Skip to contentSkip to conversation

pci: switch a ton of drivers to symbolic names

PATCH In review Michael S. Tsirkin 4 participants 5 messages Dec 10, 2009 – Dec 12, 2009

lore
Share & tools
lore
#0 lore raw

cover pci: switch a ton of drivers to symbolic names

The recent e1000 bug made the important of using symbolic macros for pci config access clear for me. So I started going over drivers and converting to symbolic constants instead of hard-coded ones. I did a large part until I run out of steam. Maybe some brave soul will take up converting the rest of them, or maybe I will: note that when converting bridges one should be careful to use bridge macros where appropriate.

Instead of testing a huge number of configurations, I compared binaries before and after conversion. Almost all of them generate exact same stripped binary before and after the change.
The only object changed was eepro100, objdump showed that the change was because gcc for some reason decides to use a bit more stack for init function after comments are added there.

This methodology was the reason that I added TODOs where I saw deviations from spec or other code ugliness, will have to be fixed separately.

I also verified separately that eepro100 still works and survives light usage.

This patcheset applies on top of e1000 bugfix I have posted previously.

Michael S. Tsirkin (17):
e1000: switch to symbolic names for pci registers
ne2000: switch to symbolic names for pci registers
rtl: switch to symbolic names for pci registers
pcnet: switch to symbolic names for pci registers
pci: add more status bits
eepro100: symbolic names for pci registers
piix: symbolic constants
cmd646: symbolic names for pci registers
vmware_vga: symbolic names for pci registers
lsi: symbolic names for pci registers
pci: add another devsel macro
es1370: symbolic names for pci registers
wdt_i6300esb: symbolic names for pci registers
ac97: symbolic names for pci registers
usb-uhci: symbolic names for pci registers
usb-uhci: symbolic names for pci registers
pci: remove unused macro

 hw/ac97.c         |   57 ++++++++++++++++++++++++++++------------------------
 hw/e1000.c        |   11 ++++++---
 hw/eepro100.c     |   49 +++++++++++++++++++++++++++++---------------
 hw/es1370.c       |   29 ++++++++++++++-------------
 hw/ide/cmd646.c   |    7 +++--
 hw/ide/piix.c     |   13 +++++++----
 hw/lsi53c895a.c   |   10 +++++---
 hw/ne2000.c       |    3 +-
 hw/pci.h          |    5 +++-
 hw/pcnet.c        |   26 +++++++++++++++--------
 hw/rtl8139.c      |   16 +++++++++-----
 hw/usb-ohci.c     |    6 +++-
 hw/usb-uhci.c     |    7 +++--
 hw/vmware_vga.c   |   20 ++++++++++--------
 hw/wdt_i6300esb.c |    2 +-
 15 files changed, 156 insertions(+), 105 deletions(-)
#1 lore raw
Quoted 19 lines · Michael S. Tsirkin

The recent e1000 bug made the important of using symbolic macros for pci config access clear for me. So I started going over drivers and converting to symbolic constants instead of hard-coded ones. I did a large part until I run out of steam. Maybe some brave soul will take up converting the rest of them, or maybe I will: note that when converting bridges one should be careful to use bridge macros where appropriate.

Instead of testing a huge number of configurations, I compared binaries before and after conversion. Almost all of them generate exact same stripped binary before and after the change.
The only object changed was eepro100, objdump showed that the change was because gcc for some reason decides to use a bit more stack for init function after comments are added there.

Quoted 4 lines

This methodology was the reason that I added TODOs where I saw deviations from spec or other code ugliness, will have to be fixed separately.

IMHO, this is a huge enhancement.

I myself was found expending huge amounts of time trying do figure out the meaning of some specific constants in the past.

+1

Signature · 5 lines
Glauber  Costa.
"Free as in Freedom"
http://glommer.net

"The less confident you are, the more serious you have to act."
#2 lore raw
In reply to Michael S. Tsirkin
Quoted 9 lines · Michael S. Tsirkin

The recent e1000 bug made the important of using symbolic macros for pci config access clear for me. So I started going over drivers and converting to symbolic constants instead of hard-coded ones. I did a large part until I run out of steam. Maybe some brave soul will take up converting the rest of them, or maybe I will: note that when converting bridges one should be careful to use bridge macros where appropriate.

Seeing as you're introducing a huge amount of churn, wouldn't it be better to come up with a sane abstraction for initializing the the PCI config data (c.f. pci_config_set_vendor_id)?

Paul

#3 lore raw
Quoted 11 lines · Paul Brook
Quoted 9 lines · Michael S. Tsirkin

The recent e1000 bug made the important of using symbolic macros for pci config access clear for me. So I started going over drivers and converting to symbolic constants instead of hard-coded ones. I did a large part until I run out of steam. Maybe some brave soul will take up converting the rest of them, or maybe I will: note that when converting bridges one should be careful to use bridge macros where appropriate.

Quoted 4 lines

Seeing as you're introducing a huge amount of churn, wouldn't it be
better to come up with a sane abstraction for initializing the the PCI config data

Yes, I am looking at a larger cleanup, but just decoding the magic numbers is a necesary intermediate step to make that one safe.

Patch removing
- /* TODO: no need to do this, BAR registration does it already */
- d->config[PCI_BASE_ADDRESS_0] = PCI_BASE_ADDRESS_SPACE_IO; will be much clearer than if it just killed this line outright:
- d->config[0x10] = 0x0001;

And the recent bug introduced in e1000 made it clear to me that it is better to submit this one meanwhile, so that one can use grep to find code affecting a register.

Quoted 3 lines

(c.f. pci_config_set_vendor_id)?

Paul

I'm not sure pci_config_set_vendor_id is an example of a good abstraction. I think good abstractions would be higher level, c.f. pci_add_capability()

I also have no idea when will a larger cleanup be done.

Signature · 1 line
MST
Acked-by on the v1 series
#4 lore raw
In reply to Michael S. Tsirkin
Quoted 9 lines · Michael S. Tsirkin

The recent e1000 bug made the important of using symbolic macros for pci config access clear for me. So I started going over drivers and converting to symbolic constants instead of hard-coded ones. I did a large part until I run out of steam. Maybe some brave soul will take up converting the rest of them, or maybe I will: note that when converting bridges one should be careful to use bridge macros where appropriate.

Nice patchset. And makes things way clear.

Acked-by Juan Quintela