linux-kernel - Re: [PATCH] esp_scsi: Clear Transfer Count registers before PIO transfers

lists.openwall.net		lists / announce owl-users owl-dev john-users john-dev passwdqc-users yescrypt popa3d-users / oss-security kernel-hardening musl sabotage tlsify passwords / crypt-dev xvendor / Bugtraq Full-Disclosure linux-kernel linux-netdev linux-ext4 linux-hardening linux-cve-announce PHC
Open Source and information security mailing list archives
Hash Suite: Windows password security audit tool. GUI, reports in PDF.
[<prev] [next>] [<thread-prev] [thread-next>] [day] [month] [year] [list]
Message-ID: <CACz-3rhr_R+_mJpg+Rvgoj=1GC5AO3YKxYxS_0ShAfQ-NiFDtg@mail.gmail.com>
Date:   Mon, 18 Nov 2019 09:13:48 +0100
From:   Kars de Jong <jongk@...ux-m68k.org>
To:     Finn Thain <fthain@...egraphics.com.au>
Cc:     "James E.J. Bottomley" <jejb@...ux.ibm.com>,
        "Martin K. Petersen" <martin.petersen@...cle.com>,
        Michael Schmitz <schmitzmic@...il.com>,
        linux-scsi@...r.kernel.org, linux-kernel@...r.kernel.org
Subject: Re: [PATCH] esp_scsi: Clear Transfer Count registers before PIO transfers

Hi Finn,

Op ma 18 nov. 2019 om 00:13 schreef Finn Thain <fthain@...egraphics.com.au>:
>
> On Sun, 17 Nov 2019, Kars de Jong wrote:
> > Are you sure this is really needed?
> >
>
> No. I think it improves robustness and correctness.
>
> I would be interested to know whether there is any measurable performance
> impact on zorro_esp.
>
> > The only [time when] the driver reads these registers is after a data
> > transfer. These are done using DMA on all Zorro boards, so I don't think
> > there's a risk of stale values from a PIO transfer there.
> >
>
> I'm not entirely sure that the chip is unaffected by stale counter values.
>
> (Stale transfer counter values are distinct from stale transfer count
> register values. Both are addressed by the patch.)

I still don't see the need to address that in the PIO transfer code.
The ESP (when in initiator mode)
doesn't use the transfer counter (registers) in PIO mode.

> If there are DMA controllers out there that can't do very short transfers
> then this objection would seem to be invalid, because the "DMA length is
> zero!" issue could be tackled using PIO.

That's the issue you fixed by limiting the transfer size to 65535
bytes, correct?
The SYM53CF92-X in my Blizzard didn't show this error for the 1 byte
transfers. It just hung up:

