The bfa driver is one of the main users of do_gettimeofday(), a function that I'm trying to remove as part of the y2038 cleanup.
The timestamps are all uses in slightly different ways, so this has turned into a rather longish series for doing something that should be simple.
The last patch in the series ("scsi: bfa: use 64-bit times in bfa_aen_entry_s ABI") is one that needs to be reviewed very carefully, and it can be skipped if the maintainers prefer to leave the 32-bit ABI unchanged, the rest are hopefully fairly straightforward.
Arnd
Arnd Bergmann (7): scsi: bfa: use ktime_get_real_ts64 for firmware timestamp scsi: bfa: use proper time accessor for stats_reset_time scsi: bfa: improve bfa_ioc_send_enable/disable data scsi: bfa: document overflow of io_profile_start_time scsi: bfa: replace bfa_get_log_time() with ktime_get_real_seconds() scsi: bfa: try to sanitize vendor netlink events scsi: bfa: use 64-bit times in bfa_aen_entry_s ABI
drivers/scsi/bfa/bfa_cs.h | 6 +++--- drivers/scsi/bfa/bfa_defs_svc.h | 3 ++- drivers/scsi/bfa/bfa_fcpim.c | 3 ++- drivers/scsi/bfa/bfa_fcpim.h | 4 ++-- drivers/scsi/bfa/bfa_ioc.c | 8 ++++--- drivers/scsi/bfa/bfa_port.c | 15 +++---------- drivers/scsi/bfa/bfa_port.h | 2 +- drivers/scsi/bfa/bfa_svc.c | 47 ++++++++++++----------------------------- drivers/scsi/bfa/bfa_svc.h | 2 +- drivers/scsi/bfa/bfad_bsg.c | 4 +--- drivers/scsi/bfa/bfad_im.h | 32 +++++++++++++++++++--------- 11 files changed, 56 insertions(+), 70 deletions(-)
BFA_TRC_TS() calculates a 32-bit microsecond timestamp using the deprecated do_gettimeofday() function. This overflows roughly every 71 minutes, so it's obviously not used as an absolute time stamp, but it seems wrong to use a time base for it that will jump during settimeofday() calls, leap seconds, or the y2038 overflow.
This converts it to ktime_get_ts64(), which has none of those problems but is not synchronized to wall-clock time.
Signed-off-by: Arnd Bergmann arnd@arndb.de --- drivers/scsi/bfa/bfa_cs.h | 6 +++--- 1 file changed, 3 insertions(+), 3 deletions(-)
diff --git a/drivers/scsi/bfa/bfa_cs.h b/drivers/scsi/bfa/bfa_cs.h index df6760ca0911..9685efc59b16 100644 --- a/drivers/scsi/bfa/bfa_cs.h +++ b/drivers/scsi/bfa/bfa_cs.h @@ -35,10 +35,10 @@
#define BFA_TRC_TS(_trcm) \ ({ \ - struct timeval tv; \ + struct timespec64 ts; \ \ - do_gettimeofday(&tv); \ - (tv.tv_sec*1000000+tv.tv_usec); \ + ktime_get_ts64(&ts); \ + (ts.tv_sec*1000000+ts.tv_nsec / 1000); \ })
#ifndef BFA_TRC_TS
We use the deprecated do_gettimeofday() function to read the current time when resetting the statistics in both bfa_port and bfa_svc. This works fine because overflow is handled correctly, but we want to get rid of do_gettimeofday() and using a non-monotonic time suffers from concurrent settimeofday calls and other problems.
This uses the ktime_get_seconds() function instead, which does what we need here.
