From 8b2103add08b79abd3ac7015b4bac744c0af534e Mon Sep 17 00:00:00 2001 From: Steve Hodgson Date: Wed, 3 Feb 2010 09:30:17 +0000 Subject: sfc: Handle firmware assertion failure while resetting This allows the driver to recover if the MC firmware has crashed due to an assertion failure. Signed-off-by: Ben Hutchings Signed-off-by: David S. Miller --- drivers/net/sfc/mcdi.c | 63 +++++++++++++++++++++++++++++-------------------- drivers/net/sfc/siena.c | 6 +++++ 2 files changed, 44 insertions(+), 25 deletions(-) (limited to 'drivers') diff --git a/drivers/net/sfc/mcdi.c b/drivers/net/sfc/mcdi.c index 9f035b9f035..e9f0e5ee451 100644 --- a/drivers/net/sfc/mcdi.c +++ b/drivers/net/sfc/mcdi.c @@ -896,29 +896,27 @@ fail: return rc; } -int efx_mcdi_handle_assertion(struct efx_nic *efx) +static int efx_mcdi_read_assertion(struct efx_nic *efx) { - union { - u8 asserts[MC_CMD_GET_ASSERTS_IN_LEN]; - u8 reboot[MC_CMD_REBOOT_IN_LEN]; - } inbuf; - u8 assertion[MC_CMD_GET_ASSERTS_OUT_LEN]; + u8 inbuf[MC_CMD_GET_ASSERTS_IN_LEN]; + u8 outbuf[MC_CMD_GET_ASSERTS_OUT_LEN]; unsigned int flags, index, ofst; const char *reason; size_t outlen; int retry; int rc; - /* Check if the MC is in the assertion handler, retrying twice. Once + /* Attempt to read any stored assertion state before we reboot + * the mcfw out of the assertion handler. Retry twice, once * because a boot-time assertion might cause this command to fail * with EINTR. And once again because GET_ASSERTS can race with * MC_CMD_REBOOT running on the other port. */ retry = 2; do { - MCDI_SET_DWORD(inbuf.asserts, GET_ASSERTS_IN_CLEAR, 0); + MCDI_SET_DWORD(inbuf, GET_ASSERTS_IN_CLEAR, 1); rc = efx_mcdi_rpc(efx, MC_CMD_GET_ASSERTS, - inbuf.asserts, MC_CMD_GET_ASSERTS_IN_LEN, - assertion, sizeof(assertion), &outlen); + inbuf, MC_CMD_GET_ASSERTS_IN_LEN, + outbuf, sizeof(outbuf), &outlen); } while ((rc == -EINTR || rc == -EIO) && retry-- > 0); if (rc) @@ -926,21 +924,11 @@ int efx_mcdi_handle_assertion(struct efx_nic *efx) if (outlen < MC_CMD_GET_ASSERTS_OUT_LEN) return -EINVAL; - flags = MCDI_DWORD(assertion, GET_ASSERTS_OUT_GLOBAL_FLAGS); + /* Print out any recorded assertion state */ + flags = MCDI_DWORD(outbuf, GET_ASSERTS_OUT_GLOBAL_FLAGS); if (flags == MC_CMD_GET_ASSERTS_FLAGS_NO_FAILS) return 0; - /* Reset the hardware atomically such that only one port with succeed. - * This command will succeed if a reboot is no longer required (because - * the other port did it first), but fail with EIO if it succeeds. - */ - BUILD_BUG_ON(MC_CMD_REBOOT_OUT_LEN != 0); - MCDI_SET_DWORD(inbuf.reboot, REBOOT_IN_FLAGS, - MC_CMD_REBOOT_FLAGS_AFTER_ASSERTION); - efx_mcdi_rpc(efx, MC_CMD_REBOOT, inbuf.reboot, MC_CMD_REBOOT_IN_LEN, - NULL, 0, NULL); - - /* Print out the assertion */ reason = (flags == MC_CMD_GET_ASSERTS_FLAGS_SYS_FAIL) ? "system-level assertion" : (flags == MC_CMD_GET_ASSERTS_FLAGS_THR_FAIL) @@ -949,20 +937,45 @@ int efx_mcdi_handle_assertion(struct efx_nic *efx) ? "watchdog reset" : "unknown assertion"; EFX_ERR(efx, "MCPU %s at PC = 0x%.8x in thread 0x%.8x\n", reason, - MCDI_DWORD(assertion, GET_ASSERTS_OUT_SAVED_PC_OFFS), - MCDI_DWORD(assertion, GET_ASSERTS_OUT_THREAD_OFFS)); + MCDI_DWORD(outbuf, GET_ASSERTS_OUT_SAVED_PC_OFFS), + MCDI_DWORD(outbuf, GET_ASSERTS_OUT_THREAD_OFFS)); /* Print out the registers */ ofst = MC_CMD_GET_ASSERTS_OUT_GP_REGS_OFFS_OFST; for (index = 1; index < 32; index++) { EFX_ERR(efx, "R%.2d (?): 0x%.8x\n", index, - MCDI_DWORD2(assertion, ofst)); + MCDI_DWORD2(outbuf, ofst)); ofst += sizeof(efx_dword_t); } return 0; } +static void efx_mcdi_exit_assertion(struct efx_nic *efx) +{ + u8 inbuf[MC_CMD_REBOOT_IN_LEN]; + + /* Atomically reboot the mcfw out of the assertion handler */ + BUILD_BUG_ON(MC_CMD_REBOOT_OUT_LEN != 0); + MCDI_SET_DWORD(inbuf, REBOOT_IN_FLAGS, + MC_CMD_REBOOT_FLAGS_AFTER_ASSERTION); + efx_mcdi_rpc(efx, MC_CMD_REBOOT, inbuf, MC_CMD_REBOOT_IN_LEN, + NULL, 0, NULL); +} + +int efx_mcdi_handle_assertion(struct efx_nic *efx) +{ + int rc; + + rc = efx_mcdi_read_assertion(efx); + if (rc) + return rc; + + efx_mcdi_exit_assertion(efx); + + return 0; +} + void efx_mcdi_set_id_led(struct efx_nic *efx, enum efx_led_mode mode) { u8 inbuf[MC_CMD_SET_ID_LED_IN_LEN]; diff --git a/drivers/net/sfc/siena.c b/drivers/net/sfc/siena.c index f8c6771e66d..0e4c13abf08 100644 --- a/drivers/net/sfc/siena.c +++ b/drivers/net/sfc/siena.c @@ -181,6 +181,12 @@ static int siena_test_registers(struct efx_nic *efx) static int siena_reset_hw(struct efx_nic *efx, enum reset_type method) { + int rc; + + /* Recover from a failed assertion pre-reset */ + rc = efx_mcdi_handle_assertion(efx); + if (rc) + return rc; if (method == RESET_TYPE_WORLD) return efx_mcdi_reset_mc(efx); -- cgit v1.2.3