Re: workaround for buggy bios_memmap?
On Tue, Mar 06, 2018 at 01:46:58PM +0900, Naoki Fukaumi wrote: > Hi tech@, > > I'm using FreeBSD bhyve to run OpenBSD/amd64 VM for testing biosboot. > (well, I know efiboot works fine on bhyve, but I want to try biosboot) > > it seems bhyve's BIOS has buggy(?) entry in bios_memmap(E820 table). > > > >> OpenBSD/amd64 BOOT 3.34 > boot> m m > Region 0: type 1 at 0x0 for 639KB > Region 1: type 2 at 0x9fc00 for 1KB > Region 2: type 2 at 0xe for 128KB > Region 3: type 1 at 0x10 for 3078904KB > Region 4: type 3 at 0xbbfbe000 for 128KB > Region 5: type 1 at 0xbbfde000 for 38836KB > Region 6: type 3 at 0xbe5cb000 for 4984KB > Region 7: type 1 at 0xbeaa9000 for 3556KB > Region 8: type 3 at 0xbee22000 for 11804KB > Region 9: type 1 at 0xbf9a9000 for 8KB > Region 10: type 3 at 0xbf9ab000 for 1864KB > Region 11: type 2 at 0xbfb7d000 for 16KB > Region 12: type 3 at 0xbfb81000 for 32KB > Region 13: type 4 at 0xbfb89000 for 16KB > Region 14: type 3 at 0xbfb8d000 for 4492KB > Region 15: type 1 at 0xbfff for 64KB > Region 16: type 2 at 0xc000 for 983040KB > Region 17: type 2 at 0xfec0 for 4KB > Region 18: type 2 at 0xfed0 for 1KB > Region 19: type 2 at 0xfee0 for 1024KB > Region 20: type 1 at 0x1 for 5242880KB > Low ram: 639KB High ram: 3078904KB > Total free memory: 8364887KB > > > in this case, "Region 16" covers PCI memory-mapped I/O space. (why?) > > in kernel, pci_init_extents() reserves there. then, PCI devices cannot > use there, it cannot be attached or it cannot work properly. > > > boot> boot > cannot open hd0a:/etc/random.seed: No such file or directory > booting hd0a:/6.3/amd64/bsd.rd: 3418091+1467392+3899896+0+593920 > [363308+90+433368+287961]=0x9fd048 > entry point at 0x1000158 > Copyright (c) 1982, 1986, 1989, 1991, 1993 > The Regents of the University of California. All rights reserved. > Copyright (c) 1995-2018 OpenBSD. All rights reserved. https://www.OpenBSD.org > > OpenBSD 6.3-beta (RAMDISK_CD) #26: Sun Mar 4 11:42:12 MST 2018 > dera...@amd64.openbsd.org:/usr/src/sys/arch/amd64/compile/RAMDISK_CD > real mem = 8549187584 (8153MB) > avail mem = 8286302208 (7902MB) > mainbus0 at root > bios0 at mainbus0: SMBIOS rev. 3.0 @ 0xbfb7f000 (14 entries) > bios0: vendor BHYVE version "1.00" date 03/14/2014 > bios0: bhyve BHYVE > acpi0 at bios0: rev 2 > acpi0: tables DSDT FACP HPET APIC MCFG SPCR > acpimadt0 at acpi0 addr 0xfee0: PC-AT compat > cpu0 at mainbus0: apid 0 (boot processor) > cpu0: AMD Ryzen 7 1700 Eight-Core Processor, 2994.31 MHz > cpu0: > FPU,VME,DE,PSE,TSC,MSR,PAE,MCE,CX8,APIC,SEP,MTRR,PGE,MCA,CMOV,PAT,PSE36,CFLUSH,MMX,FXSR,SSE,SSE2,HTT,SSE3,PCLMUL,SSSE3,FMA3,CX16,SSE4.1,SSE4.2,MOVBE,POPCNT,AES,XSAVE,AVX,F16C,RDRAND,HV,NXE,MMXX,FFXSR,PAGE1GB,LONG,LAHF,CMPLEG,EAPICSP,AMCR8,ABM,SSE4A,MASSE,3DNOWP,SKINIT,TCE,DBKP,ITSC,FSGSBASE,BMI1,AVX2,BMI2 > cpu0: 64KB 64b/line 4-way I-cache, 32KB 64b/line 8-way D-cache, 512KB > 64b/line 8-way L2 cache, 16MB 64b/line 16-way L3 cache > cpu0: ITLB 64 4KB entries fully associative, 64 4MB entries fully associative > cpu0: DTLB 64 4KB entries fully associative, 64 4MB entries fully associative > cpu0: apic clock running at 134MHz > cpu at mainbus0: not configured > cpu at mainbus0: not configured > cpu at mainbus0: not configured > ioapic0 at mainbus0: apid 4 pa 0xfec0, version 11, 32 pins > , remapped to apid 4 > acpiprt0 at acpi0: bus 0 (PC00) > pvbus0 at mainbus0: bhyve > pci0 at mainbus0 bus 0 > 0:4:0: mem address conflict 0xc000/0x400 > 0:10:0: mem address conflict 0xc0002000/0x2000 > (bhyve crashes here by calling pci_conf_write(pc, tag, reg, 0) twice ;) > > > I can see similar problem on HPE ProLiant DL20 Gen9 with BIOS boot mode. > (well, I'm sure it works well with UEFI boot mode, but I want to try...) > > I made a dirty patch for boot(8). it adds function to force-delete > reserved memory region. > > > boot> m m ~983040K@0xc000 > Region 0: type 1 at 0x0 for 639KB > Region 1: type 2 at 0x9fc00 for 1KB > Region 2: type 2 at 0xe for 128KB > Region 3: type 1 at 0x10 for 3078904KB > Region 4: type 3 at 0xbbfbe000 for 128KB > Region 5: type 1 at 0xbbfde000 for 38836KB > Region 6: type 3 at 0xbe5cb000 for 4984KB > Region 7: type 1 at 0xbeaa9000 for 3556KB > Region 8: type 3 at 0xbee22000 for 11804KB > Region 9: type 1 at 0xbf9a9000 for 8KB > Region 10: type 3 at 0xbf9ab000 for 1864KB > Region 11: type 2 at 0xbfb7d000 for 16KB > Region 12: type 3 at 0xbfb81000 for 32KB > Region 13: type 4 at 0xbfb89000 for 16KB > Region 14: type 3 at 0xbfb8d000 for 4492KB > Region 15: type 1 at 0xbfff for 64KB > Region 16: type 2 at 0xfec0 for 4KB > Region 17: type 2 at 0xfed0 for 1KB > Region 18: type 2 at 0xfee0 for 1024KB > Region 19: type 1 at 0x1 for 5242880KB > Low ram: 639KB High ram: 3078904KB > Total free memory: 8364887KB > boot> boot > cannot open hd0a:/etc/random.seed: No such file or directory > booting hd0a:/6.3/amd64/bsd.rd: 3418091+14
vmt: implement disk quiesce protocol with vfs_stall
A few days ago we ran into a situation where we wanted to clone a VM running
on VMWare to investigate some filesystem problems, but couldn't because vmt(4)
doesn't implement the protocol VMWare uses to quiesce disks to allow for
clean snapshots.
As it happens, we recently got vfs_stall(), which fits into the protocol nicely.
When the host tells us to quiesce, we just call vfs_stall(curproc, 1) and when
it says it's taken its snapshot, we call vfs_stall(curproc, 0).
I've tested this on esxi 5.5, where it lets me take disk-only snapshots of a VM
with 10gb of disk in around a second. As expected, when there are deleted
files that are still open, the filesystem is marked dirty, so when you revert
to the snapshot and boot the VM, fsck runs and cleans up the deleted files.