 sdb: RDSK (512) sdb1 (DOS^C)(res 2 spb 1) sdb2 (SWP^@)(res 0 spb 8)
sdb3 (LNX^@)(res 2 spb 1) sdb4 (LNX^@)(res 2 spb 1)
sd 1:0:0:0: [sdb] Attached SCSI disk
EXT4-fs (sdb3): mounting ext3 file system using the ext4 subsystem
scsi host1: Aborting command [(ptrval):28]
scsi host1: Current command [(ptrval):28]
scsi host1:  Active command [(ptrval):28]
scsi host1: Dumping command log
scsi host1: ent[6] CMD val[12] sreg[97] seqreg[9c] sreg2[00] ireg[08]
ss[00] event[0c]
scsi host1: ent[7] EVENT val[0d] sreg[91] seqreg[9c] sreg2[00]
ireg[08] ss[00] event[0c]
scsi host1: ent[8] EVENT val[03] sreg[91] seqreg[c4] sreg2[00]
ireg[10] ss[00] event[0d]
scsi host1: ent[9] CMD val[80] sreg[91] seqreg[c4] sreg2[00] ireg[10]
ss[00] event[03]
scsi host1: ent[10] CMD val[90] sreg[91] seqreg[c4] sreg2[00] ireg[10]
ss[00] event[03]
scsi host1: ent[11] EVENT val[05] sreg[91] seqreg[c4] sreg2[00]
ireg[10] ss[00] event[03]
scsi host1: ent[12] EVENT val[0d] sreg[93] seqreg[cc] sreg2[00]
ireg[10] ss[00] event[05]
scsi host1: ent[13] CMD val[01] sreg[93] seqreg[cc] sreg2[00] ireg[10]
ss[00] event[0d]
scsi host1: ent[14] CMD val[11] sreg[93] seqreg[cc] sreg2[00] ireg[10]
ss[00] event[0d]
scsi host1: ent[15] EVENT val[0b] sreg[93] seqreg[cc] sreg2[00]
ireg[10] ss[00] event[0d]
scsi host1: ent[16] CMD val[12] sreg[97] seqreg[cc] sreg2[00] ireg[08]
ss[00] event[0b]
scsi host1: ent[17] EVENT val[0c] sreg[97] seqreg[cc] sreg2[00]
ireg[08] ss[00] event[0b]
scsi host1: ent[18] CMD val[44] sreg[90] seqreg[cc] sreg2[00] ireg[20]
ss[00] event[0c]
scsi host1: ent[19] CMD val[01] sreg[90] seqreg[cc] sreg2[00] ireg[20]
ss[01] event[0c]
scsi host1: ent[20] CMD val[46] sreg[90] seqreg[cc] sreg2[00] ireg[20]
ss[01] event[0c]
scsi host1: ent[21] EVENT val[0d] sreg[97] seqreg[04] sreg2[00]
ireg[18] ss[00] event[0c]
scsi host1: ent[22] EVENT val[06] sreg[97] seqreg[04] sreg2[00]
ireg[18] ss[00] event[0d]
scsi host1: ent[23] CMD val[01] sreg[97] seqreg[04] sreg2[00] ireg[18]
ss[00] event[06]
scsi host1: ent[24] CMD val[10] sreg[97] seqreg[04] sreg2[00] ireg[18]
ss[00] event[06]
scsi host1: ent[25] EVENT val[0c] sreg[97] seqreg[8c] sreg2[00]
ireg[08] ss[00] event[06]
scsi host1: ent[26] CMD val[12] sreg[97] seqreg[8c] sreg2[00] ireg[08]
ss[00] event[0c]
scsi host1: ent[27] CMD val[44] sreg[90] seqreg[cc] sreg2[00] ireg[20]
ss[00] event[0c]
scsi host1: ent[28] CMD val[01] sreg[97] seqreg[9c] sreg2[00] ireg[0c]
ss[00] event[0c]
scsi host1: ent[29] CMD val[00] sreg[97] seqreg[9c] sreg2[00] ireg[0c]
ss[00] event[0c]
scsi host1: ent[30] CMD val[12] sreg[97] seqreg[9c] sreg2[00] ireg[0c]
ss[00] event[0c]
scsi host1: ent[31] CMD val[10] sreg[97] seqreg[9c] sreg2[00] ireg[10]
ss[00] event[0c]
scsi host1: ent[0] CMD val[12] sreg[97] seqreg[9c] sreg2[00] ireg[08]
ss[00] event[0c]
scsi host1: ent[1] EVENT val[0d] sreg[91] seqreg[9c] sreg2[00]
ireg[08] ss[00] event[0c]
scsi host1: ent[2] EVENT val[03] sreg[91] seqreg[c4] sreg2[00]
ireg[10] ss[00] event[0d]
scsi host1: ent[3] CMD val[80] sreg[91] seqreg[c4] sreg2[00] ireg[10]
ss[00] event[03]
scsi host1: ent[4] CMD val[90] sreg[91] seqreg[c4] sreg2[00] ireg[10]
ss[00] event[03]
scsi host1: ent[5] EVENT val[05] sreg[91] seqreg[c4] sreg2[00]
ireg[10] ss[00] event[03]

> > The only place the controller reads these registers is when a DMA
> > command is issued. The only place where that is done is in the zorro_esp
> > send_dma_command() functions.
>
> Aren't you overlooking all of the ESP_CMD_DMA flags in the core driver?

No, all those occurences are only used when calling
send_dma_command(), except the NULL/NOP DMA commands
directly after a chip reset.

Kind regards,

Kars.