- 08 Jan, 2015 24 commits
-
-
Joe Thornber authored
commit f29a3147 upstream. Overwrite causes the cache block and origin blocks to diverge, which is only allowed in writeback mode. Signed-off-by:
Joe Thornber <ejt@redhat.com> Signed-off-by:
Mike Snitzer <snitzer@redhat.com> Signed-off-by:
Greg Kroah-Hartman <gregkh@linuxfoundation.org>
-
Milan Broz authored
commit 1a71d6ff upstream. Use memzero_explicit to cleanup sensitive data allocated on stack to prevent the compiler from optimizing and removing memset() calls. Signed-off-by:
Milan Broz <gmazyland@gmail.com> Signed-off-by:
-
Darrick J. Wong authored
commit 445559cd upstream. When dm-bufio sets out to use the bio built into a struct dm_buffer to issue an IO, it needs to call bio_reset after it's done with the bio so that we can free things attached to the bio such as the integrity payload. Therefore, inject our own endio callback to take care of the bio_reset after calling submit_io's end_io callback. Test case: 1. modprobe scsi_debug delay=0 dif=1 dix=199 ato=1 dev_size_mb=300 2. Set up a dm-bufio client, e.g. dm-verity, on the scsi_debug device 3. Repeatedly read metadata and watch kmalloc-192 leak! Signed-off-by:
Darrick J. Wong <darrick.wong@oracle.com> Signed-off-by:
Mikulas Patocka <mpatocka@redhat.com> Signed-off-by:
-
Mikulas Patocka authored
commit 08d4f772 upstream. This patch fixes kmemcheck warning in switch_names. The function switch_names swaps inline names of two dentries. It swaps full arrays d_iname, no matter how many bytes are really used by the strings. Reading data beyond string ends results in kmemcheck warning. We fix the bug by marking both arrays as fully initialized. Signed-off-by:
Al Viro <viro@zeniv.linux.org.uk> Signed-off-by:
-
Peng Tao authored
commit 4bd5a980 upstream. nfs4_layoutget_release() drops layout hdr refcnt. Grab the refcnt early so that it is safe to call .release in case nfs4_alloc_pages fails. Signed-off-by:
Peng Tao <tao.peng@primarydata.com> Fixes: a47970ff ("NFSv4.1: Hold reference to layout hdr in layoutget") Signed-off-by:
Trond Myklebust <trond.myklebust@primarydata.com> Signed-off-by:
-
Jan Kara authored
commit 9234f319 upstream. f2fs_write_begin() doesn't initialize the 'dn' variable if the inode has inline data. However it uses its contents to decide whether it should just zero out the page or load data to it. Thus if we are unlucky we can zero out page contents instead of loading inline data into a page. CC: Changman Lee <cm224.lee@samsung.com> Signed-off-by:
Jan Kara <jack@suse.cz> Signed-off-by:
Jaegeuk Kim <jaegeuk@kernel.org> Signed-off-by:
-
Jan Kara authored
commit 9bd27ae4 upstream. If user specifies too low end sector for trimming, f2fs_trim_fs() will use uninitialized value as a number of trimmed blocks and returns it to userspace. Initialize number of trimmed blocks early to avoid the problem. Coverity-id: 1248809 Signed-off-by:
-
Hannes Reinecke authored
commit b6c92b7e upstream. The .eh_abort_handler needs to return SUCCESS, FAILED, or FAST_IO_FAIL. So fixup all callers to adhere to this requirement. Reviewed-by:
Robert Elliott <elliott@hp.com> Signed-off-by:
Hannes Reinecke <hare@suse.de> Signed-off-by:
Christoph Hellwig <hch@lst.de> Signed-off-by:
-
Markus Pargmann authored
commit fe08be3e upstream. The code reads the default voltage selector from its register. If the bootloader disables the regulator, the default voltage selector will be 0 which results in faulty behaviour of this regulator driver. This patch sets a default voltage selector for vddpu if it is not set in the register. Signed-off-by:
