| Message ID | 20240730175118.27105-1-brking@linux.ibm.com (mailing list archive) |
|---|---|
| State | Changes Requested |
| Headers | show |
| Series | ibmvfc: Add max_sectors module parameter | expand |
On Tue, 2024-07-30 at 12:51 -0500, Brian King wrote: > There are some scenarios that can occur, such as performing an > upgrade of the virtual I/O server, where the supported max transfer > of the backing device for an ibmvfc HBA can change. If the max > transfer of the backing device decreases, this can cause issues with > previously discovered LUNs. This patch accomplishes two things. > First, it changes the default ibmvfc max transfer value to 1MB. > This is generally supported by all backing devices, which should > mitigate this issue out of the box. Secondly, it adds a module > parameter, enabling a user to increase the max transfer value to > values that are larger than 1MB, as long as they have configured > these larger values on the virtual I/O server as well. > > Signed-off-by: Brian King <brking@linux.ibm.com> > --- > drivers/scsi/ibmvscsi/ibmvfc.c | 10 +++++++--- > drivers/scsi/ibmvscsi/ibmvfc.h | 2 +- > 2 files changed, 8 insertions(+), 4 deletions(-) > > diff --git a/drivers/scsi/ibmvscsi/ibmvfc.c > b/drivers/scsi/ibmvscsi/ibmvfc.c > index a3d1013c8307..611901562e06 100644 > --- a/drivers/scsi/ibmvscsi/ibmvfc.c > +++ b/drivers/scsi/ibmvscsi/ibmvfc.c > @@ -37,6 +37,7 @@ static unsigned int default_timeout = > IBMVFC_DEFAULT_TIMEOUT; > static u64 max_lun = IBMVFC_MAX_LUN; > static unsigned int max_targets = IBMVFC_MAX_TARGETS; > static unsigned int max_requests = IBMVFC_MAX_REQUESTS_DEFAULT; > +static unsigned int max_sectors = IBMVFC_MAX_SECTORS; > static u16 scsi_qdepth = IBMVFC_SCSI_QDEPTH; > static unsigned int disc_threads = IBMVFC_MAX_DISC_THREADS; > static unsigned int ibmvfc_debug = IBMVFC_DEBUG; > @@ -83,6 +84,9 @@ MODULE_PARM_DESC(default_timeout, > module_param_named(max_requests, max_requests, uint, S_IRUGO); > MODULE_PARM_DESC(max_requests, "Maximum requests for this adapter. " > "[Default=" > __stringify(IBMVFC_MAX_REQUESTS_DEFAULT) "]"); > +module_param_named(max_sectors, max_sectors, uint, S_IRUGO); > +MODULE_PARM_DESC(max_sectors, "Maximum sectors for this adapter. " > + "[Default=" __stringify(IBMVFC_MAX_SECTORS) "]"); > module_param_named(scsi_qdepth, scsi_qdepth, ushort, S_IRUGO); > MODULE_PARM_DESC(scsi_qdepth, "Maximum scsi command depth per > adapter queue. " > "[Default=" __stringify(IBMVFC_SCSI_QDEPTH) "]"); > @@ -1494,7 +1498,7 @@ static void ibmvfc_set_login_info(struct > ibmvfc_host *vhost) > memset(login_info, 0, sizeof(*login_info)); > > login_info->ostype = cpu_to_be32(IBMVFC_OS_LINUX); > - login_info->max_dma_len = cpu_to_be64(IBMVFC_MAX_SECTORS << > 9); > + login_info->max_dma_len = cpu_to_be64(max_sectors << 9); > login_info->max_payload = cpu_to_be32(sizeof(struct > ibmvfc_fcp_cmd_iu)); > login_info->max_response = cpu_to_be32(sizeof(struct > ibmvfc_fcp_rsp)); > login_info->partition_num = cpu_to_be32(vhost- > >partition_number); > @@ -5230,7 +5234,7 @@ static void ibmvfc_npiv_login_done(struct > ibmvfc_event *evt) > } > > vhost->logged_in = 1; > - npiv_max_sectors = min((uint)(be64_to_cpu(rsp->max_dma_len) > >> 