerpc/64s: POWER9 machine check handler")
Fixes: 201220bb0e8c ("powerpc/powernv: Machine check handler for POWER10")
Signed-off-by: Nicholas Piggin <npiggin@xxxxxxxxx>
Signed-off-by: Michael Ellerman <mpe@xxxxxxxxxxxxxx>
Link: https://lore.kernel.org/r/20210517140355.2325406-1-npiggin@xxxxxxxxx
Signed-off-by: Sasha Levin <sashal@xxxxxxxxxx>
---
arch/powerpc/kernel/mce_power.c | 48 +++++++++++++++++++++++++++------
1 file changed, 40 insertions(+), 8 deletions(-)

diff --git a/arch/powerpc/kernel/mce_power.c b/arch/powerpc/kernel/mce_power.c
index 667104d4c455..2fff886c549d 100644
--- a/arch/powerpc/kernel/mce_power.c
+++ b/arch/powerpc/kernel/mce_power.c
@@ -481,12 +481,11 @@ static int mce_find_instr_ea_and_phys(struct pt_regs *regs, uint64_t *addr,
return -1;
}

-static int mce_handle_ierror(struct pt_regs *regs,
+static int mce_handle_ierror(struct pt_regs *regs, unsigned long srr1,
const struct mce_ierror_table table[],
struct mce_error_info *mce_err, uint64_t *addr,
uint64_t *phys_addr)
{
- uint64_t srr1 = regs->msr;
int handled = 0;
int i;

@@ -695,19 +694,19 @@ static long mce_handle_ue_error(struct pt_regs *regs,
}

static long mce_handle_error(struct pt_regs *regs,
+ unsigned long srr1,
const struct mce_derror_table dtable[],
const struct mce_ierror_table itable[])
{
struct mce_error_info mce_err = { 0 };
uint64_t addr, phys_addr = ULONG_MAX;
- uint64_t srr1 = regs->msr;
long handled;

if (SRR1_MC_LOADSTORE(srr1))
handled = mce_handle_derror(regs, dtable, &mce_err, &addr,
&phys_addr);
else
- handled = mce_handle_ierror(regs, itable, &mce_err, &addr,
+ handled = mce_handle_ierror(regs, srr1, itable, &mce_err, &addr,
&phys_addr);

if (!handled && mce_err.error_type == MCE_ERROR_TYPE_UE)
@@ -723,16 +722,20 @@ long __machine_check_early_realmode_p7(struct pt_regs *regs)
/* P7 DD1 leaves top bits of DSISR undefined */
regs->dsisr &= 0x0000ffff;

- return mce_handle_error(regs, mce_p7_derror_table, mce_p7_ierror_table);
+ return mce_handle_error(regs, regs->msr,
+ mce_p7_derror_table, mce_p7_ierror_table);
}

long __machine_check_early_realmode_p8(struct pt_regs *regs)
{
- return mce_handle_error(regs, mce_p8_derror_table, mce_p8_ierror_table);
+ return mce_handle_error(regs, regs->msr,
+ mce_p8_derror_table, mce_p8_ierror_table);
}

long __machine_check_early_realmode_p9(struct pt_regs *regs)
{
+ unsigned long srr1 = regs->msr;
+
/*
* On POWER9 DD2.1 and below, it's possible to get a machine check
* caused by a paste instruction where only DSISR bit 25 is set. This
@@ -746,10 +749,39 @@ long __machine_check_early_realmode_p9(struct pt_regs *regs)
if (SRR1_MC_LOADSTORE(regs->msr) && regs->dsisr == 0x02000000)
return 1;

- return mce_handle_error(regs, mce_p9_derror_table, mce_p9_ierror_table);
+ /*
+ * Async machine check due to bad real address from store or foreign
+ * link time out comes with the load/store bit (PPC bit 42) set in
+ * SRR1, but the cause comes in SRR1 not DSISR. Clear bit 42 so we're
+ * directed to the ierror table so it will find the cause (which
+ * describes it correctly as a store error).
+ */
+ if (SRR1_MC_LOADSTORE(srr1) &&
+ ((srr1 & 0x081c0000) == 0x08140000 ||
+ (srr1 & 0x081c0000) == 0x08180000)) {
+ srr1 &= ~PPC_BIT(42);
+ }
+
+ return mce_handle_error(regs, srr1,
+ mce_p9_derror_table, mce_p9_ierror_table);
}

long __machine_check_early_realmode_p10(struct pt_regs *regs)
{
- return mce_handle_error(regs, mce_p10_derror_table, mce_p10_ierror_table);
+ unsigned long srr1 = regs->msr;
+
+ /*
+ * Async machine check due to bad real address from store comes with
+ * the load/store bit (PPC bit 42) set in SRR1, but the cause comes in
+ * SRR1 not DSISR. Clear bit 42 so we're directed to the ierror table
+ * so it will find the cause (which describes it correctly as a store
+ * error).
+ */
+ if (SRR1_MC_LOADSTORE(srr1) &&
+ (srr1 & 0x081c0000) == 0x08140000) {
+ srr1 &= ~PPC_BIT(42);
+ }
+
+ return mce_handle_error(regs, srr1,
+ mce_p10_derror_table, mce_p10_ierror_table);
}
--
2.30.2