Markus Pargmann <mpa@pengutronix.de> Signed-off-by:
Mark Brown <broonie@kernel.org> Signed-off-by:
-
Sumit.Saxena@avagotech.com authored
commit 6e755ddc upstream. This patch addresses few endianness related bug fixes. Signed-off-by:
Sumit Saxena <sumit.saxena@avagotech.com> Signed-off-by:
Kashyap Desai <kashyap.desai@avagotech.com> Reviewed-by:
Tomas Henzl <thenzl@redhat.com> Signed-off-by:
-
Sumit.Saxena@avagotech.com authored
commit 170c2387 upstream. Corrected wait_event() call which was waiting for wrong completion status (0xFF). Signed-off-by:
-
Peter Guo authored
commit 6380ea09 upstream. Fix Dell E5440 when reboot Linux, can't find o2micro sd host chip issue. Fixes: 01acf691 (mmc: sdhci-pci: add support of O2Micro/BayHubTech SD hosts) Signed-off-by:
Peter Guo <peter.guo@bayhubtech.com> Signed-off-by:
Ulf Hansson <ulf.hansson@linaro.org> Signed-off-by:
-
Baruch Siach authored
commit 0031a98a upstream. Make force_ro consistent with other sysfs entries. Fixes: 371a689f ('mmc: MMC boot partitions support') Cc: Andrei Warkentin <andrey.warkentin@gmail.com> Signed-off-by:
Baruch Siach <baruch@tkos.co.il> Signed-off-by:
-
Ulf Hansson authored
commit 903101a8 upstream. The commit, mmc: omap: clarify DDR timing mode between SD-UHS and eMMC, switched omap_hsmmc to support MMC DDR mode instead of UHS DDR50 mode. Add UHS DDR50 mode again and this time let's also keep the MMC DDR mode. Fixes: 5438ad95 (mmc: omap: clarify DDR timing mode between SD-UHS and eMMC) Reported-by:
Kishon Vijay Abraham I <kishon@ti.com> Signed-off-by:
-
James Hogan authored
commit 66dfd101 upstream. Commit f1d2736c (mmc: dw_mmc: control card read threshold) added dw_mci_ctrl_rd_thld() with an unconditional write to the CDTHRCTL register at offset 0x100. However before version 240a, the FIFO region started at 0x100, so the write messes with the FIFO and completely breaks the driver. If the version id < 240A, return early from dw_mci_ctl_rd_thld() so as not to hit this problem. Fixes: f1d2736c (mmc: dw_mmc: control card read threshold) Signed-off-by:
James Hogan <james.hogan@imgtec.com> Acked-by:
Jaehoon Chung <jh80.chung@samsung.com> Signed-off-by:
-
Dmitry Eremin-Solenikov authored
commit 1a5fb99d upstream. Some boards with TC6393XB chip require full state restore during system resume thanks to chip's VCC being cut off during suspend (Sharp SL-6000 tosa is one of them). Failing to do so would result in ohci Oops on resume due to internal memory contentes being changed. Fail ohci suspend on tc6393xb is full state restore is required. Recommended workaround is to unbind tmio-ohci driver before suspend and rebind it after resume. Signed-off-by:
Dmitry Eremin-Solenikov <dbaryshkov@gmail.com> Signed-off-by:
Lee Jones <lee.jones@linaro.org> Signed-off-by:
-
Tony Lindgren authored
commit 1b9b46d0 upstream. Commit e7cd1d1e ("mfd: twl4030-power: Add generic reset configuration") accidentally removed the compatible flag for "ti,twl4030-power" that should be there as documented in the binding. If "ti,twl4030-power" only the poweroff configuration is done by the driver. Fixes: e7cd1d1e ("mfd: twl4030-power: Add generic reset configuration") Reported-by:
"Dr. H. Nikolaus Schaller" <hns@goldelico.com> Signed-off-by:
Tony Lindgren <tony@atomide.com> Signed-off-by:
-
Sonny Rao authored