9), IBMVFC_MAX_SECTORS); > + npiv_max_sectors = min((uint)(be64_to_cpu(rsp->max_dma_len) > >> 9), max_sectors); > dev_info(vhost->dev, "Host partition: %s, device: %s %s %s > max sectors %u\n", > rsp->partition_name, rsp->device_name, rsp- > >port_loc_code, > rsp->drc_name, npiv_max_sectors); > @@ -6329,7 +6333,7 @@ static int ibmvfc_probe(struct vio_dev *vdev, > const struct vio_device_id *id) > shost->can_queue = scsi_qdepth; > shost->max_lun = max_lun; > shost->max_id = max_targets; > - shost->max_sectors = IBMVFC_MAX_SECTORS; > + shost->max_sectors = max_sectors; Would it make sense to check whether the user-provided max_sectors value is within some reasonable limits? Thanks Martin
On 8/12/24 3:22 PM, Martin Wilck wrote: > On Tue, 2024-07-30 at 12:51 -0500, Brian King wrote: >> There are some scenarios that can occur, such as performing an >> upgrade of the virtual I/O server, where the supported max transfer >> of the backing device for an ibmvfc HBA can change. If the max >> transfer of the backing device decreases, this can cause issues with >> previously discovered LUNs. This patch accomplishes two things. >> First, it changes the default ibmvfc max transfer value to 1MB. >> This is generally supported by all backing devices, which should >> mitigate this issue out of the box. Secondly, it adds a module >> parameter, enabling a user to increase the max transfer value to >> values that are larger than 1MB, as long as they have configured >> these larger values on the virtual I/O server as well. >> >> Signed-off-by: Brian King <brking@linux.ibm.com> >> --- >> drivers/scsi/ibmvscsi/ibmvfc.c | 10 +++++++--- >> drivers/scsi/ibmvscsi/ibmvfc.h | 2 +- >> 2 files changed, 8 insertions(+), 4 deletions(-) >> >> diff --git a/drivers/scsi/ibmvscsi/ibmvfc.c >> b/drivers/scsi/ibmvscsi/ibmvfc.c >> index a3d1013c8307..611901562e06 100644 >> --- a/drivers/scsi/ibmvscsi/ibmvfc.c >> +++ b/drivers/scsi/ibmvscsi/ibmvfc.c >> @@ -37,6 +37,7 @@ static unsigned int default_timeout = >> IBMVFC_DEFAULT_TIMEOUT; >> static u64 max_lun = IBMVFC_MAX_LUN; >> static unsigned int max_targets = IBMVFC_MAX_TARGETS; >> static unsigned int max_requests = IBMVFC_MAX_REQUESTS_DEFAULT; >> +static unsigned int max_sectors = IBMVFC_MAX_SECTORS; >> static u16 scsi_qdepth = IBMVFC_SCSI_QDEPTH; >> static unsigned int disc_threads = IBMVFC_MAX_DISC_THREADS; >> static unsigned int ibmvfc_debug = IBMVFC_DEBUG; >> @@ -83,6 +84,9 @@ MODULE_PARM_DESC(default_timeout, >> module_param_named(max_requests, max_requests, uint, S_IRUGO); >> MODULE_PARM_DESC(max_requests, "Maximum requests for this adapter. " >> "[Default=" >> __stringify(IBMVFC_MAX_REQUESTS_DEFAULT) "]"); >> +module_param_named(max_sectors, max_sectors, uint, S_IRUGO); >> +MODULE_PARM_DESC(max_sectors, "Maximum sectors for this adapter. " >> + "[Default=" __stringify(IBMVFC_MAX_SECTORS) "]"); >> module_param_named(scsi_qdepth, scsi_qdepth, ushort, S_IRUGO); >> MODULE_PARM_DESC(scsi_qdepth, "Maximum scsi command depth per >> adapter queue. " >> "[Default=" __stringify(IBMVFC_SCSI_QDEPTH) "]"); >> @@ -1494,7 +1498,7 @@ static void ibmvfc_set_login_info(struct >> ibmvfc_host *vhost) >> memset(login_info, 