Otherwise, the filesystem is clean.
ok?
Index: vmt.c
===
RCS file: /cvs/src/sys/dev/pv/vmt.c,v
retrieving revision 1.13
diff -u -p -u -p -r1.13 vmt.c
--- vmt.c 4 Jun 2017 05:04:24 - 1.13
+++ vmt.c 6 Mar 2018 04:29:25 -
@@ -36,6 +36,8 @@
#include
#include
#include
+#include
+#include
#include
#include
@@ -125,6 +127,13 @@
#define VM_RPC_REPLY_ERROR "ERROR Unknown command"
#define VM_RPC_REPLY_ERROR_IP_ADDR "ERROR Unable to find guest IP address"
+/* VM backup error codes */
+#define VM_BACKUP_SUCCESS 0
+#define VM_BACKUP_SYNC_ERROR 3
+#define VM_BACKUP_REMOTE_ABORT 4
+
+#define VM_BACKUP_TIMEOUT 30 /* seconds */
+
/* A register. */
union vm_reg {
struct {
@@ -167,6 +176,8 @@ struct vmt_softc {
int sc_rpc_error;
int sc_tclo_ping;
int sc_set_guest_os;
+ int sc_quiesce;
+ struct task sc_quiesce_task;
#define VMT_RPC_BUFLEN 4096
struct timeout sc_tick;
@@ -231,6 +242,14 @@ voidvmt_tclo_resume(struct vmt_softc *
voidvmt_tclo_capreg(struct vmt_softc *);
voidvmt_tclo_broadcastip(struct vmt_softc *);
+voidvmt_set_backup_status(struct vmt_softc *, const char *, int,
+ const char *);
+voidvmt_quiesce_task(void *);
+voidvmt_quiesce_done_task(void *);
+voidvmt_tclo_abortbackup(struct vmt_softc *);
+voidvmt_tclo_startbackup(struct vmt_softc *);
+voidvmt_tclo_backupdone(struct vmt_softc *);
+
int vmt_probe(void);
struct vmt_tclo_rpc {
@@ -247,7 +266,10 @@ struct vmt_tclo_rpc {
{ "Set_Option broadcastIP 1", vmt_tclo_broadcastip },
{ "ping", vmt_tclo_ping },
{ "reset", vmt_tclo_reset },
- { NULL },
+{ "vmbackup.abort",vmt_tclo_abortbackup },
+{ "vmbackup.snapshotDone", vmt_tclo_backupdone },
+{ "vmbackup.start 1", vmt_tclo_startbackup },
+{ NULL },
#if 0
/* Various unsupported commands */
{ "Set_Option autohide 0" },
@@ -809,6 +831,123 @@ vmt_tclo_broadcastip(struct vmt_softc *s
}
}
+void
+vmt_set_backup_status(struct vmt_softc *sc, const char *state, int code,
+const char *desc)
+{
+ if (vm_rpc_send_rpci_tx(sc, "vmbackup.eventSet %s %d %s",
+ state, code, desc) != 0) {
+ DPRINTF("%s: setting backup status failed\n", DEVNAME(sc));
+ }
+}
+
+void
+vmt_quiesce_task(void *data)
+{
+ struct vmt_softc *sc = data;
+ int err;
+
+ DPRINTF("%s: quiescing filesystems for backup\n", DEVNAME(sc));
+ err = vfs_stall(curproc, 1);
+ if (err != 0) {
+ printf("%s: unable to quiesce filesystems\n", DEVNAME(sc));
+ vfs_stall(curproc, 0);
+
+ vmt_set_backup_status(sc, "req.aborted", VM_BACKUP_SYNC_ERROR,
+ "vfs_stall failed");
+ vmt_set_backup_status(sc, "req.done", VM_BACKUP_SUCCESS, "");
+ sc->sc_quiesce = 0;
+ return;
+ }
+
+ DPRINTF("%s: filesystems quiesced\n", DEVNAME(sc));
+ vmt_set_backup_status(sc, "prov.snapshotCommit", VM_BACKUP_SUCCESS, "");
+}
+
+void
+vmt_quiesce_done_task(void *data)
+{
+ struct vmt_softc *sc = data;
+
+ vfs_stall(curproc, 0);
+
+ if (sc->sc_quiesce == -1)
+ vmt_set_backup_status(sc, "req.aborted", VM_BACKUP_REMOTE_ABORT,
+ "");
+
+ vmt_set_backup_status(sc, "req.done", VM_BACKUP_SUCCESS, "");
+ sc->sc_quiesce = 0;
+}
+
+void
+vmt_tclo_abortbackup(struct vmt_softc *sc)
+{
+ const char *reply = VM_RPC_REPLY_OK;
+
+ if (sc->sc_quiesce > 0) {
+ DPRINTF("%s: aborting backup\n", DEVNAME(sc));
+ sc->sc_quiesce = -1;
+ task_set(&sc->sc_quiesce_task, vmt_quiesce_done_task, sc);
+ task_add(systq, &sc->sc_quiesce_task);
+ } else {
+ DPRINTF("%s: can't abort, no backup in
workaround for buggy bios_memmap?
Hi tech@, I'm using FreeBSD bhyve to run OpenBSD/amd64 VM for testing biosboot. (well, I know efiboot works fine on bhyve, but I want to try biosboot) it seems bhyve's BIOS has buggy(?) entry in bios_memmap(E820 table). >> OpenBSD/amd64 BOOT 3.34 boot> m m Region 0: type 1 at 0x0 for 639KB Region 1: type 2 at 0x9fc00 for 1KB Region 2: type 2 at 0xe for 128KB Region 3: type 1 at 0x10 for 3078904KB Region 4: type 3 at 0xbbfbe000 for 128KB Region 5: type 1 at 0xbbfde000 for 38836KB Region 6: type 3 at 0xbe5cb000 for 4984KB Region 7: type 1 at 0xbeaa9000 for 3556KB Region 8: type 3 at 0xbee22000 for 11804KB Region 9: type 1 at 0xbf9a9000 for 8KB Region 10: type 3 at 0xbf9ab000 for 1864KB Region 11: type 2 at 0xbfb7d000 for 16KB Region 12: type 3 at 0xbfb81000 for 32KB Region 13: type 4 at 0xbfb89000 for 16KB Region 14: type 3 at 0xbfb8d000 for 4492KB Region 15: type 1 at 0xbfff for 64KB Region 16: type 2 at 0xc000 for 983040KB Region 17: type 2 at 0xfec0 for 4KB Region 18: type 2 at 0xfed0 for 1KB Region 19: type 2 at 0xfee0 for 1024KB Region 20: type 1 at 0x1 for 5242880KB Low ram: 639KB High ram: 3078904KB Total free memory: 8364887KB in this case, "Region 16" covers PCI memory-mapped I/O space. (why?) in kernel, pci_init_extents() reserves there. then, PCI devices cannot use there, it cannot be attached or it cannot work properly. boot> boot cannot open hd0a:/etc/random.seed: No such file or directory booting hd0a:/6.3/amd64/bsd.rd: 3418091+1467392+3899896+0+593920 [363308+90+433368+287961]=0x9fd048 entry point at 0x1000158 Copyright (c) 1982, 1986, 1989, 1991, 1993 The Regents of the University of California. All rights reserved. Copyright (c) 1995-2018 OpenBSD. All rights reserved. https://www.OpenBSD.org OpenBSD 6.3-beta (RAMDISK_CD) #26: Sun Mar 4 11:42:12 MST 2018 dera...@amd64.openbsd.org:/usr/src/sys/arch/amd64/compile/RAMDISK_CD real mem = 8549187584 (8153MB) avail mem = 8286302208 (7902MB) mainbus0 at root bios0 at mainbus0: SMBIOS rev. 3.0 @ 0xbfb7f000 (14 entries) bios0: vendor BHYVE version "1.00" date 03/14/2014 bios0: bhyve BHYVE acpi0 at bios0: rev 2 acpi0: tables DSDT FACP HPET APIC MCFG SPCR acpimadt0 at acpi0 addr 0xfee0: PC-AT compat cpu0 at mainbus0: apid 0 (boot processor) cpu0: AMD Ryzen 7 1700 Eight-Core Processor, 2994.31 MHz cpu0: FPU,VME,DE,PSE,TSC,MSR,PAE,MCE,CX8,APIC,SEP,MTRR,PGE,MCA,CMOV,PAT,PSE36,CFLUSH,MMX,FXSR,SSE,SSE2,HTT,SSE3,PCLMUL,SSSE3,FMA3,CX16,SSE4.1,SSE4.2,MOVBE,POPCNT,AES,XSAVE,AVX,F16C,RDRAND,HV,NXE,MMXX,FFXSR,PAGE1GB,LONG,LAHF,CMPLEG,EAPICSP,AMCR8,ABM,SSE4A,MASSE,3DNOWP,SKINIT,TCE,DBKP,ITSC,FSGSBASE,BMI1,AVX2,BMI2 cpu0: 64KB 64b/line 4-way I-cache, 32KB 64b/line 8-way D-cache, 512KB 64b/line 8-way L2 cache, 16MB 64b/line 16-way L3 cache cpu0: ITLB 64 4KB entries fully associative, 64 4MB entries fully associative cpu0: DTLB 64 4KB entries fully associative, 64 4MB entries fully associative cpu0: apic clock running at 134MHz cpu at mainbus0: not configured cpu at mainbus0: not configured cpu at mainbus0: not configured ioapic0 at mainbus0: apid 4 pa 0xfec0, version 11, 32 pins , remapped to apid 4 acpiprt0 at acpi0: bus 0 (PC00) pvbus0 at mainbus0: bhyve pci0 at mainbus0 bus 0 0:4:0: mem address conflict 0xc000/0x400 0:10:0: mem address conflict 0xc0002000/0x2000 (bhyve crashes here by calling pci_conf_write(pc, tag, reg, 