commit 0b46b8a7 upstream. This is a bug fix for using physical arch timers when the arch_timer_use_virtual boolean is false. It restores the arch_counter_get_cntpct() function after removal in 0d651e4e "clocksource: arch_timer: use virtual counters" We need this on certain ARMv7 systems which are architected like this: * The firmware doesn't know and doesn't care about hypervisor mode and we don't want to add the complexity of hypervisor there. * The firmware isn't involved in SMP bringup or resume. * The ARCH timer come up with an uninitialized offset between the virtual and physical counters. Each core gets a different random offset. * The device boots in "Secure SVC" mode. * Nothing has touched the reset value of CNTHCTL.PL1PCEN or CNTHCTL.PL1PCTEN (both default to 1 at reset) One example of such as system is RK3288 where it is much simpler to use the physical counter since there's nobody managing the offset and each time a core goes down and comes back up it will get reinitialized to some other random value. Fixes: 0d651e4e ("clocksource: arch_timer: use virtual counters") Signed-off-by:
Sonny Rao <sonnyrao@chromium.org> Acked-by:
Catalin Marinas <catalin.marinas@arm.com> Acked-by:
Daniel Lezcano <daniel.lezcano@linaro.org> Signed-off-by:
Olof Johansson <olof@lixom.net> Signed-off-by:
-
Hante Meuleman authored
commit 333c2aa0 upstream. Reviewed-by:
Arend Van Spriel <arend@broadcom.com> Reviewed-by:
Pieter-Paul Giesberts <pieterpg@broadcom.com> Signed-off-by:
Hante Meuleman <meuleman@broadcom.com> Signed-off-by:
John W. Linville <linville@tuxdriver.com> Signed-off-by:
-
Andy Lutomirski authored
commit 29fa6825 upstream. paravirt_enabled has the following effects: - Disables the F00F bug workaround warning. There is no F00F bug workaround any more because Linux's standard IDT handling already works around the F00F bug, but the warning still exists. This is only cosmetic, and, in any event, there is no such thing as KVM on a CPU with the F00F bug. - Disables 32-bit APM BIOS detection. On a KVM paravirt system, there should be no APM BIOS anyway. - Disables tboot. I think that the tboot code should check the CPUID hypervisor bit directly if it matters. - paravirt_enabled disables espfix32. espfix32 should *not* be disabled under KVM paravirt. The last point is the purpose of this patch. It fixes a leak of the high 16 bits of the kernel stack address on 32-bit KVM paravirt guests. Fixes CVE-2014-8134. Suggested-by:
Konrad Rzeszutek Wilk <konrad.wilk@oracle.com> Signed-off-by:
Andy Lutomirski <luto@amacapital.net> Signed-off-by:
Paolo Bonzini <pbonzini@redhat.com> Signed-off-by:
-
Andy Lutomirski authored
commit f647d7c1 upstream. Otherwise, if buggy user code points DS or ES into the TLS array, they would be corrupted after a context switch. This also significantly improves the comments and documents some gotchas in the code. Before this patch, the both tests below failed. With this patch, the es test passes, although the gsbase test still fails. ----- begin es test ----- /* * Copyright (c) 2014 Andy Lutomirski * GPL v2 */ static unsigned short GDT3(int idx) { return (idx << 3) | 3; } static int create_tls(int idx, unsigned int base) { struct user_desc desc = { .entry_number = idx, .base_addr = base, .limit = 0xfffff, .seg_32bit = 1, .contents = 0, /* Data, grow-up */ .read_exec_only = 0, .limit_in_pages = 1, .seg_not_present = 0, .useable = 0, }; if (syscall(SYS_set_thread_area, &desc) != 0) err(1, "set_thread_area"); return