From 4d48a542b42747c36a5937447d9c3de7c897ea50 Mon Sep 17 00:00:00 2001 From: Paul Clements Date: Wed, 11 Feb 2009 13:04:45 -0800 Subject: nbd: fix I/O hang on disconnected nbds Fix a problem that causes I/O to a disconnected (or partially initialized) nbd device to hang indefinitely. To reproduce: # ioctl NBD_SET_SIZE_BLOCKS /dev/nbd23 514048 # dd if=/dev/nbd23 of=/dev/null bs=4096 count=1 ...hangs... This can also occur when an nbd device loses its nbd-client/server connection. Although we clear the queue of any outstanding I/Os after the client/server connection fails, any additional I/Os that get queued later will hang. This bug may also be the problem reported in this bug report: http://bugzilla.kernel.org/show_bug.cgi?id=12277 Testing would need to be performed to determine if the two issues are the same. This problem was introduced by the new request handling thread code ("NBD: allow nbd to be used locally", 3/2008), which entered into mainline around 2.6.25. The fix, which is fairly simple, is to restore the check for lo->sock being NULL in do_nbd_request. This causes I/O to an uninitialized nbd to immediately fail with an I/O error, as it did prior to the introduction of this bug. Signed-off-by: Paul Clements Reported-by: Jon Nelson Acked-by: Pavel Machek Cc: [2.6.26.x, 2.6.27.x, 2.6.28.x] Signed-off-by: Andrew Morton Signed-off-by: Linus Torvalds --- drivers/block/nbd.c | 9 +++++++++ 1 file changed, 9 insertions(+) (limited to 'drivers/block') diff --git a/drivers/block/nbd.c b/drivers/block/nbd.c index 34f80fa6fed1..8299e2d3b611 100644 --- a/drivers/block/nbd.c +++ b/drivers/block/nbd.c @@ -549,6 +549,15 @@ static void do_nbd_request(struct request_queue * q) BUG_ON(lo->magic != LO_MAGIC); + if (unlikely(!lo->sock)) { + printk(KERN_ERR "%s: Attempted send on closed socket\n", + lo->disk->disk_name); + req->errors++; + nbd_end_request(req); + spin_lock_irq(q->queue_lock); + continue; + } + spin_lock_irq(&lo->queue_lock); list_add_tail(&req->queuelist, &lo->waiting_queue); spin_unlock_irq(&lo->queue_lock); -- cgit v1.2.2 From c8cbec6bdf6329279fd14696020f6b59d1d3124d Mon Sep 17 00:00:00 2001 From: Roel Kluin Date: Mon, 16 Feb 2009 13:11:55 +0100 Subject: paride/pg.c: xs(): &&/|| confusion &&/|| confusion Signed-off-by: Roel Kluin Cc: Bartlomiej Zolnierkiewicz Signed-off-by: Andrew Morton Signed-off-by: Jens Axboe --- drivers/block/paride/pg.c | 2 +- 1 file changed, 1 insertion(+), 1 deletion(-) (limited to 'drivers/block') diff --git a/drivers/block/paride/pg.c b/drivers/block/paride/pg.c index 9dfa27163001..c397b3ddba9b 100644 --- a/drivers/block/paride/pg.c +++ b/drivers/block/paride/pg.c @@ -422,7 +422,7 @@ static void xs(char *buf, char *targ, int len) for (k = 0; k < len; k++) { char c = *buf++; - if (c != ' ' || c != l) + if (c != ' ' && c != l) l = *targ++ = c; } if (l == ' ') -- cgit v1.2.2 From 82eb03cfd862a65363fa2826de0dbd5474cfe5e2 Mon Sep 17 00:00:00 2001 From: Chip Coldwell Date: Mon, 16 Feb 2009 13:11:56 +0100 Subject: cciss: PCI power management reset for kexec The kexec kernel resets the CCISS hardware in three steps: 1. Use PCI power management states to reset the controller in the kexec kernel. 