0, sizeof(*login_info)); >> >> login_info->ostype = cpu_to_be32(IBMVFC_OS_LINUX); >> - login_info->max_dma_len = cpu_to_be64(IBMVFC_MAX_SECTORS << >> 9); >> + login_info->max_dma_len = cpu_to_be64(max_sectors << 9); >> login_info->max_payload = cpu_to_be32(sizeof(struct >> ibmvfc_fcp_cmd_iu)); >> login_info->max_response = cpu_to_be32(sizeof(struct >> ibmvfc_fcp_rsp)); >> login_info->partition_num = cpu_to_be32(vhost- >>> partition_number); >> @@ -5230,7 +5234,7 @@ static void ibmvfc_npiv_login_done(struct >> ibmvfc_event *evt) >> } >> >> vhost->logged_in = 1; >> - npiv_max_sectors = min((uint)(be64_to_cpu(rsp->max_dma_len) >>>> 9), IBMVFC_MAX_SECTORS); >> + npiv_max_sectors = min((uint)(be64_to_cpu(rsp->max_dma_len) >>>> 9), max_sectors); >> dev_info(vhost->dev, "Host partition: %s, device: %s %s %s >> max sectors %u\n", >> rsp->partition_name, rsp->device_name, rsp- >>> port_loc_code, >> rsp->drc_name, npiv_max_sectors); >> @@ -6329,7 +6333,7 @@ static int ibmvfc_probe(struct vio_dev *vdev, >> const struct vio_device_id *id) >> shost->can_queue = scsi_qdepth; >> shost->max_lun = max_lun; >> shost->max_id = max_targets; >> - shost->max_sectors = IBMVFC_MAX_SECTORS; >> + shost->max_sectors = max_sectors; > > Would it make sense to check whether the user-provided max_sectors > value is within some reasonable limits? Agreed. I'll follow up with an updated version. Thanks, Brian
On 7/30/24 19:51, Brian King wrote: > There are some scenarios that can occur, such as performing an > upgrade of the virtual I/O server, where the supported max transfer > of the backing device for an ibmvfc HBA can change. If the max > transfer of the backing device decreases, this can cause issues with > previously discovered LUNs. This patch accomplishes two things. > First, it changes the default ibmvfc max transfer value to 1MB. > This is generally supported by all backing devices, which should > mitigate this issue out of the box. Secondly, it adds a module > parameter, enabling a user to increase the max transfer value to > values that are larger than 1MB, as long as they have configured > these larger values on the virtual I/O server as well. > > Signed-off-by: Brian King <brking@linux.ibm.com> > --- > drivers/scsi/ibmvscsi/ibmvfc.c | 10 +++++++--- > drivers/scsi/ibmvscsi/ibmvfc.h | 2 +- > 2 files changed, 8 insertions(+), 4 deletions(-) > Reviewed-by: Hannes Reinecke <hare@suse.de> Cheers, Hannes
diff --git a/drivers/scsi/ibmvscsi/ibmvfc.c b/drivers/scsi/ibmvscsi/ibmvfc.c index a3d1013c8307..611901562e06 100644 --- a/drivers/scsi/ibmvscsi/ibmvfc.c +++ b/drivers/scsi/ibmvscsi/ibmvfc.c @@ -37,6 +37,7 @@ static unsigned int default_timeout = IBMVFC_DEFAULT_TIMEOUT; static u64 max_lun = IBMVFC_MAX_LUN; static unsigned int max_targets = IBMVFC_MAX_TARGETS; static unsigned int max_requests = IBMVFC_MAX_REQUESTS_DEFAULT; +static unsigned int max_sectors = IBMVFC_MAX_SECTORS; static u16 scsi_qdepth = IBMVFC_SCSI_QDEPTH; static unsigned int disc_threads = IBMVFC_MAX_DISC_THREADS; static unsigned int ibmvfc_debug = IBMVFC_DEBUG; @@ -83,6 +84,9 @@ MODULE_PARM_DESC(default_timeout, module_param_named(max_requests, max_requests, uint, S_IRUGO); MODULE_PARM_DESC(max_requests, "Maximum