desc.entry_number; } int main() { int idx = create_tls(-1, 0); printf("Allocated GDT index %d\n", idx); unsigned short orig_es; asm volatile ("mov %%es,%0" : "=rm" (orig_es)); int errors = 0; int total = 1000; for (int i = 0; i < total; i++) { asm volatile ("mov %0,%%es" : : "rm" (GDT3(idx))); usleep(100); unsigned short es; asm volatile ("mov %%es,%0" : "=rm" (es)); asm volatile ("mov %0,%%es" : : "rm" (orig_es)); if (es != GDT3(idx)) { if (errors == 0) printf("[FAIL]\tES changed from 0x%hx to 0x%hx\n", GDT3(idx), es); errors++; } } if (errors) { printf("[FAIL]\tES was corrupted %d/%d times\n", errors, total); return 1; } else { printf("[OK]\tES was preserved\n"); return 0; } } ----- end es test ----- ----- begin gsbase test ----- /* * gsbase.c, a gsbase test * Copyright (c) 2014 Andy Lutomirski * GPL v2 */ static unsigned char *testptr, *testptr2; static unsigned char read_gs_testvals(void) { unsigned char ret; asm volatile ("movb %%gs:%1, %0" : "=r" (ret) : "m" (*testptr)); return ret; } int main() { int errors = 0; testptr = mmap((void *)0x200000000UL, 1, PROT_READ | PROT_WRITE, MAP_PRIVATE | MAP_FIXED | MAP_ANONYMOUS, -1, 0); if (testptr == MAP_FAILED) err(1, "mmap"); testptr2 = mmap((void *)0x300000000UL, 1, PROT_READ | PROT_WRITE, MAP_PRIVATE | MAP_FIXED | MAP_ANONYMOUS, -1, 0); if (testptr2 == MAP_FAILED) err(1, "mmap"); *testptr = 0; *testptr2 = 1; if (syscall(SYS_arch_prctl, ARCH_SET_GS, (unsigned long)testptr2 - (unsigned long)testptr) != 0) err(1, "ARCH_SET_GS"); usleep(100); if (read_gs_testvals() == 1) { printf("[OK]\tARCH_SET_GS worked\n"); } else { printf("[FAIL]\tARCH_SET_GS failed\n"); errors++; } asm volatile ("mov %0,%%gs" : : "r" (0)); if (read_gs_testvals() == 0) { printf("[OK]\tWriting 0 to gs worked\n"); } else { printf("[FAIL]\tWriting 0 to gs failed\n"); errors++; } usleep(100); if (read_gs_testvals() == 0) { printf("[OK]\tgsbase is still zero\n"); } else { printf("[FAIL]\tgsbase was corrupted\n"); errors++; } return errors == 0 ? 0 : 1; } ----- end gsbase test ----- Signed-off-by:
Ingo Molnar <mingo@kernel.org> Signed-off-by:
-
Andy Lutomirski authored
commit 0e58af4e upstream. Users have no business installing custom code segments into the GDT, and segments that are not present but are otherwise valid are a historical source of interesting attacks. For completeness, block attempts to set the L bit. (Prior to this patch, the L bit would have been silently dropped.) This is an ABI break. I've checked glibc, musl, and Wine, and none of them look like they'll have any trouble. Note to stable maintainers: this is a hardening patch that fixes no known bugs. Given the possibility of ABI issues, this probably shouldn't be backported quickly. Signed-off-by:
H. Peter Anvin <hpa@zytor.com> Cc: Konrad Rzeszutek Wilk <konrad.wilk@oracle.com> Cc: Linus Torvalds <torvalds@linux-foundation.org> Cc: Willy Tarreau <w@1wt.eu> Signed-off-by:
-
Andy Lutomirski authored
commit 41bdc785 upstream. Installing a 16-bit RW data segment into the GDT defeats espfix. AFAICT this will not affect glibc, Wine, or dosemu at all. Signed-off-by:
-
Jan Kara authored
commit f54e18f1 upstream. Rock Ridge extensions define so called Continuation Entries (CE) which define where is further space with Rock Ridge data. Corrupted isofs image can contain arbitrarily long chain of these, including a one containing loop and thus causing kernel to end in an infinite loop when traversing these entries. Limit the traversal to 32 entries which should be more than enough space to store all the Rock Ridge data. Reported-by:
P J P <ppandit@redhat.com> Signed-off-by:
-
- 16 Dec, 2014 16 commits
-
-
Greg Kroah-Hartman authored
-
Takashi Iwai authored
commit 66139a48 upstream. In snd_usbmidi_error_timer(), the driver tries to resubmit MIDI input URBs to reactivate the MIDI stream, but this causes the error when some of URBs are still pending like: WARNING: CPU: 0 PID: 0 at ../drivers/usb/core/urb.c:339 usb_submit_urb+0x5f/0x70() URB ef705c40 submitted while active CPU: 0 PID: 0 Comm: swapper/0 Not tainted 3.16.6-2-desktop #1 Hardware name: FOXCONN TPS01/TPS01, BIOS 080015 03/23/2010 c0984bfa f4009ed4 c078deaf f4009ee4 c024c884 c09a135c f4009f00 00000000 c0984bfa 00000153 c061ac4f c061ac4f 00000009 00000001 ef705c40 e854d1c0 f4009eec c024c8d3 00000009 f4009ee4 c09a135c f4009f00 f4009f04 c061ac4f Call Trace: [<c0205df6>] try_stack_unwind+0x156/0x170 [<c020482a>] dump_trace+0x5a/0x1b0 [<c0205e56>] show_trace_log_lvl+0x46/0x50 [<c02049d1>] show_stack_log_lvl+0x51/0xe0 [<c0205eb7>] show_stack+0x27/0x50 [<c078deaf>] dump_stack+0x45/0x65 [<c024c884>] warn_slowpath_common+0x84/0xa0 [<c024c8d3>] warn_slowpath_fmt+0x33/0x40 [<c061ac4f>] usb_submit_urb+0x5f/0x70 [<f7974104>] snd_usbmidi_submit_urb+0x14/0x60 [snd_usbmidi_lib] [<f797483a>] snd_usbmidi_error_timer+0x6a/0xa0 [snd_usbmidi_lib] [<c02570c0>] call_timer_fn+0x30/0x130 [<c0257442>] run_timer_softirq+0x1c2/0x260 [<c0251493>] __do_softirq+0xc3/0x270 [<c0204732>] do_softirq_own_stack+0x22/0x30 [<c025186d>] irq_exit+0x8d/0xa0 [<c0795228>] smp_apic_timer_interrupt+0x38/0x50 [<c0794a3c>] apic_timer_interrupt+0x34/0x3c [<c0673d9e>] cpuidle_enter_state+0x3e/0xd0 [<c028bb8d>] cpu_idle_loop+0x29d/0x3e0 [<c028bd23>] cpu_startup_entry+0x53/0x60 [<c0bfac1e>] start_kernel+0x415/0x41a For avoiding these errors, check the pending URBs and skip resubmitting such ones. Reported-and-tested-by:
Stefan Seyfried <stefan.seyfried@googlemail.com> Acked-by:
Clemens Ladisch <clemens@ladisch.de> Signed-off-by:
Takashi Iwai <tiwai@suse.de> Signed-off-by:
-
Takashi Iwai authored
commit fedb2245 upstream. The built-in mic boost volume gets almost muted after suspend/resume on Lenovo Ideapad S210. Bugzilla: https://bugzilla.kernel.org/show_bug.cgi?id=88121Reported-and-tested-by:
Roman Kagan <rkagan@mail.ru> Signed-off-by:
-
Takashi Iwai authored
commit f62f5eff upstream. The same fixup to enable EAPD is needed for ASUS Z99He with AD1986A codec like another ASUS machine. Reported-and-tested-by:
Dmitry V. Zimin <pfzim@mail.ru> Signed-off-by:
-
Al Viro authored
commit ca5358ef upstream. ... by not hitting rename_retry for reasons other than rename having happened. In other words, do _not_ restart when finding that between unlocking the child and locking the parent the former got into __dentry_kill(). Skip the killed siblings instead... Signed-off-by:
-
Al Viro authored
commit 946e51f2 upstream. Signed-off-by:
-
Larry Finger authored
commit 87141db0 upstream. Proper operation with the rewritten PCI mini driver requires that a flag be set when interrupts are enabled. This flag was missed. This patch is one of three needed to fix the kernel regression reported at https://bugzilla.kernel.org/show_bug.cgi?id=88951. Signed-off-by:
Larry Finger <Larry.Finger@lwfinger.net> Reported-by:
Catalin Iacob <iacobcatalin@gmail.com> Tested-by:
-
Larry Finger authored
