The following reply was made to PR system/6607; it has been noted by GNATS.
From: Marco Pfatschbacher <m...@mailq.de> To: Mark Kettenis <mark.kette...@xs4all.nl> Cc: stol...@vinson.dyndns.info, gn...@openbsd.org, b...@cvs.openbsd.org, cstol...@hushmail.com Subject: Re: system/6607: Network device em0 detaches when transferring large files Date: Thu, 19 May 2011 13:42:40 +0200 On Thu, May 19, 2011 at 01:37:15PM +0200, Mark Kettenis wrote: > Spontanious detach usually is related to issues with Active State > Power Management (ASPM). Indeed it seems there is an errata for the > Intel 82573 that affects the 82574 as well. > > I don't have time to write a diff for this right now. But perhaps > somebody else can pick this up. Could you try this? Index: if_em.c =================================================================== RCS file: /cvs/src/sys/dev/pci/if_em.c,v retrieving revision 1.256 diff -p -u -p -u -r1.256 if_em.c --- if_em.c 22 Apr 2011 10:09:57 -0000 1.256 +++ if_em.c 19 May 2011 09:59:43 -0000 @@ -2619,13 +2619,6 @@ em_initialize_receive_unit(struct em_sof E1000_WRITE_REG(&sc->hw, RXCSUM, reg_rxcsum); } - /* - * XXX TEMPORARY WORKAROUND: on some systems with 82573 - * long latencies are observed, like Lenovo X60. - */ - if (sc->hw.mac_type == em_82573) - E1000_WRITE_REG(&sc->hw, RDTR, 0x20); - /* Enable Receives */ E1000_WRITE_REG(&sc->hw, RCTL, reg_rctl); Index: if_em_hw.c =================================================================== RCS file: /cvs/src/sys/dev/pci/if_em_hw.c,v retrieving revision 1.65 diff -p -u -p -u -r1.65 if_em_hw.c --- if_em_hw.c 2 May 2011 18:16:58 -0000 1.65 +++ if_em_hw.c 19 May 2011 09:58:05 -0000 @@ -1343,6 +1343,19 @@ em_init_hw(struct em_hw *hw) gcr |= E1000_GCR_L1_ACT_WITHOUT_L0S_RX; E1000_WRITE_REG(hw, GCR, gcr); } + if ((hw->mac_type == em_82573) || (hw->mac_type == em_82574)) { + uint16_t apsm_reg; + int cap_off; + struct pci_attach_args *pa = &((struct em_osdep *)hw->back)->em_pa; + + if (pci_get_capability(pa->pa_pc, pa->pa_tag, + PCI_CAP_PCIEXPRESS, &cap_off, NULL) != 0) { + em_read_pci_cfg(hw, cap_off + PCI_PCIE_LCSR, &apsm_reg); + printf("ASPM 0x%x -> 0x%x\n", apsm_reg, apsm_reg & ~PCI_PCIE_LCSR_ASPM_L1); + apsm_reg &= ~PCI_PCIE_LCSR_ASPM_L1; + em_write_pci_cfg(hw, cap_off + PCI_PCIE_LCSR, &apsm_reg); + } + } /* * Clear all of the statistics registers (clear on read). It is * important that we do this after we have tried to establish link