2. Clear the MSI/MSI-X bits in PCI configuration space so that MSI initialization in the kexec kernel doesn't fail. 3. Use the CCISS "No-op" message to determine when the controller firmware has recovered from the PCI PM reset. [akpm@linux-foundation.org: cleanups] Signed-off-by: Mike Miller Cc: Randy Dunlap Signed-off-by: Andrew Morton Signed-off-by: Jens Axboe --- drivers/block/cciss.c | 215 ++++++++++++++++++++++++++++++++++++++++++++++++++ 1 file changed, 215 insertions(+) (limited to 'drivers/block') diff --git a/drivers/block/cciss.c b/drivers/block/cciss.c index 01e69383d9c0..d2cb67b61176 100644 --- a/drivers/block/cciss.c +++ b/drivers/block/cciss.c @@ -3390,6 +3390,203 @@ static void free_hba(int i) kfree(p); } +/* Send a message CDB to the firmware. */ +static __devinit int cciss_message(struct pci_dev *pdev, unsigned char opcode, unsigned char type) +{ + typedef struct { + CommandListHeader_struct CommandHeader; + RequestBlock_struct Request; + ErrDescriptor_struct ErrorDescriptor; + } Command; + static const size_t cmd_sz = sizeof(Command) + sizeof(ErrorInfo_struct); + Command *cmd; + dma_addr_t paddr64; + uint32_t paddr32, tag; + void __iomem *vaddr; + int i, err; + + vaddr = ioremap_nocache(pci_resource_start(pdev, 0), pci_resource_len(pdev, 0)); + if (vaddr == NULL) + return -ENOMEM; + + /* The Inbound Post Queue only accepts 32-bit physical addresses for the + CCISS commands, so they must be allocated from the lower 4GiB of + memory. */ + err = pci_set_consistent_dma_mask(pdev, DMA_32BIT_MASK); + if (err) { + iounmap(vaddr); + return -ENOMEM; + } + + cmd = pci_alloc_consistent(pdev, cmd_sz, &paddr64); + if (cmd == NULL) { + iounmap(vaddr); + return -ENOMEM; + } + + /* This must fit, because of the 32-bit consistent DMA mask. Also, + although there's no guarantee, we assume that the address is at + least 4-byte aligned (most likely, it's page-aligned). */ + paddr32 = paddr64; + + cmd->CommandHeader.ReplyQueue = 0; + cmd->CommandHeader.SGList = 0; + cmd->CommandHeader.SGTotal = 0; + cmd->CommandHeader.Tag.lower = paddr32; + cmd->CommandHeader.Tag.upper = 0; + memset(&cmd->CommandHeader.LUN.LunAddrBytes, 0, 8); + + cmd->Request.CDBLen = 16; + cmd->Request.Type.Type = TYPE_MSG; + cmd->Request.Type.Attribute = ATTR_HEADOFQUEUE; + cmd->Request.Type.Direction = XFER_NONE; + cmd->Request.Timeout = 0; /* Don't time out */ + cmd->Request.CDB[0] = opcode; + cmd->Request.CDB[1] = type; + memset(&cmd->Request.CDB[2], 0, 14); /* the rest of the CDB is reserved */ + + cmd->ErrorDescriptor.Addr.lower = paddr32 + sizeof(Command); + cmd->ErrorDescriptor.Addr.upper = 0; + cmd->ErrorDescriptor.Len = sizeof(ErrorInfo_struct); + + writel(paddr32, vaddr + SA5_REQUEST_PORT_OFFSET); + + for (i = 0; i < 10; i++) { + tag = readl(vaddr + SA5_REPLY_PORT_OFFSET); + if ((tag & ~3) == paddr32) + break; + schedule_timeout_uninterruptible(HZ); + } + + iounmap(vaddr); + + /* we leak the DMA buffer here ... no choice since the controller could + still complete the command. */ + if (i == 