0) twice ;) I can see similar problem on HPE ProLiant DL20 Gen9 with BIOS boot mode. (well, I'm sure it works well with UEFI boot mode, but I want to try...) I made a dirty patch for boot(8). it adds function to force-delete reserved memory region. boot> m m ~983040K@0xc000 Region 0: type 1 at 0x0 for 639KB Region 1: type 2 at 0x9fc00 for 1KB Region 2: type 2 at 0xe for 128KB Region 3: type 1 at 0x10 for 3078904KB Region 4: type 3 at 0xbbfbe000 for 128KB Region 5: type 1 at 0xbbfde000 for 38836KB Region 6: type 3 at 0xbe5cb000 for 4984KB Region 7: type 1 at 0xbeaa9000 for 3556KB Region 8: type 3 at 0xbee22000 for 11804KB Region 9: type 1 at 0xbf9a9000 for 8KB Region 10: type 3 at 0xbf9ab000 for 1864KB Region 11: type 2 at 0xbfb7d000 for 16KB Region 12: type 3 at 0xbfb81000 for 32KB Region 13: type 4 at 0xbfb89000 for 16KB Region 14: type 3 at 0xbfb8d000 for 4492KB Region 15: type 1 at 0xbfff for 64KB Region 16: type 2 at 0xfec0 for 4KB Region 17: type 2 at 0xfed0 for 1KB Region 18: type 2 at 0xfee0 for 1024KB Region 19: type 1 at 0x1 for 5242880KB Low ram: 639KB High ram: 3078904KB Total free memory: 8364887KB boot> boot cannot open hd0a:/etc/random.seed: No such file or directory booting hd0a:/6.3/amd64/bsd.rd: 3418091+1467392+3899896+0+593920 [363308+90+433368+287961]=0x9fd048 entry point at 0x1000158 Copyright (c) 1982, 1986, 1989, 1991, 1993 The Regents of the University of California. All rights reserved. Copyright (c) 1995-2018 OpenBSD. All rights reserved. https://www.OpenBSD.org OpenBSD 6
ipmi(4): don't panic if ipmi_sendcmd() fails
Hi tech@,
sending IPMI command can fail for some reason (e.g. during BMC reset,
and possibly BMC is just busy?), then it should be better to return
error than panic().
--- sys/dev/ipmi.c
+++ sys/dev/ipmi.c
@@ -1039,10 +1039,10 @@ ipmi_cmd_poll(struct ipmi_cmd *c)
{
mtx_enter(&c->c_sc->sc_cmd_mtx);
- if (ipmi_sendcmd(c)) {
- panic("%s: sendcmd fails", DEVNAME(c->c_sc));
- }
- c->c_ccode = ipmi_recvcmd(c);
+ if ((c->c_ccode = ipmi_sendcmd(c)))
+ printf("%s: sendcmd fails\n", DEVNAME(c->c_sc));
+ else
+ c->c_ccode = ipmi_recvcmd(c);
mtx_leave(&c->c_sc->sc_cmd_mtx);
}
@@ -1819,8 +1819,6 @@ ipmiioctl(dev_t dev, u_long cmd, caddr_t data, int flag,
struct proc *proc)
c->c_txlen = req->msg.data_len;
c->c_rxlen = 0;
ipmi_cmd(c);
-
- KASSERT(c->c_ccode != -1);
break;
case IPMICTL_RECEIVE_MSG_TRUNC:
case IPMICTL_RECEIVE_MSG:
--
FUKAUMI Naoki
Re: Patch to add 2 words to share/dict/web2
> On 2018/03/05 17:14, Kurt Mosiejczuk wrote: > > It was annoying me that "aspirational" wasn't in common spellcheck > > dictionaries and afresh@ suggested I submit a patch to at least get > > the situation corrected for OpenBSD. > > > > Before getting around to this, I noticed "supremacist" wasn't there > > either so added that. > > hmm, according to README, this file is supposed to be "Webster's Second > International Dictionary, all 234,936 words worth. The 1934 copyright > has lapsed" ... should additions go in some other file or should the > description be changed? It seems to have grown a bit already, > > $ wc -l web2 > 235971 web2 If we added the complete english language, /usr would be full. (Already a few people having trouble with multiple upgrades + the new kernel/library link kits)
Re: mfii(4): add bio(4) support
Hi, From: YASUOKA Masahiko Subject: Re: mfii(4): add bio(4) support Date: Tue, 06 Mar 2018 08:14:44 +0900 (JST) >> most parts are taken from mfi(4), plus fix for rebuilding (bioctl -R). > > Do I understand the rebuilding problem correctly? > > - mfii_makegood() and mfii_makespare() are added before sending the > rebuild command to fix the problem > - The problem doesn't happen with mfi(4) well, sorry for insufficient explanation. mfi(4) has the problem. original fix for mfi(4) is here, https://marc.info/?t=14987241292&r=1&w=2 to support rebuilding in current bio(4) framework, swapped(new) disk should be exposed to userland as "unused" disk, but mfi(4) exposes removed disk as "failed" disk which cannot be controlled any way. first fix, in mfi(i)_ioctl_disk(), is "find an unused(new) disk and expose it to userland". and, disk state of unused disk need to be "unconfigured good" and not to be "foreign". second fix, mfi(i)_makegood(), try to change disk state if needed/possible. then, mfi(i)_makespare() starts rebuilding. Best regards, -- FUKAUMI Naoki
Re: Patch to add 2 words to share/dict/web2
On 03/05/18 19:16, Stuart Henderson wrote: On 2018/03/05 17:14, Kurt Mosiejczuk wrote: It was annoying me that "aspirational" wasn't in common spellcheck dictionaries and afresh@ suggested I submit a patch to at least get the situation corrected for OpenBSD. Before getting around to this, I noticed "supremacist" wasn't there either so added that. hmm, according to README, this file is supposed to be "Webster's Second International Dictionary, all 234,936 words worth. The 1934 copyright has lapsed" ... should additions go in some other file or should the description be changed? It seems to have grown a bit already, $ wc -l web2 235971 web2 Definitely it should just grow. Having multiple files doesn't make sense, to me. English is a growing language. --STeve Andre'
Re: Patch to add 2 words to share/dict/web2
On 2018/03/05 17:14, Kurt Mosiejczuk wrote: > It was annoying me that "aspirational" wasn't in common spellcheck > dictionaries and afresh@ suggested I submit a patch to at least get > the situation corrected for OpenBSD. > > Before getting around to this, I noticed "supremacist" wasn't there > either so added that. hmm, according to README, this file is supposed to be "Webster's Second International Dictionary, all 234,936 words worth. The 1934 copyright has lapsed" ... should additions go in some other file or should the description be changed? It seems to have grown a bit already, $ wc -l web2 235971 web2
Patch to add 2 words to share/dict/web2
It was annoying me that "aspirational" wasn't in common spellcheck dictionaries and afresh@ suggested I submit a patch to at least get the situation corrected for OpenBSD. Before getting around to this, I noticed "supremacist" wasn't there either so added that. Index: share/dict/web2 === RCS file: /cvs/src/share/dict/web2,v retrieving revision 1.8 diff -u -p -r1.8 web2 --- share/dict/web2 16 Dec 2017 13:01:22 - 1.8 +++ share/dict/web2 5 Mar 2018 22:02:55 - @@ -14387,6 +14387,7 @@ aspirant aspirata aspirate aspiration +aspirational aspirator aspiratory aspire @@ -195461,6 +195462,7 @@ supraventricular supraversion supravital supraworld +supremacist supremacy suprematism supreme
Re: mfii(4): add bio(4) support
Hi,
> most parts are taken from mfi(4), plus fix for rebuilding (bioctl -R).
