SG_NONE fix and cleanup
Subsystems (9)
cover SG_NONE fix and cleanup
These two patches address some issues stemming from scsi-mq conversion.
Finn Thain (2):
atari_scsi, sun3_scsi: Set sg_tablesize to 1 instead of SG_NONE
scsi: Clean up SG_NONE
Documentation/scsi/scsi_mid_low_api.txt | 3 ++- drivers/ata/pata_arasan_cf.c | 1 - drivers/scsi/atari_scsi.c | 6 +++--- drivers/scsi/atp870u.c | 2 +- drivers/scsi/mac_scsi.c | 2 +- drivers/scsi/sun3_scsi.c | 4 ++-- drivers/usb/storage/uas.c | 1 - include/scsi/scsi_host.h | 13 ------------- 8 files changed, 9 insertions(+), 23 deletions(-)
Signature · 1 line
2.23.0
PATCH 1/2 atari_scsi, sun3_scsi: Set sg_tablesize to 1 instead of SG_NONE
Since the scsi subsystem adopted the blk-mq API, a host with zero sg_tablesize crashes with a NULL pointer dereference.
blk_queue_max_segments: set to minimum 1
scsi 0:0:0:0: Direct-Access QEMU QEMU HARDDISK 2.5+ PQ: 0 ANSI: 5
scsi target0:0:0: Beginning Domain Validation
scsi target0:0:0: Domain Validation skipping write tests
scsi target0:0:0: Ending Domain Validationblk_queue_max_segments: set to minimum 1
scsi 0:0:1:0: Direct-Access QEMU QEMU HARDDISK 2.5+ PQ: 0 ANSI: 5
scsi target0:0:1: Beginning Domain Validation
scsi target0:0:1: Domain Validation skipping write tests
scsi target0:0:1: Ending Domain Validationblk_queue_max_segments: set to minimum 1
scsi 0:0:2:0: CD-ROM QEMU QEMU CD-ROM 2.5+ PQ: 0 ANSI: 5
scsi target0:0:2: Beginning Domain Validation
scsi target0:0:2: Domain Validation skipping write tests
scsi target0:0:2: Ending Domain Validationblk_queue_max_segments: set to minimum 1blk_queue_max_segments: set to minimum 1blk_queue_max_segments: set to minimum 1blk_queue_max_segments: set to minimum 1
sr 0:0:2:0: Power-on or device reset occurred
sd 0:0:0:0: Power-on or device reset occurred
sd 0:0:1:0: Power-on or device reset occurred
sd 0:0:0:0: [sda] 10485762 512-byte logical blocks: (5.37 GB/5.00 GiB)
sd 0:0:0:0: [sda] Write Protect is off
sd 0:0:0:0: [sda] Write cache: enabled, read cache: enabled, doesn't support DPO or FUA
Unable to handle kernel NULL pointer dereference at virtual address (ptrval) Oops: 00000000 Modules linked in: PC: [<001cd874>] blk_mq_free_request+0x66/0xe2 SR: 2004 SP: (ptrval) a2: 00874520 d0: 00000000 d1: 00000000 d2: 009ba800 d3: 00000000 d4: 00000000 d5: 08000002 a0: 0087be68 a1: 009a81e0
Process kworker/u2:2 (pid: 15, task=(ptrval)) Frame format=7 eff addr=0000007a ssw=0505 faddr=0000007a wb 1 stat/addr/data: 0000 00000000 00000000
wb 2 stat/addr/data: 0000 00000000 00000000
wb 3 stat/addr/data: 0000 0000007a 00000000 push data: 00000000 00000000 00000000 00000000 Stack from 0087bd98:
00000002 00000000 0087be72 009a7820 0087bdb4 001c4f6c 009a7820 0087bdd4
0024d200 009a7820 0024d0dc 0087be72 009baa00 0087be68 009a5000 0087be7c
00265d10 009a5000 0087be72 00000003 00000000 00000000 00000000 0087be68
00000bb8 00000005 00000000 00000000 00000000 00000000 00265c56 00000000
009ba60c 0036ddf4 00000002 ffffffff 009baa00 009ba600 009a50d6 0087be74
00227ba0 009baa08 00000001 009baa08 009ba60c 0036ddf4 00000000 00000000
Call Trace: [<001c4f6c>] blk_put_request+0xe/0x14
[<0024d200>] __scsi_execute+0x124/0x174
[<0024d0dc>] __scsi_execute+0x0/0x174
[<00265d10>] sd_revalidate_disk+0xba/0x1f02
[<00265c56>] sd_revalidate_disk+0x0/0x1f02