10) { + printk(KERN_ERR "cciss: controller message %02x:%02x timed out\n", + opcode, type); + return -ETIMEDOUT; + } + + pci_free_consistent(pdev, cmd_sz, cmd, paddr64); + + if (tag & 2) { + printk(KERN_ERR "cciss: controller message %02x:%02x failed\n", + opcode, type); + return -EIO; + } + + printk(KERN_INFO "cciss: controller message %02x:%02x succeeded\n", + opcode, type); + return 0; +} + +#define cciss_soft_reset_controller(p) cciss_message(p, 1, 0) +#define cciss_noop(p) cciss_message(p, 3, 0) + +static __devinit int cciss_reset_msi(struct pci_dev *pdev) +{ +/* the #defines are stolen from drivers/pci/msi.h. */ +#define msi_control_reg(base) (base + PCI_MSI_FLAGS) +#define PCI_MSIX_FLAGS_ENABLE (1 << 15) + + int pos; + u16 control = 0; + + pos = pci_find_capability(pdev, PCI_CAP_ID_MSI); + if (pos) { + pci_read_config_word(pdev, msi_control_reg(pos), &control); + if (control & PCI_MSI_FLAGS_ENABLE) { + printk(KERN_INFO "cciss: resetting MSI\n"); + pci_write_config_word(pdev, msi_control_reg(pos), control & ~PCI_MSI_FLAGS_ENABLE); + } + } + + pos = pci_find_capability(pdev, PCI_CAP_ID_MSIX); + if (pos) { + pci_read_config_word(pdev, msi_control_reg(pos), &control); + if (control & PCI_MSIX_FLAGS_ENABLE) { + printk(KERN_INFO "cciss: resetting MSI-X\n"); + pci_write_config_word(pdev, msi_control_reg(pos), control & ~PCI_MSIX_FLAGS_ENABLE); + } + } + + return 0; +} + +/* This does a hard reset of the controller using PCI power management + * states. */ +static __devinit int cciss_hard_reset_controller(struct pci_dev *pdev) +{ + u16 pmcsr, saved_config_space[32]; + int i, pos; + + printk(KERN_INFO "cciss: using PCI PM to reset controller\n"); + + /* This is very nearly the same thing as + + pci_save_state(pci_dev); + pci_set_power_state(pci_dev, PCI_D3hot); + pci_set_power_state(pci_dev, PCI_D0); + pci_restore_state(pci_dev); + + but we can't use these nice canned kernel routines on + kexec, because they also check the MSI/MSI-X state in PCI + configuration space and do the wrong thing when it is + set/cleared. Also, the pci_save/restore_state functions + violate the ordering requirements for restoring the + configuration space from the CCISS document (see the + comment below). So we roll our own .... */ + + for (i = 0; i < 32; i++) + pci_read_config_word(pdev, 2*i, &saved_config_space[i]); + + pos = pci_find_capability(pdev, PCI_CAP_ID_PM); + if (pos == 0) { + printk(KERN_ERR "cciss_reset_controller: PCI PM not supported\n"); + return -ENODEV; + } + + /* Quoting from the Open CISS Specification: "The Power + * Management Control/Status Register (CSR) controls the power + * state of the device. The normal operating state is D0, + * CSR=00h. The software off state is D3, CSR=03h. To reset + * the controller, place the interface device in D3 then to + * D0, this causes a secondary PCI reset which will reset the + * controller." */ + + /* enter the D3hot power management state */ + pci_read_config_word(pdev, pos + PCI_PM_CTRL, &pmcsr); + pmcsr &= ~PCI_PM_CTRL_STATE_MASK; + pmcsr |= PCI_D3hot; + pci_write_config_word(pdev, pos + PCI_PM_CTRL, pmcsr); + + schedule_timeout_uninterruptible(HZ >> 1); + + /* enter the D0 power management state */ + pmcsr &= ~PCI_PM_CTRL_STATE_MASK; + pmcsr |= PCI_D0; + pci_write_config_word(pdev, pos + PCI_PM_CTRL, pmcsr); + + schedule_timeout_uninterruptible(HZ >> 1); + + /* Restore the PCI configuration space. The Open CISS + * Specification says, "Restore the PCI Configuration + * Registers, offsets 00h through 60h. It is important to + * restore the command register, 16-bits at offset 04h, + * last. Do not restore the configuration status register, + * 16-bits at offset 06h." Note that the offset is 2*i. */ + for (i = 0; i < 32; i++) { + if (i == 2 || i == 3) + continue; + pci_write_config_word(pdev, 2*i, saved_config_space[i]); + } + wmb(); + pci_write_config_word(pdev, 4, saved_config_space[2]); + + return 0; +} + /* * This is it. Find all the controllers and register them. I really hate * stealing all these major device numbers. @@ -3404,6 +3601,24 @@ static int __devinit cciss_init_one(struct pci_dev *pdev, int dac, return_code; InquiryData_struct *inq_buff = NULL; + if (reset_devices) { + /* Reset the controller with a PCI power-cycle */ + if (cciss_hard_reset_controller(pdev) || cciss_reset_msi(pdev)) + return -ENODEV; + + /* Some devices (notably the HP Smart Array 5i Controller) + need a little pause here */ + schedule_timeout_uninterruptible(30*HZ); + + /* Now try to get the controller to respond to a no-op */ + for (i=0; i<12; i++) { + if (cciss_noop(pdev) == 0) + break; + else + printk("cciss: no-op failed%s\n", (i < 11 ? "; re-trying" : "")); + } + } + i = alloc_cciss_hba(); if (i < 0) return -1; -- cgit v1.2.2 From b6d6c5175809934e04a606d9193ef04924a7a7d9 Mon Sep 17 00:00:00 2001 From: Ed Cashin Date: Wed, 18 Feb 2009 14:48:13 -0800 Subject: aoe: ignore vendor extension AoE responses The Welland ME-747K-SI AoE target generates unsolicited AoE responses that are marked as vendor extensions. Instead of ignoring these packets, the aoe driver was generating kernel messages for each unrecognized response received. This patch corrects the behavior. Signed-off-by: Ed Cashin Reported-by: Tested-by: Cc: Cc: Alex Buell Signed-off-by: Andrew Morton Signed-off-by: Linus Torvalds --- drivers/block/aoe/aoe.h | 1 + drivers/block/aoe/aoenet.c | 2 ++ 2 files changed, 3 insertions(+) (limited to 'drivers/block') diff --git a/drivers/block/aoe/aoe.h b/drivers/block/aoe/aoe.h index c237527b1aa5..5e41e6dd657b 100644 --- a/drivers/block/aoe/aoe.h +++ b/drivers/block/aoe/aoe.h @@ -18,6 +18,7 @@ enum { AOECMD_ATA, AOECMD_CFG, + AOECMD_VEND_MIN = 0xf0, AOEFL_RSP = (1<<3), AOEFL_ERR = (1<<2), diff --git a/drivers/block/aoe/aoenet.c b/drivers/block/aoe/aoenet.c index 30de5b1c647e..c6099ba9a4b8 100644 --- a/drivers/block/aoe/aoenet.c +++ b/drivers/block/aoe/aoenet.c @@ -142,6 +142,8 @@ aoenet_rcv(struct sk_buff *skb, struct net_device *ifp, struct packet_type *pt, aoecmd_cfg_rsp(skb); break; default: + if (h->cmd >= AOECMD_VEND_MIN) + break; /* don't complain about vendor commands */ printk(KERN_INFO "aoe: unknown cmd %d\n", h->cmd); } exit: -- cgit v1.2.2 From 