commit f892914c upstream. In the major update of the rtlwifi-family of drivers, one of the callback entries was missed, which leads to memory corruption. Unfortunately, this corruption never caused a kernel oops, but showed up in other parts of the system. This patch is one of three needed to fix the kernel regression reported at https://bugzilla.kernel.org/show_bug.cgi?id=88951. Signed-off-by:
-
Larry Finger authored
commit 99a82f73 upstream. In the major update of the rtlwifi-family of drivers, there was an editing mistake. Unfortunately, this particular error leads to memory corruption that silently leads to failure of the system. This patch is one of three needed to fix the kernel regression reported at https://bugzilla.kernel.org/show_bug.cgi?id=88951. Signed-off-by:
-
Daniel Borkmann authored
[ Upstream commit 7f19fc5e ] For netlink, we shouldn't be using arch_fast_hash() as a hashing discipline, but rather jhash() instead. Since netlink sockets can be opened by any user, a local attacker would be able to easily create collisions with the DPDK-derived arch_fast_hash(), which trades off performance for security by using crc32 CPU instructions on x86_64. While it might have a legimite use case in other places, it should be avoided in netlink context, though. As rhashtable's API is very flexible, we could later on still decide on other hashing disciplines, if legitimate. Reference: http://thread.gmane.org/gmane.linux.kernel/1844123 Fixes: e341694e ("netlink: Convert netlink_lookup() to use RCU protected hash table") Cc: Herbert Xu <herbert@gondor.apana.org.au> Signed-off-by:
Daniel Borkmann <dborkman@redhat.com> Acked-by:
Thomas Graf <tgraf@suug.ch> Acked-by:
Hannes Frederic Sowa <hannes@stressinduktion.org> Signed-off-by:
David S. Miller <davem@davemloft.net> Signed-off-by:
-
Valdis.Kletnieks@vt.edu authored
[ Upstream commit 69204cf7 ] commit 46e5da40 (net: qdisc: use rcu prefix and silence sparse warnings) triggers a spurious warning: net/sched/sch_fq_codel.c:97 suspicious rcu_dereference_check() usage! The code should be using the _bh variant of rcu_dereference. Signed-off-by:
Valdis Kletnieks <valdis.kletnieks@vt.edu> Acked-by:
Eric Dumazet <edumazet@google.com> Acked-by:
John Fastabend <john.r.fastabend@intel.com> Signed-off-by:
-
David Vrabel authored
[ Upstream commit 11d3d2a1 ] Commit 97a6d1bb (xen-netfront: Fix handling packets on compound pages with skb_linearize) attempted to fix a problem where an skb that would have required too many slots would be dropped causing TCP connections to stall. However, it filled in the first slot using the original buffer and not the new one and would use the wrong offset and grant access to the wrong page. Netback would notice the malformed request and stop all traffic on the VIF, reporting: vif vif-3-0 vif3.0: txreq.offset: 85e, size: 4002, end: 6144 vif vif-3-0 vif3.0: fatal error; disabling device Reported-by:
Anthony Wright <anthony@overnetdata.com> Tested-by:
David Vrabel <david.vrabel@citrix.com> Signed-off-by:
-
Eric Dumazet authored
[ Upstream commit 0f85feae ] When I cooked commit c3658e8d ("tcp: fix possible NULL dereference in tcp_vX_send_reset()") I missed other spots we could deref a NULL skb_dst(skb) Again, if a socket is provided, we do not need skb_dst() to get a pointer to network namespace : sock_net(sk) is good enough. Reported-by:
Dann Frazier <dann.frazier@canonical.com> Bisected-by:
-
Daniel Borkmann authored