[<0036ddf4>] strlen+0x0/0x22
[<00227ba0>] device_add+0x3da/0x604
[<0036ddf4>] strlen+0x0/0x22
[<00267e64>] sd_probe+0x30c/0x4b4
[<0002da44>] process_one_work+0x0/0x402
[<0022b978>] really_probe+0x226/0x354
[<0022bc34>] driver_probe_device+0xa4/0xf0
[<0002da44>] process_one_work+0x0/0x402
[<0022bcd0>] __driver_attach_async_helper+0x50/0x70
[<00035dae>] async_run_entry_fn+0x36/0x130
[<0002db88>] process_one_work+0x144/0x402
[<0002e1aa>] worker_thread+0x0/0x570
[<0002e29a>] worker_thread+0xf0/0x570
[<0002e1aa>] worker_thread+0x0/0x570
[<003768d8>] schedule+0x0/0xb8
[<0003f58c>] __init_waitqueue_head+0x0/0x12
[<00033e92>] kthread+0xc2/0xf6
[<000331e8>] kthread_parkme+0x0/0x4e
[<003768d8>] schedule+0x0/0xb8
[<00033dd0>] kthread+0x0/0xf6
[<00002c10>] ret_from_kernel_thread+0xc/0x14
Code: 0280 0006 0800 56c0 4400 0280 0000 00ff <52b4> 0c3a 082b 0006 0013 6706 2042 53a8 00c4 4ab9 0047 3374 6640 202d 000c 670cDisabling lock debugging due to kernel taint
Avoid this by setting sg_tablesize = 1.
References: commit 68ab2d76e4be ("scsi: cxlflash: Set sg_tablesize to 1 instead of SG_NONE")
--- drivers/scsi/atari_scsi.c | 6 +++--- drivers/scsi/mac_scsi.c | 2 +- drivers/scsi/sun3_scsi.c | 4 ++-- 3 files changed, 6 insertions(+), 6 deletions(-)
drivers/scsi/atari_scsi.c +3 -3
@@ -742,7 +742,7 @@ static int __init atari_scsi_probe(struct platform_device *pdev) atari_scsi_template.sg_tablesize =SG_ALL ; }else { atari_scsi_template.can_queue =1 ; - atari_scsi_template.sg_tablesize =SG_NONE ; + atari_scsi_template.sg_tablesize =1 ; }if (setup_can_queue >0 )
@@ -751,8 +751,8 @@ static int __init atari_scsi_probe(struct platform_device *pdev)if (setup_cmd_per_lun >0 ) atari_scsi_template.cmd_per_lun = setup_cmd_per_lun; -/* Leave sg_tablesize at 0 on a Falcon! */ -if (ATARIHW_PRESENT (TT_SCSI ) && setup_sg_tablesize >=0 ) +/* Don't increase sg_tablesize on Falcon! */ +if (ATARIHW_PRESENT (TT_SCSI ) && setup_sg_tablesize >0 ) atari_scsi_template.sg_tablesize = setup_sg_tablesize;if (setup_hostid >=0 ) {
drivers/scsi/mac_scsi.c +1 -1
@@ -464,7 +464,7 @@ static int __init mac_scsi_probe(struct platform_device *pdev) mac_scsi_template.can_queue = setup_can_queue;if (setup_cmd_per_lun >0 ) mac_scsi_template.cmd_per_lun = setup_cmd_per_lun; -if (setup_sg_tablesize >=0 ) +if (setup_sg_tablesize >0 ) mac_scsi_template.sg_tablesize = setup_sg_tablesize;if (setup_hostid >=0 ) mac_scsi_template.this_id = setup_hostid &7 ;
drivers/scsi/sun3_scsi.c +2 -2
@@ -501,7 +501,7 @@ static struct scsi_host_template sun3_scsi_template = { .eh_host_reset_handler = sun3scsi_host_reset, .can_queue = 16, .this_id = 7, - .sg_tablesize = SG_NONE, + .sg_tablesize = 1, .cmd_per_lun = 2, .dma_boundary = PAGE_SIZE - 1, .cmd_size = NCR5380_CMD_SIZE,
@@ -523,7 +523,7 @@ static int __init sun3_scsi_probe(struct platform_device *pdev) sun3_scsi_template.can_queue = setup_can_queue;if (setup_cmd_per_lun >0 ) sun3_scsi_template.cmd_per_lun = setup_cmd_per_lun; -if (setup_sg_tablesize >=0 ) +if (setup_sg_tablesize >0 ) sun3_scsi_template.sg_tablesize = setup_sg_tablesize;if (setup_hostid >=0 ) sun3_scsi_template.this_id = setup_hostid &7 ;
Signature · 1 line
2.23.0
PATCH 2/2 scsi: Clean up SG_NONE
In reply to Finn ThainRemove SG_NONE and a related misleading comment. Update documentation.