Signed-off-by: Arnd Bergmann arnd@arndb.de --- drivers/scsi/bfa/bfa_port.c | 15 +++------------ drivers/scsi/bfa/bfa_port.h | 2 +- drivers/scsi/bfa/bfa_svc.c | 15 ++++----------- drivers/scsi/bfa/bfa_svc.h | 2 +- 4 files changed, 9 insertions(+), 25 deletions(-)
diff --git a/drivers/scsi/bfa/bfa_port.c b/drivers/scsi/bfa/bfa_port.c index da1721e0d167..079bc77f4102 100644 --- a/drivers/scsi/bfa/bfa_port.c +++ b/drivers/scsi/bfa/bfa_port.c @@ -96,14 +96,11 @@ bfa_port_get_stats_isr(struct bfa_port_s *port, bfa_status_t status) port->stats_busy = BFA_FALSE;
if (status == BFA_STATUS_OK) { - struct timeval tv; - memcpy(port->stats, port->stats_dma.kva, sizeof(union bfa_port_stats_u)); bfa_port_stats_swap(port, port->stats);
- do_gettimeofday(&tv); - port->stats->fc.secs_reset = tv.tv_sec - port->stats_reset_time; + port->stats->fc.secs_reset = ktime_get_seconds() - port->stats_reset_time; }
if (port->stats_cbfn) { @@ -124,16 +121,13 @@ bfa_port_get_stats_isr(struct bfa_port_s *port, bfa_status_t status) static void bfa_port_clear_stats_isr(struct bfa_port_s *port, bfa_status_t status) { - struct timeval tv; - port->stats_status = status; port->stats_busy = BFA_FALSE;
/* * re-initialize time stamp for stats reset */ - do_gettimeofday(&tv); - port->stats_reset_time = tv.tv_sec; + port->stats_reset_time = ktime_get_seconds();
if (port->stats_cbfn) { port->stats_cbfn(port->stats_cbarg, status); @@ -471,8 +465,6 @@ void bfa_port_attach(struct bfa_port_s *port, struct bfa_ioc_s *ioc, void *dev, struct bfa_trc_mod_s *trcmod) { - struct timeval tv; - WARN_ON(!port);
port->dev = dev; @@ -494,8 +486,7 @@ bfa_port_attach(struct bfa_port_s *port, struct bfa_ioc_s *ioc, /* * initialize time stamp for stats reset */ - do_gettimeofday(&tv); - port->stats_reset_time = tv.tv_sec; + port->stats_reset_time = ktime_get_seconds();
bfa_trc(port, 0); } diff --git a/drivers/scsi/bfa/bfa_port.h b/drivers/scsi/bfa/bfa_port.h index 26dc1bf14c85..0c3b200243ca 100644 --- a/drivers/scsi/bfa/bfa_port.h +++ b/drivers/scsi/bfa/bfa_port.h @@ -36,7 +36,7 @@ struct bfa_port_s { bfa_port_stats_cbfn_t stats_cbfn; void *stats_cbarg; bfa_status_t stats_status; - u32 stats_reset_time; + time64_t stats_reset_time; union bfa_port_stats_u *stats; struct bfa_dma_s stats_dma; bfa_boolean_t endis_pending; diff --git a/drivers/scsi/bfa/bfa_svc.c b/drivers/scsi/bfa/bfa_svc.c index e640223bab3c..dd7d1e6bc2d8 100644 --- a/drivers/scsi/bfa/bfa_svc.c +++ b/drivers/scsi/bfa/bfa_svc.c @@ -3047,7 +3047,6 @@ bfa_fcport_attach(struct bfa_s *bfa, void *bfad, struct bfa_iocfc_cfg_s *cfg, struct bfa_fcport_s *fcport = BFA_FCPORT_MOD(bfa); struct bfa_port_cfg_s *port_cfg = &fcport->cfg; struct bfa_fcport_ln_s *ln = &fcport->ln; - struct timeval tv;
fcport->bfa = bfa; ln->fcport = fcport; @@ -3060,8 +3059,7 @@ bfa_fcport_attach(struct bfa_s *bfa, void *bfad, struct bfa_iocfc_cfg_s *cfg, /* * initialize time stamp for stats reset */ - do_gettimeofday(&tv); - fcport->stats_reset_time = tv.tv_sec; + fcport->stats_reset_time = ktime_get_seconds(); fcport->stats_dma_ready = BFA_FALSE;
/* @@ -3295,9 +3293,7 @@ __bfa_cb_fcport_stats_get(void *cbarg, bfa_boolean_t complete) union bfa_fcport_stats_u *ret;
if (complete) { - struct timeval tv; - if (fcport->stats_status == BFA_STATUS_OK) - do_gettimeofday(&tv); + time64_t time = ktime_get_seconds();