5a74db06cc8d36a325913aa4968ae169f997a466 Mon Sep 17 00:00:00 2001 From: Philippe De Muyter Date: Wed, 18 Feb 2009 14:48:36 -0800 Subject: floppy: request and release only the ports we actually use The floppy driver requests an I/O port it doesn't need, and sometimes this causes a conflict with a motherboard device reported by PNPBIOS. This patch makes the floppy driver request and release only the ports it actually uses. It also factors out the request/release stuff and the io-ports list so they're all in one place now. The current floppy driver uses only these ports: 0x3f2 (FD_DOR) 0x3f4 (FD_STATUS) 0x3f5 (FD_DATA) 0x3f7 (FD_DCR/FD_DIR) but it requests 0x3f2-0x3f5 and 0x3f7, which includes the unused port 0x3f3. Some BIOSes report 0x3f3 as a motherboard resource. The PNP system driver reserves that, which causes a conflict when the floppy driver requests 0x3f2-0x3f5 later. Philippe reported that this conflict broke the floppy driver between 2.6.11 and 2.6.22. His PNPBIOS reports these devices: $ cat 00:07/id 00:07/resources # motherboard device PNP0c02 state = active io 0x80-0x80 io 0x10-0x1f io 0x22-0x3f io 0x44-0x5f io 0x90-0x9f io 0xa2-0xbf io 0x3f0-0x3f1 io 0x3f3-0x3f3 $ cat 00:03/id 00:03/resources # floppy device PNP0700 state = active io 0x3f4-0x3f5 io 0x3f2-0x3f2 Reference: http://lkml.org/lkml/2009/1/31/162 Signed-off-by: Bjorn Helgaas Signed-off-by: Philippe De Muyter Reported-by: Philippe De Muyter Tested-by: Philippe De Muyter Cc: Adam M Belay Cc: Robert Hancock Signed-off-by: Andrew Morton Signed-off-by: Linus Torvalds --- drivers/block/floppy.c | 79 +++++++++++++++++++++++++++++++++----------------- 1 file changed, 52 insertions(+), 27 deletions(-) (limited to 'drivers/block') diff --git a/drivers/block/floppy.c b/drivers/block/floppy.c index cf29cc4e6ab7..83d8ed39433d 100644 --- a/drivers/block/floppy.c +++ b/drivers/block/floppy.c @@ -558,6 +558,8 @@ static void process_fd_request(void); static void recalibrate_floppy(void); static void floppy_shutdown(unsigned long); +static int floppy_request_regions(int); +static void floppy_release_regions(int); static int floppy_grab_irq_and_dma(void); static void floppy_release_irq_and_dma(void); @@ -4274,8 +4276,7 @@ static int __init floppy_init(void) FDCS->rawcmd = 2; if (user_reset_fdc(-1, FD_RESET_ALWAYS, 0)) { /* free ioports reserved by floppy_grab_irq_and_dma() */ - release_region(FDCS->address + 2, 4); - release_region(FDCS->address + 7, 1); + floppy_release_regions(fdc); FDCS->address = -1; FDCS->version = FDC_NONE; continue; @@ -4284,8 +4285,7 @@ static int __init floppy_init(void) FDCS->version = get_fdc_version(); if (FDCS->version == FDC_NONE) { /* free ioports reserved by floppy_grab_irq_and_dma() */ - release_region(FDCS->address + 2, 4); - release_region(FDCS->address + 7, 1); + floppy_release_regions(fdc); FDCS->address = -1; continue; } @@ -4358,6 +4358,47 @@ out_put_disk: static DEFINE_SPINLOCK(floppy_usage_lock); +static const struct io_region { + int offset; + int size; +} io_regions[] = { + { 2, 1 }, + /* address + 3 is sometimes reserved by pnp bios for motherboard */ + { 4, 2 }, + /* address + 6 is reserved, and may be taken by IDE. + * Unfortunately, Adaptec doesn't know this :-(, */ + { 7, 1 }, +}; + +static void floppy_release_allocated_regions(int fdc, const struct io_region *p) +{ + while (p != io_regions) { + p--; + release_region(FDCS->address + p->offset, p->size); + } +} + +#define ARRAY_END(X) (&((X)[ARRAY_SIZE(X)])) + +static int floppy_request_regions(int fdc) +{ + const struct io_region *p; + + for (p = io_regions; p < ARRAY_END(io_regions); p++) { + if (!request_region(FDCS->address + p->offset, p->size, "floppy")) { + DPRINT("Floppy io-port 0x%04lx in use\n", FDCS->address + p->offset); + floppy_release_allocated_regions(fdc, p); + return -EBUSY; + } + } + return 0; +} + +static void floppy_release_regions(int fdc) +{ + floppy_release_allocated_regions(fdc, ARRAY_END(io_regions)); +} + static int floppy_grab_irq_and_dma(void) { unsigned long flags; @@ -4399,18 +4440,8 @@ static int floppy_grab_irq_and_dma(void) for (fdc = 0; fdc < N_FDC; fdc++) { if (FDCS->address != -1) { - if (!request_region(FDCS->address + 2, 4, "floppy")) { - DPRINT("Floppy io-port 0x%04lx in use\n", - FDCS->address + 2); - goto cleanup1; - } - if (!request_region(FDCS->address + 7, 1, "floppy DIR")) { - DPRINT("Floppy io-port 0x%04lx in use\n", - FDCS->address + 7); - goto cleanup2; - } - /* address + 6 is reserved, and may be taken by IDE. - * Unfortunately, Adaptec doesn't know this :-(, */ + if (floppy_request_regions(fdc)) + goto cleanup; } } for (fdc = 0; fdc < N_FDC; fdc++) { @@ -4432,15 +4463,11 @@ static int floppy_grab_irq_and_dma(void) fdc = 0; irqdma_allocated = 1; return 0; -cleanup2: - release_region(FDCS->address + 2, 4); -cleanup1: +cleanup: fd_free_irq(); fd_free_dma(); - while (--fdc >= 0) { - release_region(FDCS->address + 2, 4); - release_region(FDCS->address + 7, 1); - } + while (--fdc >= 0) + floppy_release_regions(fdc); spin_lock_irqsave(&floppy_usage_lock, flags); usage_count--; spin_unlock_irqrestore(&floppy_usage_lock, flags); @@ -4501,10 +4528,8 @@ static void floppy_release_irq_and_dma(void) #endif old_fdc = fdc; for (fdc = 0; fdc < N_FDC; fdc++) - if (FDCS->address != -1) { - release_region(FDCS->address + 2, 4); - release_region(FDCS->address + 7, 1); - } + if (FDCS->address != -1) + floppy_release_regions(fdc); fdc = old_fdc; } -- cgit v1.2.2 From 3d92e8f3ae9ba21cac30370eb254ed9dc20df043 Mon Sep 17 00:00:00 2001 From: Geert Uytterhoeven Date: Sun, 22 Feb 2009 09:38:47 +0100 Subject: m68k: atari - Rename "mfp" to "st_mfp" http://kisskb.ellerman.id.au/kisskb/buildresult/72115/: | net/mac80211/ieee80211_i.h:327: error: syntax error before 'volatile' | net/mac80211/ieee80211_i.h:350: error: syntax error before '}' token | net/mac80211/ieee80211_i.h:455: error: field 'sta' has incomplete type | distcc[19430] ERROR: compile net/mac80211/main.c on sprygo/32 failed This is caused by | # define mfp ((*(volatile struct MFP*)MFP_BAS)) in arch/m68k/include/asm/atarihw.h, which conflicts with the new "mfp" enum in net/mac80211/ieee80211_i.h. Rename "mfp" to "st_mfp", as it's a way too generic name for a global #define. Signed-off-by: Geert Uytterhoeven Signed-off-by: Linus Torvalds --- drivers/block/ataflop.c | 4 ++-- 1 file changed, 2 insertions(+), 2 deletions(-) (limited to 'drivers/block') diff --git a/drivers/block/ataflop.c b/drivers/block/ataflop.c index 69e1df7dfa14..4234c11c1e4c 100644 --- a/drivers/block/ataflop.c +++ b/drivers/block/ataflop.c @@ -1730,7 +1730,7 @@ static int __init fd_test_drive_present( int drive ) timeout = jiffies + 2*HZ+HZ/2; while (time_before(jiffies, timeout)) - if (!(mfp.par_dt_reg & 0x20)) + if (!(st_mfp.par_dt_reg & 0x20)) break; status = FDC_READ( FDCREG_STATUS ); @@ -1747,7 +1747,7 @@ static int __init fd_test_drive_present( int drive ) /* dummy seek command to make WP bit accessible */ FDC_WRITE( FDCREG_DATA, 0 ); FDC_WRITE( FDCREG_CMD, FDCCMD_SEEK ); - while( mfp.par_dt_reg & 0x20 ) + while( st_mfp.par_dt_reg & 0x20 ) ; status = FDC_READ( FDCREG_STATUS ); } -- cgit v1.2.2 From 5e4c91c84b194b26cf592779e451f4b5be777cba Mon Sep 17 00:00:00 2001 From: Jens Axboe Date: Mon, 23 Feb 2009 08:53:35 +0100 Subject: cciss: shorten 30s timeout on controller reset If reset_devices is set for kexec, then cciss will delay 30 seconds since the old 5i controller _may_ need that long to recover. Replace the long sleep with incremental sleep and tests to reduce the 30 seconds to worst case for 5i, so that other controllers will proceed quickly. Reviewed-by: Mike Miller Signed-off-by: Jens Axboe --- drivers/block/cciss.c | 10 +++++++--- 1 file changed, 7 insertions(+), 3 deletions(-) (limited to 'drivers/block') diff --git a/drivers/block/cciss.c b/drivers/block/cciss.c index d2cb67b61176..b5a061114630 100644 --- a/drivers/block/cciss.c +++ b/drivers/block/cciss.c @@ -3611,11 +3611,15 @@ static int __devinit cciss_init_one(struct pci_dev *pdev, schedule_timeout_uninterruptible(30*HZ); /* Now try to get the controller to respond to a no-op */ - for (i=0; i<12; i++) { + for (i=0; i<30; i++) { if (cciss_noop(pdev) == 0) break; - else - printk("cciss: no-op failed%s\n", (i < 11 ? "; re-trying" : "")); + + schedule_timeout_uninterruptible(HZ); + } + if (i == 30) { + printk(KERN_ERR "cciss: controller seems dead\n"); + return -EBUSY; } } -- cgit v1.2.2 From 9e973e64ac6dc504e6447d52193d4fff1a670156 Mon Sep 17 00:00:00 2001 From: Jens Axboe Date: Tue, 24 Feb 2009 08:10:09 +0100 Subject: xen/blkfront: use blk_rq_map_sg to generate ring entries On occasion, the request will apparently have more segments than we fit into the ring. Jens says: > The second problem is that the block layer then appears to create one > too many segments, but from the dump it has rq->nr_phys_segments == > BLKIF_MAX_SEGMENTS_PER_REQUEST. I suspect the latter is due to > xen-blkfront not handling the merging on its own. It should check that > the new page doesn't form part of the previous page. The > rq_for_each_segment() iterates all single bits in the request, not dma > segments. The "easiest" way to do this is to call blk_rq_map_sg() and > then iterate the mapped sg list. That will give you what you are > looking for. > Here's a test patch, compiles but otherwise untested. I