This patch does not affect behaviour as zero initialization is redundant.
--- Documentation/scsi/scsi_mid_low_api.txt | 3 ++- drivers/ata/pata_arasan_cf.c | 1 - drivers/scsi/atp870u.c | 2 +- drivers/usb/storage/uas.c | 1 - include/scsi/scsi_host.h | 13 ------------- 5 files changed, 3 insertions(+), 17 deletions(-)
Documentation/scsi/scsi_mid_low_api.txt +2 -1
@@ -1084,7 +1084,8 @@ of interest: commands to the adapter. this_id - scsi id of host (scsi initiator) or -1 if not known sg_tablesize - maximum scatter gather elements allowed by host. - 0 implies scatter gather not supported by host + Set this to SG_ALL or less to avoid chained SG lists. + Must be at least 1. max_sectors - maximum number of sectors (usually 512 bytes) allowed in a single SCSI command. The default value of 0 leads to a setting of SCSI_DEFAULT_MAX_SECTORS (defined in
drivers/ata/pata_arasan_cf.c +0 -1
@@ -219,7 +219,6 @@ struct arasan_cf_dev {static struct scsi_host_template arasan_cf_sht = {ATA_BASE_SHT (DRIVER_NAME ), - .sg_tablesize =SG_NONE , .dma_boundary =0xFFFFFFFFUL , };
drivers/scsi/atp870u.c +1 -1
@@ -1680,7 +1680,7 @@ static struct scsi_host_template atp870u_template = { .bios_param = atp870u_biosparam /* biosparm */, .can_queue = qcnt /* can_queue */, .this_id = 7 /* SCSI ID */, - .sg_tablesize = ATP870U_SCATTER /*SG_ALL*/ /*SG_NONE*/, + .sg_tablesize = ATP870U_SCATTER /*SG_ALL*/, .max_sectors = ATP870U_MAX_SECTORS, };
drivers/usb/storage/uas.c +0 -1
@@ -879,7 +879,6 @@ static struct scsi_host_template uas_host_template = { .eh_abort_handler = uas_eh_abort_handler, .eh_device_reset_handler = uas_eh_device_reset_handler, .this_id = -1, - .sg_tablesize = SG_NONE, .skip_settle_delay = 1, .dma_boundary = PAGE_SIZE - 1, };
include/scsi/scsi_host.h +0 -13
@@ -23,19 +23,6 @@ struct scsi_host_cmd_pool;struct scsi_transport_template ; -/* -* The various choices mean: -* NONE: Self evident. Host adapter is not capable of scatter-gather. -* ALL: Means that the host adapter module can do scatter-gather, -* and that there is no limit to the size of the table to which -* we scatter/gather data. The value we set here is the maximum -* single element sglist. To use chained sglists, the adapter -* has to set a value beyond ALL (and correctly use the chain -* handling API. -* Anything else: Indicates the maximum number of chains that can be -* used in one scatter-gather request. -*/ -#define SG_NONE 0#define SG_ALL SG_CHUNK_SIZE#define MODE_UNKNOWN 0x00
Signature · 1 line
2.23.0
Michael Schmitz
maintainerPATCH 1/1 scsi: scsi-lib.c: increase cmd_size by sizeof(struct scatterlist)
In reply to Finn ThainIn scsi_mq_setup_tags(), cmd_size is calculated based on zero size for the scatter-gather list in case the low level driver uses SG_NONE in its host template.
cmd_size is passed on to the block layer for calculation of the request size, and we've seen NULL pointer dereference errors from the block layer in drivers where SG_NONE is used and a mq IO scheduler is active, apparently as a consequence of this (see commit 68ab2d76e4be for the cxflash driver, and a recent patch by Finn Thain converting the three m68k NFR5380 drivers to avoid setting SG_NONE).
Try to avoid these errors by accounting for at least one sg list entry when caculating cmd_size, regardless of whether the low level driver set a zero sg_tablesize.
Tested on 030 m68k with the atari_scsi driver - setting sg_tablesize to SG_NONE no longer results in a crash when loading this driver.