list_for_each_safe(qe, qen, &fcport->stats_pending_q) { bfa_q_deq(&fcport->stats_pending_q, &qe); @@ -3312,7 +3308,7 @@ __bfa_cb_fcport_stats_get(void *cbarg, bfa_boolean_t complete) bfa_fcport_fcoe_stats_swap(&ret->fcoe, &fcport->stats->fcoe); ret->fcoe.secs_reset = - tv.tv_sec - fcport->stats_reset_time; + time - fcport->stats_reset_time; } } bfa_cb_queue_status(fcport->bfa, &cb->hcb_qe, @@ -3373,13 +3369,10 @@ __bfa_cb_fcport_stats_clr(void *cbarg, bfa_boolean_t complete) struct list_head *qe, *qen;
if (complete) { - struct timeval tv; - /* * re-initialize time stamp for stats reset */ - do_gettimeofday(&tv); - fcport->stats_reset_time = tv.tv_sec; + fcport->stats_reset_time = ktime_get_seconds(); list_for_each_safe(qe, qen, &fcport->statsclr_pending_q) { bfa_q_deq(&fcport->statsclr_pending_q, &qe); cb = (struct bfa_cb_pending_q_s *)qe; diff --git a/drivers/scsi/bfa/bfa_svc.h b/drivers/scsi/bfa/bfa_svc.h index ea2278bc78a8..7e8fb6231d49 100644 --- a/drivers/scsi/bfa/bfa_svc.h +++ b/drivers/scsi/bfa/bfa_svc.h @@ -505,7 +505,7 @@ struct bfa_fcport_s { struct list_head stats_pending_q; struct list_head statsclr_pending_q; bfa_boolean_t stats_qfull; - u32 stats_reset_time; /* stats reset time stamp */ + time64_t stats_reset_time; /* stats reset time stamp */ bfa_boolean_t diag_busy; /* diag busy status */ bfa_boolean_t beacon; /* port beacon status */ bfa_boolean_t link_e2e_beacon; /* link beacon status */
In bfa_ioc_send_enable, we use the deprecated do_gettimeofday() function to read the current time. This is not a problem, since the firmware interface is already limited to 32-bit timestamps, but it's better to use ktime_get_seconds() and document what the limitation is.
I noticed that I did the same change in commit a5af83925363 ("bna: avoid writing uninitialized data into hw registers") for the ethernet driver. That commit also changed the "disable" funtion to initialize the data we pass to the firmware properly, so I'm doing the same thing here.
Signed-off-by: Arnd Bergmann arnd@arndb.de --- drivers/scsi/bfa/bfa_ioc.c | 8 +++++--- 1 file changed, 5 insertions(+), 3 deletions(-)
diff --git a/drivers/scsi/bfa/bfa_ioc.c b/drivers/scsi/bfa/bfa_ioc.c index 256f4afaccf9..117332537763 100644 --- a/drivers/scsi/bfa/bfa_ioc.c +++ b/drivers/scsi/bfa/bfa_ioc.c @@ -1809,13 +1809,12 @@ static void bfa_ioc_send_enable(struct bfa_ioc_s *ioc) { struct bfi_ioc_ctrl_req_s enable_req; - struct timeval tv;
bfi_h2i_set(enable_req.mh, BFI_MC_IOC, BFI_IOC_H2I_ENABLE_REQ, bfa_ioc_portid(ioc)); enable_req.clscode = cpu_to_be16(ioc->clscode); - do_gettimeofday(&tv); - enable_req.tv_sec = be32_to_cpu(tv.tv_sec); + /* unsigned 32-bit time_t overflow in y2106 */ + enable_req.tv_sec = be32_to_cpu(ktime_get_real_seconds()); bfa_ioc_mbox_send(ioc, &enable_req, sizeof(struct bfi_ioc_ctrl_req_s)); }
@@ -1826,6 +1825,9 @@ bfa_ioc_send_disable(struct bfa_ioc_s *ioc)
bfi_h2i_set(disable_req.mh, BFI_MC_IOC, BFI_IOC_H2I_DISABLE_REQ, bfa_ioc_portid(ioc)); + disable_req.clscode = cpu_to_be16(ioc->clscode); + /* unsigned 32-bit time_t overflow in y2106 */ + disable_req.tv_sec = be32_to_cpu(ktime_get_real_seconds()); bfa_ioc_mbox_send(ioc, &disable_req, sizeof(struct bfi_ioc_ctrl_req_s)); }
io_profile_start_time() gets read using do_gettimeofday() and passed down as a 32-bit value through multiple functions. This will overflow in y2038 or y2106, depending on whether it gets interpreted as unsigned in the end.