spent more > time figuring out how to enable XEN than to code it up, so YMMV! > Probably the sg list wants to be put inside the ring and only > initialized on allocation, then you can get rid of the sg on stack and > sg_init_table() loop call in the function. I'll leave that, and the > testing, to you. [Moved sg array into info structure, and initialize once. -J] Signed-off-by: Jens Axboe Signed-off-by: Jeremy Fitzhardinge --- drivers/block/xen-blkfront.c | 30 +++++++++++++++--------------- 1 file changed, 15 insertions(+), 15 deletions(-) (limited to 'drivers/block') diff --git a/drivers/block/xen-blkfront.c b/drivers/block/xen-blkfront.c index 918ef725de41..b6c8ce254359 100644 --- a/drivers/block/xen-blkfront.c +++ b/drivers/block/xen-blkfront.c @@ -40,6 +40,7 @@ #include #include #include +#include #include #include @@ -82,6 +83,7 @@ struct blkfront_info enum blkif_state connected; int ring_ref; struct blkif_front_ring ring; + struct scatterlist sg[BLKIF_MAX_SEGMENTS_PER_REQUEST]; unsigned int evtchn, irq; struct request_queue *rq; struct work_struct work; @@ -204,12 +206,11 @@ static int blkif_queue_request(struct request *req) struct blkfront_info *info = req->rq_disk->private_data; unsigned long buffer_mfn; struct blkif_request *ring_req; - struct req_iterator iter; - struct bio_vec *bvec; unsigned long id; unsigned int fsect, lsect; - int ref; + int i, ref; grant_ref_t gref_head; + struct scatterlist *sg; if (unlikely(info->connected != BLKIF_STATE_CONNECTED)) return 1; @@ -238,12 +239,13 @@ static int blkif_queue_request(struct request *req) if (blk_barrier_rq(req)) ring_req->operation = BLKIF_OP_WRITE_BARRIER; - ring_req->nr_segments = 0; - rq_for_each_segment(bvec, req, iter) { - BUG_ON(ring_req->nr_segments == BLKIF_MAX_SEGMENTS_PER_REQUEST); - buffer_mfn = pfn_to_mfn(page_to_pfn(bvec->bv_page)); - fsect = bvec->bv_offset >> 9; - lsect = fsect + (bvec->bv_len >> 9) - 1; + ring_req->nr_segments = blk_rq_map_sg(req->q, req, info->sg); + BUG_ON(ring_req->nr_segments > BLKIF_MAX_SEGMENTS_PER_REQUEST); + + for_each_sg(info->sg, sg, ring_req->nr_segments, i) { + buffer_mfn = pfn_to_mfn(page_to_pfn(sg_page(sg))); + fsect = sg->offset >> 9; + lsect = fsect + (sg->length >> 9) - 1; /* install a grant reference. */ ref = gnttab_claim_grant_reference(&gref_head); BUG_ON(ref == -ENOSPC); @@ -254,16 +256,12 @@ static int blkif_queue_request(struct request *req) buffer_mfn, rq_data_dir(req) ); - info->shadow[id].frame[ring_req->nr_segments] = - mfn_to_pfn(buffer_mfn); - - ring_req->seg[ring_req->nr_segments] = + info->shadow[id].frame[i] = mfn_to_pfn(buffer_mfn); + ring_req->seg[i] = (struct blkif_request_segment) { .gref = ref, .first_sect = fsect, .last_sect = lsect }; - - ring_req->nr_segments++; } info->ring.req_prod_pvt++; @@ -622,6 +620,8 @@ static int setup_blkring(struct xenbus_device *dev, SHARED_RING_INIT(sring); FRONT_RING_INIT(&info->ring, sring, PAGE_SIZE); + sg_init_table(info->sg, BLKIF_MAX_SEGMENTS_PER_REQUEST); + err = xenbus_grant_ring(dev, virt_to_mfn(info->ring.sring)); if (err < 0) { free_page((unsigned long)sring); -- cgit v1.2.2