[PATCH net-next v8 00/11] r8152: refactor and extend RTL8157/8159 support

[Chih Kai Hsu <hsu.chih.kai@realtek.com>]

RTL8157 and RTL8159 have different init, enable, up/down, hw_phy_cfg,
and PHY access sequences from RTL8156. This series first fixes the
up/down, unload, change_mtu and flow control settings that these chips
shared with RTL8156 but need to be different, then separates the shared
callbacks by chip generation and adds the missing RTL8157/RTL8159
functionality: TGPHY register access via phy_read/phy_write pointers in
struct rtl_ops, and UPS support via r8157_ups_en().

v8:
- Move the fixes to the front of the series, so that they can be
  backported without the refactoring, and add Fixes tags:
  - Patch #1 (was v7 #5): fix up and down register settings, now
    applied to the shared rtl8156_up()/rtl8156_down() before they are
    split.
  - Patch #2 and #3 (were v7 #7): split rtl8157_unload() and
    rtl8157_change_mtu() into separate patches. Reword the commit
    messages.
  - Patch #4 (was v7 #9): fix flow control for
    RTL8156/RTL8156B/RTL8157/RTL8159. It is now applied to the shared
    rtl8156_init() and rtl8156_enable(), and goes ahead of the
    refactoring. Rework the commit message: describe what was missing
    on each chip, and note that the 500 and 1250 Mbit/s rates the
    hardware can report disarm the pause limit.

- Patch #5 (was v7 #1): the RTL8156B-only part of r8156_init() is now
  the AUTO_SPEEDUP clear, as the flow control setup moved to patch #4.
  Update the commit message.

- Patch #6 (was v7 #2): RTL_VER_12 no longer gets
  r8156b_u2phy_backup().

- Patch #7 (was v7 #3): rebase on the new order, with the VER_17 cases
  in rtl_fc_pause_pkt_en(), rtl8156_up() and rtl8156_down() split into
  QFN68 and QFN100.

- Patch #8 (was v7 #4): rebase on the new order. The register fixes
  and rtl8157_change_mtu() now come from patches #1 to #3, so this
  patch is a pure split with no functional change intended. Add
  rtl8156b_enable() for RTL8156B (VER_12/13/15), so that
  rtl8156_enable() serves RTL8156 only and neither needs version
  checks.

- Patch #9 (was v7 #6): rebase on the new order.

- Patch #10 (was v7 #8):
  - Log a TGPHY busy timeout ratelimited, so that a stuck busy bit does
    not flood the log.
  - Propagate the PHY access failure also in
    rtl8152_get_link_ksettings(), r8153_get_eee() and
    r8152_mdio_read(), and stop on a failed read in rtl8153_in_nway(),
    r8156_mdio_force_mode(), r8156_eee_en() and rtl8152_fw_phy_fixup().
  - rtl8152_get_pauseparam() reports no flow control, and
    r8153_phy_status() keeps polling, when a read fails.
  - rtl8152_set_speed() leaves the cached advertising and a pending PHY
    reset untouched when a write fails.
  - read_mii_word() reports a failed read as 0, as the MII library
    takes the return value for register content.
  - rtl_phy_patch_request() returns the errno of a failed write and
    releases PATCH_REQUEST when a later step fails.
  - Rewrite the commit message and list what is left to a follow-up
    series.

- Patch #11 (was v7 #10): in r8157_ups_en(), stop waiting for the
  backup restore, and return, as soon as the device is marked
  RTL8152_INACCESSIBLE.

v7:
- Patch #1: split r8156_init() by chip generation rather than per
  chip. RTL8156 and RTL8156B share r8156_init(), and RTL8157 and
  RTL8159 share r8157_init(), since their sequences are near-identical.
  The few differences are selected with switch (tp->version).
  r8156b_init() and r8159_init() are gone.

- Patch #2: apply the new init writes to the merged r8156_init() and
  r8157_init().

- Patch #3:
  - The RTL8159 QFN68/QFN100 differences become cases of the version
    switches in r8157_init(), and the QFN68-only u2phy backup restore
    write falls through to the common u3phy backup write.
  - Report the number of retries (i - 1) in the "Needed %d retries to
    read package" warning.
  - Update the commit message.

- Patch #4:
  - Share the exit-OOB and enter-OOB sequences of RTL8156, RTL8156B,
    RTL8157 and RTL8159 through r8156_exit_oob() and r8156_enter_oob(),
    as RTL8152B and RTL8153 do.
  - Serve RTL8156 and RTL8156B (VER_10/11/12/13/15) with a single
    rtl8156_enable() instead of adding rtl8156b_enable(). The
    r8156_fc_parameter() call stays behind a version check, so it is
    no longer dropped.
  - Retitle and update the commit message.

- Patch #5: adapt to the shared r8156_exit_oob() and r8156_enter_oob().
  Update the commit message.

- Patch #7: r8156_exit_oob() now calls tp->rtl_ops.change_mtu() instead
  of rtl8156_change_mtu(), so that rtl8157_up() gets
  rtl8157_change_mtu(). Update the commit message.

- Patch #9: apply rtl_fc_pause_pkt_en() in the merged r8156_init().
  AUTO_SPEEDUP is still cleared there for RTL8156B only. Update the
  commit message.