This changes do_gettimeofday() to ktime_get_real_seconds() and pushes the point at which it overflows to where we actually assign it to the bfa_fcpim_del_itn_stats_s structure, with an appropriate comment.
Signed-off-by: Arnd Bergmann arnd@arndb.de --- drivers/scsi/bfa/bfa_fcpim.c | 3 ++- drivers/scsi/bfa/bfa_fcpim.h | 4 ++-- drivers/scsi/bfa/bfad_bsg.c | 4 +--- 3 files changed, 5 insertions(+), 6 deletions(-)
diff --git a/drivers/scsi/bfa/bfa_fcpim.c b/drivers/scsi/bfa/bfa_fcpim.c index 5f53b3276234..2c85f5b1f9c1 100644 --- a/drivers/scsi/bfa/bfa_fcpim.c +++ b/drivers/scsi/bfa/bfa_fcpim.c @@ -468,7 +468,7 @@ bfa_ioim_profile_start(struct bfa_ioim_s *ioim) }
bfa_status_t -bfa_fcpim_profile_on(struct bfa_s *bfa, u32 time) +bfa_fcpim_profile_on(struct bfa_s *bfa, time64_t time) { struct bfa_itnim_s *itnim; struct bfa_fcpim_s *fcpim = BFA_FCPIM(bfa); @@ -1478,6 +1478,7 @@ bfa_itnim_get_ioprofile(struct bfa_itnim_s *itnim, return BFA_STATUS_IOPROFILE_OFF;
itnim->ioprofile.index = BFA_IOBUCKET_MAX; + /* unsigned 32-bit time_t overflow here in y2106 */ itnim->ioprofile.io_profile_start_time = bfa_io_profile_start_time(itnim->bfa); itnim->ioprofile.clock_res_mul = bfa_io_lat_clock_res_mul; diff --git a/drivers/scsi/bfa/bfa_fcpim.h b/drivers/scsi/bfa/bfa_fcpim.h index e93921dec347..ec8f863540ae 100644 --- a/drivers/scsi/bfa/bfa_fcpim.h +++ b/drivers/scsi/bfa/bfa_fcpim.h @@ -136,7 +136,7 @@ struct bfa_fcpim_s { struct bfa_fcpim_del_itn_stats_s del_itn_stats; bfa_boolean_t ioredirect; bfa_boolean_t io_profile; - u32 io_profile_start_time; + time64_t io_profile_start_time; bfa_fcpim_profile_t profile_comp; bfa_fcpim_profile_t profile_start; }; @@ -310,7 +310,7 @@ bfa_status_t bfa_fcpim_port_iostats(struct bfa_s *bfa, struct bfa_itnim_iostats_s *stats, u8 lp_tag); void bfa_fcpim_add_stats(struct bfa_itnim_iostats_s *fcpim_stats, struct bfa_itnim_iostats_s *itnim_stats); -bfa_status_t bfa_fcpim_profile_on(struct bfa_s *bfa, u32 time); +bfa_status_t bfa_fcpim_profile_on(struct bfa_s *bfa, time64_t time); bfa_status_t bfa_fcpim_profile_off(struct bfa_s *bfa);
#define bfa_fcpim_ioredirect_enabled(__bfa) \ diff --git a/drivers/scsi/bfa/bfad_bsg.c b/drivers/scsi/bfa/bfad_bsg.c index 72ca2a2e08e2..01fc51a70356 100644 --- a/drivers/scsi/bfa/bfad_bsg.c +++ b/drivers/scsi/bfa/bfad_bsg.c @@ -2094,13 +2094,11 @@ bfad_iocmd_fcpim_cfg_profile(struct bfad_s *bfad, void *cmd, unsigned int v_cmd) { struct bfa_bsg_fcpim_profile_s *iocmd = (struct bfa_bsg_fcpim_profile_s *)cmd; - struct timeval tv; unsigned long flags;
- do_gettimeofday(&tv); spin_lock_irqsave(&bfad->bfad_lock, flags); if (v_cmd == IOCMD_FCPIM_PROFILE_ON) - iocmd->status = bfa_fcpim_profile_on(&bfad->bfa, tv.tv_sec); + iocmd->status = bfa_fcpim_profile_on(&bfad->bfa, ktime_get_real_seconds()); else if (v_cmd == IOCMD_FCPIM_PROFILE_OFF) iocmd->status = bfa_fcpim_profile_off(&bfad->bfa); spin_unlock_irqrestore(&bfad->bfad_lock, flags);
The bfa_get_log_time() returns a 64-bit timestamp that does not suffer from the y2038 overflow on 64-bit systems. However, on 32-bit architectures the timestamp will jump from 0x000000007fffffff to 0xffffffff80000000 in y2038 and produce wrong results.