Do I understand the rebuilding problem correctly?
- mfii_makegood() and mfii_makespare() are added before sending the
rebuild command to fix the problem
- The problem doesn't happen with mfi(4)
On Mon, 05 Mar 2018 21:49:16 +0900 (JST)
Naoki Fukaumi wrote:
> Hi tech@,
>
> This patch adds bio(4) support for mfii(4).
> # with "mfii(4): use MFII_FUNCTION_PASSTHRU_IO for MFI commands"
>
> most parts are taken from mfi(4), plus fix for rebuilding (bioctl -R).
>
> --- sys/dev/ic/mfireg.h
> +++ sys/dev/ic/mfireg.h
> @@ -139,6 +139,9 @@
> #define MR_DCMD_CONF_GET 0x0401
> #define MR_DCMD_CFG_ADD 0x0402
> #define MR_DCMD_CFG_CLEAR0x0403
> +#define MR_DCMD_CFG_MAKE_SPARE 0x0404
> +#define MR_DCMD_CFG_FOREIGN_SCAN 0x04060100
> +#define MR_DCMD_CFG_FOREIGN_CLEAR0x04060500
> #define MR_DCMD_BBU_GET_STATUS 0x0501
> #define MR_DCMD_BBU_GET_CAPACITY_INFO0x0502
> #define MR_DCMD_BBU_GET_DESIGN_INFO 0x0503
> @@ -1228,3 +1231,13 @@ struct mfi_pr_properties {
> uint32_texec_freq;
> uint32_tclear_freq;
> } __packed;
> +
> +/* We currently don't know the full details of the following struct */
> +struct mfii_foreign_scan_cfg {
> + char data[24];
> +};
> +
> +struct mfii_foreign_scan_info {
> + uint32_t count; /* Number of foreign configs found */
> + struct mfii_foreign_scan_cfg cfgs[8];
> +};
> --- sys/dev/pci/mfii.c
> +++ sys/dev/pci/mfii.c
> @@ -22,10 +22,14 @@
> #include
> #include
> #include
> +#include
> #include
> #include
> #include
> +#include
> +#include
>
> +#include
> #include
> #include
>
> @@ -307,7 +311,35 @@ struct mfii_softc {
> struct mfii_pd_softc*sc_pd;
> struct scsi_iopool sc_iopool;
>
> + /* save some useful information for logical drives that is missing
> + * in sc_ld_list
> + */
> + struct {
> + uint32_tld_present;
> + charld_dev[16]; /* device name sd? */
> + } sc_ld[MFI_MAX_LD];
> +
> + /* scsi ioctl from sd device */
> + int (*sc_ioctl)(struct device *, u_long, caddr_t);
> +
> + uint32_tsc_ld_cnt;
> +
> + /* bio */
> + struct mfi_conf *sc_cfg;
> struct mfi_ctrl_infosc_info;
> + struct mfi_ld_list sc_ld_list;
> + struct mfi_ld_details *sc_ld_details; /* array to all logical disks */
> + int sc_no_pd; /* used physical disks */
> + int sc_ld_sz; /* sizeof sc_ld_details */
> +
> + /* mgmt lock */
> + struct rwlock sc_lock;
> +
> + /* sensors */
> + struct ksensordev sc_sensordev;
> + struct ksensor *sc_bbu;
> + struct ksensor *sc_bbu_status;
> + struct ksensor *sc_sensors;
> };
>
> #ifdef MFII_DEBUG
> @@ -355,13 +387,15 @@ struct cfdriver mfii_cd = {
>
> void mfii_scsi_cmd(struct scsi_xfer *);
> void mfii_scsi_cmd_done(struct mfii_softc *, struct mfii_ccb *);
> +int mfii_scsi_ioctl(struct scsi_link *, u_long, caddr_t, int);
> +int mfii_ioctl_cache(struct scsi_link *, u_long, struct dk_cache *);
>
> struct scsi_adapter mfii_switch = {
> mfii_scsi_cmd,
> scsi_minphys,
> NULL, /* probe */
> NULL, /* unprobe */
> - NULL /* ioctl */
> + mfii_scsi_ioctl
> };
>
> void mfii_pd_scsi_cmd(struct scsi_xfer *);
> @@ -409,9 +443,11 @@ int mfii_load_mfa(struct mfii_softc
> *, struct mfii_ccb *,
>
> int mfii_mfa_poll(struct mfii_softc *, struct mfii_ccb *);
>
> -int mfii_mgmt(struct mfii_softc *, struct mfii_ccb *,
> - u_int32_t, const union mfi_mbox *,
> - void *, size_t, int);
> +int mfii_mgmt(struct mfii_softc *, uint32_t, uint32_t,
> + size_t, void *, const union mfi_mbox *);
> +int mfii_do_mgmt(struct mfii_softc *, struct mfii_ccb *,
> + uint32_t, uint32_t, size_t, void *,
> + const union mfi_mbox *);
> void mfii_empty_done(struct mfii_softc *, struct mfii_ccb *);
>
> int mfii_scsi_cmd_io(struct mfii_softc *,
> @@ -445,6 +481,42 @@ void mfii_aen_pd_remove(struct
> mfii_softc *,
> void mfii_aen_pd_state_change(struct mfii_softc *,
> const struct mfi_evtarg_pd_state *);
>
> +#if NBIO > 0
> +int mfii_ioctl(struct device *, u_long, caddr_t);
> +int mfii_bio_getitall(struct mfii_softc *);
> +int mfii_ioctl_inq(struct mfi
Re: handle updates via Adj-RIB-Out
Claudio,
How best to test this change proposal? Should this maybe be tested on
one of the yycix route servers?