[ Upstream commit 9772b54c ] To accomodate for enough headroom for tunnels, use MAX_HEADER instead of LL_MAX_HEADER. Robert reported that he has hit after roughly 40hrs of trinity an skb_under_panic() via SCTP output path (see reference). I couldn't reproduce it from here, but not using MAX_HEADER as elsewhere in other protocols might be one possible cause for this. In any case, it looks like accounting on chunks themself seems to look good as the skb already passed the SCTP output path and did not hit any skb_over_panic(). Given tunneling was enabled in his .config, the headroom would have been expanded by MAX_HEADER in this case. Reported-by:
Robert Święcki <robert@swiecki.net> Reference: https://lkml.org/lkml/2014/12/1/507 Fixes: 594ccc14 ("[SCTP] Replace incorrect use of dev_alloc_skb with alloc_skb in sctp_packet_transmit().") Signed-off-by:
Vlad Yasevich <vyasevich@gmail.com> Acked-by:
Neil Horman <nhorman@tuxdriver.com> Signed-off-by:
-
Eric Dumazet authored
[ Upstream commit 5f478b41 ] mvneta_tx() dereferences skb to get skb->len too late, as hardware might have completed the transmit and TX completion could have freed the skb from another cpu. Fixes: 71f6d1b3 ("net: mvneta: replace Tx timer with a real interrupt") Signed-off-by:
-
willy tarreau authored
[ Upstream commit aebea2ba ] The mvneta driver sets the amount of Tx coalesce packets to 16 by default. Normally that does not cause any trouble since the driver uses a much larger Tx ring size (532 packets). But some sockets might run with very small buffers, much smaller than the equivalent of 16 packets. This is what ping is doing for example, by setting SNDBUF to 324 bytes rounded up to 2kB by the kernel. The problem is that there is no documented method to force a specific packet to emit an interrupt (eg: the last of the ring) nor is it possible to make the NIC emit an interrupt after a given delay. In this case, it causes trouble, because when ping sends packets over its raw socket, the few first packets leave the system, and the first 15 packets will be emitted without an IRQ being generated, so without the skbs being freed. And since the socket's buffer is small, there's no way to reach that amount of packets, and the ping ends up with "send: no buffer available" after sending 6 packets. Running with 3 instances of ping in parallel is enough to hide the problem, because with 6 packets per instance, that's 18 packets total, which is enough to grant a Tx interrupt before all are sent. The original driver in the LSP kernel worked around this design flaw by using a software timer to clean up the Tx descriptors. This timer was slow and caused terrible network performance on some Tx-bound workloads (such as routing) but was enough to make tools like ping work correctly. Instead here, we simply set the packet counts before interrupt to 1. This ensures that each packet sent will produce an interrupt. NAPI takes care of coalescing interrupts since the interrupt is disabled once generated. No measurable performance impact nor CPU usage were observed on small nor large packets, including when saturating the link on Tx, and this fixes tools like ping which rely on too small a send buffer. If one wants to increase this value for certain workloads where it is safe to do so, "ethtool -C $dev tx-frames" will override this default setting. This fix needs to be applied to stable kernels starting with 3.10. Tested-By:
Maggie Mae Roxas <maggie.mae.roxas@gmail.com> Signed-off-by:
Willy Tarreau <w@1wt.eu> Signed-off-by:
-