The ktime_get_real_seconds() function does the same thing as bfa_get_log_time() without that problem, so we can simply remove the former use ktime_get_real_seconds() instead.
Signed-off-by: Arnd Bergmann arnd@arndb.de --- drivers/scsi/bfa/bfa_svc.c | 32 ++++++++++---------------------- 1 file changed, 10 insertions(+), 22 deletions(-)
diff --git a/drivers/scsi/bfa/bfa_svc.c b/drivers/scsi/bfa/bfa_svc.c index dd7d1e6bc2d8..9d20d2c92e8c 100644 --- a/drivers/scsi/bfa/bfa_svc.c +++ b/drivers/scsi/bfa/bfa_svc.c @@ -288,18 +288,6 @@ plkd_validate_logrec(struct bfa_plog_rec_s *pl_rec) return 0; }
-static u64 -bfa_get_log_time(void) -{ - u64 system_time = 0; - struct timeval tv; - do_gettimeofday(&tv); - - /* We are interested in seconds only. */ - system_time = tv.tv_sec; - return system_time; -} - static void bfa_plog_add(struct bfa_plog_s *plog, struct bfa_plog_rec_s *pl_rec) { @@ -320,7 +308,7 @@ bfa_plog_add(struct bfa_plog_s *plog, struct bfa_plog_rec_s *pl_rec)
memcpy(pl_recp, pl_rec, sizeof(struct bfa_plog_rec_s));
- pl_recp->tv = bfa_get_log_time(); + pl_recp->tv = ktime_get_real_seconds(); BFA_PL_LOG_REC_INCR(plog->tail);
if (plog->head == plog->tail) @@ -6141,13 +6129,13 @@ bfa_fcdiag_lb_is_running(struct bfa_s *bfa) /* * D-port */ -#define bfa_dport_result_start(__dport, __mode) do { \ - (__dport)->result.start_time = bfa_get_log_time(); \ - (__dport)->result.status = DPORT_TEST_ST_INPRG; \ - (__dport)->result.mode = (__mode); \ - (__dport)->result.rp_pwwn = (__dport)->rp_pwwn; \ - (__dport)->result.rp_nwwn = (__dport)->rp_nwwn; \ - (__dport)->result.lpcnt = (__dport)->lpcnt; \ +#define bfa_dport_result_start(__dport, __mode) do { \ + (__dport)->result.start_time = ktime_get_real_seconds(); \ + (__dport)->result.status = DPORT_TEST_ST_INPRG; \ + (__dport)->result.mode = (__mode); \ + (__dport)->result.rp_pwwn = (__dport)->rp_pwwn; \ + (__dport)->result.rp_nwwn = (__dport)->rp_nwwn; \ + (__dport)->result.lpcnt = (__dport)->lpcnt; \ } while (0)
static bfa_boolean_t bfa_dport_send_req(struct bfa_dport_s *dport, @@ -6581,7 +6569,7 @@ bfa_dport_scn(struct bfa_dport_s *dport, struct bfi_diag_dport_scn_s *msg)
switch (dport->i2hmsg.scn.state) { case BFI_DPORT_SCN_TESTCOMP: - dport->result.end_time = bfa_get_log_time(); + dport->result.end_time = ktime_get_real_seconds(); bfa_trc(dport->bfa, dport->result.end_time);
dport->result.status = msg->info.testcomp.status; @@ -6628,7 +6616,7 @@ bfa_dport_scn(struct bfa_dport_s *dport, struct bfi_diag_dport_scn_s *msg) case BFI_DPORT_SCN_SUBTESTSTART: subtesttype = msg->info.teststart.type; dport->result.subtest[subtesttype].start_time = - bfa_get_log_time(); + ktime_get_real_seconds(); dport->result.subtest[subtesttype].status = DPORT_TEST_ST_INPRG;
bfa_aen_entry_s is passed to user space in a netlink message, but is defined using a 'struct timeval' and an 'enum' that are not only different between architectures, but also between 32-bit user space and 64-bit kernels they may run on, as well as depending on the particular C library that defines timeval.