I'll let it run on my home router, if that doesn't cause issues in a
week or so; we can consider rs2.yycix.ca
Kind regards,
Job
On Fri, Mar 02, 2018 at 04:55:23PM +0100, Claudio Jeker wrote:
> On Wed, Feb 07, 2018 at 05:52:09AM +0100, Claudio Jeker wrote:
> > This diff changes the way bgpd does updates. Instead of having its own
> > special update queue/tree it uses a regular RIB (Adj-RIB-Out) to store all
> > updates to be sent. Stuff that has been sent is linked to the prefixes
> > queue. On the peer there are also queues for updates and withdraws.
> > The whole update code becomes a lot simpler but also results in the bulk
> > of the diff. Other changes include the bgpctl show rib handling (we can
> > just walk the Adj-RIB-Out now). Last but not least the EOR records are
> > also now a magic rde_aspath (flag F_ATTR_EOR) which is added to the update
> > queue.
> >
> > This diff is still very large and the changes are intrusive so reviews and
> > testing is very welcome.
>
> No news on this? Anyone?
>
> --
> :wq Claudio
>
> Index: rde.c
> ===
> RCS file: /cvs/src/usr.sbin/bgpd/rde.c,v
> retrieving revision 1.377
> diff -u -p -r1.377 rde.c
> --- rde.c 7 Feb 2018 00:02:02 - 1.377
> +++ rde.c 7 Feb 2018 00:02:18 -
> @@ -80,8 +80,6 @@ void rde_dump_rib_as(struct prefix *,
>int);
> void rde_dump_filter(struct prefix *,
>struct ctl_show_rib_request *);
> -void rde_dump_filterout(struct rde_peer *, struct prefix *,
> - struct ctl_show_rib_request *);
> void rde_dump_upcall(struct rib_entry *, void *);
> void rde_dump_prefix_upcall(struct rib_entry *, void *);
> void rde_dump_ctx_new(struct ctl_show_rib_request *, pid_t,
> @@ -2262,71 +2260,33 @@ rde_dump_rib_as(struct prefix *p, struct
> }
>
> void
> -rde_dump_filterout(struct rde_peer *peer, struct prefix *p,
> -struct ctl_show_rib_request *req)
> +rde_dump_filter(struct prefix *p, struct ctl_show_rib_request *req)
> {
> - struct bgpd_addr addr;
> - struct rde_aspath *asp, *fasp;
> - enum filter_actions a;
> + struct rde_aspath *asp;
>
> - if (up_test_update(peer, p) != 1)
> + if (req->peerid && req->peerid != prefix_peer(p)->conf.id)
> return;
> + if (p->flags & F_PREFIX_USE_PEER)
> + return; /* pending withdraw, skip */
>
> - pt_getaddr(p->re->prefix, &addr);
> asp = prefix_aspath(p);
> - a = rde_filter(out_rules, &fasp, peer, asp, &addr,
> - p->re->prefix->prefixlen, asp->peer);
> - if (fasp)
> - fasp->peer = asp->peer;
> - else
> - fasp = asp;
> -
> - if (a == ACTION_ALLOW)
> - rde_dump_rib_as(p, fasp, req->pid, req->flags);
> -
> - if (fasp != asp)
> - path_put(fasp);
> -}
> -
> -void
> -rde_dump_filter(struct prefix *p, struct ctl_show_rib_request *req)
> -{
> - struct rde_peer *peer;
> - struct rde_aspath *asp;
> -
> - if (req->flags & F_CTL_ADJ_IN ||
> - !(req->flags & (F_CTL_ADJ_IN|F_CTL_ADJ_OUT))) {
> - asp = prefix_aspath(p);
> - if (req->peerid && req->peerid != asp->peer->conf.id)
> - return;
> - if (req->type == IMSG_CTL_SHOW_RIB_AS &&
> - !aspath_match(asp->aspath->data, asp->aspath->len,
> - &req->as, req->as.as))
> - return;
> - if (req->type == IMSG_CTL_SHOW_RIB_COMMUNITY &&
> - !community_match(asp, req->community.as,
> - req->community.type))
> - return;
> - if (req->type == IMSG_CTL_SHOW_RIB_EXTCOMMUNITY &&
> - !community_ext_match(asp, &req->extcommunity, 0))
> - return;
> - if (req->type == IMSG_CTL_SHOW_RIB_LARGECOMMUNITY &&
> - !community_large_match(asp, req->large_community.as,
> - req->large_community.ld1, req->large_community.ld2))
> - return;
> - if ((req->flags & F_CTL_ACTIVE) && p->re->active != p)
> - return;
> - rde_dump_rib_as(p, asp, req->pid, req->flags);
> - } else if (req->flags & F_CTL_ADJ_OUT) {
> - if (p->re->active != p)
> - /* only consider active prefix */
> - return;
> - if (req->peerid) {
> - if ((peer = peer_get(req->peerid)) != NULL)
> - rde_dump_filterout(peer, p, req);
> - return;
> - }
> - }
> + if (req->type == IMSG_CTL_SHOW_RIB_AS &&
> + !aspath_match(asp->aspath->data, asp->aspath->len
Re: A patch that fixes /usr/games/hack
The following patch is a slight improvement of the patch that I posted
on Thursday. In particular, this patch places saved games exactly where
the man page says they should be placed.
This patch, together with the patch that prevents segmentation faults
(http://marc.info/?l=openbsd-tech&m=152020888711960), should be all that
is needed to make /usr/games/hack playable on OpenBSD 6.2 without
setting the setgid bit.
Remember to create the /usr/share/games/hack directory before trying to
install this patched version of hack.
--- games/hack/config.h.origFri Oct 16 07:37:46 2015
+++ games/hack/config.h Mon Mar 5 02:31:18 2018
@@ -77,10 +77,10 @@
#define WIZARD "bruno"/* the person allowed to use the -D option */
#define RECORD "record"/* the file containing the list of topscorers */
#defineNEWS"news" /* the file containing the latest hack news */
-#defineHELP"help" /* the file containing a description of the
commands */
-#defineSHELP "hh"/* abbreviated form of the same */
-#defineRUMORFILE "rumors"/* a file with fortune cookies
*/
-#defineDATAFILE"data" /* a file giving the meaning of symbols
used */
+#defineHELP"/usr/share/games/hack/help"/* the file containing
a description of the commands */
+#defineSHELP "/usr/share/games/hack/hh" /* abbreviated form of
the same */
+#defineRUMORFILE "/usr/share/games/hack/rumors" /* a file with
fortune cookies */
+#defineDATAFILE"/usr/share/games/hack/data"/* a file
giving the meaning of symbols used */
#defineFMASK 0660/* file creation mask */
#defineHLOCK "perm" /* an empty file used for locking purposes */
#define LLOCK "safelock" /* link to previous */
@@ -93,7 +93,7 @@
* (This might be preferable for security reasons.)
* #define DEF_PAGER ".../mydir/mypager"
*/
-#defineDEF_PAGER _PATH_PAGER
+/* #define DEF_PAGER _PATH_PAGER */
/*
* If you define MAIL, then the player will be notified of new mail
@@ -120,7 +120,7 @@
#ifdef QUEST
#define HACKDIR _PATH_QUEST
#else /* QUEST */
-#define HACKDIR_PATH_HACK
+/* #define HACKDIR _PATH_HACK */
#endif /* QUEST */
/*
@@ -130,7 +130,7 @@
* since the user might create files in a directory of his choice.
* Of course SECURE is meaningful only if HACKDIR is defined.