--- drivers/scsi/scsi_lib.c | 3 ++- 1 files changed, 2 insertions(+), 1 deletions(-)
drivers/scsi/scsi_lib.c +2 -1
@@ -1867,7 +1867,8 @@ int scsi_mq_setup_tags(struct Scsi_Host *shost) {unsigned int cmd_size, sgl_size; - sgl_size =scsi_mq_inline_sgl_size (shost); + sgl_size = max_t(unsigned int, sizeof(struct scatterlist), +scsi_mq_inline_sgl_size (shost)); cmd_size =sizeof (struct scsi_cmnd ) + shost->hostt->cmd_size + sgl_size;if (scsi_host_get_prot (shost)) cmd_size += sizeof(struct scsi_data_buffer) +
Signature · 1 line
1.7.0.4
Re: [PATCH] scsi: scsi-lib.c: increase cmd_size by sizeof(struct scatterlist)
Quoted 15 lines · Michael Schmitz
In
scsi_mq_setup_tags(),cmd_sizeis calculated based on zero size for the scatter-gather list in case the low level driver usesSG_NONEin its host template.
cmd_sizeis passed on to the block layer for calculation of the request size, and we've seen NULL pointer dereference errors from the block layer in drivers whereSG_NONEis used and a mq IO scheduler is active, apparently as a consequence of this (see commit68ab2d76e4befor the cxflash driver, and a recent patch by Finn Thain converting the three m68k NFR5380 drivers to avoid settingSG_NONE).Try to avoid these errors by accounting for at least one sg list entry when caculating
cmd_size, regardless of whether the low level driver set a zerosg_tablesize.
Quoted 5 lines
Tested on 030 m68k with the
atari_scsidriver - settingsg_tablesizetoSG_NONEno longer results in a crash when loading this driver.Signed-off-by Michael Schmitz <[email protected]>CC Finn Thain <[email protected]>
No objections from me. I still like the patch series I sent but it's a matter of taste. I'm happy to leave it to the scsi maintainers to figure out the best fix for this regression.
Quoted 19 lines
--- drivers/scsi/scsi_lib.c | 3 ++- 1 files changed, 2 insertions(+), 1 deletions(-)drivers/scsi/scsi_lib.c +2
-1@@ -1867,7 +1867,8 @@ int scsi_mq_setup_tags(struct Scsi_Host *shost) {unsigned int cmd_size, sgl_size; - sgl_size =scsi_mq_inline_sgl_size (shost); + sgl_size = max_t(unsigned int, sizeof(struct scatterlist), +scsi_mq_inline_sgl_size (shost)); cmd_size =sizeof (struct scsi_cmnd ) + shost->hostt->cmd_size + sgl_size;if (scsi_host_get_prot (shost)) cmd_size += sizeof(struct scsi_data_buffer) +
Martin K. Petersen
AppliedRe: [PATCH] scsi: scsi-lib.c: increase cmd_size by sizeof(struct scatterlist)
In reply to Michael SchmitzMichael,
Quoted 3 lines
In
scsi_mq_setup_tags(),cmd_sizeis calculated based on zero size for the scatter-gather list in case the low level driver usesSG_NONEin its host template.
Applied to 5.4/scsi-fixes. I picked your version since it is small.
I still intend to queue Finn's SG_NONE cleanups for 5.5. In a way these are orthogonal.
Thanks!
Signature · 1 line
Martin K. Petersen Oracle Linux Engineering
Michael Schmitz
maintainerRe: [PATCH] scsi: scsi-lib.c: increase cmd_size by sizeof(struct scatterlist)
Thanks Martin,
Am 06.11.2019 um 18:46 schrieb Martin K. Petersen:
Quoted 8 lines
Michael,
Quoted 3 lines
In
scsi_mq_setup_tags(),cmd_sizeis calculated based on zero size for the scatter-gather list in case the low level driver usesSG_NONEin its host template.Applied to 5.4/scsi-fixes. I picked your version since it is small.
I realize it's more of a band-aid and doesn't address the real issue perhaps. Maybe it gives someone with a better clue about the block layer some idea though (I got out of my depth rather quickly).
Quoted 2 lines
I still intend to queue Finn's
SG_NONEcleanups for 5.5. In a way these are orthogonal.
Either zero or one work for sg_tablesize in my case, so that's perfectly fine by me.
Cheers,
Michael
Quoted 3 lines
Thanks!
Martin K. Petersen
AppliedFinn,
Quoted 1 line
These two patches address some issues stemming from scsi-mq conversion.
Applied to 5.5/scsi-queue, thanks!
Signature · 1 line
Martin K. Petersen Oracle Linux Engineering