This changes the in-kernel definition to no longer use the timeval type directly but instead use two open-coded 'unsigned long' members. This keeps the existing ABI, but making the variable unsigned also helps make it work after y2038, until it overflows in 2106.
Since the macro becomes overly complex at this point, I'm changing it to an inline function for readability.
I'm not changing the 32-bit user-space ABI at this point, to keep the changes separate, I deally this would be defined using the same binary layout for all architectures.
Signed-off-by: Arnd Bergmann arnd@arndb.de --- drivers/scsi/bfa/bfa_defs_svc.h | 3 ++- drivers/scsi/bfa/bfad_im.h | 32 ++++++++++++++++++++++---------- 2 files changed, 24 insertions(+), 11 deletions(-)
diff --git a/drivers/scsi/bfa/bfa_defs_svc.h b/drivers/scsi/bfa/bfa_defs_svc.h index e81707f938cb..df1e874015c4 100644 --- a/drivers/scsi/bfa/bfa_defs_svc.h +++ b/drivers/scsi/bfa/bfa_defs_svc.h @@ -1455,7 +1455,8 @@ struct bfa_aen_entry_s { enum bfa_aen_category aen_category; u32 aen_type; union bfa_aen_data_u aen_data; - struct timeval aen_tv; + unsigned long aen_tv_sec; + unsigned long aen_tv_usec; u32 seq_num; u32 bfad_num; }; diff --git a/drivers/scsi/bfa/bfad_im.h b/drivers/scsi/bfa/bfad_im.h index c81ec2a77ef5..7f7616c52814 100644 --- a/drivers/scsi/bfa/bfad_im.h +++ b/drivers/scsi/bfa/bfad_im.h @@ -131,16 +131,28 @@ struct bfad_im_s { } while (0)
/* post fc_host vendor event */ -#define bfad_im_post_vendor_event(_entry, _drv, _cnt, _cat, _evt) do { \ - do_gettimeofday(&(_entry)->aen_tv); \ - (_entry)->bfad_num = (_drv)->inst_no; \ - (_entry)->seq_num = (_cnt); \ - (_entry)->aen_category = (_cat); \ - (_entry)->aen_type = (_evt); \ - if ((_drv)->bfad_flags & BFAD_FC4_PROBE_DONE) \ - queue_work((_drv)->im->drv_workq, \ - &(_drv)->im->aen_im_notify_work); \ -} while (0) +static inline void bfad_im_post_vendor_event(struct bfa_aen_entry_s *entry, + struct bfad_s *drv, int cnt, + enum bfa_aen_category cat, + enum bfa_ioc_aen_event evt) +{ + struct timespec64 ts; + + ktime_get_real_ts64(&ts); + /* + * 'unsigned long aen_tv_sec' overflows in y2106 on 32-bit + * architectures, or in 2038 if user space interprets it + * as 'signed'. + */ + entry->aen_tv_sec = ts.tv_sec; + entry->aen_tv_usec = ts.tv_nsec / NSEC_PER_USEC; + entry->bfad_num = drv->inst_no; + entry->seq_num = cnt; + entry->aen_category = cat; + entry->aen_type = evt; + if (drv->bfad_flags & BFAD_FC4_PROBE_DONE) + queue_work(drv->im->drv_workq, &drv->im->aen_im_notify_work); +}
struct Scsi_Host *bfad_scsi_host_alloc(struct bfad_im_port_s *im_port, struct bfad_s *);
bfa_aen_entry_s is passed through a netlink socket that can be read by either 32-bit or 64-bit processes, but the data format is different between the two on current implementations.