*/
-#define SECURE /* do setuid(getuid()) after chdir() */
+/* #define SECURE */ /* do setuid(getuid()) after chdir() */
/*
* If it is desirable to limit the number of people that can play Hack
--- games/hack/hack.main.c.orig Sat Jan 9 21:54:11 2016
+++ games/hack/hack.main.c Mon Mar 5 16:17:31 2018
@@ -62,6 +62,7 @@
*/
#include
+#include
#include
#include
@@ -538,6 +539,8 @@
}
/* warn the player if he cannot write the record file */
+ /* warn the player if he cannot read the permanent lock file */
+ /* warn the player if he cannot create the save directory */
/* perhaps we should also test whether . is writable */
/* unfortunately the access systemcall is worthless */
if(wr) {
@@ -545,11 +548,20 @@
if(dir == NULL)
dir = ".";
- if((fd = open(RECORD, O_RDWR)) < 0) {
+ if((fd = open(RECORD, O_RDWR | O_CREAT, FMASK)) < 0) {
printf("Warning: cannot write %s/%s", dir, RECORD);
getret();
} else
(void) close(fd);
+ if((fd = open(HLOCK, O_RDONLY | O_CREAT, FMASK)) < 0) {
+ printf("Warning: cannot read %s/%s", dir, HLOCK);
+ getret();
+ } else
+ (void) close(fd);
+ if(mkdir("save", 0770) && errno != EEXIST) {
+ printf("Warning: cannot create %s/save", dir);
+ getret();
+ }
}
}
#endif
--- games/hack/Makefile.origMon Jul 10 21:30:37 2017
+++ games/hack/Makefile Mon Mar 5 02:31:18 2018
@@ -27,12 +27,7 @@
${HOSTCC} ${CFLAGS} ${LDFLAGS} ${LDSTATIC} -o ${.TARGET}
${.CURDIR}/${.PREFIX}.c ${LDADD}
beforeinstall:
- ${INSTALL} ${INSTALL_COPY} -o root -g games -m 660 /dev/null \
- ${DESTDIR}/var/games/hackdir/perm
- ${INSTALL} ${INSTALL_COPY} -o root -g games -m 660 /dev/null \
- ${DESTDIR}/var/games/hackdir/record
${INSTALL} ${INSTALL_COPY} -o ${BINOWN} -g ${BINGRP} -m 444
${.CURDIR}/help \
- ${.CURDIR}/hh ${.CURDIR}/data ${DESTDIR}/var/games/hackdir
- rm -f ${DESTDIR}/var/games/hackdir/bones*
+ ${.CURDIR}/hh ${.CURDIR}/data ${.CURDIR}/rumors
${DESTDIR}/usr/share/games/hack
.include
--- games/hack/hack.6.orig Fri Sep 9 15:32:28 2016
+++ games/hack/hack.6 Mon Mar 5 15:41:06 2018
@@ -75,13 +75,6 @@
.Op Fl d Ar directory
.Op Ar playername ...
.Sh DESCRIPTION
-.Bf -symbolic
-.Nm
-is currently unus
Re: SA_RESETHAND race (or the mpv(1) hanging story)
On Fri, Mar 02, 2018 at 02:18:46PM +0100, Hiltjo Posthuma wrote:
> On Wed, Feb 21, 2018 at 02:15:07PM +0100, Martin Pieuchot wrote:
> > Diff below fixes a race in signal handling that can be triggered when
> > a multi-threaded process installs a signal handler with SA_RESETHAND.
> >
> > I described a case of this problem on bugs@ a month ago:
> > https://marc.info/?l=openbsd-bugs&m=151672599508049&w=2
> >
> > The race occurs when one of the threads is doing a NOLOCK syscall, like
> > futex(2):
> >
> > CPU0:
> > Before returning to userland this thread will call CURSIG() without
> > KERNEL_LOCK(), issignal() will look at `ps_sigact', see that there's
> > a handler and return the signal number. After that the thread will
> > call postsig() and spin on the KERNEL_LOCK().
> >
> > CPU1:
> > In the meantime another thread, with the KERNEL_LOCK(), also looked at
> > `ps_sigact' and is already processing the signal. Since the handler
> > has been installed with SA_RESETHAND it will reset the action to
> > SIG_DFL then release the KERNEL_LOCK().
> >
> > CPU0:
> > After grabbing the KERNEL_LOCK() inside postsig(), the thread tries
> > to process an already processed signal, see that the actions is
> > SIG_DFL and call sigexit() while all other threads are SSTOP'ed.
> >
> >
> > Diff below fixes that by serializing changes to `ps_sigact' via the
> > KERNEL_LOCK(). I also introduce postsig_done(), gypped from FreeBSD,
> > to assert that the lock is held and to prepare for finer grained
> > locking.
> >
> > Tests, oks?
> >
> > Index: kern/kern_sig.c
> > ===
> > RCS file: /cvs/src/sys/kern/kern_sig.c,v
> > retrieving revision 1.215
> > diff -u -p -r1.215 kern_sig.c
> > --- kern/kern_sig.c 19 Feb 2018 08:59:52 - 1.215
> > +++ kern/kern_sig.c 21 Feb 2018 13:09:54 -
> > @@ -82,6 +82,7 @@ void proc_stop(struct proc *p, int);
> > void proc_stop_sweep(void *);
> > struct timeout proc_stop_to;
> >
> > +void postsig(struct proc *, int);
> > int cansignal(struct proc *, struct process *, int);
> >
> > struct pool sigacts_pool; /* memory pool for sigacts structures */
> > @@ -747,6 +748,27 @@ pgsignal(struct pgrp *pgrp, int signum,
> > }
> >
> > /*
> > + * Recalculate the signal mask and reset the signal disposition after
> > + * usermode frame for delivery is formed.
> > + */
> > +void
> > +postsig_done(struct proc *p, int signum, struct sigacts *ps)
> > +{
> > + int mask = sigmask(signum);
> > +
> > + KERNEL_ASSERT_LOCKED();
> > +
> > + p->p_ru.ru_nsignals++;
> > + atomic_setbits_int(&p->p_sigmask, ps->ps_catchmask[signum]);
> > + if ((ps->ps_sigreset & mask) != 0) {
> > + ps->ps_sigcatch &= ~mask;
> > + if (signum != SIGCONT && sigprop[signum] & SA_IGNORE)
> > + ps->ps_sigignore |= mask;
> > + ps->ps_sigact[signum] = SIG_DFL;
> > + }
> > +}
> > +
> > +/*
> > * Send a signal caused by a trap to the current thread
> > * If it will be caught immediately, deliver it with correct code.
> > * Otherwise, post it normally.
> > @@ -780,16 +802,9 @@ trapsignal(struct proc *p, int signum, u
> > p->p_sigmask, code, &si);
> > }
> > #endif
> > - p->p_ru.ru_nsignals++;
> > (*pr->ps_emul->e_sendsig)(ps->ps_sigact[signum], signum,
> > p->p_sigmask, trapno, code, sigval);
> > - atomic_setbits_int(&p->p_sigmask, ps->ps_catchmask[signum]);
> > - if ((ps->ps_sigreset & mask) != 0) {
> > - ps->ps_sigcatch &= ~mask;
> > - if (signum != SIGCONT && sigprop[signum] & SA_IGNORE)
> > - ps->ps_sigignore |= mask;
> > - ps->ps_sigact[signum] = SIG_DFL;
> > - }
> > + postsig_done(p, signum, ps);
> > } else {
> > p->p_sisig = signum;
> > p->p_sitrapno = trapno; /* XXX for core dump/debugger */
> > @@ -1330,9 +1345,8 @@ proc_stop_sweep(void *v)
> > * from the current set of pending signals.