- Patch #10: replace r8156_ups_en() with r8157_ups_en() in the merged
  r8157_init() only. Update the commit message.

v6:
- Patch #1: make it a pure split of r8156_init() per chip, plus the
  wait_autoload_done() helper, with no functional change. The new
  init writes it used to carry move to patch #2. Also convert
  r8152b_init() to wait_autoload_done().

- Patch #2 (new): add the new init writes for RTL8156B/8157/8159,
  taken out of patch #1.

- Patch #3:
  - Fix the "Needed %d retries" warning in __rtl_get_pkg_det() to
    trigger only when a retry was actually needed (i != 1).
  - Reword the commit message.

- Patch #6:
  - r8159_hw_phy_cfg() ends with rtl_eee_enable(tp, tp->eee_en) so
    that the configured state is applied in both directions.
  - Reword the commit message.

- Patch #8:
  - A TGPHY busy timeout no longer sets RTL8152_INACCESSIBLE.
    wait_tgphy_cmd_ready() logs and returns the poll error, as the
    ADV command channel does, and returns -ENODEV only if the device
    is already marked inaccessible.
  - Drop the RTL8152_INACCESSIBLE exceptions from the LAN_ON
    WARN_ON_ONCE() checks in hw_phy_cfg, which are no longer needed.
  - Narrow the commit message to what the patch does and list what is
    left to a follow-up series: the remaining callers, the MII library
    callbacks, and the recovery from a TGPHY timeout.

- Patch #9:
  - Explain the symptom, that RTL8157/RTL8159 could keep sending pause
    frames indefinitely, and add Fixes tags for RTL8156, RTL8157 and
    RTL8159.
  - Keep the FC patch in r8156b_init() split from the AUTO_SPEEDUP
    clearing, and call rtl_fc_pause_pkt_en() right before it.

- Patch #10: note in the commit message that a backup-restore timeout
  in r8157_ups_en() follows what r8159_init() does for the same wait
  (mark the device inaccessible and log, without queuing a reset), and
  that recovery is left to a follow-up series.

v5:
- Patch #1: in r8157_init()/r8159_init(), call
  r8156b_wait_loading_flash() before wait_autoload_done(). Document
  in the commit message that the "Bulk In End" bit is cleared in
  r8157_init(), and drop the claim about the FC-timer block.

- Patch #2: accept the package read only if the full word was returned,
  warn when a retry was needed, and print an unknown package as
  0x%04x. Say "firmware" instead of "PHY tuning" in the commit
  message.

- Patch #3: make it a pure split with no functional change. Its
  register fixes move to the new patch #4.

- Patch #4 (new): fix up and down register settings for
  RTL8156/8156B/8157/8159.

- Patch #5: document the new RTL8159 PHY tuning parameters in the
  commit message.

- Patch #6: add Fixes tags.

- Patch #7:
  - A TGPHY busy timeout now sets RTL8152_INACCESSIBLE and returns
    -ENODEV.
  - sram_write() returns the data-write result.
  - The PHY firmware version checks, the BMCR reset poll,
    rtl8152_set_speed(), the pause parameters and the PHY patch
    handshake handle PHY access failures.
  - SIOCGMIIREG returns 0 with val_out set on success, or the error.
  - Commit message rewritten.

- Patch #8: drop the "no link" case, since the caller checks the link.
  An unknown speed now disarms the pause limit. Commit message
  rewritten.

- Patch #9: wait for the backup-restore flag with a timeout in
  r8157_ups_en(). Rename USB_FW_USE_VER to USB_FW_USB_VER.

v4:
- Patch #1: make wait_autoload_done() bail out immediately on
  RTL8152_INACCESSIBLE instead of waiting out the full 10 second
  timeout.
- Patch #6: propagate PHY access failures instead of ignoring them.
  ocp_reg_read()/ocp_reg_write() and every helper built on top of them
  (sram_read/write, sram2_read/write, the ocp_reg_*_bits and MDIO bit
  helpers) now return int and stop silently operating on stale data
  when a TGPHY access times out.

v3:
Rebase on top of latest net-next to resolve merge conflicts.

v2:
- For patch #1, use read_poll_timeout() to replace the manual
  for-loop polling.
- For patch #2 and #3, rewrite the commit message.

Chih Kai Hsu (11):
  r8152: fix RTL8156/RTL8156B/RTL8157/RTL8159 up and down settings
  r8152: add rtl8157_unload() for RTL8157 and RTL8159
  r8152: add rtl8157_change_mtu() for RTL8157 and RTL8159
  r8152: fix flow control for RTL8156/RTL8156B/RTL8157/RTL8159
  r8152: split r8156_init by generation and factor out
    wait_autoload_done
  r8152: add new init writes for RTL8156/8156B/8157/8159
  r8152: split RTL_VER_17 into QFN68 and QFN100 package variants
  r8152: split rtl8156_enable/up/down into per-generation functions
  r8152: split r8157_hw_phy_cfg into RTL8157 and RTL8159 variants
  r8152: add TGPHY register access for RTL8157 and RTL8159
  r8152: enable UPS for RTL8157 and RTL8159 QFN68

 drivers/net/usb/r8152.c | 1778 +++++++++++++++++++++++++++++----------
 1 file changed, 1341 insertions(+), 437 deletions(-)

-- 
2.34.1