Originally, this was using a 'struct timeval', which also suffers from getting redefined with a new libc implementation.
With this patch, the layout gets fixed to having two 64-bit members for the time, making it the same on 32-bit kernels and 64-bit kernels running either compat or native user space including x32.
Provided that the new header file gets used to recompile any 32-bit application binaries, this will fix running those on a 64-bit kernel (with or without this patch) e.g. in a container environment, and it will make binaries work that will be built against a future 32-bit glibc that uses a 64-bit time_t, and avoid the y2038 overflow there.
However, this also breaks compatibility with any existing 32-bit binary running on a native 32-bit kernel, those must be recompiled against the new header, which in turn makes them incompatible with older kernels unless the same change gets applied there.
Obviously this patch should only be applied when the benefits outweigh the possible breakage. I'm posting it under the assumption that there are no open-source tools using the netlink interface, and that users of the binaries provided by qlogic for SLES10/11 and RHEL5/6 are not actually being used on new future systems with 32-bit x86 kernels.
Signed-off-by: Arnd Bergmann arnd@arndb.de --- --- drivers/scsi/bfa/bfa_defs_svc.h | 4 ++-- 1 file changed, 2 insertions(+), 2 deletions(-)
diff --git a/drivers/scsi/bfa/bfa_defs_svc.h b/drivers/scsi/bfa/bfa_defs_svc.h index df1e874015c4..3d0c96a5c873 100644 --- a/drivers/scsi/bfa/bfa_defs_svc.h +++ b/drivers/scsi/bfa/bfa_defs_svc.h @@ -1455,8 +1455,8 @@ struct bfa_aen_entry_s { enum bfa_aen_category aen_category; u32 aen_type; union bfa_aen_data_u aen_data; - unsigned long aen_tv_sec; - unsigned long aen_tv_usec; + u64 aen_tv_sec; + u64 aen_tv_usec; u32 seq_num; u32 bfad_num; };
This stuff look ok - but I have a bigger question and that is if bfa is still alive at all. Everytime I look at it I have doubts if it works at all, so I wonder if we need to keep it on life support.
The bfa drivers are in deep maintenance mode for some time now. We have not had issues reported on them for a while, but it would be good to continue to have them.
The series looks ok to me too.
Thanks, Anil
-----Original Message----- From: Christoph Hellwig [mailto:hch@lst.de] Sent: 10 November 2017 22:55 To: Arnd Bergmann arnd@arndb.de Cc: Gurumurthy, Anil Anil.Gurumurthy@cavium.com; Kalluru, Sudarsana Sudarsana.Kalluru@cavium.com; James E . J . Bottomley jejb@linux.vnet.ibm.com; Martin K . Petersen martin.petersen@oracle.com; linux-scsi@vger.kernel.org; linux-kernel@vger.kernel.org; y2038@lists.linaro.org; hch@lst.de; hare@suse.com; jthumshirn@suse.de Subject: Re: [PATCH 0/7] scsi: bfa: do_gettimeofday removal
This stuff look ok - but I have a bigger question and that is if bfa is still alive at all. Everytime I look at it I have doubts if it works at all, so I wonder if we need to keep it on life support.
On Fri, 2017-11-10 at 16:37 +0100, Arnd Bergmann wrote:
In bfa_ioc_send_enable, we use the deprecated do_gettimeofday() function to read the current time. This is not a problem, since the firmware interface is already limited to 32-bit timestamps, but it's better to use ktime_get_seconds() and document what the limitation is.
I noticed that I did the same change in commit a5af83925363 ("bna: avoid writing uninitialized data into hw registers") for the ethernet driver. That commit also changed the "disable" funtion to initialize the data we pass to the firmware properly, so I'm doing the same thing here.