> > */
> > void
> > -postsig(int signum)
> > +postsig(struct proc *p, int signum)
> > {
> > - struct proc *p = curproc;
> > struct process *pr = p->p_p;
> > struct sigacts *ps = pr->ps_sigacts;
> > sig_t action;
> > @@ -1341,12 +1355,8 @@ postsig(int signum)
> > union sigval sigval;
> > int s, code;
> >
> > -#ifdef DIAGNOSTIC
> > - if (signum == 0)
> > - panic("postsig");
> > -#endif
> > -
> > - KERNEL_LOCK();
> > + KASSERT(signum != 0);
> > + KERNEL_ASSERT_LOCKED();
> >
> > mask = sigmask(signum);
> > atomic_clearbits_int(&p->p_siglist, mask);
> > @@ -1366,7 +1376,7 @@ postsig(int signum)
> > #ifdef KTRACE
> > if (KTRPOINT(p, KTR_PSIG)) {
> > siginfo_t si;
> > -
> > +
> > initsiginfo(&si, signum, trapno, code, sigval);
> > ktrpsig(p, signum, action, p->p_fla
Re: update ifstated parser
On Mon, Feb 26, 2018 at 05:10:43PM -0600, Michael Graves wrote:
> Hello
>
> I use ifstated(8) to track the state of the the external interface that is
> configured via dhcp and based upon the state, (re)configure a VXLAN
> interface.
> The ifstated.conf currently looks like
>
> ===
> exif="em0"
> vxif="vxlan0"
>
> init-state state_down
>
> state state_up {
> init {
> run "ifconfig vlxna0 up"
> }
> if ( "ifconfig em0 | grep -q inet" every 60 )
> run "sleep 30 && ifconfig vxlan0 tunnel `ifconfig em0 | \
> sed -nre 's/.*inet ([^ ]+).*/\1/p'` \
> `dig +short name-of-remote-device`"
> if $exif.link.down
> set-state state_down
> }
>
> state state_down {
> init {
> run "ifconfig vxlan0 down"
> }
> if $exif.link.up
> set-state state_up
> }
> ===
>
> The problem I ran into is that when I tried to substitute the vxlan0 and em0
> entries with exif and vxif in the 'run' statements no macro expansion
> occurred.
> This patch allows macro expansion within the 'run' and 'if' statements.
>
> I appreciate any feedback.
> Regards
> Index: parse.y
> ===
> RCS file: /cvs/src/usr.sbin/ifstated/parse.y,v
> retrieving revision 1.47
> diff -u -p -r1.47 parse.y
> --- parse.y 21 Aug 2017 17:38:55 - 1.47
> +++ parse.y 26 Feb 2018 22:47:11 -
> @@ -509,9 +509,10 @@ int
> yylex(void)
> {
> u_char buf[8096];
> - u_char *p, *val;
> + u_char *p, *p1, *val;
> int quotec, next, c;
> int token;
> + size_t x;
>
> top:
> p = buf;
> @@ -575,6 +576,35 @@ top:
> } else if (c == '\0') {
> yyerror("syntax error");
> return (findeol());
> + } else if (c == '$') {
> + p1 = p;
> + while (1) {
> + if ((c = lgetc(0)) == EOF)
> + return (0);
> + if (p1 + 1 >= buf + sizeof(buf) - 1) {
> + yyerror("string too long");
> + return (findeol());
> + }
> + if (isalnum(c) || c == '_') {
> + *p1++ = c;
> + continue;
> + }
> + *p1 = '\0';
> + lungetc(c);
> + break;
> + }
> + val = symget(p);
> + if (val == NULL) {
> + yyerror("macro '%s' not defined", buf);
> + return (findeol());
> + }
> + x = strlcpy(p,val,(buf-p));
> + if (x >= (buf-p)) {
> + yyerror("string too long");
> + return (findeol());
> + }
> + p += x;
> + continue;
> }
> if (p + 1 >= buf + sizeof(buf) - 1) {
> yyerror("string too long");
Hey Michael,
Thank you for your email. I have been playing with your diff and will send you
some comments shortly. This might be worth future consideration.
For now, I think we should update the man page to clearly state that macro
expansion does not take place inside quotes as is currently done in the other
man pages.
Ok?
Index: ifstated.conf.5
===
RCS file: /cvs/src/usr.sbin/ifstated/ifstated.conf.5,v
retrieving revision 1.11
diff -u -p -r1.11 ifstated.conf.5
--- ifstated.conf.5 3 Mar 2018 02:57:17 - 1.11
+++ ifstated.conf.5 5 Mar 2018 18:39:06 -
@@ -55,6 +55,8 @@ instead of using the first state defined
Macros can be defined that will later be expanded in context.
Macro names must start with a letter, digit, or underscore,
and may contain any of those characters.
+Macros are not expanded inside quotes.
+.Pp
Macro names may not be reserved words like, for example,
.Ar state
or
mfii(4): add bio(4) support
Hi tech@,
This patch adds bio(4) support for mfii(4).
# with "mfii(4): use MFII_FUNCTION_PASSTHRU_IO for MFI commands"
most parts are taken from mfi(4), plus fix for rebuilding (bioctl -R).
--- sys/dev/ic/mfireg.h
+++ sys/dev/ic/mfireg.h
@@ -139,6 +139,9 @@
#define MR_DCMD_CONF_GET 0x0401
#define MR_DCMD_CFG_ADD0x0402
#define MR_DCMD_CFG_CLEAR 0x0403
+#define MR_DCMD_CFG_MAKE_SPARE 0x0404
+#define MR_DCMD_CFG_FOREIGN_SCAN 0x04060100
+#define MR_DCMD_CFG_FOREIGN_CLEAR 0x04060500
#define MR_DCMD_BBU_GET_STATUS 0x0501
#define MR_DCMD_BBU_GET_CAPACITY_INFO 0x0502
#define MR_DCMD_BBU_GET_DESIGN_INFO0x0503
@@ -1228,3 +1231,13 @@ struct mfi_pr_properties {
uint32_texec_freq;
uint32_tclear_freq;
} __packed;
+
+/* We currently don't know the full details of the following struct */
+struct mfii_foreign_scan_cfg {
+ char data[24];
+};
+
+struct mfii_foreign_scan_info {
+ uint32_t count; /* Number of foreign configs found */
+ struct mfii_foreign_scan_cfg cfgs[8];
+};
--- sys/dev/pci/mfii.c
+++ sys/dev/pci/mfii.c
@@ -22,10 +22,14 @@
#include
#include
#include
+#include
#include
#include
#include
+#include
+#include
+#include
#include
#include
@@ -307,7 +311,35 @@ struct mfii_softc {
struct mfii_pd_softc*sc_pd;
struct scsi_iopool sc_iopool;
+ /* save some useful information for logical drives that is missing
+* in sc_ld_list
+*/
+ struct {
+ uint32_tld_present;
+ charld_dev[16]; /* device name sd? */
+ } sc_ld[MFI_MAX_LD];
+
+ /* scsi ioctl from sd device */
+ int (*sc_ioctl)(struct device *, u_long, caddr_t);
+
+ uint32_tsc_ld_cnt;
+
+ /* bio */
+ struct mfi_conf *sc_cfg;
struct mfi_ctrl_infosc_info;
+ struct mfi_ld_list sc_ld_list;
+ struct mfi_ld_details *sc_ld_details; /* array to all logical disks */
+ int sc_no_pd; /* used physical disks */
+ int sc_ld_sz; /* sizeof sc_ld_details */
+
+ /* mgmt lock */
+ struct rwlock sc_lock;
+
+ /* sensors */
+ struct ksensordev sc_sensordev;
+ struct ksensor *sc_bbu;
+ struct ksensor *sc_bbu_status;
+ struct ksensor *sc_sensors;
};
#ifdef MFII_DEBUG
@@ -355,13 +387,15 @@ struct cfdriver mfii_cd = {
void mfii_scsi_cmd(struct scsi_xfer *);
void mfii_scsi_cmd_done(struct mfii_softc *, struct mfii_ccb *);
+intmfii_scsi_ioctl(struct scsi_link *, u_long, caddr_t, int);
+intmfii_ioctl_cache(struct scsi_link *, u_long, struct dk_cache *);
struct scsi_adapter mfii_switch = {
mfii_scsi_cmd,
scsi_minphys,
NULL, /* probe */
NULL, /* unprobe */
- NULL /* ioctl */
+ mfii_scsi_ioctl
};
void mfii_pd_scsi_cmd(struct scsi_xfer *);
@@ -409,9 +443,11 @@ intmfii_load_mfa(struct mfii_softc
*, struct mfii_ccb *,
intmfii_mfa_poll(struct mfii_softc *, struct mfii_ccb *);
-intmfii_mgmt(struct mfii_softc *, struct mfii_ccb *,
- u_int32_t, const union mfi_mbox *,
- void *, size_t, int);
+intmfii_mgmt(struct mfii_softc *, uint32_t, uint32_t,
+ size_t, void *, const union mfi_mbox *);
+intmfii_do_mgmt(struct mfii_softc *, struct mfii_ccb *,
+ uint32_t, uint32_t, size_t, void *,
+ const union mfi_mbox *);
void mfii_empty_done(struct mfii_softc *, struct mfii_ccb *);
intmfii_scsi_cmd_io(struct mfii_softc *,
@@ -445,6 +481,42 @@ void mfii_aen_pd_remove(struct
mfii_softc *,
void mfii_aen_pd_state_change(struct mfii_softc *,
const struct mfi_evtarg_pd_state *);
+#if NBIO > 0
+intmfii_ioctl(struct device *, u_long, caddr_t);
+intmfii_bio_getitall(struct mfii_softc *);
+intmfii_ioctl_inq(struct mfii_softc *, struct bioc_inq *);
+intmfii_ioctl_vol(struct mfii_softc *, struct bioc_vol *);
+intmfii_ioctl_disk(struct mfii_softc *, struct bioc_disk *);
+intmfii_ioctl_alarm(struct mfii_softc *, struct bioc_alarm *);
+intmfii_ioctl_blink(struct mfii_softc *sc, struct bioc_blink *);
+intmfii_ioctl_setstate(struct mfii_softc *,
+ struct bioc_setstate *);
+intmfii_ioctl_patrol(struct mfii_soft
mfii(4): use MFII_FUNCTION_PASSTHRU_IO for MFI commands
Hi tech@,
while trying to add bio(4) support to mfii(4), we noticed that SAS2208
based card with cache memory hangs on boot.