requests for this adapter. " "[Default=" __stringify(IBMVFC_MAX_REQUESTS_DEFAULT) "]"); +module_param_named(max_sectors, max_sectors, uint, S_IRUGO); +MODULE_PARM_DESC(max_sectors, "Maximum sectors for this adapter. " + "[Default=" __stringify(IBMVFC_MAX_SECTORS) "]"); module_param_named(scsi_qdepth, scsi_qdepth, ushort, S_IRUGO); MODULE_PARM_DESC(scsi_qdepth, "Maximum scsi command depth per adapter queue. " "[Default=" __stringify(IBMVFC_SCSI_QDEPTH) "]"); @@ -1494,7 +1498,7 @@ static void ibmvfc_set_login_info(struct ibmvfc_host *vhost) memset(login_info, 0, sizeof(*login_info)); login_info->ostype = cpu_to_be32(IBMVFC_OS_LINUX); - login_info->max_dma_len = cpu_to_be64(IBMVFC_MAX_SECTORS << 9); + login_info->max_dma_len = cpu_to_be64(max_sectors << 9); login_info->max_payload = cpu_to_be32(sizeof(struct ibmvfc_fcp_cmd_iu)); login_info->max_response = cpu_to_be32(sizeof(struct ibmvfc_fcp_rsp)); login_info->partition_num = cpu_to_be32(vhost->partition_number); @@ -5230,7 +5234,7 @@ static void ibmvfc_npiv_login_done(struct ibmvfc_event *evt) } vhost->logged_in = 1; - npiv_max_sectors = min((uint)(be64_to_cpu(rsp->max_dma_len) >> 9), IBMVFC_MAX_SECTORS); + npiv_max_sectors = min((uint)(be64_to_cpu(rsp->max_dma_len) >> 9), max_sectors); dev_info(vhost->dev, "Host partition: %s, device: %s %s %s max sectors %u\n", rsp->partition_name, rsp->device_name, rsp->port_loc_code, rsp->drc_name, npiv_max_sectors); @@ -6329,7 +6333,7 @@ static int ibmvfc_probe(struct vio_dev *vdev, const struct vio_device_id *id) shost->can_queue = scsi_qdepth; shost->max_lun = max_lun; shost->max_id = max_targets; - shost->max_sectors = IBMVFC_MAX_SECTORS; + shost->max_sectors = max_sectors; shost->max_cmd_len = IBMVFC_MAX_CDB_LEN; shost->unique_id = shost->host_no; shost->nr_hw_queues = mq_enabled ? min(max_scsi_queues, nr_scsi_hw_queues) : 1; diff --git a/drivers/scsi/ibmvscsi/ibmvfc.h b/drivers/scsi/ibmvscsi/ibmvfc.h index 745ad5ac7251..c73ed2314ad0 100644 --- a/drivers/scsi/ibmvscsi/ibmvfc.h +++ b/drivers/scsi/ibmvscsi/ibmvfc.h @@ -32,7 +32,7 @@ #define IBMVFC_DEBUG 0 #define IBMVFC_MAX_TARGETS 1024 #define IBMVFC_MAX_LUN 0xffffffff -#define IBMVFC_MAX_SECTORS 0xffffu +#define IBMVFC_MAX_SECTORS 2048 #define IBMVFC_MAX_DISC_THREADS 4 #define IBMVFC_TGT_MEMPOOL_SZ 64 #define IBMVFC_MAX_CMDS_PER_LUN 64
There are some scenarios that can occur, such as performing an upgrade of the virtual I/O server, where the supported max transfer of the backing device for an ibmvfc HBA can change. If the max transfer of the backing device decreases, this can cause issues with previously discovered LUNs. This patch accomplishes two things. First, it changes the default ibmvfc max transfer value to 1MB. This is generally supported by all backing devices, which should mitigate this issue out of the box. Secondly, it adds a module parameter, enabling a user to increase the max transfer value to values that are larger than 1MB, as long as they have configured these larger values on the virtual I/O server as well. Signed-off-by: Brian King <brking@linux.ibm.com> --- drivers/scsi/ibmvscsi/ibmvfc.c | 10 +++++++--- drivers/scsi/ibmvscsi/ibmvfc.h | 2 +- 2 files changed, 8 insertions(+), 4 deletions(-)