Signed-off-by: Arnd Bergmann arnd@arndb.de
drivers/scsi/bfa/bfa_ioc.c | 8 +++++--- 1 file changed, 5 insertions(+), 3 deletions(-)
diff --git a/drivers/scsi/bfa/bfa_ioc.c b/drivers/scsi/bfa/bfa_ioc.c index 256f4afaccf9..117332537763 100644 --- a/drivers/scsi/bfa/bfa_ioc.c +++ b/drivers/scsi/bfa/bfa_ioc.c @@ -1809,13 +1809,12 @@ static void bfa_ioc_send_enable(struct bfa_ioc_s *ioc) { struct bfi_ioc_ctrl_req_s enable_req;
- struct timeval tv;
bfi_h2i_set(enable_req.mh, BFI_MC_IOC, BFI_IOC_H2I_ENABLE_REQ, bfa_ioc_portid(ioc)); enable_req.clscode = cpu_to_be16(ioc->clscode);
- do_gettimeofday(&tv);
- enable_req.tv_sec = be32_to_cpu(tv.tv_sec);
- /* unsigned 32-bit time_t overflow in y2106 */
- enable_req.tv_sec = be32_to_cpu(ktime_get_real_seconds());
The byte order conversion should also logically be cpu_to_be32(), not be32_to_cpu().
Ben.
bfa_ioc_mbox_send(ioc, &enable_req, sizeof(struct bfi_ioc_ctrl_req_s)); } @@ -1826,6 +1825,9 @@ bfa_ioc_send_disable(struct bfa_ioc_s *ioc) bfi_h2i_set(disable_req.mh, BFI_MC_IOC, BFI_IOC_H2I_DISABLE_REQ, bfa_ioc_portid(ioc));
- disable_req.clscode = cpu_to_be16(ioc->clscode);
- /* unsigned 32-bit time_t overflow in y2106 */
- disable_req.tv_sec = be32_to_cpu(ktime_get_real_seconds());
bfa_ioc_mbox_send(ioc, &disable_req, sizeof(struct bfi_ioc_ctrl_req_s)); }
On Mon, Nov 13, 2017 at 3:08 PM, Ben Hutchings ben.hutchings@codethink.co.uk wrote:
On Fri, 2017-11-10 at 16:37 +0100, Arnd Bergmann wrote:
bfi_h2i_set(enable_req.mh, BFI_MC_IOC, BFI_IOC_H2I_ENABLE_REQ, bfa_ioc_portid(ioc)); enable_req.clscode = cpu_to_be16(ioc->clscode);
do_gettimeofday(&tv);enable_req.tv_sec = be32_to_cpu(tv.tv_sec);
/* unsigned 32-bit time_t overflow in y2106 */enable_req.tv_sec = be32_to_cpu(ktime_get_real_seconds());The byte order conversion should also logically be cpu_to_be32(), not be32_to_cpu().
Right, I had not noticed that. I just tried fixing this, looking at what sparse thinks of the file as a whole. I found it basically hopeless to fix the endianess warnings in this driver, it seems completely random here whether they are used correctly, in the opposite direction as expected, or just flip data in place as in
attr->part[i].part_off = be32_to_cpu(f->part[i].part_off);
I'd rather not touch that part. IIRC I had at some point spent a day trying to clean up the "warning: symbol 'bfa_flash_sem_get' was not declared. Should it be static?" sparse warnings for some driver, before giving up for similar reasons.
Arnd
Arnd,
The bfa driver is one of the main users of do_gettimeofday(), a function that I'm trying to remove as part of the y2038 cleanup.
The timestamps are all uses in slightly different ways, so this has turned into a rather longish series for doing something that should be simple.
The last patch in the series ("scsi: bfa: use 64-bit times in bfa_aen_entry_s ABI") is one that needs to be reviewed very carefully, and it can be skipped if the maintainers prefer to leave the 32-bit ABI unchanged, the rest are hopefully fairly straightforward.
Applied to 4.16/scsi-queue, thanks!
Will drop #7 if something breaks.
-
Arnd Bergmann -
Ben Hutchings -
Christoph Hellwig -
Gurumurthy, Anil -
Martin K. Petersen