https://marc.info/?t=14773842861&r=1&w=2
after investigation, we made conclusion that MPII SCSI PASSTHRU feature
need to be used for sending MFI commands (such as DCMD_*) instead of
using MFA request.
this patch shouldn't change any functionality/behavior, but this is
required for upcoming bio(4) support.
tested on: SAS2208, SAS3008, and SAS3108
--- sys/dev/pci/mfii.c
+++ sys/dev/pci/mfii.c
@@ -44,6 +44,8 @@
#define MFII_OSTS_INTR_VALID 0x0009
#define MFII_RPI 0x6c /* reply post host index */
+#define MFII_OSP2 0xb4 /* outbound scratch pad 2 */
+#define MFII_OSP3 0xb8 /* outbound scratch pad 3 */
#define MFII_REQ_TYPE_SCSI MPII_REQ_DESCR_SCSI_IO
#define MFII_REQ_TYPE_LDIO (0x7 << 1)
@@ -51,20 +53,42 @@
#define MFII_REQ_TYPE_NO_LOCK (0x2 << 1)
#define MFII_REQ_TYPE_HI_PRI (0x6 << 1)
-#define MFII_REQ_MFA(_a) htole64((_a) | MFII_REQ_TYPE_MFA)
-
#define MFII_FUNCTION_PASSTHRU_IO (0xf0)
#define MFII_FUNCTION_LDIO_REQUEST (0xf1)
-struct mfii_request_descr {
+#define MFII_MAX_CHAIN_UNIT0x0040
+#define MFII_MAX_CHAIN_MASK0x03E0
+#define MFII_MAX_CHAIN_SHIFT 5
+
+#define MFII_256K_IO 128
+#define MFII_1MB_IO(MFII_256K_IO * 4)
+
+#define MFII_CHAIN_FRAME_MIN 1024
+
+struct mfii_scsi_io_request_descr {
u_int8_tflags;
u_int8_tmsix_index;
u_int16_t smid;
-
u_int16_t lmid;
u_int16_t dev_handle;
} __packed;
+struct mfii_mfa_io_request_descr {
+ u_int32_t flags:8;
+ u_int32_t addr1:24;
+ u_int32_t addr2;
+} __packed;
+
+union mfii_request_descr {
+ struct mfii_scsi_io_request_descr scsi;
+ struct mfii_mfa_io_request_descrmfa;
+ struct {
+ u_int32_t lo;
+ u_int32_t hi;
+ } __packed u;
+ u_int64_t word;
+};
+
#define MFII_RAID_CTX_IO_TYPE_SYSPD(0x1 << 4)
#define MFII_RAID_CTX_TYPE_CUDA(0x2 << 4)
@@ -180,6 +204,10 @@ struct mfii_ccb {
u_int64_t ccb_request_dva;
bus_addr_t ccb_request_offset;
+ void*ccb_mfi;
+ u_int64_t ccb_mfi_dva;
+ bus_addr_t ccb_mfi_offset;
+
struct mfi_sense*ccb_sense;
u_int64_t ccb_sense_dva;
bus_addr_t ccb_sense_offset;
@@ -189,7 +217,7 @@ struct mfii_ccb {
bus_addr_t ccb_sgl_offset;
u_int ccb_sgl_len;
- struct mfii_request_descr ccb_req;
+ union mfii_request_descr ccb_req;
bus_dmamap_tccb_dmamap;
@@ -250,6 +278,7 @@ struct mfii_softc {
struct mutexsc_ccb_mtx;
struct mutexsc_post_mtx;
+ u_int sc_max_fw_cmds;
u_int sc_max_cmds;
u_int sc_max_sgl;
@@ -259,6 +288,7 @@ struct mfii_softc {
struct mfii_dmamem *sc_reply_postq;
struct mfii_dmamem *sc_requests;
+ struct mfii_dmamem *sc_mfi;
struct mfii_dmamem *sc_sense;
struct mfii_dmamem *sc_sgl;
@@ -280,6 +310,32 @@ struct mfii_softc {
struct mfi_ctrl_infosc_info;
};
+#ifdef MFII_DEBUG
+#define DPRINTF(x...) do { if (mfii_debug) printf(x); } while(0)
+#define DNPRINTF(n,x...) do { if (mfii_debug & n) printf(x); } while(0)
+#defineMFII_D_CMD 0x0001
+#defineMFII_D_INTR 0x0002
+#defineMFII_D_MISC 0x0004
+#defineMFII_D_DMA 0x0008
+#defineMFII_D_IOCTL0x0010
+#defineMFII_D_RW 0x0020
+#defineMFII_D_MEM 0x0040
+#defineMFII_D_CCB 0x0080
+uint32_t mfii_debug = 0
+/* | MFII_D_CMD */
+/* | MFII_D_INTR */
+/* | MFII_D_MISC */
+/* | MFII_D_DMA */
+/* | MFII_D_IOCTL */
+/* | MFII_D_RW */
+/* | MFII_D_MEM */
+/* | MFII_D_CCB */
+ ;
+#else
+#define DPRINTF(x...)
+#define DNPRINTF(n,x...)
+#endif
+
intmfii_match(struct device *, void *, void *);
void mfii_attach(struct device *, struct device *, void *);
intmfii_detach(struct device *, int);
@@ -356,6 +412,7 @@ int mfii_mfa_poll(struct mfii_softc *,
struct mfii_ccb *);
intmfii_mgmt(struct mfii_softc *, struct mfii_ccb *,
u_int32_t, const union mfi_mbox *,
void
